Record for a UF thesis. Title & abstract won't display until thesis is accessible after 2013-08-31.

MISSING IMAGE

Material Information

Title:
Record for a UF thesis. Title & abstract won't display until thesis is accessible after 2013-08-31.
Physical Description:
Book
Language:
english
Creator:
Scheick,Benjamin J
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Business Administration, Finance, Insurance and Real Estate
Committee Chair:
Ling, David C
Committee Co-Chair:
Naranjo, Andy
Committee Members:
Nimalendran, Mahendrara
Tucker, Jenny

Subjects

Subjects / Keywords:
Finance, Insurance and Real Estate -- Dissertations, Academic -- UF
Genre:
Business Administration thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Statement of Responsibility:
by Benjamin J Scheick.
Thesis:
Thesis (Ph.D.)--University of Florida, 2011.
Local:
Adviser: Ling, David C.
Local:
Co-adviser: Naranjo, Andy.
Electronic Access:
INACCESSIBLE UNTIL 2013-08-31

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2011
System ID:
UFE0043225:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 INVESTOR SENTIMENT, CREDIT AVAILABILITY, INFORMATION DISSEMINATION, AND ASSET PRICE MOVEMENTS By BENJAMIN J. SCHEICK A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE R EQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2011

PAGE 2

2 2011 Benjamin J. Scheick

PAGE 3

3 To the many teachers who went the extra mile to cultivate my passion for learning, my parents who have always encouraged me and made it possible for me to follow my dream s my brothers who have always shown me how to make the most of the talents we have been blessed with all of my family and friends for their continued support, and my loving wife whose selfless sacrifice, patience and understanding has made this journey the happiest of my life

PAGE 4

4 ACKNOWLEDGMENTS There are a number of individuals that I would like to thank, without whose guidance and encouragement none of this would be possible. I would like to thank members of my supervisory committee: David Ling (committee chair), Andy Naranjo (committee co chair), Mahendrarajah Nimalendran, and Jennifer Wu Tucker. Jenny, I am extremely grateful for the confidence you instilled in me to ask interesting questions. Nimal, I am th ankful for the fresh viewpoints that you have consistently provided in your feedback. Andy, it is with sincere gratitude that I thank you for your continued guidance, critical feedback, devout encouragement and all the words of wisdom you shared throughout our many conversations. Most of all, I would like to thank you David for taking me under your wing and fostering my development not only as an academic, but more importantly as a person. I am forever grateful for the numerous hours you have given me of yo ur precious time, the unwavering patience you have shown throughout our projects together, and most importantly the experiences that you have exposed me to throughout my time here at the University of Florida. I would like to extend a special thanks to Wa yne Archer, who too has played an integral role in my development. Working with you has truly been a pleasure and I am grateful for the opportunities that y ou have provided me with over the past few years I also would like to thank Shawn Howton, who opene d my eyes to all that academia has to offer. I would like to thank all faculty members, administrative staff, as well as current and past Ph.D. students in the Department of Finance at the University of Florida for their guidance and support. Finally, I wo uld like to extend sincere thank s to my parents, family and wife for always believing in me.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 7 LIST OF FIGURES ................................ ................................ ................................ ......................... 9 ABSTRACT ................................ ................................ ................................ ................................ ... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 12 2 INVESTOR SENTIMENT AND ASSET PRICING IN PUBLIC AND PRIVATE MARKETS ................................ ................................ ................................ ............................. 16 Measuring Investor Sentiment ................................ ................................ ................................ 21 An Indirect Index of Real Estate Sentiment ................................ ................................ .... 22 A Direct Measure of Real Estate Sentiment ................................ ................................ .... 24 An Indir ect Index of Stock Market Sentiment ................................ ................................ 27 Empirical Methodology ................................ ................................ ................................ .......... 29 Short Run Regressions ................................ ................................ ................................ .... 29 Long Horizon Regressions ................................ ................................ .............................. 30 Data and Descriptive Statistics ................................ ................................ ............................... 32 Return Data Sources and Definitions ................................ ................................ .............. 32 Control Variables ................................ ................................ ................................ ............. 34 Descriptive Statistics ................................ ................................ ................................ ....... 34 Short Run Regressi on Results ................................ ................................ ................................ 36 Dynamic Relations amongst Sentiment Measures ................................ .......................... 36 Dynamic Relations between Returns and Sentiment ................................ ....................... 37 Stock Market Sentiment Effects ................................ ................................ ...................... 42 Long Horizon Regression Results ................................ ................................ .......................... 43 Long Run R elation between Returns and Sentiment ................................ ...................... 43 Long Run Stock Market Sentiment Effects ................................ ................................ .... 47 Summary and Conclusion ................................ ................................ ................................ ....... 48 3 CREDIT AVAILABILITY AND ASSET PRICING SPIRALS IN ILLIQUID MARKETS ................................ ................................ ................................ ............................. 62 Background Literature and Research Development ................................ ............................... 65 Empirical Methodology ................................ ................................ ................................ .......... 70 Data and Descriptive Statistics ................................ ................................ ............................... 71 Liquidity Data Sources and De finitions ................................ ................................ .......... 71 Asset Pricing Data Sources and Definitions ................................ ................................ .... 74 Control Variables ................................ ................................ ................................ ............. 75

PAGE 6

6 Descriptive Statistics ................................ ................................ ................................ ....... 76 Empirical Results ................................ ................................ ................................ .................... 80 Dynamic Relations in Illiquid Private Markets ................................ ............................... 80 Dynamic Relations in Liquid Public Markets ................................ ................................ 83 Impulse Response Functions: Credit Availability and Asset Price Movements ............. 85 Leverage and Liquidity Portfolio Sorts: Public Market ................................ .................. 86 Using Returns on REIT Preferred Stock as a Robustness Check ................................ .... 87 Credit Availability, Asset Prices, and Investor Sentiment ................................ .............. 89 Summary and Conclusion ................................ ................................ ................................ ....... 93 4 VOLU NTARY DISCLOSURE, INVESTOR SENTIMENT, AND RETURN COMOVEMENT ................................ ................................ ................................ .................. 109 Background Literature ................................ ................................ ................................ .......... 112 Data and Descriptive Statistics ................................ ................................ ............................. 115 Voluntary Disclosure ................................ ................................ ................................ ..... 115 Sentiment Portfolio ................................ ................................ ................................ ........ 117 Return and Contro l Variable Data Sources. ................................ ................................ .. 118 Descriptive Statistics ................................ ................................ ................................ ..... 119 Empirical Methodology ................................ ................................ ................................ ........ 122 Bivariate Regression Approach ................................ ................................ ..................... 122 Multivariate Regressions ................................ ................................ ............................... 123 Empirical Results ................................ ................................ ................................ .................. 124 Bivariate Regressions ................................ ................................ ................................ .... 124 Multivariate Regressions ................................ ................................ ............................... 127 Matched Sample Regressions ................................ ................................ ........................ 128 Multivariate Regressions by Disclosure Type ................................ ............................... 130 Additional Robustness Checks ................................ ................................ ...................... 133 Summary and Conclusion ................................ ................................ ................................ ..... 134 5 CONCLUSION ................................ ................................ ................................ ..................... 151 APPENDIX: BOOTSTRAPPING WITH LONG HORIZON REGRESSIONS ........................ 154 Bootstrap Simulation Procedure ................................ ................................ ........................... 1 54 Impact of Bootstrap Simulation Adjustments ................................ ................................ ....... 158 LIST OF REFERENCES ................................ ................................ ................................ ............. 161 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 168

PAGE 7

7 LIST OF TABLES Table page 2 1 De scriptive s tatistics: sentiment proxies ................................ ................................ ............ 53 2 2 Descriptive statistics: aggregate sentiment measures ................................ ........................ 54 2 3 Descripti ve statistic s and correlations: aggregate sentiment measures .............................. 55 2 4 Descriptive statistics: return series and control variables ................................ .................. 56 2 5 Dynam ic relations amongst sentiment measures ................................ ............................... 57 2 6 Dynamic relations between public market returns and sentiment ................................ ..... 58 2 7 Dynamic rela tions between private market returns and sentiment ................................ .... 59 2 8 Long horizon regression results: coefficient estimates and statistical significance .......... 60 2 9 Economic significance of long horizon regression results ................................ ................ 61 3 1 Descriptive statistics : credit availability, market liquidity, price returns, and controls ..... 98 3 2 Contemporaneous correlations between credit availabi lity and asset price movements ... 99 3 3 Credit availability and private commercial real estate : individual lags ........................... 100 3 4 Credit availability and private commercial real estate: cumulative effect and joint significance ................................ ................................ ................................ ...................... 101 3 5 Credit availability and public commercial real estate : individual lags ............................ 102 3 6 Credit availability and public commercial real estate: cumulativ e effect and joint significance ................................ ................................ ................................ ...................... 103 3 7 Credit availability and public commercial real estate : leverage and liquidity sorts ........ 104 3 8 Credit availability and public comme rcial real estate: preferred equity individual lags 105 3 9 Credit availability and public commercial real estate: preferred equity cumulative effect and joint significance ................................ ................................ ............................. 105 3 10 Credit availability and investor sentiment using a direct sentiment measure .................. 107 3 11 Credit availability and investor sentiment usin g an i ndirect sentiment measure ............. 108 4 1 Descriptive statistics : voluntary disclosure ................................ ................................ ...... 137

PAGE 8

8 4 2 Industry classifications: voluntary discl osure and sentiment portfolios ......................... 137 4 3 Size classifications : voluntary disclosure and sentiment portfolios ................................ 139 4 4 Portfolio da ily returns : voluntary disclosure and sentiment portfolios ............................ 140 4 5 Bivariate regression results: limited disclosure ................................ ............................... 141 4 6 Biva riate regression results: isolated disclosure ................................ .............................. 142 4 7 Multivariate regres sion results: limited disclosure ................................ .......................... 143 4 8 Multivariate r egression results: isolated disclosure ................................ ......................... 144 4 9 Bivariate regression results: matched sample ................................ ................................ .. 145 4 10 Multivariate reg ression res ults: matched sample ................................ ............................. 146 4 11 Multivariate regression results by disclosure type: limited disclosure ............................ 147 4 12 Multivariate regre ssion results by discl osure type: isolated disclosure ........................... 148 4 13 Multivariate regression results for robustness checks: restricted limite d disclosure ....... 149 4 14 Multivariate regression results for robus tness checks: first disclosure ............................ 150

PAGE 9

9 LIST OF FIGURES Figure page 2 1 Investor se ntiment indices ................................ ................................ ................................ .. 50 2 2 Investor sentiment and returns ................................ ................................ ........................... 51 2 3 Impulse response functions : investor sentiment and returns ................................ ............. 52 3 1 Credit availability and market liquidity ................................ ................................ ............. 95 3 2 Credit availability and asset prices ................................ ................................ .................... 96 3 3 Impulse response functions : credit availability and price returns ................................ ...... 97 4 1 Investor sentiment and voluntary disclosure ................................ ................................ .... 136 A 1 Empirical distribution of bootstrap test statistics and coefficient estimates .................... 160

PAGE 10

10 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fu lfillment of the Requirements for the Degree of Doctor of Philosophy INVESTOR SENTIMENT, CREDIT AVAILABILITY, INFORMATION DISSEMINATION, AND ASSET PRICE MOVEMENTS By Benjamin J. Scheick August 2011 Chair: David Ling Cochair: Andy Naranjo Major: Business Administration In models of efficient financial markets, security prices represent fair valuations by rational investors, incorporate the perceived debt and equity financing costs of the marginal investor, and reflect all available public information. Ho wever, the rapid rise of asset prices in the technology sector and real estate market, liquidity crises such as those of Long Term Capital Management and Lehman Brothers, and the use of misleading information in accounting scandals such as those of Enron a nd WorldCom, have weakened support among economists for asset pricing models based on these fundamental determinants of value. In the following three studies, I provide evidence that behavioral factors, market frictions, and the strategic disclosure of inf ormation significantly impact asset prices. In my first study, I examine the differential impact of investor sentiment on short and long run returns in public and private commercial real estate markets. In the short run, I find evidence of sentiment induc ed mispricing in both public and private real estate markets. In the long run, however, I find a distinct difference in the time it takes prices to revert to fundamental value, as assets trading in the private market are susceptible to prolonged periods of sentiment induced mispricing due to inherent limits to arbitrage in these markets.

PAGE 11

11 In my second study, I examine the relation between credit availability, market liquidity and asset prices in both private and public commercial real estate markets. I find that changes in credit availability impact asset price movements in both private and public markets. However, I find the underlying liquidity with which these assets trade to be a key determinant of the likelihood of an asset pricing spiral, in which chan ges in asset prices lead to further changes in credit availability when assets are relatively illiquid In my final study, I examine whether the strategic disclosure of earnings guidance by firm management caters to misguided investor expectations, partic ularly when investors are overly optimistic. I find that firms issuing voluntary earnings guidance during periods of high investor confidence covary more with stocks that are highly sensitive to changes in investor sentiment and less with market fundamenta ls following the information disclosure.

PAGE 12

12 CHAPTER 1 INTRODUCTION Efficient financial markets assume that security prices represent fair valuations by rational investors, explicitly incorporate the perceived debt and equity financing costs of the marginal investor, and reflect all available public information. Any deviations in price that result from temporary changes in these fundamental determinants of value should be quickly arbitraged away by rational investors who take immediate advantage of this prof it opportunity. Therefore, in an efficient market, asset price movements should be insensitive to the actions of sentiment based investors, changes in the availability of funding for investment, and the dissemination of misleading information. In reality, however, arbitrageurs face limitations in their ability to enter the market at precisely the time when it may be most advantageous for them to do so. For example, in response to speculative driven investment, taking the opposite side of a transaction is n o longer risk free, since sentiment based investors introduce additional risk as they continue to bid asset prices away from their fundamental values. Furthermore, arbitrageurs may face significant funding constraints, inhibiting their ability to obtain fi nancing for investment when it is needed most. If behavioral influences and market frictions cause prices to deviate significantly from fundamental values for prolonged periods of time, then this raises an additional concern that firms may release value re levant information in such a manner that takes advantage of this mispricing, further exacerbating the underlying issue. Over the past decade, the rapid rise of asset prices in the technology sector and real estate market, liquidity crises such as those of Long Term Capital Management and Lehman Brothers, and the use of misleading information that was at the heart of accounting scandals such as those of Enron and WorldCom, have further weakened support among economists for asset pricing

PAGE 13

13 models based on the assumptions of efficient markets, rational expectations, and costless arbitrage. In the following three studies, I examine the impact that behavioral factors, market frictions, and information dissemination have on asset price movements in both private and public markets. In particular, I study the roles that investor sentiment, credit availability and voluntary disclosure of earnings guidance play in driving asset prices away from their fundamental values. In my first study, I examine the differential imp act of investor sentiment on both short and long run returns in public and private commercial real estate markets. Using vector autoregressive models to capture the short run dynamics between returns and investor sentiment, I find a positive relation betw een investor sentiment and subsequent quarter returns in both public and private real estate markets. The magnitude of this short run effect is larger in public markets than in private markets, which is consistent with private market investors being better informed and more sophisticated. This result also provides evidence on the relative importance of sentiment driven investor demand and limits to arbitrage in asset mispricing. I further find a negative relation between investor sentiment and subsequent lo ng horizon public market returns, consistent with prices reverting to their fundamental values over the long run. In contrast, I find sustained periods of sentiment induced mispricing in private real estate markets, consistent with the significant limits t o arbitrage that characterize this investment environment. In my second study, I examine the relation between the availability of credit, market liquidity and asset price movements in both private and public commercial real estate markets. Given the relati ve illiquidity and significant use of leverage in acquisitions within commercial real estate markets, theory predicts that funding constraints are likely to play a significant role in asset price determination. Using vector autoregressive models to capture the short run dynamics

PAGE 14

14 between fluctuations in credit availability and price changes, I find that a tightening in credit availability is negatively related to subsequent price movements in both the private property and public REIT markets, consistent with significant leverage effects. I also find that assets trading in illiquid segments of the commercial real estate market are highly susceptible to a spiral effect, in which changes in asset prices lead to further changes in the availability of credit. In p articular, I document a feedback effect of lagged price changes on subsequent capital availability in the private commercial real estate market, the lowest liquidity quartiles of the public commercial real estate market, and the relatively illiquid market for REIT preferred shares. These results suggest that while leverage plays a significant role in determining the degree to which credit availability affects asset prices, the underlying liquidity with which these assets trade is a key factor in determining the likelihood of a liquidity spiral, with lower liquidity creating the market setting for a spiral effect. In my fin al study, I examine whether firm managers cater their strategic disclosure of earnings guidance to take advantage of misguided investor e xpectations, particularly when investors are overly optimistic. If firms issue earnings guidance that is consistent with misguided investor beliefs, voluntary disclosure of such information may contribute to sentiment induced mispricing, rather than aid in resolving sentiment driven overvaluation. To identify whether company issued earnings guidance impacts future asset price movements, I examine shifts in return comovement around the disclosure event. My analysis is the first to utilize voluntary disclosur e events to examine sentiment based return comovement. I find that firms issuing voluntary earnings guidance during periods of high investor confidence increase their return comovement with stocks that are highly sensitive to changes in investor sentiment and decrease their comovement with market fundamentals following the disclosure. Furthermore, I provide

PAGE 15

15 evidence that this effect is concentrated around the issuance of neutral earnings guidance, which fundamental value. This result provides further support for the sentiment based theory of comovement. Taken together, these three studies provide evidence that behavioral factors, such as investor sentiment, market frictions, such as a la ck of liquidity or changes in the credit supply, and the strategic issuance of information, such as the voluntary disclosure of earnings guidance, can have significant impacts on asset price movements. Furthermore, these results indicate that certain market conditions, such as significant limits to arbitrage, funding constraints in relatively illiquid markets, or periods of investor optimism may act as a catalyst for non fundamental based pricing effects.

PAGE 16

16 CHAPTER 2 INVESTOR SENTIMENT A ND ASSET PRICING IN PUBLIC AND PRIVAT E MARKETS Real estate is a key input in the production process of most firms and plays a central role in economic growth and business cycles in countries around the world. Consequently, it is important to understand the sources and dynamics of real estate valuation. Although it is clear that fundamental factors influence real estate pricing, the recent well documented bubble in U.S. decline in U.S. commercial real estate prices that followed the onset of the financial crisis in 2007 have further weakened support among financial economists and practitioners for traditional asset pricing models based on the assumptions of efficient markets, rational expectations, and costless arbitrage. A growing behavioral economics and finance literature acknowledges that economic be. 1 Rather, the emerging behavioral approach recognizes t he bounded rationality and psychological biases of investors who often rely on and are influenced by computational shortcuts, heuristics, frame dependence, and intuition when making decisions in a complicated and uncertain world with market frictions. 2 As a result, changes in asset prices may be driven by more than changes in market fundamentals as shown theoretically by De Long, Shleifer, Summers, and Waldmann (1990) and Shleifer and Vishny (1997), among others. Recent empirical research in behavioral fina nce also suggests that this movement away from fundamentals encompasses the influences of investor sentiment on asset valuation, where 1 2 Barber and Odean (2008), for example, find that unlike institutional investors, individual investors are more prone to buy stocks that have recently been in the news.

PAGE 17

17 investor sentiment is de fined by Baker and Wurgler (2006, 2007 ) as a misguided belief about the growth in future cash flo ws or investment risks (or both) based on the current information set. The empirical literature on investor sentiment and asset pricing has largely focused on public stock markets. This focus is understandable given the difficulties associated with obtaini ng return information on private equity investments since private equity has historically been exempt from public disclosure requirements (Kaplan and Schoar, 2005). Nevertheless, private investment markets provide an appealing testing ground for examining pricing role. Relative to more liquid public markets, private investment markets exhibit significant information asymmetries and illiquidity. The lack of continuous price revelation in private markets suggests the potential impact of investor s entiment on market values may be revealed with significant lags. Sentiment induced asset mispricing arises from a combination of sentiment driven investor demand and limits to arbitrage. Thus, the degree to which private markets are affected by investor s entiment is not ex ante clear. To the extent that private market investors are better informed and more sophisticated, asset prices could potentially be less prone to the influence of investor sentiment in these markets. However, the illiquidity, informati on asymmetries, and more limited price revelation inherent in private markets may allow sentiment to play a more persistent role in pushing asset prices away from their fundamental values. Moreover, the inability to short sell in private markets impedes th e opportunity for informed arbitrageurs to counteract mispricing. 3 In contrast, price revelation occurs more rapidly in public stock markets where the ability of informed investors to short sell exists, albeit with limits. Therefore, the reversion of 3 The limited arbitrage literature suggests that assets that are more difficult to arbitrage have potentially larger initial mispricing and then larger subsequent abnormal returns as prices revert to fundamentals (e.g., Pontiff, 1996, 2006; and Wurgler and Zhuravskaya, 2002). However, with more sophisticated investors, it is not ex ante clear that the limited arbitrage e ffect would lead to greater initial mispricing and greater subsequent reversals.

PAGE 18

18 price s to fundamental values should occur more quickly in public markets. Despite the potential importance of investor sentiment in the price formation process, no previous research has directly investigated the relative importance of sentiment in public and pr ivate asset markets or the relative importance of sentiment driven investor demand and limits to arbitrage in determining asset prices within each market. In this paper, I examine the relation between investor sentiment and returns in both public and priva te commer cial real estate markets. In my empirical tests, I use both direct (survey based) and indirect (derived from multiple indirect sentiment proxies) real estate sentiment measures. I first examine the t ime varying relation between these two sentiment measures, their relation with stock market sentiment, and their unconditional relation with public and private real estate returns. I then test the conditional short and long run effects of real estate investor sentiment on both public and private real e state returns. This analysis of private and public markets provides evidence on the relative importance of sentiment driven investor demand and limits to a rbitrage in asset mispricing. I also provide a unique side by side a nd long run impact on similar assets that are owned and traded in two distinct investment environments. The commercial real estate market provides an appealing testing ground for examining roperty markets exhibit the segmentation, information asymmetries, and illiquidity that characterize other private markets. Second, unlike other private markets, several representative total return indices for private commercial real estate are available, permitting the calculation of time weighted returns that can be compared directly to corresponding returns in public real estate markets. Finally, the underlying properties held by the publ icly traded real estate firms I analyze are similar to the

PAGE 19

19 property holdings of the institutional real estate investors whose private market returns I also track. Since I also control for the additional equity characteristic embedded in the public real n public and private real estate markets can be ascribed to differences in the characteristics of these two markets, not to fundamental differences in the types of assets owned. I nitial stock and real estate market sentiment results provide evidence sugges ting that investors view commercial real estate and the general stock market as distinct asset classes. The time varying results also suggest that as the recent subprime mortgage crisis unfolded, sentiment in the commercial real estate market turned sharpl y downward in 2007. Real estate investor sentiment appears to have led the significant decline in property prices that occurred after the most recent peak. In contrast, stock market sentiment was slower to decline and fell less precipitously during this la tter period. Using vector autoregressive (VAR) models in which commercial real estate returns and sentiment are specified as endogenous variables in a two equation system that also includes exogenous control variables, I first address two questions: Do ch anges in investor sentiment predict short run returns? And, second, do returns predict short run changes in sentiment? Daniel, Hirshleifer, and Subrahmanyam (1998) suggest that overconfident investors form optimistic expectations about the future value of an asset and tend to disregard information that contradicts these beliefs due to self attribution bias. Welch (1992) also establishes a cascade model in which investors base their decisions on observations of previous market demand and ultimately ignore th eir own private information. Froot, Scharfstein, and Stein (1992) posit that short horizon investors often focus on recent market demand rather than focusing on

PAGE 20

20 long arra y of rumors, vague and incomplete information, and uncertainties. 4 In each case, investors act on noisy information, creating momentum that ultimately pushes prices away from fundamental value over short horizons. Therefore, I expect to observe persistenc e in my measures of sentiment as the expectations of sentiment based investors are influenced by prior periods of high or low sentiment. I also expect a positive relation between sentiment and subsequent short run returns as sentiment driven demand, accomp anied by limits to arbitrage, temporarily drive prices away from their fundamental value. However, it is plausible for the magnitude of the short run sentiment effect on prices to be smaller (or possibly zero) in private markets than in public markets to t he extent that potentially sophisticated private market investors offset the greater limits to arbitrage inherent in private markets. I next examine whether a negative relation exists between investor sentiment and subsequent long horizon returns in public and private commercial real estate markets. If excessive investor optimism (pessimism) leads to market overvaluation (undervaluation), then periods of high (low) sentiment should be followed by low (high) cumulative long run returns as the market price re verts to its fundamental value in the long run. Moreover, given the greater limits to arbitrage, short sale constraints, information externalities, and delays in information transmission that characterize private real estate markets, I expect the impact of investor sentiment on market values in private real estate to be more persistent in pushing asset prices away from their fundamental values over time. 4 Lamont and Thaler (2003a) provide an alternative explanation in which the marginal market participant is a sentiment induced investor. They argue that if optimists are will ing to bid up the prices of some stocks and not enough investors are willing to meet that demand by selling short, the optimists will set the price.

PAGE 21

21 In my short run analysis, I find a positive relation between investor sentiment and subsequent quarter r eturns in both public and private real estate markets. That is, sentiment based investment drives prices away from fundamental value in the short run, resulting in a short term continuation of returns. For a given change in sentiment, the magnitude of this short run effect is larger in the public than in the private real estate market. This result is consistent with private market investors being better informed and more sophisticated, and consequently provides evidence on the relative importance of market participant demand and arbitra ge costs in asset mispricing. I also find that the effects of sentiment on asset prices are conditional on the level of sentiment. In particular, if sentiment is more than one standard deviation above average, subsequent publi c and private real estate returns are on average lower, which is consistent with Baker and Wurgler (2007). Using long horizon regressions, I also provide evidence that limits to arbitrage play an important role in determining the time it takes for prices to revert to fundamental values. In public real estate markets, periods of sentiment induced mispricing are quickly followed by price reversals. For example, I find that an increase in inve stor sentiment results in a 2.9 % decrease in public real estate mar ket returns over the following year. In contrast, private real estate markets are more susceptible to prolonged periods of sentiment induced mispricing. I find that an increase in inve stor sentiment results in a 3.9 % increase in private real estate market returns over the subsequent year, and that this mispricing continues to persist over longer horizons. Measuring Investor Sentiment Prior research uses several approaches to quantify investor sentiment. One stream of research focuses on direct sentiment mea sures, such as survey based measures developed to capture the outlook of market participants. Qiu and Welch (2005) provide a comparison of several direct survey based measures of investor sentiment. An alternative stream of research

PAGE 22

22 uses multiple indirect sentiment proxies for investor sentiment. Although no single measure is a pure indicator of investor sentiment, each imperfect proxy is likely to contain a sentiment component. Baker and Wurgler (2006, 2007), for example, utilize principal component analys is to develop an indirect measure of investor sentiment from multiple indirect proxies. 5 I employ both indirect and direct measures of investor sentimen t in my analysis. An Indirect Index of Real Estate Sentiment Following the framework of Ba ker and Wurgle r (2006, 2007), I use principal component analysis to construct an indirect quarterly sentiment index based on the common variation in seven underlying proxies of investor sentiment in commercial real estate markets: (i) the average REIT stock price premiu m to net asset value (NAV), (ii) the percentage of properties sold each quarter from the National Council of Real Estate Investment Fiduciaries (NCREIF) Property Index (NPI), (iii) the number of REIT IPOs, (iv) the average first day returns on REIT IPOs, ( v) the share of net REIT equity issues relative to total net REIT equity and debt issues, (vi) net commercial mortgage flows as a percentage of GDP, and (vii) net capital flows to dedicated REIT mutual funds. Lee, Shleifer, and Thaler (1991) suggest that closed end fund discounts represent movements in stock prices away from fundamental values. Similarly, REIT price premiums estimated net asset value of the underlyi ng properties that comprise the REIT. Stock price deviations from NAV may, in part, reflect the price impact of sentiment based trading during periods of investor opti mism or pessimism. Therefore, I obtain the average quarterly U.S. REIT price premium to N AV from Green Street Advisors, a prominent buy side REIT advisory firm 5 Baker, Wurgler, and Yuan (2009) also utilize this methodology to create local and global sentiment indice s across six major international stock markets.

PAGE 23

23 Baker and Stein (2004) argue that aggregate market liquidity can serve as a sentiment proxy. In a market with short sale constraints, sentiment driven investors are more likely to par ticipate when they are optimistic; therefore liquidity will likely increase during period s of investor overconfidence. I use the percentage of properties sold from the NPI each quarter as a proxy for aggregate liquidity in the private commercial real estat e market. The market timing of IPOs and secondary equity offerings have been used to measure investor sentiment in the general stock market (e.g., Ritter, 1991; Baker and Wurgler, 2000). Similarly, the number of REIT IPOs, the average first day returns on REIT IPOs, and the share of net REIT equity issues relative to the total capital raised by REITs may identify periods of sentiment induced mispricing in commercial real estate markets. The number of REIT IPOs and average first day returns are constructed using data provided by the National Association of Real Estate Investment Trusts (NAREIT). The share of REIT equity issues relative to total REIT equity and debt offerings is constructed from data obtained from the Federal Reserve Flow of Funds Accounts. Clayton, Ling, and Naranjo (2009) argue that net commercial mortgage flows are widely viewed by industry participants as a barometer of investor sentiment, in part because of the association between past real estate cycles and excessive mortgage flows du ring periods in which default risk may have been underpriced by lenders. Therefore, periods of increased commercial mortgage flows may reflect the influence of investor sentiment. Quarterly commercial mortgage flows are obtained from the Federal Reserve Fl ow of Funds Accounts and then scaled to be a percentage of GDP. Finally, Brown et al. (2002) and Frazzini and Lamont (2008) suggest that flows into and out of mutual funds proxy for investor sentiment. Therefore, shifts in capital flows to dedicated

PAGE 24

24 REIT m utual funds may indicate periods of investor over or under confidence. The quarterly flow of investment capital into, and out of, dedicated REIT mutual funds is obtained from AMG Data Services. The top panel of Table 2 1 contains summary statistics for e ach indirect commerci al real estate sentiment proxy Note that there is substantial vari ation both within and across the indirect proxies. Along with this variation, however, there is also substantial persistence in the levels and changes in the indirect p roxies. Utilizing quarterly data from 1992:Q2 to 2008:Q4, I generate a composite indirect real estate sentiment index (INDRES) based on the first principal component of the contemporaneous levels of each of the seven sentiment proxies. 6 INDRES is standardi zed to have a mean of zero and unit variance. The quarterly serial correlation of INDRES is 0.73; the serial correlation of quarterly changes in INDRES is 0.29. A Direct Measure of Real Estate Sentiment In addition to indirect measures of investor sentime nt, the equity market sentiment literature has also used various survey based measures to capture investor sentiment. For stock investment newsletters deemed to be bullish minus the percentage categorized as bearish, 7 Brown and Cliff (2004, 2005) relate the bull bear spread to deviations from fundamental values and examine both short and long run effects of sentiment on s tock returns. T he authors find that the bull bear spread is highly correlated with contemporaneous stock returns but has little short run predictive power (Brown and Cliff, 2004). 6 I detrend the commercial mortgage flow series using the prior 2 year rolling average before including it in my principal component analysis. 7 Other direct measures of investor sentiment include, for exampl e, the Michigan Consumer Confidence Index and the UBS/GALLUP Index of Investor Optimism.

PAGE 25

25 However, taking a longer term perspective of two to three years, periods of high sentiment are followed by low returns as stock prices mean revert (Brown and Cliff, 2005). Along similar lines, I employ survey data published by the Real Estate Research Corporation (RERC) in its quarterly Real Estate Report as a direct measure of in vestor sentiment in commercial real estate markets. RERC surveys institutional real estate investors, appraisers, lenders, and managers throughout the United States to gather information on current investment criteria, such as required rates of return on e quity, expected rental growth rates, and current respondents are asked to rank current investment conditions for multiple property types, both nationally and similar in spirit to the bull bear spread in that it captures movements in the pro portion of participants in commercial real estate markets who are bullish relative to those less optimistic about current investment opportunities. 8 The middle panel of Table 2 1 contains summary statistics on the property type components of the national level RERC investor sentiment index. Note that over the 1992:Q2 to 2008:Q4 sample period the consensus opinion of survey respondents was that apartment and industrial warehouse properties, with an average investment conditions rank of 6.3, were considered to be the most desirable, followed by neighborhood retail properties. In contrast, retail power centers, with a mean investment conditions ranking of 5.0, were deemed the least desirable investments of the eight property types over the study period. Simila r to the indirect 8 RERC also collects other investment condition variables in their survey, such as the percentage of respondents who give a buy recommendation and the percentage who give a sell recommendation. However, these variables are only available for a shorter sub sample beginning in the latter half of the 1990 buy sell recommendation and investment conditions variables is high over the sho rter sub sample (0.77 for the direct measure that I use).

PAGE 26

26 the sample period. For example, the investment desirability of suburban office properties ranged from a low of 2.8 to a high of 7.5. It is als o important to note that RERC investment conditions display substantial positive serial correlation across quarters, with changes displaying significant negative serial correlation. The direct measure of commercial real estate sentiment (DRES) is construct ed from the first principal component extracted from quarterly RERC investment condition survey responses pertaining to the eight RERC property types. 9 DRES is standardized to have a mean of zero and unit variance. The quarterly serial correlation of DRES is 0.81; the serial correlation of quarterly changes in DRES is 0.35. Panel A of Figure 2 1 plots DRES against INDRES over the sample period. Overall, the correlation between the two sentiment indices is 0.48 as shown in Table 2 3 During the early to mid 1990s, as the commercial real estate market was emerging from a downturn in the late 1980s, INDRES (the dashed line) was somewhat more volatile than DRES (the solid line). After peaking at a higher level than DRES in early 1998, INDRES dropped more precip itously during the subsequent slowdown that occurred in commercial real estate markets in the late 1990s and early 2000s. The private commercial real estate market began what became a prolonged bull market around 2003. It is interesting to note that the in direct measure of commercial real estate sentiment stabilized and then t urned upward sooner than did the survey based measure of sentiment. The significant and sustained run up in commercial property prices finally peaked in late 2007 in most U.S. markets. However, both measures of sentiment leveled out and then began to decline much earlier than transaction prices. That is, investor 9 The correlation between an equally weighted average investment condition across the eight RERC property types and DRES is 0.93 over the sample period.

PAGE 27

27 sentiment appears to have led the significant decline in property prices that occurred after the most recent peak. An Indirec t Index of Stock Market Sentiment F ollow ing Baker and Wurgle construct an indirect measure of investor sentiment for the general stock market. In particular, I utilize principal component analysis to generate a quarterly sentim ent index based on the common variation in six underlying proxies of investor sentiment in the stock market: (i) the closed end fund discount, (ii) share turnover on the NYSE, (iii) the number of IPOs, (iv) the average first day returns on IPOs, (v) the sh are of equity issues in total equity and debt issues, and (vi) the dividend premium. I definitio ns consistent with their study The closed end fund discount is defined as the difference between the net asset values (NAVs) of closed end stock fund shares and their market prices as reported in the Wall Street Journal. 10 Share turnover on the NYSE is defined as the total volume of NYSE Group Shares divided by shares outstanding as reported in the NYSE Fact Book. 11 I obtain the number of IPOs and the average first website. The share of equity issues in total equity and debt issues is defined as gross equity issuance divided by gross equity plus gross long term debt issuance as reported in the Statistical Supplement to the Federal Reserve Bulletin. The dividend premium is defined as the log difference of the average market to book ratios of dividend payers and non payers (Baker and Wur gler, 2004). The bottom panel of Table 2 1 contains summary statistics for each indirect 10 I compute the value weighted average discou nt on closed end funds classified as General Equity Funds in the Wall Street Journal. 11 The turnover measure is expressed as the natural log of the turnover ratio and is detrend ed using a 5 year moving average.

PAGE 28

28 sentiment proxy pertaining to the general stock market. S ummary statistics for these variables are of similar magnitude to those reported in Baker and Wurgler (2007). Utilizing data from 1965 to 2008, I generate a composite indirect stock market sentiment index (INDSMS) based on the first principal component of the contemporaneous levels or lags of each of the six sentiment proxies. The index is standardized to have a m ean of zero and un it variance for the period 1965 to 2008. Table 2 2 and Table 2 3 contain descriptive statistics and correlations respectively, for the two real estate sentiment indices as well as the stock market s entiment index over the 1992:Q2 to 2008 :Q4 sample period. The mean of INDSMS is 0.43 over the sample period. The quarterly serial correlation of INDSMS is 0.76 when expressed in levels and 0.14 when expressed in changes. Panel B of Figure 2 1 plots INDSMS against INDRES over the 1992:Q2 to 20 08:Q4 sample period. Overall, the correlation between the two indices is 0.194 (Table 2 3 ), suggesting that investors view commercial real estate and the general stock market as distinct asset classes. Note that the divergence between the two indices begi nning in 1999 coincides with the internet bubble, where investor stock market sentiment was very optimistic and real estate market sentiment was more pessimistic. During this period, many investors were shifting their holdings out of value oriented investm ents, including commercial real estate, into high growth technology stocks. As the tech nology bubble burst and the Federal Reserve acted to avoid a recession in the wake of 9/11, real estate and other value investments became popular alternatives for inves tors seeking safer investment options. However, as the recent subprime mortgage crisis unfolded, sentiment in the commercial real estate market turned sharply downward in early 2007. In contrast, stock market sentiment was slower to decline and fell less p recipitously during this latter period.

PAGE 29

29 Empirical Methodology Short Run Regressions To capture short term sentiment and return dynamics, I employ vector autoregressive (VAR) models. In its simplest form, a VAR model is composed of a system of regressions w here lagged values, as well as exogenous control variables. In more technical terms, a vector autoregression model is the unconstrained reduced form of a dynami c simultaneous equations model. An unrestricted p th order Gaussian VAR model can be represented as: Y t 1 Y t 1 2 Y t 2 k Y t p t (2 1) where Y t is a vector of variables, is a p x 1 vector of intercepts, 1 2 k are p x p matrice s of parameters with all eigenvalues of having moduli less than one so that the VAR is stationary, and e t is a vector of uncorrelated structural shocks [ NID(0, )]. In a bivariate framework consisting of only sentiment and returns as endogenous variable s, the diagonal coefficients of represent conditional momentum in sentiment and returns, while the off diagonal coefficients of represent conditional positive feedback trading (sentiment following returns) and conditional anticipation effects (returns following sentiment). The off diagonal elements of capture the price impact effect of sentiment on retu rns. I obtain maximum likelihood estimates of and using iterated least squares. The number of quarterly lags is chosen based on examination of the AIC, SBIC, and the likelihood ratio selection criteria for various choices of p. It is important to note that the inclusion of lagged returns in the private market return equation controls for the well documented autoregressive nature of NCREIF returns.

PAGE 30

30 I first use an unconstrained VAR system to examine the dynamic relation between direct and indirect indices of commercia l real estate sentiment over the 1992:Q2 to 2008:Q4 sample period. I then examine the relation be tween indirect measures of real estate and sto ck market sentiment. Finally, I examine the relations between sentiment and real estate returns in public and private markets. To measure returns on commercial real estate I use public U.S. equity REIT and private NCREIF returns. T o measure investo r sentiment, I use both INDRES and DRES. To contro l for stock market sentiment, I utilize the indirect general stock market index (INDSMS). Simila r to Brown and Cliff (2005), I specify sentiment measures in both levels and changes within the regression fra mework To control for other potential sources of variat ion in returns and sentiment, I also include lagged values of several control variables that have been shown to matter in the asset pricing literature as explained further in the data section. Long Ho rizon Regressions Following Brown regress future k period quarterly returns on a vector of control variables, z t and a measure of investor sentiment, S t (r t+1 t+k t t t (2 2) wher e r t+1 ... r t+k are quarterly log returns, k is the number of quarters over which the investment horizon spans include the same set of control variables in the long horizon framework as I employ in the short run VAR framewor k. If periods of optimistic (pessimistic) sentiment lead initially to overvaluation (undervaluation), then periods of high (low) sentiment should be followed by low (high) cumulative long run returns as prices revert to their fundamental values over time. Thus, a negative coefficient on S t in the long horizon regressions is consistent with the price reversion that occurs following the initial impact of p ositive c oefficient on S t indicating a continuation of short term investment returns as asset

PAGE 31

31 prices either continue to move away from fundamental value or are slower to revert over longer horizons. Several econometric issues arise from the use of long horizon re gressions. The first issue stems from the presence of overlapping observations in the dependent variable of the regression specification. Because the dependent variable consists of average long horizon returns calculated over consecutive quarters, there wi ll be a moving average process in the error term. Thus, the use of OLS will lead to standard errors that are biased downwards. Conventional corrections such as the methodology proposed by Hansen and Hodrick (1980) or a related procedure outlined by Newey a nd West (1987) are not appropriate in the present context because both procedures have been shown to exhibit poor finite sample properties, especially when serial correlation is high, as is the case with overlapping return observations. Moreover, a number of studies have documented a sizeable bias in the error term as the sample size decreases and autocorrelation in the error term increases. 12 A second issue that arises within a small sample setting is the potential for finite sample bias in the coefficient estimate of a persistent independent variable. Stambaugh (1999) shows that a persistent explanatory variable will be predetermined but not strictly exogenous; thus, coefficient estimates may suffer from significant finite sample bias. Although an OLS esti mate is consistent and asymptotically normally distributed under the predetermined assumption, it is not necessarily unbiased in finite samples. In empirical applications, this bias reduces to zero as the sample size approaches infinity. However, the coeff icient estimate on a persistent investor sentiment measure may exhibit finite sample bias using quarterly observations that decrease in number as the length of the return horizon increases. 12 See Nelson and Kim (1993), Goetzmann and J orion (1993), and Hansen and Tuypens (2004) for further discussion of the effects of small sample bias in long horizon return regressions.

PAGE 32

32 To address the econometric issues of potentially biased coefficie nt estimates and standard errors that arise from the use of long horizon regressions, I use a bootstrap simulation procedure similar to Brown and Cliff (2005). See Appendix for details. Data and Descriptive Statistics Return Data Sources and Definitions P u blic commercial real estate returns are obtained from the CRSP/Ziman database which is produced jointly by the Center for Research in Security Prices at the University of Chicago l REITs that have traded on the NYSE, Amex, and Nasdaq exchanges since 1980. Ownership of REITs on each of these exchanges has historically consisted of a combination of retail and institutional investors. It is interesting to note that the institutional o wnership of REITs was on average lower than that of all other publicly traded stocks on each of these exchan ges during the first half of the sample period (Chan, Erickson, and Wang, 2003), but has grown substantially in more recent years. Daily and monthly return indices are computed with both equal and value weighting for all U.S. REITs and for subsets defined by type of REIT (equity, mortgage, and hybrid) and by type of property. I use the value weighted aggregate U.S. equity REIT index to construct a qu arterly return series; thus REITs that invest significantly in mortgages are excluded. R eturn data for private real estate markets is provided by NCREIF, a not for profit institutional real estate industry association. Established in 1982, NCREIF collec ts, processes, validates, and disseminates information on the risk/return characteristics of commercial real the NCREIF Property Index (NPI), tracks the quart erly total return performance of a large pool of individual commercial real estate properties acquired in the private market for investment

PAGE 33

33 purposes only. 13 To be included in the NPI, the data contributing member least 60% leased and b e wholly owned or in a joint venture structure. Although levered properties are included in the NPI, investment performance is reported by NCREIF on an unlevered basis. The property composition of the NPI changes quarterly as data contributing NCREIF membe rs buy and sell properties. However, all historical property level data remain in the database and index. The average holding period of the properties included in the NCREIF terly return is weighted by its estimated market value relative to the total market value of the properties that comprise the NPI. The NCREIF NPI is the only source of consistently collected information on the total returns earned by investors in private U.S. commercial real estate markets and is therefore widely used as a benchmark return index. Nevertheless, the NPI has several shortcomings (Geltner and Ling, 2007). 14 For example, unless the underlying property is sold during the quarter, the NPI uses cha the construction of the NPI should not affect the validity of my empiri cal results as the short run VAR methodology includes lagged values of the index as explanatory values and the long horizon regressions make use of bootstrapping adjustments Ho wever, as a robustness check, I also use a variant of the NPI that is based on the transaction prices of index properties that sold during the quarter and obtain similar results. 13 At the end of the fourth quarter of 2008, the NPI database contained 6,287 properties with an estimated market valu e of $305 billion. 14 Also see Pagliari, Scherer, and Monopoli (2005) for a detailed description of the NCREIF NPI index.

PAGE 34

34 Control Variables To control for other potential sources of variation in returns and sentiment in both the VAR and long horiz on regression specifications, I include the following macroeconomic variables that have been shown to affect asset returns: the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSP), the spread between yields on BAA r ated and AAA rated corporate bonds (DEFAULTSP), and the rate of inflation (INFLA) (e.g., Chen, Roll, and Ross, 1986; Ferson and Harvey, 1991; Fama and French, 1993; Fama and Schwert, 1977; Sharpe, 2002). I also include the three Fama French risk factors: M KT, SMB, HML, augmented by the return momentum factor, UMD (e.g., Fama and French 1996; Liew and Vassalou, 2000; Lettau and Ludvigson, 2001; Jegadeesh and Titman, 1993; and Carhart, 1997). 15 These factors also control for the equity characteristic embedded within the public market real estate returns (e.g., Riddiough, Moriarty, and Yeatman, 2005, and Pagliari, Scherer, and Monopoli, 2005). Descriptive Statistics Table 2 4 reports descriptive statistics for the two primary aggregate r eal estate return series and control variables. As documented in the real estate literature, average total returns on publicly traded REITs typically exceed returns on similar institutional quality assets owned and managed in private markets, albeit with greater volatility. Howeve r, according to NAREIT, equity REITs produced total returns of 17.8 % in 2007 and 37.8 % in 2008, both of which are significantly below the comparable total returns on the NPI Index. 16 As a result, quarterly REIT returns averaged jus t 0.90 % from 1992:Q2 to 2008:Q4 sample period. The corresponding 15 In robustness tests, I s additional contr ol variables. The magnitude of the coefficient estimates and statistical significance of the sentiment measures remain similar. 16 The aggregate NAREIT equity with the CRSP/Ziman equity index.

PAGE 35

35 average NPI return is 2.3% per quarter. At 2.0 % the standard deviation of quarterly NPI returns is som ewhat lower than the 2.8 % standard deviation of REIT returns. The quarterly serial correlation of REITRET over t he sample period is 0.04. In contrast, the serial correlation of What about the contemporaneous correlations between sentiment and returns? In Panels A and B of Figure 2 2, I plot REIT total returns against INDRES and DRES, respectively, over the sample period. Sentiment is measured on the left vertical axis, while quarterly REIT returns are measur ed on the right. Recall that sentiment indices are constructed to have a mean of zero and unit variance. In Panels C and D of Figure 2 2, I plot NPI returns against INDRES and DRES, respectively. Inspection of Panels A and B of Figure 2 2 does not reveal a consistent contemporaneous univariate relation between either of the sentiment measures and REITRET. However, there are several periods, including 2006 to 2008, during which the sentiment measures and REIT returns do appear to move together. This comovement is reflected in a contemporaneous correlation of 0.46 between REITRET and IN DRES over the full sample period. The contemporaneous correlation between REITRET and DRES (Panel B of Figure 2 2 ) is 0.34. Panels C and D of Figure 2 2 reveal that, relative to Panels A and B, private real estate returns appear to better track movements in sentiment than do REIT returns. This is confirmed by a contemporaneous correlation between INDRES and NPIRET of 0.43 (Panel C); the correlation between DRES and NPIRET is 0.57 (Panel D). I also observe this positive relation between private market retur ns an d lagged direct sentiment in both the VAR and long horizon regression models.

PAGE 36

36 Short Run Regression Results Dynamic Relations amongst Sentiment Measures Table 2 5 provides estimates of unconstrained VAR models with two measures of investor sentiment sp ecified as endogenous variables and the following exogenous control variables: TBILL, TERMSP, DEFAULTSP, INFL, MKT, SMB, HML, and UMD. The sample period is 1992:Q2 to 2008:Q4. The first set of VARs (reported in columns 1 through 4) quantifies the relation between INDRES and DRES, expressed in levels and changes, respectively, over the 1992:Q2 to 2008:Q4 sample period. Previously, I documented a strong positive However, inspection of Table 2 5 rev eals that lagged DRES does not explain variation in indirect sentiment regardless of whether INDRES is specified in levels or changes. In the INDRES equation, the estimated coefficient on INDRES t 1 is positive and highly significant, consistent with the hi gh serial correlation of INDRES. Changes in INDRES (column 3) are negatively related to lagged I find similar results; that is, lagged INDRES does not explain current levels or changes in DRES. Rather, the emphasize that, although not reported in Table 2 5, macroeconomic control variables do not explain the variation in either sentiment measure. That is, b oth INDRES and DRES capture investor sentiment unrelated to market fundamentals. 17 The second set of VARs (columns 5 through 8 in Table 2 5) examine the dynamic relation between INDRES and general stock market sentiment (INDSMS) Previously, I documented a contemporaneous correlation of 0.194 between INDRES and INDSMS, suggesting investors 17 In all speci fications, I fail to reject the null hypothesis that the set of exogenous controls in the sentiment VARs are jointly insignificant.

PAGE 37

37 view commerc ial real estate and the general stock market as distinct markets. In the INDRES and I find that lagged stock market sentiment, INDSMS t 1 does not explain the variation in real estate sentiment, regardless of whether sentiment is specifi ed in levels or changes (columns 5 and 7). Similarly, I find no role for INDRES in explaining the variat ion in INDSMS over time. These results provide additional evidence that commercial real estate is an asset class separate and distinct from the general stock market. Dynamic Relations between Returns and Sentiment Table 2 6 and Table 2 7 contain re sults from the estimation of unrestricted VAR models with both real estate returns and changes in sentiment specified as endogenous variables. While the use of levels may capture the impact of sentiment conditional on the current state of investor beliefs, investors are more likely to be concerned with changes in sentiment as asset prices reflect updates in investor expectations. For example, if investor sentimen t decreases, even if the current sentiment level is high, investors may interpret this as negative information (Brown and Cliff, 2004). Although not separately tabulated, specifications utilizing the level of sentiment produce similar results. Table 2 6 r eport s results from the joint estimation of public REIT returns (REITRET) and changes in the two measures of investor sentiment in the commercial real estate market; Table 2 7 contains the corresponding results from the joint VAR estimation of private NPI returns (NPIRET) and c hanges in investor sentiment. I first report results from the estimation of VAR models that include the following exogenous control variables: TBILL, TERMSP, DEFAULTSP, INFL, MKT, SMB, HML, and UMD. In examining aggregate return predi ctability, Baker and Wurgler (2007) show that in periods where sentiment is more than one standard deviation above its average, subsequent equity returns are, on average, lower. To is olate high sentiment periods, I also report specifications that include a sentiment indicator variable for

PAGE 38

38 periods in which sentiment is more than one standard deviation above its average, SENT HIGH t 1 Turning first to results using INDRES without the high senti ment indicator variable (Table 2 6, column 1), I find that curren t quarter REIT returns are positively influenced by REIT returns in quarter t 2, but not by returns in quarter t t 1 and t 2 cannot be distinguished from zero. However, with the addition of the high sentiment dum my variable (column 3 t 1 is positive and statistically significant. Moreover, the coefficient on SENT_HIGH t 1 is negative ( 0.028) and significant (p value=0.010), suggesting that when sentiment is high, subsequent REIT returns are on average lower, consistent with Baker and Wurgler (2007). If DRES is used in place of INDRES (columns 5 through 8 ), prior REIT returns continue to predict current REIT returns with a two quarter lag. Moreover, c hanges in DRES over the prior quarter are positively associated with REIT returns in the curr ent quarter. The addition of a high senti ment indicator variable ( column 7) slightly increases the magnitude and significance of the coeffi t 1 The estimated coefficient on SENT_HIGH t 1 is again negative, but cannot be distinguished from zero in this specification. Overall, the results in Table 2 6 provide support for the hypothesis that quarterly REIT returns are driven, in par t, by changes in investor sentiment over the prior quarter. Turning be seen in column 2 of Table 2 6 changes in INDRES display no relation to lagged REIT returns. Thus, although sentiment appears to predict REIT returns with a one quarter lag, prior REIT returns do not predict investor sentiment. This absence of a feedback loop between returns and sentiment is also found when DRES is used in place of INDRES (columns 6 and 8). The

PAGE 39

39 t 1 t 2 INDRES (columns 2 and 4), are negative and significant, indicating that increases in sentiment predict decreases in subsequent quarters. This result is consistent w ith the negative serial correlation of INDRES reported in Table 2 2 If SENT_HIGH t 1 is included in the sp ecification (column 4 t 1 remains negative and significant. Moreover, the estimated coefficient on SE NT_HIGH t 1 is negative and significant (p value=0.028). Thus, both average REIT returns and changes in sentiment are smaller in magnitude following periods of high sentiment. The VAR results obtained using R EIT returns and DRES (columns 5 through 8) are ve ry similar to those obtained from the joint REITRET and INDRES estimations. That is, increases in DRES predict significant decreases in DRES in the following quarter. Moreover, changes in investor sentiment are related to the level of sentiment in the prior qu arter. Table 2 7 contains VAR results from the joint estimation of private real estate returns (NPIRET) and changes in investor sentiment. When using INDRES (columns 1 through 4), the estimated coefficients on both NPIRET t 1 and NPIRET t 2 are positive and highly significant. This result is expected given the previously discussed autocorrelation in the NPI return series. Controlling for the smoothed nature of NPI returns with lagged NPI returns, the estimated coefficient on t 1 is positive and significant. The magnitude of this short run effect is approximately half as large as the corresponding short run public market sent iment effect observed in Table 2 6. Similar to the public market results, the addition of a high se ntiment indicator variable (column 3 of Table 2 7 ) increases the estimated magnitude and significance of t 1 t 1 remains half as large as its magnitude in the public real estate market. Thus, although limits to a rbitrage in private markets are substantial, the short run price impact of sentiment is smaller than the impact in public markets. This result

PAGE 40

40 reflects, in part, the short run dominance of institutional investors with in the private real estate market. The estimated coefficient on SENT_HIGH t 1 is negative ( 0.009) and significant (p value=0.074). This suggests that, similar to public market returns, private market returns are lower, on average, when sentiment is extremely high in the prior quarter. When DRE S is used in place of INDRES (columns 5 through 8 ), changes in DRES over the prior quarter are positively associated with returns in the current quarter (p value=0.009). The addition of SENT_HIGH t 1 to the sp ecification (column 7 ) marginally increases the t 1 However, similar to the public market results, the estimated coefficient on SENT_HIGH t 1 cannot be distinguished from zero. In summary, the results in Table 2 7 provide strong support for the hypot hesis that changes in investor sentiment are predictive of private market real estate returns in the subsequent quarter. Finally, turn ing intly with the NPIRET equations, inspection of column 2 in Table 2 7 reveals that t 1 and t 2 are negative and significant, indicating that increases in indirect sentiment predict decreases in sentiment in subsequent quart ers. However, in contrast to the public market results, changes in INDRE S are negatively and significantly related to lagged NPI returns. That is, higher NPI returns in quarter t 1 predict smaller changes in sentiment in quarter t. This result also holds in the presence of a high sentiment dummy (column 4). When using NPI retu rns and DRES as the endogenous variables in the VAR specification (columns 6 and 8 ), I find that increases in DRES predict significant decreases in DRES in the following quarter. Moreover, changes in sentiment are associated with the level of sentiment in the prior quarter (column 8). However, changes in DRES are not related to lagged NPI returns.

PAGE 41

41 Taken together, the results reported in Tables 2 6 and 2 7 strongly suggest that changes in investor sentiment are predictive of returns in the following quarter in both public and private com mercial real estate markets. The public market results however, are in contrast to Brown and Cliff (2004) who find no evidence that sentiment predicts subsequent short run returns in public equity markets. Moreover, the magn itude of the short run effect of sentiment on returns is greater in public markets than in private markets, particularly when using the direct measure of investor sentiment. In public markets, I find no evidence that past returns predict changes in sentime nt. Interestingly, in private markets I do find some evidence of a feedback loop between re turns and indirect sentiment. As a robustness check, I also use the Transaction Based NCREIF Index (TBI) in place of the NPI. The TBI is a statistical index of pric e appreciation and total returns constructed using only those properties in the NCREIF database that sold during the quarter. Thus, the TBI is constructed to avoid the potential smoothing problem associated with the NPI. Though not tabulated, the use of th e TBI index in place of the NPI strengthens the magnitude and significance of the sentiment coefficients using both DRES and INDRES, particularly when sentiment is measured in levels. Further evidence on the impact of sentiment on public and private market real estate returns is provided by the VAR generalized impulse response functions displayed in Figure 2 3. Panels A and B depict the response of quarterly REIT returns to a one standard deviation change in indirect (Panel A) and direct real estate sentime nt (Panel B). The middle curve in each figure represents the estimated diffusion of quarterly REIT returns to the shock in sentiment. The remain ing two curves represent the 95 % confidence interval around the estimated response.

PAGE 42

42 Panels A and B of Figure 2 3 reveal an initial increase in REIT returns in response to a shock in investor sentiment, which quickly dissipates over subsequent quarters. Panels C and D display the response of NPI returns to a one standard deviation change in indirect and direct real estate sentiment, respectively. In contrast to the REIT results, the response of private market returns to an innovation in sentiment appears to persist over subsequent quarters. In addition, the magnitude of the initial increase in returns is noticeably s maller than what is observe d in the public market results, consistent with private market investors being better informed and more sophisticated. The long run impacts of sentiment on returns are addressed in detail in the subsequent long h orizon regression analysis Stock Market Sentiment Effects As displayed in Table 2 3, the indirect measure of general stock market sentiment i s negatively correlated with the indirect measure of real estate sentiment ( 0.194) and displ ays limited correlation with the correlations suggest commercial real estate is often viewed by investors as an asset class separate and distinct from the general stock mar ket. To further test this hypothesis, I estimate a set of VAR models in which real estate returns, real estate sentiment, and stock market sentiment are treated as endogenous variables in a three equati on VAR model. In addition to the three endogenous va r iables, I retain the full set of exogenous control variables used previously. Although not reported, the inclusion of general stock mark et sentiment does not affect my real estate VAR results. Private market real estate returns remain positively and signif icantly related to lagged real estate sentiment, both indirect and direct. In addition, the relation between lagged real estate sentiment and public market real estate returns is consistent with prior results. It is important to reiterate that real estate sentiment

PAGE 43

43 and stock market sentiment are not predictive of each other. These results are virtually unchanged if direct real estate sentiment is used in place of indirect real estate sentiment. Long Horizon Regression Results Long Run Relation between Retur ns and Sentiment Shleifer and Vishny (1997) theorize that investor sentiment may have a more prolonged impact on pricing in markets with considerable limits to arbitrage, such as private markets. As the impact of sentiment on pricing deepens, precisely the time when returns to arbitrage would be the greatest, informed investors are unable to take immediate advantage of the mispricing. Consequently, prices may continue to move further away from fundamental values. In the longer run, however, sentiment should eventually be negatively related to subsequent long horizon returns as price reversion occurs. I n highly segmented and illiquid private markets however, it is also more difficult for participants to determine property values. The lack of transactions may cause the effects of sentiment to be revealed more slowly and to be more persistent. Therefore, a positive relation between sentiment and returns may alternatively exist in the longer run as prices are slower to revert to fundamental value. By examining t he empirical relation between investor sentiment and long horizon returns in two markets that share s imilar underlying assets, I am able to shed additional light on the differential effect of investor sentiment on asset pricing in both public and private m arkets. Table 2 8 reports bias adjusted coefficient estimates and bootstrapped p values for long horizon regressions corresponding to t he specification in Equation 2 2 18 In the long run regressions, sentiment is expressed in levels a s I am interested in d epicting longer term patterns in subsequent returns conditional on the state of sentiment in prior periods. Focusing first on the 18 The reported results are robust to the inclusion of additional exogenous controls, including lagged returns, liquidity, a nd dividend yields.

PAGE 44

44 REITRET regressions using the indirect sentiment measure (column 1), the estimated coefficient on INDRES is negative and signi ficant across the one two three and four year horizons. The magnitude and significance of the estimated coefficients also decrease as the return horizon increases beyond the second year. These results, in conjunction with the short run indir ect senti ment results in Table 2 6 are consistent with predictions from the theoretical literature in which the initial sentiment induced mispricing is subsequently reversed in the long run. Recall, i n periods of extremely high sentiment, the reversal begins in th e subsequent period when using INDRES as reported in Table 2 6. However, in the long horiz on REITRET regressions that make use of the direct sentim ent measure (column 2 of Table 2 8 ), there is a one year continuation in returns following periods of sentim ent induced mispricing. This res ult is again consistent with the short run direct senti ment results reported in Table 2 6 where there is a positive short run sentiment impact and no significant high sentiment period reversal. However, the positive relatio n between long horizon REIT returns and DRES is reversed after one year as the coefficient estimate on DRES for a two year horizon cannot be distinguished from zero. Furthermore, the estimated coefficient on DRES is negative and significant for both the th ree and four year return horizons. Thus, the price reversal after the first year is substantial enough to turn the relation between sentiment and REIT returns negative for longer holding periods. The long horizon NPIRET regression results using indirect and direct sentiment are reported in columns 3 and 4 of Table 2 8 (1997) theor y of limits to arbitrage and the hypothesis that private markets suffer from more limited price revelation in the long run, private real es tate markets appear to be more susceptible to prolonged periods of sentiment induced mispricing. 19 The results in Table 2 8 document a positive and 19 T here is also a branch of literature theorizing that delays in information transmission among investors may cause prices to move away from intrinsic value. Daniel, Hirshleifer, and Subrahmanyam (1998) suggest that overreaction

PAGE 45

45 statistically significant relation between investor sentiment and long horizon NPI returns using both measure s of sentiment. That is, following periods of high (low) sentiment, private market returns increase (decrease). These sentiment effects on returns are consistent with the previous short run results (Table 2 7) where I also observe a consistently positive r elation between sentiment and subsequent private market returns, although there is a marginally significant short run reversal effect in periods of very high sentiment when using the indirect sentiment measure. Due to limits to arbitrage, the sentiment ind uced price changes are not reversed horizon returns negative, though the coefficient estimates on sentiment do decrease in magnitude as the return horizon increases. However, if I estimate the margin al sentiment effect per year, I find that it is insignificant in years two through four and becomes significantly neg ative in year five. Overall, the long horizon results strongly suggest that sentiment induced pricing bubbles in pr ivate real estate markets are slow to develop. Moreover, during this particular sample period, they deflate slowly rather than bursting. The long horizon private market results are consistent with Lamont and Thaler (2003b) and Scheinkman and Xiong (2003) who argue that short sales restrictions can make arbitrage costly and lead to discrepancies in prices of economically equivalent assets. These results are also consistent with Froot and Dabora (1999) who report differences between the prices of pairs of la ownership habitats. They surmise that among other factors, country specific sentiment shocks might affect relative prices. Of course, an alternative explanation for the positive long horizon to pr ivate information and underreaction to public information by informed investors tend to produce short term continuation of investment returns, but long term reversals as public information is eventually incorporated into asset prices. Thus, investor sentim ent may cause prices to diverge significantly from fundamental values before a reversion ultimately occurs, particularly when there are significant limits to arbitrage.

PAGE 46

46 returns I find is that commercial real estate assets were consistently under priced at acquisition during our sample period. However, given that REITs were also purchasing and selling their assets in the same private market, this explanation does not seem plausible. Table 2 9 reports the economic magnitude of a one standard deviation increase in sentiment on returns over the indicated horizon. Following Brown and Cliff (2005), I take the bias adjusted coefficient estimates reporte d in Table 2 8 and multiply by the number of quarters in the specified return horizon a nd the standard deviation of the appropriate sentiment variable. Focusing first on the public market results, a one standard deviation increase in INDRES is associated w ith a 2.9 % decrease in returns over the subsequent one year period. The longer horizon economic impact rea ches a negative trough of 11.1 % at the two year return horizon and then rises thereafter. Using DRES and REITRET (column 2 of Table 2 9 ), a one standa rd deviation increase i n DRES is associated with a 3.8 % increase in returns over the subsequent one year period. This continuation of returns is subsequently reversed at longer horizons as a one standard deviation shock to sentiment is associated with a r eduction i n returns of 3.6, 9.7, and 12.3 % over the two, three, and four year horizons respectively. This increasing negative impact is consistent with the market offsetting both a large contemporaneous sentiment effect as well as any continuation of retur ns that persists into the subsequent quarter(s). In the private real estate market, a one standard deviation increase in INDRES is associated with a 3.9 % increase in returns over the next year. Using DRES, the magnitude of the initial increase in returns f ollowing a standard deviation shock to sentiment is slightly larger than is the case when using INRES. Over the subsequent one ye ar period, returns increase 5.5 % Over longer horizons, returns continue to increase, although the marginal change in returns d ecreases over time. These results are consistent with limits to arbitrage fostering persistence in

PAGE 47

47 sentiment induced mispricing in private real estate markets and provide evidence in regards to the relative importance of price revelation and limits to arbi trage in the long run effect of investor sentiment on asset prices. While private markets may have more well informed institutional investors, significant limits to arbitrage prevent these investors from taking swift action against short run asset misprici ng. In summary, the impact of investor sentiment on similar underlying assets depends in part on whether the investments are owned and managed in private or public markets. In both markets, excessive optimism (pessimism) drives prices above (below) fundam ental value in the short run. However, the speed of price revelation and the degree to which investors face restrictions in capitalizing on asset mispricing, play important roles in determining the time it takes for prices to revert to intrinsic values ove r the long run. In public real estate markets, periods of sentiment induced mispricing are followed by quicker price reversals. In private real estate markets, on the other hand, the initial sentiment induced price response is more persistent as the impact of investor sentiment slowly dissipates over time. Long Run Stock Market Sentiment Effects As discussed previously in the VAR analysis, I find that the inclusion of stock mark et sentiment does not affect the short run results. To examine if stock market s entiment impacts real estate returns in the long run, I implement long hori zon regressions that include the indirect real estate sentiment measure, INDRES, as well as the indirect measure of general stock market sentiment INDSMS.I retain the full set of e xogenous control variables used previously. Although not tabulated, the inclusion of stock mark et sentiment does not affect the long run results. Coefficient estimates and statistical significance of the real estate sentiment variable are virtually unchang ed from the results reported in Table 2 8 while the long run impact of stock market sentiment on real estate returns is not statistically different from zero.

PAGE 48

48 Summary and Conclusion Sentiment is the component of investor expectations not based on a carefu l and complete theories of De Long et al. (1990) and Shleifer and Vishny (1997), a growing empirical literature has begun to focus on measuring and quantifyin g the effects of investor sentiment on asset pricing. Although results vary, a number of recent articles document a significant role for sentiment in the valuation of assets in public stock markets. The focus on public markets in the existing literature i s understandable given the difficulties associated with obtaining return information on investments that trade in private markets. Nevertheless, private investment markets are large in size and provide an appealing pricing role. Relative to more liquid public markets, private investment markets exhibit significant information asymmetries and illiquidity. Moreover, the inability to short sell in private markets impedes the opportunity for informed arbitrageurs to coun teract mispricing. I posit that investor sentiment plays a more persistent role in pushing asset prices away from their fundamental values in private markets because of increased illiquidity and limits to arbitrage. No previous research, however, has dire ctly investigated the relative importance of sentiment in public and private asset markets. This paper provides a contribution to the investor sentiment literature by examining the short and long run relation between sentiment and the pricing of similar underlying assets that are owned and traded in two distinct investment environments. In addition, this study provides evidence of the relative importance of sentiment driven investor demand and limits to arbitrage in asset mispricing within both the short and long run.

PAGE 49

49 Using vector autoregressive (VAR) models, I find evidence of a positive relation between investor sentiment and subsequent quarter returns in both public and private real estate markets. The magnitude of this short run effect is larger in the public real estate market, which is consistent with private market investors being better informed and more sophisticated. Using long horizon regressions, I also provide evidence that periods of sentiment induced mispricing are followed by quicker pric e reversals in public real estate markets. In contrast, private real estate markets are more susceptible to prolonged periods of sentiment induced mispricing. These results support the hypothesis that limits to arbitrage and delays in price revelation play important roles in determining the time it takes for prices to revert to fundamental values.

PAGE 50

50 A B Figure 2 1. Investor s entiment i ndices A) This figure plots levels of direct (DRES) and indirect (INDRES) investor sentiment indices for the commercia l real estate market from 1992 : Q2 to 2008 : Q4. B) This figure plots levels of indirect investor sentiment indices for the commercial real estate market (INDRES) and the general stock market (INDSMS) from1992 : Q2 to 2008 : Q4.

PAGE 51

51 A B C D Figure 2 2 Investor s entiment and r eturns. A) This figure plots levels of the indirect (INDRES) investor sentiment index for the commercial real estate market and the aggregate return index for publicly traded REITs (REITRET) from 1992 : Q2 to 2008 : Q4. B) This figure plots levels of the direct (DRES) investor sentiment index for the commercial real estate market and the aggregate return index for publicly traded REITs (REITRET) from 1992 : Q2 to 2008 : Q4. C) This figure plots levels of the indirect (INDRES) in vestor sentiment index for the commercial real estate market and the aggregate return index for the private commercial real estate market (NPIRET) from 1992 : Q2 to 2008 : Q4. D) This figure plots levels of the direct (DRES) investor sentiment index for the co mmercial real estate market and the aggregate return index for the private commercial real estate market (NPIRET) from 1992 : Q2 to 2008 : Q4.

PAGE 52

52 A B C D Figure 2 3. Impulse response f unctions : investor sentiment and returns This figure represen ts the generalized impulse response functions corresponding to estimated VAR models from Table 2 6 and Table 2 7 with sentiment measured in changes and including the high sentiment indicator. A) This figure depicts the percentage response in the aggregate return index for publicly traded REITs (REITRET) to a one standard deviation change in the indirect (INDRES) investor sentiment index for the commercial real estate market B) This figure depicts the percentage response in the aggregate return index for pu blicly traded REITs (REITRET) to a one standard deviation change in the direct (DRES) investor sentiment index for the commercial real estate market. C) This figure depicts the percentage response in the aggregate return index for private commercial real e state (NPIRET) to a one standard deviation change in the indirect (INDRES) investor sentiment index for the commercial real estate market. This figure depicts the percentage response in the aggregate return index for private commercial real estate (NPIRET) to a one standard deviation change in the direct (DRES) investor sentiment index for the commercial real estate market.

PAGE 53

53 Table 2 1. Descriptive statistics: sentiment p roxies. Serial Correlation Sentiment Proxy Mean Median St. Dev Min Max Levels Changes INDRES REIT NAV Premium 0.033 0.031 0.115 0.244 0.302 0.81 *** 0.31 ** Percentage of Properties Sold 0.022 0.023 0.013 0.002 0.057 0.60 *** 0.48 *** Number of REIT IPOs 2.493 1.000 4.069 0.000 20.000 0.83 *** 0.06 Avg. First Day Ret. REIT IPO 0.039 0.025 0.052 0.025 0.268 0.06 0.58 *** Share of REIT Equity Issues 0.221 0.234 0.411 2.044 1.674 0.34 *** 0.77 *** Commercial Mortgage Flows 0.002 0.004 0.011 0.029 0.024 0.60 *** 0.41 *** REIT Mutual Fund Flows 0.304 0.160 1.295 4.513 3.605 0.40 *** 0.11 DRES Apartment 6.300 6.400 0.700 3.900 7.600 0.61 *** 0.38 *** Industrial R&D 5.100 5.100 0.800 3.000 6.700 0.80 *** 0.30 ** Industrial Warehouse 6.300 6.300 0.600 4.500 7.700 0.72 *** 0.40 *** CBD Office 5.500 5.700 1.100 2.800 7.300 0.88 *** 0.19 Suburban Office 5.300 5.300 1.200 2.800 7.500 0.92 *** 0.20 Neighborhood Retail 5.900 6.000 0.700 3.400 7.200 0.60 *** 0.38 *** Power Center 5.000 5.000 1.100 2.700 6.800 0.85 *** 0.38 *** Regional Malls 5.400 5.400 0.800 2.900 6.700 0.59 *** 0.44 *** INDSMS Closed end Fund Discount 0.071 0.067 0.036 0.013 0.191 0.92 *** 0.19 NYSE Share Turnover 0.136 0.136 0.129 0.249 0.543 0.33 *** 0.51 *** Number of IPOs 29.42 23.00 23.10 0.000 106.0 0.80 *** 0.27 *** Avg. First Day Ret. IPOs 0.193 0.140 0.212 0.199 1.162 0.76 *** 0.44 *** Share of Equity Issues 0.118 0.102 0.073 0.015 0.539 0.61 *** 0.33 *** Dividend Premium 0.147 0.137 0.122 0.602 0.128 0.9 2 *** 0.06 This table reports descriptive statistics for each component of the indirect real estate sentiment index (INDRES), the direct real estate sentiment index (DRES), and the indirect stock market sentiment index (INDSMS). Mean, median, standard de viation, minimum, maximum, and serial correlation of levels and changes are reported. Descriptive statistics are reported in decimal form. The sample period spans 1992:Q2 to 2008:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively

PAGE 54

54 Table 2 2. Descriptive statistics: aggregate sentiment measures. Serial Correlation Sentiment Measure Mean Median St Dev Min Max Levels Changes INDRES 0.000 0.070 1.000 2.146 3.051 0.73 *** 0.29 ** DRES 0.000 0.117 1.000 3.508 1.7 62 0.81 *** 0.35 *** INDSMS 0.426 0.343 0.651 0.758 2.799 0.76 *** 0.14 This table reports descriptive statistics for each aggregate sentiment index: the indirect real estate sentiment index (INDRES), the direct real estate sentiment index (DRES), and the indirect stock market sentiment index (INDSMS). Mean, median, standard deviation, minimum, maximum, and serial correlation of levels and changes are reported. INDRES is the first principal component extracted from seven underlying proxies of investor sentiment in commercial real estate markets : i) the REIT price premium to net asset value, (ii) the percentage of properties sold from the NCREIF Property Index, (iii) the number of REIT IPOs, (iv) the average first day returns on REIT IPOs (v) the share of net REIT equity issues in total net REIT equity and debt issues, (vi) commercial mortgage flows as a percentage of GDP, and (vii) capital flows to dedicated REIT mutual funds. DRES is the fir st principal component extracted from investment condition survey responses pertaining to eight property types that are published quarterly by the Real Estate Research Corporation (RERC) in the ir Real Estate Report INDSMS is the first principal component extracted from six underlying proxie s of investor sentiment in the stock market: (i) the average difference between the net asset values of closed end fund stock shares and their market prices, (ii) share turnover on the NYSE, (iii) the number of IPOs, (iv) the average first day returns on IPOs, (v) the share of equity issues in total equity and debt issues, and (vi) the dividend premium. Each index is standardized to have a mean of zero and unit variance for the period over which it was generated. Descriptive statistics are reported in decimal form. The sample period spans 1992:Q2 to 2008 :Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 55

55 Table 2 3. Descript ive statistics and correlations: aggregate sentiment measures. Sentiment Measure INDRES DRES INDSMS INDRES 1.000 DRES 0.47 8 *** 1.000 INDSMS 0.194 0.275 *** 1.000 This table reports correlations between each aggregate sentiment index: the indirect real estate sentiment index (INDRES), the direct real estate sentiment index (DRES), and the indirect stock market sentiment index (INDSMS). INDRES is the first principal component extracted from seven underlying proxies of investor sentiment in commercial real estate markets : i) the REIT price premium to net asset value, (ii) the percentage of properties sold from the NCREIF Property Index, (iii) the number of REIT IPOs, (iv) the average first day returns on REIT IPOs (v) the share of net REIT equity issues in total net REIT equity and debt issues, (vi) commercial mortgage flows as a percentage of GDP, and (vii) capital flows to dedicated REIT mutual funds. DRES is the fir st principal component extracted from investment condition survey responses pertaining to eight property types that are published quarterly by the Real Estate Research Corporation (RERC) in the ir Real Estate Report INDSMS is the first principal component extracted from six underlying proxie s of investor sentiment in the stock market: (i) the average difference between the net asset values of closed end fund stock shares and their market prices, (ii) share turnover on the NYSE, (iii) the number of IPOs, (iv) the average first day returns on IPOs, (v) the share of equity issues in total equity and debt issues, and (vi) the dividend premium. The sample period spans 1992:Q2 to 2008:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 56

56 Table 2 4 Descriptive statistics: return series and control variables. Variabl e Mean Median Std Dev Min Max Serial Correlation REITRET 0.009 0.012 0.028 0.125 0.058 0.04 NPIRET 0.023 0.025 0.020 0.083 0.054 0.72 *** TBILL 0.038 0.044 0.016 0.003 0.062 0.96 *** TERMSP 0.016 0.015 0.012 0.006 0 .036 0.93 *** DEFAULTSP 0.009 0.008 0.003 0.006 0.030 0.83 *** INFLA 0.006 0.006 0.009 0.039 0.025 0.13 MKT 0.004 0.012 0.082 0.223 0.196 0.01 SMB 0.007 0.002 0.055 0.108 0.191 0.01 HML 0.004 0.004 0 .077 0.320 0.206 0.16 UMD 0.027 0.019 0.080 0.201 0.260 0.07 This table reports descriptive statistics for two quarterly return series, as well as macroeconomic/risk control variables. REITRET measures returns in public commercial r eal estate markets and is obtained from the CRSP/ZIMAN database. It is the quarterly value weighted aggregate return index for U.S. equity REITs. NPIRET measures returns in private commercial real estate markets and is provided by the National Council of R eal Estate Investment Fiduciaries (NCREIF). It is a quarterly index tracking the total return performance of a large pool of individual commercial real estate properties acquired in the private market for investment purposes only. Macroeconomic/risk contro l variables include the annualized yield on three month U.S. Treasury securities (TBILL), the annual slope of the Treasury term structure of interest rates (TERMSP), the annual spread between yields on BAA rated and AAA rated corporate bonds (DEFAULTSP), q uarterly inflation (INFLA), as well as the three Fama French risk factors: MKT, SMB, and HML augmented by a return momentum factor, UMD Mean, median, standard deviation, minimum, maximum, and serial correlation are reported. Descriptive statistics are rep orted in decimal form. The sample period spans 1992:Q2 to 2008:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 57

57 Table 2 5 Dynamic relations amongst sentiment measures. Direct and Indirect Real Estate Sentiment Indi rect Real Estate and Stock Market Sentiment Sentiment in Levels Sentiment in Changes Sentiment in Levels Sentiment in Changes End. Variables INDRES DRES DRES DRES INDRES INDSMS INDRES INDSMS Constant 0.447 1.098 1.507 0.957 0.722 0.010 1.252 0.149 (0.654) (0.185) (0.129) (0.217) (0.468) (0.986) (0.203) (0.815) DRES t 1 0.245 0.447 *** 0.099 0.420 *** (0.148) (0.001) (0.567) (0.002) DRES t 2 0.197 0.303 ** 0.143 0.010 (0.246) (0.032) (0.416) (0.945) INDRES t 1 0.372 *** 0.058 0.407 *** 0.006 0.625 *** 0.080 0.293 ** 0.045 (0.010) (0.626) (0.003) (0.958) (0.000) (0.224) (0.020) (0.583) INDRES t 2 0.1 01 0.039 0.240* 0.087 (0.426) (0.710) (0.064) (0.389) INDSMS t 1 0.106 0.543 *** 0.048 0.179 (0.554) (0.000) (0.799) (0.144) Adjusted R 2 0.56 0.70 0.04 0.14 0.53 0.60 0.03 0.04 This table presents results obtained from estimating unrestricted vector autoregressive (VAR) models with investor sentiment measures specified as endogenous variables. INDRES is the indirect real estate sentiment index, DRES is the direct real estate sentiment index, and INDSMS is the indirect stock market sentiment index. Bivariate models with exogenous controls are estima ted. Exogenous control variables include the yield on three month U.S. Treasury securities (TBILL), the slope o f the Treasury term structure of interest rates (TERMSP), the spread between yields on BAA rated and AAA rated corporate bonds (DEFAULTSP), inflation (INFLA), as well as the three Fama French risk factors: MKT, SMB, and HML augmented by a return momentum f actor, UMD The lag length of the VAR is chosen by looking at the AIC, SBIC, and the likelihood ratio for various choices of p Two lags prov ide the best fit for comparing real estate sentiment measures, while one lag is deemed appropriate for comparison o f real estate and stock market sentiment measures. The sample period spans 1992:Q2 to 2008:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 58

58 Table 2 6 Dynamic relations between public market returns and sentiment. INDRES (Indirect Sentiment) DRES (Direct Sentiment) End. Variables REITRET SENT REITRET SENT REITRET SENT REITRET SENT Constant 0.024 1.437 0.049 0.914 0.012 0.893 0.006 0.685 (0.537) (0.147) (0.21 1) (0.353) (0.772) (0.251) (0.875) (0.369) REITRET t 1 0.162 5.108 0.088 6.728 0.112 3.145 0.083 1.992 (0.503) (0.400) (0.706) (0.254) (0.611) (0.464) (0.708) (0.636) REITRET t 2 0.554 *** 2.530 0.691 *** 5.512 0.416 ** 1.038 0.440 ** 2.015 (0.010) (0.638) (0.001) (0.304) (0.020) (0.766) (0.014) (0.556) t 1 0.009 0.448 *** 0.012 ** 0.367 ** 0.013 0.392 *** 0.014 ** 0.327 ** (0.155) (0.004) (0.039) (0.017) (0.072) (0.004) (0.047) (0.016) SENT t 2 0.004 0.270* 0.002 0.226 0.008 0.007 0.009 0.027 (0.533) (0.081) (0.756) (0.132) (0.268) (0.959) (0.219) (0.841) SENT_HIGH t 1 0.028 *** 0.597 ** 0.009 0.377 ** (0.010) (0.028) (0.340) (0.0 47) Adjusted R 2 0.04 0.04 0.11 0.09 0.04 0.14 0.03 0.17 This table presents results obtained from estimating two unrestricted vector autoregressive (VAR) models with investor sentim ent and public market return m easures specified as endogenous variables. INDRES is the indirect real estate sentiment index and DRES is the direct real estate sentiment index. REITRET measures returns in public commercial real estate markets and is obtained from the CRSP/ZIMAN database It is the quarterly value weighted aggregate return index for U.S. equity REITs. Exogenous control variables include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSP), the sprea d between yields on BAA rated and AAA rated corporate bonds (DEFAULTSP), inflation (INFLA), as well as the three Fama French risk factors: MKT, SMB, and HML augmented by a return momentum factor, UMD The high sentiment dummy variable, SENT_HIGH, takes a v alue of one for periods in which the sentiment measure is greater than one standard deviation above its mean value over the sample period. The lag length of the VAR is chosen by looking at the AIC, SBIC, and the likelihood ratio for various choices of p T wo lags provide the best fit for estimation. The sample period spans 1992:Q2 to 2008:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 59

59 Table 2 7 Dynamic relations between private market returns and sentiment. INDRES (Indirect Sentiment) DRES (Direct Sentiment) End. Variables NPI RET SENT NPIRET SENT NPIRET SENT NPIRET SENT Constant 0.002 0.063 0.007 0.188 0.004 0.768 0.006 0.614 (0.931) (0.964) (0.767) (0.892) (0.850) (0.505) (0.793) (0.583) NPIRET t 1 0.452 ** 24.131 ** 0.516 *** 20.922 ** 0.512 *** 4.364 0.520 *** 5.043 (0.011) (0.018) (0.004) (0.042) (0.002) (0.586) (0.002) (0.516) NPIRET t 2 0.811 *** 10.607 0.895 *** 14.858 0.67 8 *** 2.312 0.666 *** 3.415 (0.000) (0.363) (0.000) (0.207) (0.000) (0.800) (0.000) (0.699) t 1 0.004 0.376 *** 0.006 ** 0.283 0.007 *** 0.418 *** 0.008 *** 0.342 *** (0.096) (0.005) (0.022) (0.051) (0.009) (0.002) (0.005) (0.010) SENT t 2 0.001 0.298 ** 0.003 0.213 0.003 0.025 0.003 0.020 (0.686) (0.019) (0.272) (0.119) (0.313) (0.855) (0.239) (0.884) SENT_HIGH t 1 0.009 0.436 0.004 0.380** (0.074) (0.123) (0.286) (0.043) Adjusted R 2 0.60 0.11 0.61 0.12 0.62 0.13 0.62 0.17 This table presents results obtained from estimating two unrestricted vector autoregressiv e (VAR) models with investor sentiment and private market return measures specified as endogenous variables. INDRES is the indirect real estate sentiment index and DRES is the direct real estate sentiment index. NPIRET measures returns in private commercia l real estate markets and is provided by the National Council of Real Estate Investment Fiduciaries (NCREIF). It is a quarterly index tracking the total return performance of a la rge pool of individual commercial real estate properties acquired in the priv ate market for investment purposes only. Exogenous control variables include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSP), the spread between yields on BAA rated and AAA rate d corporate bonds (DEFAULTSP), inflation (INFLA), as well as the three Fama French risk factors: MKT, SMB, and HML augmented by a return momentum factor, UMD The high sentiment dummy variable, SENT_HIGH, takes a value of one for periods in which the senti ment measure is greater than one standard deviation above its mean value over the sample period. The lag length of the VAR is chosen by looking at the AIC, SBIC, and the likelihood ratio for various choices of p Two lags provide the best fit for estimatio n. The sample period spans 1992:Q2 to 2008:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 60

60 Table 2 8 Long horizon regression results: coefficient estimates and statistical significance REITRET NPIRET Return Horizon INDRES DRES INDRES DRES One year 0.007 0.010 0.010 *** 0.014 *** (0.067) (0.096) (0.000) (0.000) Two year 0.014 *** 0.004 0.003 ** 0.008 *** (0.000) (0.313) (0.017) (0.000) Three year 0.007 *** 0.008 ** 0.004 *** 0.006 *** (0.002) (0.026) (0.001) (0.000) Four year 0.004 ** 0.008 ** 0.003 *** 0.005 *** (0.034) (0.018) (0.001) (0.002) This table presents results for the long horizon regressions. The long horizon regression specification is as follows: (r t+1 t+k t t t where r t+1 t+k are log returns, k is the number of quarters over which the particular horizon spans t is the set of control variables, and S t is a measure of investor sentiment in commercial real estate. INDRES is the indirect real estate sentiment index and DRES is the direct real estate sentiment index. REITRET measures returns in public commercial real estate markets and is obtained from the CRSP/ZIMAN database. It is the quarterly value weight ed aggregate return index for U.S. equity REITs. NPIRET measures returns in private commercial real estate markets and is provided by the National Council of Real Estate Investment Fiduciaries (NCREIF). It is a quarterly index tracking the total return per formance of a large pool of individual commercial real estate properties acquired in the private market for investment purposes only. The set of control variables includes the yield on the three month Treasury bill, the slope of the Treasury term structure of interest rates, the spread between yields on BAA rated and AAA rated corporate bonds, inflation, the three Fama French risk factors (MKT, SMB, and HML) and a return momentum factor ( UMD ). Bias adjusted coefficient estimates and their associated p value s utilize adjustments derived from the bootstrap simulation procedure documented in the Appendix. 10,000 bootstrap simulations were run for each long horizon return series within each property type. The sample period spans 1992:Q2 to 2008:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 61

61 Table 2 9 Economic significance of long horizon regression results REITRET NPIRET Return Horizon INDRES DRES INDRES DRES One year 0.029 0.038 0.03 9 0.055 Two year 0.111 0.036 0.027 0.066 Three year 0.087 0.097 0.042 0.076 Four year 0.072 0.123 0.056 0.075 This table presents results for the economic significance of coefficient estimates from the long horizon regressions. The long horizon regression specification is as follows: (r t+1 t+k t t t where r t+1 t+k are log returns, k is the number of quarters over which the particular t is the set of control variables, and S t is a measure of investor sentiment in comme rcial real estate. INDRES is the indirect real estate sentiment index and DRES is the direct real estate sentiment index. REITRET measures returns in public commercial real estate markets and is obtained from the CRSP/ZIMAN database. It is the quarterly va lue weighted aggregate return index for U.S. equity REITs. NPIRET measures returns in private commercial real estate markets and is provided by the National Council of Real Estate Investment Fiduciaries (NCREIF). It is a quarterly index tracking the total return performance of a large pool of individual commercial real estate properties acquired in the private market for investment purposes only. The set of control variables includes the yield on the three month Treasury bill, the slope of the Treasury term structure of interest rates, the spread between yields on BAA rated and AAA rated corporate bonds, inflation, the three Fama French risk factors (MKT, SMB, and HML) and a return momentum factor ( UMD ). To calculate the economic magnitude of a one standard deviation shock to sentiment I take the bias adjusted coefficient estimates reported in Table 2 8 and multiply by the number of quarters in the specified return horizon and the standard deviation of the appropriate sentiment variable. Results are reported in decimal form. The sample period spans 1992:Q2 to 2008:Q4.

PAGE 62

62 CHAPTER 3 CR EDIT AVAILABILITY AN D ASSET PRICING SPIR ALS IN ILLIQUID MARK ETS When asset prices crashed in the late 1980s, a branch of theoretical literature focusing on credit cycles emerged th willingness or ability to provide credit (Bernanke and Gertler, 1989; Kiyotaki and Moore, 1997). In response to the latest credit crunch, however, the relation between asset prices, the availability of credit, and market liquidity has garnered increased attention (e.g., Geanakoplos, 2003; Garleanu and Pedersen, 2007; Longstaff and Wang, 2008; Brunnermeier and Pedersen 2009). Moreover, a number of recent empirical papers testing these the ories have emerged, including studies examining how changes in credit availability impact bank balance sheets (Adrian and Shin, 2010), hedge fund performance (Dudley and Nimalendran, 2010; Boyson, Stahel, and Stulz, 2010), and market liquidity in the gener al stock market (Hameed, Wang, and Viswanathan, 2010). Overall, these studies find that credit constraints affect asset values at times when public markets are relatively illiquid or when assets are highly leveraged. These recent empirical papers focus on brief periods of illiquidity and credit tightening in relatively liquid markets. However, the existing literature does not address the relation between the availability of credit and asset prices in markets that are persistently illiquid and composed of hi ghly leveraged assets, such as the private commercial real estate market. The relative illiquidity and significant use of leverage in this market setting allows one to better isolate and measure the impact of credit availability on asset prices in both per iods of credit tightening and easing. Furthermore, no prior research has tested the relative impact of changes in credit availability on assets with claims on similar cash flows that are traded in markets with different liquidity. However, tests using both private and public real estate markets create a clearer picture of the

PAGE 63

63 interaction between credit availability, market liquidity, and asset price movements in markets with significantly different degrees of liquidity. In this study I test the relation be tween chang es in credit availability market liquidity, and asset prices in both private and public commercial real estate markets. I first examine the time varying relation between changes in credit avail ability and market liquidity. I simultaneously test the conditional short run impact of changes in credit availability on asset prices in bot h private and public markets. I then test whether changes in asset prices reinforce changes in the availability of credit, thereby creating a spiral effect The comm ercial real estate market provides an appealing testing ground for examining the dynamic relation between credit availability, market liquidity, and changes in asset prices. First, the private commercial real estate market is a relatively illiquid market c onsisting of highly leveraged assets. 1 Therefore, one would expect asset prices in this market to be relatively sensitive to changes in credit availability not only when credit conditions are tightened, but also when lending standards are eased. Second, un like other private markets, several representative price return indices for private commercial real estate are available, permitting the calculation of time weighted price returns that can be compared directly to corresponding price changes in public real estate markets. Third, the underlying properties held by the publicly traded real estate firms I analyze are similar to the property holdings of the institutional real estate investors whose private market returns I also track. Controlling for the security characteristic embedded in public real estate market returns, disparities in the impact of credit availability on price changes in private and public real estate markets can be ascribed to differences in the liquidity characteristics of these two markets. 1 As a conservative estimate of leverage, the American Council of Life Insurer s reports that the average loan to value ratio on commercial real estate properties was approxima tely 68% over the period 1992Q2 to 2008Q4.

PAGE 64

64 Using vector autoregressive (VAR) models, I address two questions. First, do changes in credit availability affect asset price returns over and above the impact of other fundamental control variables? When investors find it difficult to obtain credit for acquisitions or refinancing, market liquidity decreases, which, in turn, puts downward pressure on prices. In contrast when access to credit is eased, an increase in the use of leverage for acquisitions will put upward pressu re on asset prices. Therefore I test whether changes in credit availability, after controlling for the impact of market liquidity, impact asset prices. Second, do price changes affect changes in future credit availability over and above the impact of other fundamental control variabl es? As prices fall in an ill iquid market the risk of financing an additional transaction rises and credit m arkets tighten further. On the other hand as asset prices increase, lenders may further ease credit standards to take advantage of this timely p rof it opportunity. Consequently, I also test the hypothesis that subsequent price changes will lead to further changes in credit availability for assets that are traded i n relatively illiquid markets. I find that credit availability is a significant determin ant of subsequent price movements in both private and public commercial real estate markets In particular, a tightening (easing) in bank lending standards is negatively (positively) related to subsequent price movements in both the private property and pu blic REIT markets, even after controlling for the impact of market liquidity and other fundamentals. These results suggest that leverage is a key factor in determining credit market effects on pricing. I also find evidence of a spiral effect, in which chan ges in price movements rei nforce future changes in credit availability. However, this spiral effect is concentrated in relatively illiquid segments of the commercial real estate market. In particular, I document a feedback effect of lagged price changes on subsequent credit availability in the private commercial real estate market, the lowest liquidity quartiles of the public

PAGE 65

65 commercial real estate market, and the relatively illiquid market for REIT preferred shares. Consistent with the theoretical predicti ons of Brunnermeier and Pedersen (2009) and Geanakoplos (2003), these results suggest that while leverage plays a significant role in determining the effects of credit availability on pricing, the underlying liquidity with which assets trade is a key facto r in determining the likelihood of a liquidity spiral. In short, a lack of liquidity creates the market setting for a spiral effect. Anecdotal evidence and recent theoretical work by Shleifer and Vishny (2010) also suggest that investor sentiment plays a role in credit market effects. Therefore, as an additional robustness check I also include a proxy for investor sentiment as a fourth endogenous variable in my VAR specifications. Consistent with Shleifer and Vishny (2010), I find that changes in investor sentiment predict future changes in credit availability I also find weak evidence of a feedback effect between changes in credit availability and future changes in investor sentiment. These results suggest banks respond to increasing investor sentiment by easing credit standards, thereby making credit more readily available to potential investors when they are most optimistic. However, during a market downturn in which investors are becoming increasingly pessimistic, banks tend to tighten lending standards which can have a destabilizing effect on asset prices. This raises an interesting policy question pertaining to the extent to which lenders and their regulators have the ability to reduce the probability that an asset pricing bubble emerges by restrictin g the amount of credit provided during boom periods or ease the severity of a downturn by making credit available when markets are distressed. Background Literature and Research Development Recent theoretical work in the asset pricing literature makes an i mportant distinction between two types of liquidity: market liquidity and funding liquidity (Brunnermeier and Pedersen, 2009; Brunnermeier, 2009). Market liquidity refers to the ease with which an investor

PAGE 66

66 can find another party to take the opposite side o f a transaction. Funding liquidity refers to the ease with which an investor can obtain capital from a financier I n other words funding liquidity is the degree to which investors have access to credit In an efficient market, asset price movements should be invariant to changes in credit availability when markets are relatively liquid. Any deviations in price that result from a temporary credit shock should be quickly arbitraged away (Fama, 1970). However, arbitrageurs may at times face funding constraint s and be unable to provide market liquidity when it is needed the most (Shleifer and Vishny, 1997). When funding constraints are so severe that investors can no longer maintain their existing positions, asset values drop significantly and markets become il liquid, as was evident in the Long Term Capital Management (LTCM) crisis of 1998. Recent developments in financial markets further suggest that shocks to market conditions can be severe enough to cause market liquidity to fluctuate significantly over time. Brunnermeier and Pedersen (2009) establish a theoretical basis for the endogenous variation of market conditions and credit requirements. In particular, they focus on the relation between credit constraints and asset pricing, the amplifying effects of the ir interaction, and differences in these impacts across high and low leverage securities during periods of market illiquidity. When credit availability is tight, traders become reluctant to take on positions, ghly leveraged assets. This lack of credit availability lowers market liquidity. Under certain conditions, decreased market liquidity increases the risk of financing an acquisition, thus further tightening the availability of funds for investment. This is what Brunnermeier and Pedersen (2009) refer to as a margin spiral. When credit constraints are tightened in periods of market illiquidity, the effect may be destabilizing and lead to a second type of liquidity spiral that has an amplifying effect on asset prices If credit constraints continue

PAGE 67

67 to be tightened as asset prices fall, investors may be forced to de lever their position by selling a ssets into an illiquid market. In short, declining asset prices and tighter credit constraints may reinforce one ano ther causing a spiral to ensue. Brunnermeier and Pedersen (2009) refer to this as a loss spiral. Therefore, when asset markets are relatively illiquid, prices may be driven more in the short run by changes in the availability of credit than by movements in fundamentals. Geanakoplos (2003) presents a similar collateral based theory of such spirals. When investors can readily obtain credit to purchase or refinance an asset, optimistic investors are able to hold a larger fraction of the capital stock. Geanakop los (2003) calls these optimistic investors to levels that exceed fundamental values. These price increases, in turn, increase the confidence and risk toleran ce of lenders, thereby fueling a relaxation of underwriting standards. This increased availability of credit may put additional upward pressure on asset prices, further increasing the confidence of lenders. Said differently, access to leverage is pro cycli cal. However, if a negative credit shock occurs that increases lender uncertainty (for example, the bankruptcy of Lehman Brothers in 2008), lenders may tighten credit standards in response. This tightening in credit markets may, in turn, force highly lever aged optimists to de lever their current positions and use less leverage on new acquisitions, thereby putting downward pressure on prices. Since it is now harder to borrow money, optimistic buyers are less able to obtain financing for investment. Therefore asset markets become less liquid and the proportion of decrease s This decrease in market liquidity puts downward pressure on prices and further increases the risk of providing financing. Therefore, asset prices will be sen sitive to changes in the availability of credit in markets in which transactions are highly leveraged.

PAGE 68

68 Previous empirical research has examined the role of market liquidity in asset pricing. Amihud (2002 ) finds that aggregate market liquidity is an importa nt factor in determining a variable in asset pricing. However few empirical analyses have tested the interaction between market liquidity, credit constraints, and asset prices. Hameed, Kang, and Viswanathan (2010) find that periods of significant negative stock market returns cause market liquidity to decrease, par ticularly during times in which funding constraints are tight. Dudley and Nimalendran (2010) provide evidence that returns on leveraged hedge funds are highly sensitive to changes in funding risk, especially during periods of market illiquidity. Finally, A drian and Shin (2010) find that the amount of leverage carried on bank balance sheets is sensitive to asset price movements of the underlying collateral, especially during a financial market downturn. Overall, the extant theoretical literature predicts th at the sensitivity of market liquidity and asset prices to the availability of funds for investment is most pronounced for assets that are relatively illiquid and highly leveraged. However, no research to date has tested the dynamic relation between the av ailability of credit and asset pricing for highly leveraged assets that trade in relatively illiquid markets. By testing the impact of changes in credit availability on asset prices within the relatively illiquid and highly leveraged private co mmercial rea l estate market, this study provides an experimental setting that allows one to better isolate and measure the potential impact of credit market availability on asset prices in both periods o f credit easing and tightening, thereby providing a unique contri bution to the literature. I further extend this contribution by measuring the relative impact of credit availability on prices of similar assets traded in markets with different liquidity. Tests using a parallel market

PAGE 69

69 with varying degrees of liquidity cre ate a clearer picture of the interaction between credit availability, liquidity, and asset price movements. Prior empirical research has shown that differences in market liquidity can lead to discrepancies in prices of economically equivalent assets traded in different markets. For example, Froot and Dabora (1999) find that pairs of large differences in trading environments. Similarly, Chan, Hong, and Subrahmanyam (200 8) show that market liquidity is an important determinant of the price difference between ADRs (American Depository Receipts) and their underlying shares. However, no prior research has examin ed the role of credit availability (funding liquidity) within th is context. Using relatively liquid REIT common shares, liquidity and leverage sorted REIT portfolios, and the relatively illiquid mark et for REIT preferred shares, I examine the relative effects of changes in credit availability on the prices of similar a ssets traded in markets with varying liquidity. This framework also provides an appealing setting to test whether the relative likelihood of a spiral effect is dependent on the underlying liquidity with which these assets trade. Anecdotal evidence also su ggests the expansion of credit availability during the real estate boom of the early to mid 2000s was in part driven by the response of creditors to increasing investor optimism and speculative demand for these assets. If banks cater their lending decision s to shifts in investor sentiment (Shleifer and Vishny, 2010) a feedback loop may be created between changes in credit availability and changes in investor sentiment. With the anecdotal evidence and th eoretical motivation in mind, I further extend my anal ysis of credit market effects by including a proxy for investor sentiment as a f ourth endogenous variable in the VAR specifications.

PAGE 70

70 Empirical Methodology Brunnermeier and Pedersen (2009) suggest that liquidity based price impacts can be tested empirically by examining short run price changes, whereas the impact of fundamental volatility is more likely to be evident in long run price movements. To capture the short term dynamics between credit availability, market liquid ity, and asset price changes, I emplo y vector autoregressive (VAR) models. In its simplest form, a VAR model is composed of a system of regressions where two or more dependent variables are expressed as linear functions of their own ontrol variables. In more technical terms, a vector autoregression model is the unconstrained reduced form of a dynamic simultaneous equations model. An unrestricted p th order Gaussian VAR model can be represented as: Y t 1 Y t 1 2 Y t 2 k Y t p t (3 1) where Y t is a vector of variables, is a p x 1 vector of intercepts, 1 2 k are p x p matrices of parameters with all eigenvalues of having moduli less than one so that the VAR is stationary, a nd e t is a vector of uncorrelated structural shocks [ NID(0, )]. I obtain maximum likelihood estimates of and using iterated least squares. The number of quarterly lags is chosen based on examination of the AIC, SBIC, and the likelihood ratio selectio n criteria for various choices of p. I use an unconstrained VAR system to examine the dynamic relation between credit availability, market liquidity, and asset prices in private and public commerc ial real estate markets over the 1992:Q2 to 2009:Q4 sample per iod. I utilize data from the Federal Reserve comm ercial real estate loans, as my primary mea sure of credit availability. As a proxy for market

PAGE 71

71 liquidity in the private c ommercial real estate market I utilize the percentage of properties sold from the National Council of Real Estate Investment Fiduciaries (NCREIF) Property Index each quarter. To capture market liquidity for publicly traded equity REITs, I utilize a price impact measure based on the methodology of Amihud (2002) 2 To capture price changes in private and p ublic commercial real estate, I utilize the percentage price change in the quarterly MIT/NCREIF Transaction Based Index (TBI) and the appreciation component of the FTSE NAREIT U.S. equ ity REIT index, respectively. I include lagged values of several macroeconomic variables and risk proxies that have been shown to matter in the asset pricing literature to control for other potential sources of variation in mark et liquidity credit availability and asset prices over time as described further in the data section. Data and Descriptive Statistics Liquidity Data Sources and Definitions In a relatively illiquid market, such as private commercial real estate, transacti on frequency is a key indicator of market liquidity (Fisher et al., 2004). When there are more potential buyers in the market, property owners can sell more assets, or sell any given asset more quickly, with less of an impact on competitively determined ma rket values. Conversely, when market liquidity is low, asset turnover will be lower and the price impact of a transaction will be relatively high. I measure aggregate market liquidity in the private commercial real estate market using the percentage of pr operties sold (PROPSOLD) from the NCREIF NPI index. Established in 1982, NCREIF is a not for profit institutional real estate industry association that collects, processes, validates, and disseminates information on the risk/return characteristics of comme rcial real 2 Results from Augmented Dickey Fuller Tests suggest the use of first differences of measures of credit availabi lity and mar ket liquidity in the VAR specifications.

PAGE 72

72 the NCREIF Property Index (NPI), tracks property level returns on a large pool of commercial real estate assets acquired in the private market for investment purposes only. The property composition of the NPI changes quarterly as data contributing NCREIF members buy and sell properties. However, all historical property level data remain in the database and index. An increase in PROPSOLD suggests that market liquidity in the private commercial real estate market is increasing. As a robustness check, I also construct an additional measure of market liquidity for the private commercial real estate market, TBI_LIQ. TBI_LIQ is the difference between the va riable liquidity TBI supply index and the constant liquidity TBI demand index. The contemporaneous correlation between PROPSOLD and TBI_LIQ is 0.79. In a relatively liquid market, such as the market for publicly traded shares of REIT common equity, market liquidity measures are directly observable. Using daily price and share (CRSP), I construct a price impact measure of market liquidity for REIT common equity based on the methodology of Amihud (2002). REIT_LIQ is a value weighted index constructed from the daily ratio of absolute stock return to dollar volume, averaged over the quarter in which shares were traded. I use the additive inverse of the Amihud (2002) meas ure so that an increase in REIT_LIQ represents an increase in market liquidity withi n the public REIT market. In my empirical analysis, I utilize the natural log of REIT_LIQ. As a robustness check, I also construct a share turnover measure for publicly tra ded common shares of equity REITs using daily trading volume and shares outstanding from CRSP. I define REIT_TURN as total trading volume in a quarter divided by total shares outstanding as of the end of the quarter. 3 3 Barinov (2010) suggests that share turnover proxies for firm specific uncertainty rather than market liquidity. Therefore, I utilize my price impact measure, REIT_LIQ, as the primary measure of market liquidity fo r the public real estate market within the empirical analysis

PAGE 73

73 Credit availability is defined as the ease with which an investor can obtain capital for acquisitions and the refinancing of existing assets. Because bank lending standards are the criteria by which banks evaluate the risk of providing credit to potential borrowers, changes in these standards reflect the relative ease with which investors may obtain funds. All else equal, a tightening in lending standards would reduce the supply of funds available to investors and therefore decrease the availability of credit To measure credit availability in the commercial real estate market, I obtain a quarterly survey of approximately sixty large domestic banks and twenty four U.S. branches or agencies of foreign banks urvey on Bank Lending Pract ices The purpose of the survey is to collect information on credit availability, with a focus on changes in lending practices within domestic loan markets. The sample of respondents is geographically diverse as participating banks come from all 12 Federal Reserve Districts. The banks must also have more than $3 bi llion in assets and more than 5% of their loan portfolio comprised of commercial and industrial (C&I) loans. While the survey was originally created in 1967, there have been several breaks in the time series and some variation in the wording of its questions. Since the second quarter of 1990, however, the survey has maintained consistency in its core set of questions. To construct a measure of changes in credit availability, I focus on responses to the one of the following options: tightened considerably, tightened somewhat, remained essentially unchanged, eased somewhat, or eased considerably. I define TIGHTEN as t he net percentage of loan officers reporting a tightening of credit conditions. More specifically, it is the sum of the

PAGE 74

74 number of respondents who selecte divided by th e total number of respondents. An increase in TIGHTEN indicates that the availability of funds for commercial real estate investment is declining. Lown, Morgan, and Rohatgi (2000) show that lending by U.S. banks slows substantially following a report of tightened lending standards in the Senior Loan Officer Survey and that reported changes in lendi ng standards are highly correlated with other measures of credit availability. An additional measure of credit availability within commercial real estate markets, RERC_CAPITAL, is constructed from survey data published by the Real Estate Research Corporat ion (RERC) in its quarterly Real Estate Report. RERC surveys institutional real estate investors, appraisers, lenders, and managers throughout the United States to gather information on current investment criteria, such as required rates of return on equit y, expected rental growth rates, and current investment conditions, including the availability of capital. RERC survey respondents are asked to rank the current availability of capital for investment on a scale of 1 to I transform the RERC measure to be consistent with the definition of TIGHTEN by taking the additive inverse, centering the series on zero, and expressing values as a percentage. An increase in RERC_CAPITAL therefore indicates that respondents believe capital availability has decreased over the prior quarter Asset Pricing Data Sources and Definitions To measure price changes in the private commercial real estate market, I utilize the TBI (Transa ctions Based Index of Industrial Commercial Property Investment Performance) price index The TBI is a hedonic price index based on a "representative property" that mirrors the average characteristics of the NCREIF properties. The TBI price index estimates quarterly

PAGE 75

75 market price changes based on the verifiable sales prices of properties sold from the NCREIF database each quarter. 4 I utilize the natural log of the quarterly percentage change in the aggregate price i ndex (TBIRET) as a measure of asset price m ovements in the private commercial real e state market. For robustness, I also utilize the natural log of the quarterly capital return component of the leveraged NCREIF index (NCREIF_LEV) as an additional measure of asset price movements in the private comm ercial real estate market. The correlation between the capital return component of the leveraged NCREIF and the change in the TBI price index is 0.62. Pricing data for common shares of publicly traded commercial real estate is obtained from the National A ssociation of Real Estate Investment Trusts (NAREIT). Members of NAREIT include REITs that own, operate and finance income producing real estate. NAREIT publishes the FTSE NAREIT Equity Index, a market capitalization weighted index measuring returns on REI Ts that meet minimum size and liquidity crit eria and are listed on the NYSE / Amex or Nasdaq. I utilize the natural log of the quarterly appreciation component of the return on the FTSE NAREIT Equity Index (REIT RET) as a measure of asset price movements in t he public commercial real estate market. Control Variables I include the following set of control variables to capture other potential sources of variation in prices, credit availability, and market liquidi ty in the VAR regression specifications: the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSPREAD), the spread between yields on BAA rated and AAA rated corporate bonds (DEFSPREAD), the rate of inflation (INFLA), and the excess retur n on the 4 Details of the index methodology are described in Fisher, Geltner, and Pollakowski (200 7).

PAGE 76

76 public stock market (MKT) (e.g., Chen, Roll, and Ross, 1986; Ferson and Harvey, 1991; Fama and French, 1993; Fama and Schwert, 1977 ; Sharpe, 2002). In addition, I include the remaining Fama French risk factors, SMB and HML, augmented by the retur n momentum factor, UMD (e.g., Fama and French 1996; Liew and Vassalou, 2000; Lettau and Ludvigson, 2001; Jegadeesh and Titman, 1993; and Carhart, 1997). These factors also control for the public equity cha racteristics of the public market real estate retur ns. Prior research has shown that dividend (current) yields are also a significant predictor of subsequent asset price changes (Ghysels, Plazzi and Valkanov, 2007; Fama and French, 1988). Therefore, I include the dividend yield on equity REITs (DIVYLD) as an ad ditional control variable in the REIT pricing equations. In the p rivate market specifications, I use the aggregate annua l net rent and its price) to proxy for current yield. Dividend yields and capitalization rates are obtained from NAREIT and the American Council of Life Insurers (ACLI), respectively. Descriptive Statistics Table 3 1 reports descriptive statistics for each measure of credit availability, marke t liquidity, price returns, and control variables over the 1992 : Q2 to 2009 : Q4 sample p eriod. The starting point of the sample period is dictated by the availability of survey data obtained from the Real Estate Research Corporation (RERC). When responding t o the Senior Loan Officer Survey, loan officers are instructed to report changes in lending standards over the previous quarter regardless of how current credit conditions compare to long term norms. Nevertheless, TIGHTEN, displays a high degree of autocor relation (0.91), indicating persistence in the direction of changes in lending standards over time. RERC_CAPITAL displays a similar le vel of autocorrelation (0.95). T hough highly autocorrelated, both measures capture significant peaks and troughs in credit availability cycle s For example, TIGHTEN ranges from a minimum of

PAGE 77

77 0.24, indicating that the net percentage of respondents reported an easing in lending standards, to a high of 0.87, denoting a period in which most banks were tightening lending standards RERC_CAPITAL displays similar var iation ranging from a low of 0.911 to a high of 0.822 The contemporaneous correlation between RERC_CAPITAL a nd TIGHTEN is 0.65 (Table 3 2 ) In reference to the latest credit cycle, both proxies capture a substantial loo sening in credit standards over the 2001 to 2005 real estate boom and a sharp increase in credit standards during the market downturn in 2007 and 2008. The second panel of Table 3 1 report s descriptive statistics for measures of market liquidity in private and public commercial real estate marke ts. On average, only 2.1 % of the properties held in the NCREIF databas e are sold each quarter over the sample period, indicating that private commercial real estate markets are extremely illiquid. PROPSOLD also displ ays a high degree of autocorrelation (0.64), indicating persistence in periods of relative illiquidity. The average daily ratio of absolute REIT stock returns to dollar volume is 1.149. REIT_LIQ displays significant volatility over the sample period. That is, even relatively liquid public markets experience significant declines in market liquidity when investors face difficulties obtaining funds for investment. Panels A and B of Figure 3 1 display TIGHTEN and aggregate measures of market liquidity in the p rivate commercial real estate market (PROPSOLD) and public commerci al real estate market (REIT_LIQ ), respectively. Following a period of large declines in commercial real estate values and a credit crunch in the late 1980s and early 1990 s, lending standard s for commercial real estate loans were eased slightly in late 1993 and remained relatively loose for several years. During this period the public REIT market was growing significantly with increased involvement from institutions. Institutional investors, who had difficulty disposing of

PAGE 78

78 properties within the private market durin g the downturn of the late 1980 s, were shifting funds into the liquid mar ket for publicly traded REITs. As funds continued to flow into the public commercial real estate sector, mar ket liquidity in the underlying property market also increased significantly over this period. Lending standards for commercial real estate loans began to tighten again in the late 1990 s, a period in which many investors rotated out of value oriented asset s, including commercial real estate, and into high growth technology stocks. Not surprisingly, a significant shock to market liquidity in private commercial real estate occurred. Following an eas ing of lending standards in 2002 commercial real estate pric es and market liquidity significantly increased However, as the credit cris is began to unfold in late 2007 to 2008, lending standards on commercial real estate loans were tightened considerably. Market liquidity plunged in subsequent quarters within both private and public real estate markets The third panel of Table 3 1 report s descriptive statistics for each price return series. Consistent with prior literature, the average quarterly return on publicly traded equity REITs (1.5%) is greater than returns on similar institutional quality assets owned and managed in private markets (TBIRET = 0.8%). However, price changes in the public REIT market are substantially more volatile than those in the underlying private property market 5 The final panel of Table 3 1 report s descriptive statistics for each control variable The annual yield on three month Treas ury bills (TBILL) averaged 3.60 % over the sample per iod, ranging from a low of 0.10 % to a high of 6.2 % The slope of the Treas ury term structure averaged 1. 7 % on an annual basis, although TERMSPREAD varied significantly over the sample period. The me an default risk premium is 0.90 % per year, but DEFSPREAD ranged from a low of just 0.60 % to a high of 3.0 % Average quarterly i nflation (INFLA) is 0.60 % although 5 Mean and standard deviation of private market re turns using NCREIF_LEV are 0. 1 % and 5 0 % respect ively.

PAGE 79

79 inflation also displayed cons iderable time variation over the sample period. The mean stock ma rket risk premium (MKT) is 1.50 % per quarter, but ranges from a low of 22.3 % to a high of 20.6 % SMB, HML, and UMD averaged 0.90 % 0.80 % and 1.7 % per quarter, respectively, and also displayed substantial vola tility over the sample period. The average cap rate (CAPRT) fo r commercial properties over the sample period is 8.60 % ra nging from a low of 6.50 % to a high of 10.3 % The mean dividend yield (DIVYLD) for equ ity REITs is 6.3 0 % and range s from 3.60% to 9.40 %. Table 3 2 reports contemporaneous correlations among measures of credit availability and price appreciation 0.47) 0.28). That is, a tightening of lending standards for commercial real estate loans is unconditionally correlated with a decline in asset values within both privat e and public co mmercial real estate markets. I find similar results using RERC_CAPITAL. Panel A of Figure 3 2 displays TIGHTEN in relation to TBIRET. Beginning in 2002, when lending standards were eased considerably, prices in private commercial real esta te experienced a significant boom. However, as credit availability decreased in 2005, prices began to spiral as returns declined consistently over the next two years. Thus, unconditionally, changes in lending standards and price changes appear to reinforce each other in the private market. Panel B of Figure 3 2 plots TIGHTEN against REIT returns (REITRET). Although REIT returns appear to respond to changes in lending standards, they quickly revert in subsequent periods. For example, while REIT prices exper ienced a significant decline in the fourth quarter of 2008, prices rebounded quickly in the first quarter of 2009.

PAGE 80

80 Empirical Results Dynamic Relations in Illiquid Private Markets Table 3 3 provides estimates of an unconstrained VAR model for the private co mmercial real estate market in which the percentage change in the TBI price index (TBIRET), change s in credit availability ( TIGHTEN), and changes in aggregate market liquidity ( PROPSOLD) are specified as endogenous vari ables. Although the coefficient est imates are not reported, I also include lagged values of the following exogenous control variables: TBILL, TERMSPREAD, DEFSPREAD, INFL, MKT, SMB, HML, UMD, and CAPRT. The sample period is 1992:Q2 to 2009:Q4. Utilizing t his specification I test whether cha nges in credit availability affect subsequent market liquidity and whether a feedback effect occurs in which changes in market liquidity affect the subsequent availability of credit, controlling for lagged fundamentals and other factors. As previously disc ussed, when access to credit markets is tightened, investors find it more difficult to finance acquisitions or to refinance previously acquired assets. This is especially true of highly levered assets. The lack of credit availability subsequently lowers ma rket liquidity. Under certain conditions, reduced market liquidity increases the risk of financing a transaction, thus leading to further tightening in the availability of credit to these investors (Brunnermeier and Pedersen, 2009). When credit conditions continue to tighten in may e nsue ), I find that an increase in credit tightening predicts a decrease in market liquidity with eithe r a two quarter (p value = 0.079) or three quarter lag (p value = 0.073). That is, as the percentage of banks tightening commercial real estate lending standards increases, the percentage of properties sold in the private commercial real estate market decl ines in subsequent quarters. Although not tabulated, I find a

PAGE 81

81 similar relation between changes in the availability of capi tal and changes in subsequent market an alternative measure of credit availability If shifts in market liquidity further reduce the availability of credit, a li quidity spiral may arise. In the the estimated coefficient on the two quarter lag Thus, consistent with the theory of Brunnermeier and Pedersen (2009), preliminary results suggest that a tightening in lending standards ca n lead to a liquidity (margin) spiral in the private commercial real estate market as decreasing market liquidity further reduces the supply of credit. I find similar results as my measure of market liquidity in the private real estate marke t. The VAR specification also allows for a test of whether changes in credit availability affect subsequent price movements in private real estate markets and whether a feedback effect occurs in which price changes affect the subsequent availability of cre dit. When changes in lending standards and mark there may be a magnification in the price impact of subsequent transactions in the form of a loss spiral Both types of liquidity spirals reinforce each other, amplifying the total effect beyond what their individual impacts would be (Brunnermeier and Pedersen, 2009; Brunnermeier, 2009; Geanakoplos, 2003). at the two and three quarter lags are ne gative and highly significant. This strongly suggests that a tightening of lending standards leads to price declines over subsequent periods. This result is consistent with the theoretical predications of Brunnermeier and Pedersen (2009) and Geanakoplos (2 003). When assets can be readily acquired using significant leverage, the most optimistic investors are able to own and control a larger percentage of the current stock of assets than they would be able to otherwise. However, as the capital available to th ese optimistic buyers

PAGE 82

82 declines, they are less able to acquire or refinance assets, thereby leaving more pessimistic investors as the marginal buyers. As a result, asset valuations will likely decline. This important The estimated coefficient on lagged TBIRET in the TBIRET equation is negative ( 0.235) and significant (p value = 0.076), indicating that price increases in the prior quarter predict a subsequent reversal in the following quarte r. However, if changes in asset values impact the availability of capital to make future transactions, prices may spiral in one direction. Returning to the find a negative ( 1.079) and highly significant (p value = 0.000) That is, declining asset prices in a relatively illiquid market increase the risk of financing another trade, which leads banks to further tighten the supply of available credit for future transac tions. Consistent with the theory of Brunn ermeier and Pedersen (2009), these results imply that a decrease in credit availability can lead to a liquidity (loss) spiral in the private commercial real estate market as a decrease in asset prices affects the s upply of credit in the subsequent quarter. Although not tabulated, I find similar resu lts when using NCREIF_LEV as an alternative measure of price changes in the private commercial real estate market. In the NCREIF_LEV equation, the estimated coefficients nt. In again document a feedback effect in which the estimated coefficient To examine t he cumulativ e effect of all four lags of the endogenous variables, I sum the estimated lagged coefficients and test whether the four lags are jointly significant. The summed coefficients and p values associated with tests of joint significance are reported in Table 3 4 Controlling for aggregate market liquidity, a tightening of lending standards over the prior year

PAGE 83

83 predicts price declines in the private market. Moreover, declining prices are associated with a subsequent tightening of lending standards. Red uced market liquidity over the prior four quarters is also associated with a tightening of credit standards. Dynamic Relations in Liquid Public Markets In contrast to their underlying properties, the equity REITs I examine trade frequently on a number of m ajor stock exchanges. Although changes in the availability of credit may still impact market liquidity in these markets, it is less likely th at subsequent changes in credit availability shocks to market liquidity create profit opportunities for speculators who anticipate that prices will return to fundamentals once liquidity has been restored. As long as arbitrageurs are far from hitting their funding constraint, then asset prices will be less susceptible to a liquidity (loss) spiral, even if a tightening in credit markets cau ses market liquidity to decline Using a VAR analysis I first test whether changes in the availability of credit affect subsequent market liquidity and whether a fee dback effect occurs in which changes in market liquidity affect subsequent credit availability in the relatively liquid market for publicly traded equity REITs I then examine whether a loss spiral ensues, thereby magnifying the overall price impact of a c hange in credit availability Table 3 5 provides estimates of an unconstrained VAR model for the REIT market in which t he appreciation component of the REIT return index (REITRET), changes in credit availability ( TIGHTEN), and changes in aggregate REIT market liquidity ( ) are specified as endogenous variables. As in the specifica tions for the private market, I include the following set of exogenous control variables: TBILL, TERMSPREAD, DEFSPREAD, INFL, MK T, SMB, HML, and UMD. I also include the dividend yield (DIVYLD) on equity REITs, rather than the aggregate cap rate for commercial properties.

PAGE 84

84 and two quarter lags are negative and significant. As investors find it difficult to obtain credit, the price impact of a subsequent transaction increases. Thus, credit tightening is associated with decreased liquidity in the REIT market. In the equa tion (Column 2), I find that the two Thus, I find further evidence of a liquidity (margin) spiral within the public real estate market. Turning to the REITRET equation (Column 1), the estimated c oefficients on both the three and four with the private market results, a tightening in lending standards leads to subsequent declines in REIT prices. I again find similar resul credit availability. Overall, these results strongly suggest that changes in credit availability are an important determinant of asset price movements for highly leveraged assets that trade in relatively liquid markets as well. sensitivity of subsequent changes in credit availability to changes in asset prices is larger for assets that are highly leveraged and relatively il liquid. Because sustained periods of illiquidity are far less common in public markets, one would expect to find less of a feedback effect between asset prices and the availability of credit. Returning to the (Column 2), I do not find a significant relation between prior period REIT returns and changes in credit availability. Therefore, I do not find evidence of a loss spiral using aggregate data for the public real estate market. To exam ine the cumulative effect of all four lags of the endogenous variables, I again sum the estimated lagged coefficients and test whether the four lags are jointly significant. The

PAGE 85

85 summed coefficients and p values associated with tests of joint significance are r eported in Table 3 6 Controlling for aggregate market liquidity, a tightening of lending standards over the prior year predicts significant price declines in the REIT market (p value = 0.001). However, in contrast to the private market results, declining REIT prices do not predict a subsequent tightening of lending standards (Column 2). Furthermore, increases in REIT prices over the prior year predict increased liquidity in the REIT market (Column 3). In contrast, increases in credit tightening significant ly reduce REIT mark et liquidity (p value = 0.001) Impulse Response Functions: Credit Availability and Asset Price Movements Further evidence regarding the impact of changes in credit availability on asset price movements in both private and public commerc ial real estate markets is provided by the VAR generalized impulse response functions displayed in Figure 3 3 Panels A and B depict the response of quarterly price changes in TBIRET and REITRET, respectively, to a one standard deviation change in my prima ry measure of credit availability (TIGHTEN). The solid line in each figure represents the estimated diffusion of quarterly price changes to the shock in credit availability. Panels A and B of Figure 3 3 reveal an initial decrease in asset prices in both pr ivate and public markets in response to a shock to credit availability. However, there is a significant delay in the time it takes for asset prices to revert following a shock to credit availability within the relatively illiquid private market. Panels C and D display the response of TIGHTEN to a one standard deviation change in private and public market price changes, respectively. In contrast to the REIT results (Panel D), subsequent credit availability is significantly impacted by a shock to asset price s within the private market (Panel C). As asset prices fall, credit conditions tighten, reinforcing the initial impact of a change in credit availability on asset prices within an illiquid market.

PAGE 86

86 Leverage and Liquidity Portfolio Sorts: Public Market In pr ivate markets, I document a spiral effect between changes in credit availability and property prices. In the REIT market, however, movements in share prices do not appear to reinforce changes in the availability of credit. However, it is possible for such a feedback relationship to exist within a particular cross section of the REIT market. Brunnermeier and availability and asset price movements is stronger for capital i ntensive assets that are relatively illiquid. Therefore, one might expect to find evidence of a spiral effect among REITs with high leverage and low share turnover. To test this hypothesis, I obtain leverage and share tu rnover data for each REIT in the ag gregate index using data from SNL and CRSP, respectively. REITs are first sorted into quartiles based on both leverage and liquidity I then construct value weighted price return portfolios for each subset of the dat a. For example, to construct the High Le verage Low Liquidity portfolio, I create a value weighted portfolio of REITs that fall into both the highest leverage quartile and lowest share turnover qua rtile within each quarter of the sample period. On average, the leverage ratios for the high and low leverage portfol ios are approxi mately 60% and 20 %, respectively. For the liquidity sorts, average quart erly share turnover is 8 % for the low l iquidity quartile and 46 % for the high liquidity quartile. I follow the empirical methodology detailed previously utilizing price returns and price impact measures of market liquidity for ea ch portfolio, in addition to the aggregate measure of credit availability, in separate VAR specifications. I include the same set of exogenous control variables as specified prev iously in the earlier REIT market analysis. Table 3 7 presents results from the portfolio analys is. For ease of presentation, I report the sum of the coefficient estimates across all four lags and the p value pertaining to the joint

PAGE 87

87 significance of those f our l ags. Focusing first on the REITRET equations (Columns 1 and 4), I document a negative and statistically significant relation between changes in credit availability p ortfolio specification. Moreover, the magnitudes of the estima increasing in leverage. In other words, share prices of more highly leveraged REITs are more sensitive to changes in cr edit availability However, I document a feedback effect between changes in REIT price movements and the subsequent availability of credit only for relatively illiquid REITs (Columns 2 and 5). Consistent with my initial hypothesis, the estimated coefficient on REITRET is negative ( 1.127) and highly significant (p value = 0.005) for the High Leverage Low Liquidity portfolio. In addition, I find evidence of a feedback effect between changes in REIT prices and credit avail ability for the Low Leverage Low Liquidity portfolio. However, I do not find evidence of a feedback effect in the High Liquidity portfolios, regardless of leverage ratios. Therefore, the spiral effect is concentrated in the low liquidity REIT portfolios. C (2009) liquidity based theory, relative illiquidity is a necessary condition for a spiral effect to occur. This result is also consistent with my previous results from the private market in which similar assets tr ade in a relatively illiquid environment. Using Returns on REIT Preferred Stock as a Robustness Check Within public stock markets there is also the opportunity to examine two types of equity securities issued by the same firm, each with claims on similar cash flow streams, yet which trade at different prices and with different liquidity. By examining the relation between credit availability and the pri ces of REIT preferred shares, I provide an alternative test of the sensitivity of asset prices to changes in the availability of capital within a relatively illiquid public market. Having documented the existence of loss spirals in the private market as well as the low liquidity

PAGE 88

88 segme nt of the public REIT market, I further hypothesize that asset prices of REIT preferred shares may be susceptible to a spiral effect due to the low liquidity environment in which they trade. I use data from Thomson Reuters DataStream to construct a price index for REIT preferred s ecurities. More specifically, I obtain historical s hare prices, shares outstanding, trading volume, and market capitalization data for all RE IT preferred securities over the sample period. I then construct a quarterly value weighted index based on the percentage price change of each REIT preferred security (REITPREF). 6 Preferred shares are often considered a hybrid security because they exhibit both bond and equity like characteristics. Nevertheless, the price movements of REIT preferred shares are significantly positively correlated with the returns on e Similarly, the returns on REIT preferred shares are greater, on average, and more volatile than the returns on private commercial real estate. Table 3 8 provides estimates of an unconstrained VAR model for the REIT pref erred market in which the percentage price change (REITPREF), changes in credit availability ( ) and changes in aggregat ) are speci fied as endogenous variables. I include th e same set of controls as in prior REIT speci fications. Focusing on the REITPREF equation (Column 1), the estimated coefficient on the three quarter 0.193 ) and hig hly significant (p value = 0.001 ). That is, when access to credit markets is tightened, prices of REIT preferred securities decrease over subsequent quar ters. Although not tabulated, I find a similar relation between 6 The correlation between my REIT preferred index and the MSCI REIT Preferred Index is 0.98 for the overlapping sample period (2005 to 2009).

PAGE 89

89 I also find evidence of a feedback effect between changes in REIT preferred share prices and the subsequent availability of credit. The estimated coefficient on the one quarter lag of REITPREF is negative ( 0.401 ) an d significant (p value = 0.026 ), suggesting that changes in credit availability are sensitive to asset price movements in these relativel y illiquid public is even stronger (p value 0.015 ). Consistent with Brun nermeier and Pedersen (2009), I find that when fu nding constraints increase in a relatively illiquid market, subsequent changes in asset prices affect the future availability of capital, thus magnifying the overall price impact of a funding shock. To exam ine the cumulative effect of all four lags of the endogenous variables, I again sum the estimated lag ged coefficients and test whether the four lags are jointly significant. The summed coefficients and p values associated with tests of joint significance are reported in Table 3 9 Controlling for aggregate market liquidity, a tightening in credit over the prior year predicts significant price declines in the REIT preferred market (p value = 0.000). Similar to the private market and low liquidity portfolio results, reductions in the prices of preferred shares predict a subsequent decrease in the availabil it y of credit (p value = 0.017) In other words, changes in credit availability are destabilizing in a relatively illiquid market, as changes in asset prices affect the future availability of credit, thus magnifying the overall price impact of a reduction in credit availability These results provide further evidence of a spiral effect within relatively illiquid segments of the public market Credit Availability, Asset Prices, and Investor Sentiment Anecdotal evidence suggests the expansion of credit availabi lity during the real estate boom of the early to mid 2000s was in part driven by the response of creditors to increasing investor optimism and speculative demand for these assets. A recent article in the Economist

PAGE 90

90 (2010) characterizes the emergence of asse valuations, the two classic symptoms of a bubble are rapid growth in private sector credit and an between c hanges in the availability of credit and changes in investor sentiment will provide further insight into the underlying factors that fueled the rapid price appreciation and subsequent collapse of the commercial real estate market. More formally, Shleifer a nd Vishny (2010) develop a theoretical model in which banks cater their financing decisions to shifts in investor sentiment. During periods of high sentiment, banks increase their mortgage originations (traditional lending), particularly when securitizatio n of these underlying loans is profitable. Banks continue to pursue such profits during a boom period in order to take advantage of attractive money making opportunities in the secondary market. When the asset pricing bubble bursts, banks forgo lending opp ortunities and credit markets tighten. If inve stors are unable to access credit precisely at the time it may be most lending standards will have a destabilizing effect on prices, similar to the theoretical predictions of Brunnermeier and Pedersen (2009) and Geanakoplos (2003). If banks cater their lending decisions to shifts in investor sentiment, a feedback loop may be created between changes in credit availabili ty and changes in investor sentiment. In my first study, I construct both a direct and indirect investor sentiment measure for the commercial rea l estate market. For the direct measure of investor sentiment (DRES), I employ survey data published by the Rea l Estate Research Corporation (RERC) in its quarterly Real Estate Report. RERC surveys institutional real estate investors, appraisers, lenders, and managers throughout the United States to gather information on current investment criteria, such as

PAGE 91

91 require multiple property types, both nationally and by metropolitan area, on a scale of 1 to 10, with 1 DRES is constructed from the first principal component extracted from quarterly RERC investment condition survey responses pertaining to the ei gh t RERC property types. For the indirect m easure of sentiment (INDRES), I follow the framework of Baker and Wurgler (2006, 2007) and use principal component analysis to construct an indirect quarterly sentiment index based on the common variation in seven u nderlying proxies of investor sentiment in commercial real estate markets: (i) the average REIT stock price premium to net asset value (NAV), (ii) the percentage of properties sold each quarter from the National Council of Real Estate Investment Fiduciarie s (NCREIF) Property Index (NPI), (iii) the number of REIT IPOs, (iv) the average first day returns on REIT IPOs, (v) the share of net REIT equity issues relative to total net REIT equity and debt issues, (vi) net commercial mortgage flows as a percentage o f GDP, and (vii) net capital flows to ded icated REIT mutual funds. In the subsequent empirical analysis, I utilize the first difference of these series. To examine the dynamic relation between changes in investor sentiment and changes i n the availability o f credit, I add changes in investor sentiment as a f ourth endogenous variable to my primary VAR specifications. These results are reported in Table s 3 10 and 3 11 For ease of presentation, I again report the sum of the coefficient estimates across all fou r quarterly lags and the p value pertaining to the joint sign ificance of those four lags. I report results utilizing aggregate private market return series (TBIRET); however, the results are robust to the u se of

PAGE 92

92 aggregate REIT returns. I include the same s et of exogenous control variable s as specified previously in my earlier private market analysis. Table 3 10 reports results using the Focusing first on the TBI RET specification (Column 1), I find that even afte r controlling for the influence of investor sentiment on asset prices, the estimated coefficient on credit availability t 1 to t 4 ) remains negative ( 0.217) and statistically significant (p value = 0.037). In addition, lagged changes in sentiment t 1 to t 4 ) predict subsequent price changes in the private market, even after controlling for the impact of changes in credit availability and market liquidity. As investors become increasingly optimistic, they bid up prices of these assets in the short run. I s reported in Table 3 11 Turning to the ), I continue to observe a feedback effect between changes in property prices and subsequent credit availability after controlling fo r the impact of investor sentiment. Moreover, increasing sentiment predicts l ooser underwriting standards. I Vishny (2010), changes in investor sentiment are a significant d eterminant of future credit availability. In particular, these results suggest banks respond to increasing investor sentiment by easing their credit standards, thereby making credit more readily available to potential investors when they are most optimisti c. However, during a market downturn in which investors are becoming increasingly pessimistic, banks tend to tighten their lending standards, which can have a destabilizing effect on asset prices. This raises an interesting policy implication pertaining to whether lenders have the ability to reduce the probability that an asset pricing bubble emerges by restricting the amount of credit they provide during boom periods or ease the severity of a downturn by making credit available when distressed assets are u ndervalued.

PAGE 93

93 effect between changes in credit availability and future changes in investor sentiment. However, if I Table 3 11) the coefficient on lagged changes in tightening on sentiment is no longer statistically significant. Summary and Conclusion Changes in the availability of credit are a significant determinant of asset price movements in both private and public commercial real estate markets. As credit market s tighten, investors are less able to make new acquisitions in highly leveraged assets and may be forced to de lever their existing positions by selling assets into an illiquid market, causing prices to decline. When highly leveraged assets trade in relati vely illiquid markets, declining asset values may trigger a liquidity (loss) spiral, thereby magnifying the overall price impact of a shock to credit availability. Commercial real estate markets provide an appealing testing ground for examining the dynami c relation between credit availability, market liquidity, and changes in asset prices. The private commercial real estate market is a relatively illiquid market consisting of highly leveraged assets. Unlike assets traded in more liquid public markets, whic h tend to experience only brief periods of illiquidity, commercial properties trade in relatively illiquid markets. Therefore, asset prices in this market may be relatively sensitive to changes in credit availability. An appealing feature of commercial re al estate markets is that assets trade in both a relatively illiquid direct private market and in liquid public stock markets, in the form of securitized portfolios of properties (i.e., REITs). Since the underlying properties held by the publicly traded re al est ate firms I analyze are similar to the property holdings of the institutional real estate investors whose private market returns I also track, disparities in the impact of funding liquidity on price changes in private and public real estate markets c an be ascribed to differences

PAGE 94

94 in the characteristics of these two markets. Therefore, I examine the dynamic relation between changes in credit availability and changes in asset prices within segments of the REIT market that are also relatively illiquid. Us ing vector autoregressive (VAR) models, I find that asset values of highly leveraged assets, such as commercial real estate, are sensitive to the availability of credit, even when controlling for aggregate market liquidity. As banks tighten (ease) their le nding standards on commercial real estate loans, asset prices decline (rise) in both private and public commercial real estate markets. These results suggest that leverage is a key factor in determining cred it market effects on pricing. I also provide evid ence that assets trading in illiquid segments of the commercial real estate market are highly susceptible to a spiral effect, in which changes in asset prices lead to further changes in the availability of credit, thus magnifying the overall price impact o f a funding shock as a spiral ensues. These results suggest that while leverage plays a significant role in determining credit market availability pricing effects, the underlying liquidity with which these assets trade is a key factor in determining the li keliho od of a liquidity (loss) spiral with lower liquidity creating the market setting for a spiral effect.

PAGE 95

95 A B Figure 3 1. Credit availability and market liquidity. A) This figure plots credit availability and aggregate market liquidity measures in the private commercial real estate market over the sample period 1992:Q2 to 2009:Q4. The measure of credit availability, TIGHTEN, is the net percentage of loan officers reporting a tightening of lending standards on commercial real estate loans in the Fed Officer Opinion Survey on Bank Lending Practices. The measure of aggregate market liquidity in the private commercial real estate market, PROPSOLD, is the percentage of properties sold from the NCREIF NPI index. B) This fig ure plots credit availability and aggregate market liquidity measures in the public commercial real estate market over the sample period 1992:Q2 to 2009:Q4. The measure of credit availability, TIGHTEN, is the net percentage of loan officers reporting a tig htening of lending Officer Opinion Survey on Bank Lending Practices. The measure of aggregate market liquidity in the public commercial real estate market, REIT_ LIQ, is a price impact measure based on the methodology of Amihud (2002) for equity REITs

PAGE 96

96 A B Figure 3 2. Credit availability and asset prices. A) This figure plots credit availability and asset price movements in the private commercial real estate market ove r the sample period 1992:Q2 to 2009:Q4. The measure of credit availability, TIGHTEN, is the net percentage of loan officers reporting a tightening of lending standards on commercial on Survey on Bank Lending Practices. The measure of price changes in the private commercial real estate market, TBIRET, is the natural log of the quarterly percentage change in the aggregate TBI price index produced by the MIT Center for Real Estate in ass ociation with the National Council of Real Estate Investment Fiduciaries (NCREIF). B) This figure plots credit availability and asset price movements in the public commercial real estate market over the sample period 1992:Q2 to 2009:Q4. The measure of cred it availability, TIGHTEN, is the net percentage of loan officers reporting a tightening of lending standards on commercial real estate loans in the Practices. The measure of price c hanges in the public commercial real estate market, REITRET, is the natural log of the quarterly appreciation component of the return on the FTSE NAREIT Equity Index.

PAGE 97

97 A B C D Figure 3 3. Impulse response functions : credit availability and price returns This figure plots the generalized impulse response functions corresponding to estimated VAR models in Table 3 3 and Table3 5 with credit availability measured in changes. The sample period spans 1992:Q2 to 2009:Q4. A) This figure depict s the percentage response in the aggregate price index for private commercial real estate, TBIRET, to a one standard deviation change in credit availability, TIGHTEN B) This figure depicts the percentage response in the aggregate price index for pubic com mercial real estate, REITRET to a one standard deviation change in credit availability, TIGHTEN C) This figure depicts the percentage response in the measure of credit availability, TIGHTEN, to a one standard deviation change in the aggregate price index for private commercial real estate, TBIRET D) This figure depicts the percentage response in the measure of credit availability, TIGHTEN, to a one standard deviation change in the aggregate price index for public commercial real estate, REITRET

PAGE 98

98 Tab le 3 1 Descriptive statistics : credit availability, market liquidity, price returns, and controls Variable Mean Median St Dev Min Max Serial Correlation Credit Availability TIGHTEN 0.161 0.089 0.263 0.237 0.870 0.91 *** RERC_CAPIT AL 0.338 0.467 0.440 0.911 0.822 0.95 *** Market Liquidity PROPSOLD 0.021 0.020 0.013 0.001 0.057 0.64 *** REIT_LIQ 1.149 1.259 1.005 1.233 2.789 0.9 6 *** Price Returns TBIRET 0.008 0.010 0.046 0.179 0.17 8 0.18 REITRET 0.015 0.020 0.106 0.400 0.320 0.16 Controls TBILL 0.036 0.039 0.018 0.001 0.062 0.97 *** TERMSPREAD 0.017 0.016 0.012 0.006 0.036 0.93 *** DEFSPREAD 0.009 0.008 0.005 0.006 0.030 0.85 *** INFLA 0.006 0.006 0.009 0.039 0.025 0.14 MKT 0.015 0.021 0.086 0.223 0.206 0.08 SMB 0.009 0.004 0.056 0.108 0.191 0.03 HML 0.008 0.004 0.082 0.320 0.239 0.16 UMD 0.017 0.019 0.097 0.398 0.260 0.14 CAPRT 0. 086 0.091 0.010 0.065 0.103 0.95 *** DIVYLD 0.063 0.065 0.014 0.036 0.094 0.87 *** This table reports descriptive statistics for measures of credit availability, market liquidity, price returns, and control variables. Mean, median, standar d deviation, minimum, maximum, and serial correlation are reported. The primary measure of credit availability, TIGHTEN, is the net percentage of loan officers reporting a tightening of lending standards on commercial real estate loans in the Federal Reser the availability of capital measure published by the Real Estate Research Corporation (RERC) in its quarterly Real Estate Report. Market liquidity in the private commercial real estate market, PROPSOLD, is the percentage of properties sold from the NCREIF NPI index. Market liquidity in the public commercial real estate market, REIT_ LIQ is a price impact measure based on the methodology of Amihud (2002) for the public equity REIT market. The mea sure of price changes in the private commercial real estate market, TBIRET, is the natural log of the quarterly percentage change in the aggregate TBI price index. The measure of price changes in the public commercial real estate market, REITRET, is the na tural log of the quarterly appreciation component of the return on the FTSE NAREIT Equity Index. Control variables include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSPREAD), t he spread between yields on BAA rated and AAA rated corporate bonds (DEFSPREAD), inflation (INFLA), Fama French factors, MKT, SMB, HML, augmented by momentum (UMD), the aggregate capitalization rate for commercial properties (CAPRT), and the dividend yield on equity REITs (DIVYLD). The sample period spans 1992:Q2 to 2009:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 99

99 Table 3 2. Contemporaneous correlations between credit availability and asset price movements. TIGHTEN RER C_CAPITAL TBIRET REITRET TIGHTEN 1.000 0.646 *** 0.467 *** 0.284 ** RERC_CAPITAL 1.000 0.579 *** 0.228 TBIRET 1.000 0.261 ** REITRET 1.000 This table reports contemporaneous correlations between measures of credit availability and price return indices for both private and public commercial real estate markets. The primary measure of credit availability, TIGHTEN, is the net pe rcentage of loan officers reporting a Senior Loan Officer Opinion Survey. RERC_CAPITAL is the availability of capital measure published by the Real Estate Resear ch Corporation (RERC) in its quarterly Real Estate Report. The measure of price changes in the private commercial real estate market, TBIRET, is the natural log of the quarterly percentage change in the aggregate TBI price index. The measure of price chang es in the public commercial real estate market, REITRET, is the natural log of the quarterly appreciation component of the return on the FTSE NAREIT Equity Index. The sample period spans 1992:Q2 to 2009:Q4. ***, **, and represent 1%, 5%, and 10% signific ance levels respectively.

PAGE 100

100 Table 3 3. Credit availability and private commercial real estate : individual lags End. Variables TBIRET Constant 0.169 ** 0.852 *** 0.009 ( 0.015) (0.000) (0.591) TBIRET t 1 0.235 1.079 *** 0.016 (0.076) (0.000) (0.615) TBIRET t 2 0.165 0.185 0.030 ( 0.180) (0.424) (0.322) TBIRET t 3 0.228 0.118 0.035 (0.150) (0.692) (0.356) TBIRET t 4 0.103 0.622 ** 0.013 (0. 466) (0.019) (0.714) t 1 0.013 0.195 ** 0.015 (0.779) (0.021) (0.179) t 2 0.099 ** 0.175 0.020 (0.038) (0.051) (0.079) t 3 0.099 ** 0.248 *** 0.021 (0.043) (0.007) (0.073) t 4 0.056 0.292 *** 0.015 (0.295) (0.004) (0.241) t 1 0.043 1.387 0.850 *** (0.935) (0.165) (0.000) t 2 0.760 2.408 ** 0 .638 *** (0.233) (0.045) (0.000) t 3 1.073 0.195 0.408 ** (0.107) (0.876) (0.012) t 4 0.092 0.863 0.093 (0.867) (0.403) (0.485) Adjusted R 2 0.22 0.55 0.30 This table presents results obtained from estimating unrestricted vector autoregressive (VAR) models with credit av ailability, market liquidity and price return measures specified as endogenous variables. The lag length of the VAR is chosen by looking at the AIC, SBIC, and the likelihood ratio for various choices of p with four lags providing the best fit. Augmented Di ckey Fuller tests suggest the use of changes in credit availability and market liquidity within the VAR system. Coefficients and p values are reported for each lag length of the endogenous variables. TIGHTEN, is the net percentage of loan officers reportin g a tightening of lending standards on on Bank Lending Practices. PROPSOLD, is the percentage of properties sold from the NCREIF NPI index. TBIRET, is the natura l log of the quarterly percentage change in the aggregate TBI price index. Control variables include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSPREAD), the spread between yiel ds on BAA rated and AAA rated corporate bonds (DEFSPREAD), inflation (INFLA), the Fama French factors: MKT, SMB, HML, augmented by momentum (UMD), and the aggregate capitalization rate for commercial properties (CAPRT). The sample period spans 1992:Q2 to 2 009:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 101

101 Table 3 4. Credit availability and private commercial real estate: cumulative eff ect and joint significance End. Variables TBIRET TBIRET t 1 to t 4 0.195 2.004 *** 0.002 (0.074) (0.000) (0.740) t 1 to t 4 0.243 0.910 *** 0.072 (0.074) (0.002) (0.128) t 1 to t 4 1.969 2.737 ** 1.989 *** (0.312) (0.035) (0.000) Adjusted R 2 0.22 0.55 0.30 This table presents results obtained from estimating unrestricted vector au toregressive (VAR) models with credit availability, market liquidity and price return measures specified as endogenous variables. The lag length of the VAR is chosen by looking at the AIC, SBIC, and the likelihood ratio for various choices of p with four l ags providing the best fit. Augmented Dickey Fuller tests suggest the use of changes in credit availability and market liquidity within the VAR system. Summed coefficients and p values from a joint significance test are reported for each endogenous variabl e. TIGHTEN, is the net percentage of loan officers reporting a tightening of Officer Opinion Survey on Bank Lending Practices. PROPSOLD, is the percentage of prope rties sold from the NCREIF NPI index. TBIRET, is the natural log of the quarterly percentage change in the aggregate TBI price index. Control variables include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term struct ure of interest rates (TERMSPREAD), the spread between yields on BAA rated and AAA rated corporate bonds (DEFSPREAD), inflation (INFLA), the Fama French factors: MKT, SMB, HML, augmented by momentum (UMD), and the aggregate capitalization rate for commerci al properties (CAPRT). The sample period spans 1992:Q2 to 2009:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 102

102 Table 3 5. Credit availability and public commercial real estate : individual lags End. Variables REIT RET Constant 0.041 0.6 17 *** 0.291 (0.780 ) (0.000) (0.473 ) REITRET t 1 0.111 0.054 1.203 ** (0.614 ) (0.786 ) (0.047 ) REITRET t 2 0.118 0.171 0 591 (0.491 ) (0.266 ) (0.209 ) REITRET t 3 0.074 0.150 0.817 ** (0.580 ) (0.215 ) (0.027 ) REITRET t 4 0. 131 0.102 0 746 ** (0.324 ) (0.390 ) (0.041 ) t 1 0.055 0.2 93 ** 0.503 (0.615 ) (0.003 ) (0.094 ) t 2 0. 101 0.2 95 * 1.256 *** (0.372 ) (0.004 ) (0. 000 ) t 3 0. 305 ** 0.206 0 253 (0.017 ) (0.073 ) (0.473 ) t 4 0. 540 *** 0. 298 * 0. 372 (0.000) (0.011 ) (0.299 ) LIQ t 1 0.0 19 0.049 0. 318 ** (0.694 ) (0.265 ) (0.018 ) LIQ t 2 0. 037 0.106 ** 0. 303 ** (0.442 ) (0.015 ) (0.022 ) LIQ t 3 0.0 21 0.044 0.325 *** (0.636 ) (0.270 ) (0 .008 ) LIQ t 4 0.080 0.021 0. 168 (0.065 ) (0.595 ) (0.158 ) Adjusted R 2 0.30 0.42 0.11 This table presents results obtained from estimating unrestricted vector autoregressive (VAR) models with credit availability, market liquidity and price return measures spe cified as endogenous variables. The lag length of the VAR is chosen by looking at the AIC, SBIC, and the likelihood ratio for various choices of p with four lags providing the best fit. Augmented Dickey Fuller tests suggest the use of changes in credit ava ilability and market liquidity within the VAR system. Coefficients and p values are reported for each lag length of the endogenous variables. TIGHTEN, is the net percentage of loan officers reporting a tightening of lending standards on commercial real est on Bank Lending Practices. REIT_ LIQ is a price impact measure based on the methodology of Amihud (2002) for the public equity REIT market REITRET is the natural log of the quarter ly appreciation component of the return on the FTSE NAREIT Equity Index. Control variables include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSPREAD), the spread between yields on BAA rated and AAA rated corporate bonds (DEFSPREAD), inflation (INFLA), the Fama French factors: MKT, SMB, HML, augmented by momentum (UMD), the dividend yield on equity REITs (DIVYLD), and market liquidity in the private commercial real estate market (PROPSOLD). The sample period spans 1992:Q2 t o 2009:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 103

103 Table 3 6. Credit availability and public commercial real estate: cumulative eff ect and joint significance End. Variables REITRET LIQ REITRET t 1 to t 4 0.435 0.369 3 .3 56 (0.865 ) (0.352 ) (0.068 ) t 1 to t 4 1 000 *** 1 092 * 2 384 *** (0.001) (0.001 ) (0.001 ) t 1 to t 4 0. 076 0.178 1.114 ** (0.165 ) (0.115 ) (0.028 ) Adjusted R 2 0.30 0.42 0.11 This table presents re sults obtained from estimating unrestricted vector autoregressive (VAR) models with credit availability, market liquidity and price return measures specified as endogenous variables. The lag length of the VAR is chosen by looking at the AIC, SBIC, and the likelihood ratio for various choices of p with four lags providing the best fit. Augmented Dickey Fuller tests suggest the use of changes in credit availability and market liquidity within the VAR system. Summed coefficients and p values from a joint signi ficance test are reported for each endogenous variable. TIGHTEN, is the net percentage of loan officers reporting a tightening of Officer Opinion Survey on Bank Le nding Practices. REIT_ LIQ is a price impact measure based on the methodology of Amihud (2002) for the public equity REIT market REITRET is the natural log of the quarterly appreciation component of the return on the FTSE NAREIT Equity Index. Control varia bles include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSPREAD), the spread between yields on BAA rated and AAA rated corporate bonds (DEFSPREAD), inflation (INFLA), the Fama F rench factors: MKT, SMB, HML, augmented by momentum (UMD), the dividend yield on equity REITs (DIVYLD), and market liquidity in the private commercial real estate market (PROPSOLD). The sample period spans 1992:Q2 t o 2009:Q4. ***, **, and represent 1%, 5 %, and 10% significance levels respectively.

PAGE 104

104 Table 3 7. Credit availability and public commercial real estate: leverage and liquidity sorts. Low Liquidity Low Leverage High Leverage End. Variables REITRET TIGHTEN REITRET TIGHTEN T_LIQ REITRET t 1 to t 4 0.132 1 062 * 6.070 0.740 1.127 * 0. 639 * (0.288 ) (0.004 ) (0.154 ) (0.151 ) (0.005 ) (0.021 ) t 1 to t 4 0. 527 * 0.87 4 *** 2 310 1. 744 *** 0.4 52 0. 904 ** (0.006 ) (0.002) (0.128 ) (0.000) (0.137 ) (0.028 ) t 1 to t 4 0.074 0.038 0 829 *** 0 .0 88 ** 0.083 0.478 *** (0.169) (0.704 ) (0.000) (0.004 ) (0.750 ) (0.000) Adjusted R 2 0.09 0.44 0.4 7 0.43 0.43 0.18 High Liquidity REITRET t 1 to t 4 0.004 0.480 0.233 0.486 0.383 1.020 (0.730 ) (0.265 ) (0.605 ) (0.265 ) (0.108 ) (0.311 ) t 1 to t 4 0. 965 *** 0. 789 * 2 056 2.169 * 0.7 10 ** 0 145 (0.000 ) (0.023 ) (0.332 ) (0.000 ) (0.026 ) (0.165 ) LIQ t 1 to t 4 0. 065 0.059 1. 245 * 0. 493 0.074 0 894 *** (0.146 ) (0.326 ) (0.000 ) (0.052 ) (0.764 ) (0.003 ) Adjusted R 2 0.22 0.38 0 .2 2 0.29 0.37 0.06 This table presents results obtained from estimating unrestricted VAR models on a series of REIT portfolios. Securities are a ssigned to portfolios based on quartile sorts on leverage ratios and share t urnover. The lag length of the VAR is chosen by looking at AIC, SBIC, and the likelihood ratio for various choices of p with four lags providing the best fit. Augmented Dickey Fuller tests sugges t the use of changes in credit availability and market li quid ity in the VAR system. Summed coefficients and p values from a joint significance test are reported for each endogenous variable. TIGHTEN, is the net percentage of loan officers reporting a tightening of lending standards on commercial real estate loans in REIT_ LIQ is a price impact measure based on the methodology of Amihud (2002) for a value weighted portfolio of equity REIT securities in each quartile REITRET is th e natural log of the quarterly appreciation component of the return on a value weighted portfolio of REIT securities in each quartile Control variables include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term struc ture of interest rates (TERMSPREAD), the spread between yields on BAA rated and AAA rated corporate bonds (DEFSPREAD), inflation (INFLA), the Fama French factors: MKT, SMB, HML, augmented by momentum (UMD), the dividend yield on equity REITs (DIVYLD), and market liquidity in the private commercial real estate market (PROPSOLD). The sample period spans 1992:Q2 to 2009:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 105

105 Table 3 8. Credit availability and public commercial real es tate: preferred equity individual lags End. Variables REITPREF Constant 0.168 ** 0.66 2 *** 0.949 (0.028 ) (0.000) (0.279 ) REITPREF t 1 0. 226 * 0. 401 * 0.3 24 (0.038 ) (0.026 ) (0.796 ) REITPREF t 2 0.070 0. 318 0.298 (0.555 ) (0.104 ) (0.826 ) REITPREF t 3 0.2 90 * 0. 429 ** 1 514 (0.007 ) (0.016 ) (0.223 ) REITPREF t 4 0.180 0.077 0.022 (0.090 ) (0.662 ) (0.985 ) t 1 0.035 0.2 70 *** 1.025 (0.536 ) (0.004 ) (0.115 ) t 2 0.031 0.2 62 *** 0.718 (0.596 ) (0.006 ) (0.280 ) EN t 3 0.19 3 *** 0.106 1.259 (0.001 ) (0.280 ) (0.065 ) t 4 0.4 00 *** 0.2 24 * 0. 471 (0.000) (0.050 ) (0.553 ) IQ t 1 0.002 0.029 0. 261 ** (0.886 ) (0.127 ) (0.047 ) LIQ t 2 0.022 0. 023 0. 164 (0.054 ) (0.223 ) (0.214 ) _LIQ t 3 0.047 *** 0.010 0.136 (0.000 ) (0.552 ) (0.225 ) LIQ t 4 0. 030 * 0.016 0.496 *** (0.002 ) (0.327 ) (0.000 ) Adjusted R 2 0.54 0.40 0.27 This table presents results obtained from estimating unrestricted vector autoregressive (VAR) models with credit availability, market liquidity and price return measures specified as endogenous variab les. The lag length of the VAR is chosen by looking at the AIC, SBIC, and the likelihood ratio for various choices of p with four lags providing the best fit. Augmented Dickey Fuller tests suggest the use of changes in credit availability and market liquid ity within the VAR system. Coefficients and p values are reported for each lag length of the endogenous variables. TIGHTEN is the net percentage of loan officers reporting a tightening of lending standards on commercial real estate loans in the Federal Res on Bank Lending Practices. REITPR_ LIQ is a price impact measure based on the methodology of Amihud (2002) for REIT preferred shares. REITPREF is a quarterly value weighted price index for REIT preferred secur ities constructed using data from Thomson Reuters Datastream Control variables include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSPREAD), the spread between yields on BAA rat ed and AAA rated corporate bonds (DEFSPREAD), inflation (INFLA), the Fama French factors: MKT, SMB, HML, augmented by momentum (UMD), the dividend yield on equity REITs (DIVYLD), and market liquidity in the private commercial real estate market (PROPSOLD). The sample period spans 1992:Q2 t o 2009:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 106

106 Table 3 9. Credit availability and public commercial real estate: preferred equity cumulative effect and joint significance. End. Var iables REITPREF LIQ REITPREF t 1 to t 4 0. 186 ** 1. 225 ** 1.466 (0.005 ) (0.017 ) (0.790 ) t 1 to t 4 0. 659 *** 0.8 61 *** 2.037 (0.000) (0.003 ) (0.117 ) t 1 to t 4 0.0 53 * 0.046 0. 457 *** (0.000 ) (0.420 ) (0.000 ) Adjusted R 2 0.54 0.40 0.27 This table presents results obtained from estimating u nrestricted vector autoregressive (VAR) models with credit availability, market liquidity and price return measures specified as endogenous variables. The lag length of the VAR is chosen by looking at the AIC, SBIC, and the likelihood ratio for various cho ices of p with four lags providing the best fit. Augmented Dickey Fuller tests suggest the use of changes in credit availability and market liquidity within the VAR system. Summed coefficients and p values from a joint significance test are reported for ea ch endogenous variable. TIGHTEN, is the net percentage of loan officers reporting a tightening of Officer Opinion Survey on Bank Lending Practices. REITPR_ LIQ is a price impact measure based on the methodology of Amihud (2002) for REIT preferred shares. REITPREF is a quarterly value weighted price index for REIT preferred securities constructed using data from Thomson Reuters Datastream Control variables include th e yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSPREAD), the spread between yields on BAA rated and AAA rated corporate bonds (DEFSPREAD), inflation (INFLA), the Fama French factors: MKT, SMB, HML, augmented by momentum (UMD), the dividend yield on equity REITs (DIVYLD), and market liquidity in the private commercial real estate market (PROPSOLD). The sample period spans 1992:Q2 t o 2009:Q4. ***, **, and represent 1%, 5%, and 10% sign ificance levels respectively.

PAGE 107

107 Table 3 10. Credit availability and investor sentiment using a direct sentiment measure End. Variables TBIRET TBIRET t 1 to t 4 0.114 ** 1.451 ** 0.014 0.836 (0.036) (0.039) (0.715) (0.248) t 1 to t 4 0.217 ** 0.953 *** 0.066 1.831 (0.037) (0.000) (0.136) (0.053) t 1 to t 4 1.613 4.577 ** 1.789 *** 1.558 (0.199) (0.045) (0.000) (0.054) t 1 to t 4 0.076 ** 0.167 ** 0.001 0.347 *** (0.049) (0.029) (0.585) (0.006) Adjusted R 2 0.14 0.50 0.23 0.07 This table presents results obtained from estimating unrestricted vector autoregressive (VAR) models with credit availability, market liquidity, pric e return, and investor sentiment measures specified as endogenous variables. The lag length of the VAR is chosen by looking at the AIC, SBIC, and the likelihood ratio for various choices of p with four lags providing the best fit. Augmented Dickey Fuller t ests suggest the use of changes in credit availability, market liquidity, and investor sentiment within the VAR system. Summed coefficients and p values from a joint significance test are reported for each endogenous variable. TIGHTEN, is the net percentag e of loan officers reporting a tightening of lending standards on commercial real estate loans in the PROPSOLD is the percentage of properties sold from the NCREIF NPI in dex. TBIRET is the natural log of the quarterly percentage change in the aggregate TBI price index. DRES is a direct survey measure of investor sentiment for the commercial real estate market obtained from the eal Estate Report. Control variables include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSPREAD), the spread between yields on BAA rated and AAA rated corporate bonds (DEFSPREAD ), inflation (INFLA), the Fama French factors: MKT, SMB, HML, augmented by momentum (UMD), and the aggregate capitalization rate for commercial properties (CAPRT). The sample period spans 1992:Q2 t o 2009:Q4. ***, **, and represent 1%, 5%, and 10% signifi cance levels respectively.

PAGE 108

108 Table 3 11 Credit availability and investor sentiment using an indirect sentiment measure. End. Variables TBIRET TBIRET t 1 to t 4 0.122 *** 1.538 *** 0.011 0.038 (0.005) (0.006) (0.824) (0.622) t 1 to t 4 0.249 *** 0.902 *** 0.076 2.698 (0.010) (0.001) (0.064) (0.376) t 1 to t 4 2.006 0.100 2.285 *** 79.151 ** (0.647) (0.118) (0.000) (0. 033) t 1 to t 4 0.020 *** 0.085 0.007 0.159 *** (0.000) (0.087) (0.228) (0.000) Adjusted R 2 0.35 0.56 0.29 0.06 This table pr esents results obtained from estimating unrestricted vector autoregressive (VAR) models with credit availability, market liquidity, price return, and investor sentiment measures specified as endogenous variables. The lag length of the VAR is chosen by look ing at the AIC, SBIC, and the likelihood ratio for various choices of p with four lags providing the best fit. Augmented Dickey Fuller tests suggest the use of changes in credit availability, market liquidity, and investor sentiment within the VAR system. Summed coefficients and p values from a joint significance test are reported for each endogenous variable. TIGHTEN, is the net percentage of loan officers reporting a tightening of lending standards on commercial real estate loans in the Federal Reserve Bo PROPSOLD is the percentage of properties sold from the NCREIF NPI index. TBIRET is the natural log of the quarterly percentage change in the aggregate TBI price index. INDRES is an indirec t measure of investor sentiment for the commercial real estate market constructed from a number of sentiment proxies. Control variables include the yield on three month U.S. Treasury securities (TBILL), the slope of the Treasury term structure of interest rates (TERMSPREAD), the spread between yields on BAA rated and AAA rated corporate bonds (DEFSPREAD), inflation (INFLA), the Fama French factors: MKT, SMB, HML, augmented by momentum (UMD), and the aggregate capitalization rate for commercial properties (C APRT). The sample period spans 1992:Q2 t o 2009:Q4. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 109

109 CHAPTER 4 VOLUNTARY DISCLOSURE INVESTOR SENTIMENT AND RETURN COMOVEM ENT The decisions of major firms such as Coca Cola Peps i Co Berkshire Hathaway, and AT&T to reduce or eliminate practices of providing voluntary earnings guidance to its investors created much of a stir on Wall Street over the past decade. Historically, investors relied on these voluntary disclosures as an op and its potential for future growth. However, there has been significant controversy as to whether the use of such announcements is in fact beneficial to shareholder wealth. On one hand, proponents for voluntary disclosure believe that management issued earnings guidance increases profitability. On the other hand, critics believe that earnings guidance allows m anagement to mislead investors in an attempt to facilitate higher asset valuations, particularly at times when investors are optimistic. This paper adds to the growing debate on whether the strategic disclosure of earnings guidance by firm management cate rs to misguided investor beliefs. Several recent empirical studies have examined the effects of strategic disclosure on the time series behavior of stock returns (Goto, Watanabe, Xu, 2009; Jiang, Xu, and Yao, 2009; Haggard Martin, and Pereira, 2008). Moreo ver, a few studies have begun to investigate the relation between voluntary disclosure practices and investor sentiment (Bergman and Roychowdhury, 2008; S eybert and Yang, 2011; Brown et al., 2011). While most of these studies have identified voluntary disc losure as a means for reducing information asymmetry, correcting overvaluation, and improving stock price informativeness, very few have provided evidence that managers issue earnings guidance to take advantage of misguided investor beliefs. Most prior stu dies have been limited by the fact that they have chosen to either aggregate the number of disclosures over a

PAGE 110

110 particular period (disclosure frequency) or examine the disclosure over extremely short event horizons (event study). In doing so, the literature fails to address the influence that selective disclosure has on longer term asset price movements. Using a return comovement approach, I am able to provide new evidence contrary to these previous findings and consistent with managers catering their disclos ures to take advantage of sentiment induced mispricing. The main contribution of this analysis focuses on determining whether firm management strategically issues voluntary earnings guidance in order to take advantage of misguided investor beliefs, particu larly during periods of high investor optimism, and whether such disclosures have a significant impact on future asset pr ice movements. In particular, I attempt to answer the following research question. Do firms that strategically disclose during high sen timent periods increase their comovement with stocks that are highly sensitive to shifts in investor sentiment (i.e., those firms that experience significant increases in their stock prices when investor sentiment rises) following the announcement? If firm managers strategically issue voluntary earnings guidance that is consistent with misguided investor beliefs, following the disclosure event one would expect these stocks to covary mo re with firms that are most sensitive to changes in investor sentiment an d less with market fun damentals. As a byproduct of this analysis, I also provide evidence that adds to the literature on the determinants of retu rn comovement. In particular, I provide further empirical support for the sentiment based theory originally exa mined by Barberis, Shleifer, and Wurgler (2005). I test my main hypothes is using a two step approach. I begin by identifying those firms that are the most likely to strategically provide earnings guidance in order to take advantage of sentiment induced bel iefs. Because volunta focus on voluntary

PAGE 111

111 (LD), defined as those firms that have not provided voluntary earnings guidance in the 12 m disc lose during only one year of my sample period and do not disclose in future years. I hypothesize that these firms are the most likely to strategically discl ose since they do not have a hi story of repeat disclosures. The sample period of this study spans January 1998 to December 2008. To identify whether company issued guidance impacts future asset price movements, I examine shifts in return comovement around the disclosure event. In particular, I utilize bivariate and multivariate regressions similar to Barberis, Shleifer, and Wurgler (2005) and Green and Hwang (2009). I estimate regressions for each firm using daily returns over the 12 month period ending one month prior to the disclosure and the 12 month period staring one month after the disclosure. The sentiment portfolio is constructed using a sentiment beta appr oach similar to Glushkov (2006) For both the LD and ID samples, I find that firms issuing volu ntary earnings guidance during high sentiment periods experience a significant increase in return comovement with the sentiment portfolio and a significant decrease with market fundamentals following the disclosure event. Prior to the disclosure event, the se firms exhibit negative or zero comovement with sentiment sensitive stocks. However, subsequent to the disclosure, disclosing firms experience significant positive comovement with the sentiment portfolio. This is consistent with managers issuing earning s guidance to take advantage of misguided investor beliefs during times of inve stor confidence. Furthermore, I find evidence that this effect is concentrated around the issuance of neutral earnings guidance, which by nature contains no new information abou t a

PAGE 112

112 ent based theory of comovement. Background Literature Several empirical studies have emerged examining how strategic disclosure of information affects the properties of time series r eturns. Goto, Watanabe, and Xu (2009) document significant shifts in the properties of stock returns for firms that undergo changes in disclosure environments, focusing on firms that choose to cross list a security on a foreign exchange. Jiang, Xu, and Yao (2009) analyze how the quality of information disclosures impact idiosyncratic incentives to either disclose or withhold information from their investors. This branch of literature has evolved considerably from the full disclosure theories of Grossman and Hart (1980) and Grossman (1981). Verrechia (1983) and Dye (1985) were the first to develop ater their disclosure decisions based on the perceived cost of providing information. Subsequent work has focused on identifying the source of this cost. One aspect that has received increased attention is how the sophistication of investors and their abil ity (or inability) to interpret information affects the decision to disclose and the timing of the disclosure (Dye, 1998; Fishman and Hagerty, 2003). One possibility is that firms cater their disclosure policies to take advantage of misguided investor bel iefs at times when less sophisticated traders are likely to be the marginal investor. misguided beliefs. Subsequent research defines these misguided beliefs about future cash flows and / or investment risks as investor sentiment (Baker and Wurgler 2006, 2007). Therefore, when investor sentiment

PAGE 113

113 is high, managers may utilize this opportunity to issue earnings guidance in an attempt to take advantage of investor optimism and higher asset valuations. Several recent empirical studies have begun to examine whether firms cater their voluntary disclosure decisions to the level of investor sentiment and how this impacts asset prices. Brown horizon adjusted (pro forma) earnings metrics increases with the level of sentiment. Furthermore, the authors provide evidence that managers tend to pro vide more favorable earnings metrics when investor sentiment increases. This study is limited by the fact that the authors focus only on short horizon guidance and fail to examine any asset pricing implications of these disclosures. Bergman and Roychowdhur y (2008) find that during high sentiment periods, managers reduce the frequency of their long horizon fails to address the implications of those disclosures that a re in fact made during these high sentiment periods. Seybert and Yang (2011) examine these disclosures more closely, focusing upon stock price movements in the days immediately surrounding the event. The authors provide evidence that earnings guidance aids in resolving sentiment driven overvaluation over short event windows. Because the authors focus on both short and long horizon guidance, as well as include all disclosures in their sample period, they overlook the events that are most likely designed to t returns over such a short horizon, the authors are unable to document changes in the patterns of asset price movements over time. Another way to identify changes in asset pr ice movements around a corporate event is to test whether there is a significant shift in return comovement with a particular portfolio of stocks (e.g. stocks in the same industry, size deciles, index, etc.) following the event (Barberis, Shleifer,

PAGE 114

114 and Wur gler, 2005; Greenwood, 2008; Kumar and Lee, 2006; Green and Hwang, 2009; Goto, Watanabe, and Xu, 2009). Of particular interest is the work of Barberis, Shleifer, and Wurgler (2005) who distinguish between two theories that attempt to explain why asset valu es move together over time: the fundamental and sentiment based theories of comovement. According to the fundamental based theory, in a frictionless market consisting of rational investors, any damentals. The sentiment based theory, on the other hand, posits that the comovement in returns can be delinked from fundamentals in markets with frictions or those that consist of irrational investors. 1 Haggard, Martin, and Pereira (2008) are the first to utilize voluntary disclosure as a crite rion for return comovement. The authors examine whether firms with more forthcoming disclosure policies, based upon rankings of disclosure quality, covary more or less with market and indust ry returns. However, they do not analyze return comovement around specific disclosure events nor do they explicitly test the theory of sentiment based return comovement. This study is the first to utilize voluntary disclosure events to examine sentiment base d return comovement. Ov erall, I bridge the gap between the voluntary disclosure, investor sentiment and return comovement literature by examining whether strategically issued earnings guidance increases return comovement amongst firms that disclose and those firms that are most sensitive to c hanges in investor sentiment. I provide new evidence, contrary to Seybert and Yang (2011), supporting the hypothesis that strategically issued earnings guidance contributes to sentiment induced mispricing. 1 Veldkamp (2006) and Jin and Myers (2006) develop similar theories regarding information based comovement within a rational expectations framework, where the revelation of new information takes the place of sentiment.

PAGE 115

115 Data and Descriptive Statistics Volu ntary Disclosure Since there is no regulation governing its issuance, management issued earnings guidance represents a voluntary form of information disclosure. At any point in time, firm management may decide to issue a press release or conduct an intervi ew to provide investors with information regarding whether a firm expects to meet, beat, or fall short of analyst earnings estimates. These disclosures may be disseminated in the form of a specific numerical value, a range of estimates, or even a text stat ement. Since the scope of this study focuses primarily on identifying the date on which a voluntary disclosure is made, rather than point estimates of the earnings guidance themselves, I rely upon the Thomson First Call Company Issued Guidance (CIG) databa se as my source of voluntary disclosure data. The CIG database contains management issued earnings guidance dating as far back as 1994. However, prior literature has documented significant shortcomings of this database in the early years of its existence, particularly its inability to capture all issuances of earnings guidance. Since First Call originally was created as a proprietary provider of sell side analyst research, the collection of management issued earnings guidance was not its primary focus. Ther efore, it is possible that this database initially suffered from significant sample selection bias, only capturing company issued earnings guidance for those firms with analyst coverage (Chuk, Matsumoto and Miller, 2009). However, Anilowski, Feng and Skinn er (2007) document a company issued earnings guidance in a more systematic fashion. In order to mit igate sample selection bias, the sample period begins in January of 1998 and runs through December 2008. Bergman and Roychowdhury (2008) suggest that strategic disclosure is likely to be concentrated in long horizon earnings guidance. The authors argue that managers are more likely

PAGE 116

116 to be able to maintain optimistic ea rnings valuations in their long horizon disclosures rather than their short term guidance since they do not face the same short run incentives to meet analyst estimates regarding imminent earni ngs announcements. Therefore, I focus specifically on long hori zon earnings guidance. Following Ber gman and Roychowdhury (2008), I classify management forecasts that were issued more than 90 days before the estimate period end date as long horizon earnings guidance. Prior research documents that earnings guidance pra if a firm initiates a disclosure policy, it may decide to continue to disclose from year to year or from period to period within a year. Most studies that have examined the role of voluntary disclosure in asset prici ng have chosen to either aggregate the disclosures over a particular period (disclosure frequency) or examine the disclosure over short horizons (event study). In doing so, the literature fails to address the influence that selective disclosure has on long er term asset price movements. Rather than focus on the frequency or increment al impact of each disclosure, I attempt to identify situations in which firms strategically disclose to take advantage of misguided investor beliefs. Prior research has shown th at follow up disclosures will typically walk up or walk down the initial guidance estimate (Bergman and Roychowdhury, 2008), suggesting that the incremental value of subsequent guidance is of secondary importance. In addition, the initial voluntary disclos ure is an attention grabbing event that may ultimately attract investors to the stock (Barber and Odean, 2008). Therefore, I begin by collecting the first long horizon ll firms in the CIG database. 2 I then further pa rtition the 2 I exclude real estate investment trusts (REITs) from my analysis for several reasons including the tendency of analysts to focus on funds fr om operations ( FFO ) estimates rather than earnings ( EPS ) for these firms, as well as empirical evidence that REIT returns appear to respond to real estate market sentiment, rather than stock market sentiment, as reported in my first study.

PAGE 117

117 data into two groups for my main analysis. The first group, wh ich I refer to as the the 12 months prior to the event. 3 The second group, which I refer to as the (ID) sample, consists of firms that disc lose during only one year of the sample period and do not disclose in future years. To address the potential complica tion of a look ahead bias in the ID sample, I also ex amine the first time that a firm discloses, even if they do disclose ag ain, in a series of robustness checks. Sentiment Portfolio Baker and Wurgler (2006, 2007) and Glushkov (2006) empirically examine whether certain stocks are more sensitive to sentiment by testing whether their prices co move with an index of sentiment changes. In other words, a stock with a positive sentiment beta will experience a contemporaneous increase in price when investor sentiment increases. To identify firms that are highly sens itive to c hanges in investor sentiment, I utilize a sentiment beta approach similar to Gl ushkov (2006). In particular, I estimate an OLS regression of monthly stock returns on changes in investor sentiment and a series of fundamental control variables. The specification is as follows: (4 1) where R i,t t (2006, 2007) sentiment index 4 MKT t is the excess return on the value weighted market portfolio, SMB t is the return on the small minus big size portfolio, HML t is the ret urn on the 3 Christensen e t provided guidance in the 2 years prior to the event. Results are stronger in terms of the magnitude and precision of the coefficient estimates in the more restrictive case. 4 I obtain the investor sentiment index for the general stock market data thro ugh 2008.

PAGE 118

118 high minus low book to market portfolio (Fama and French 1993, 1996), MOM t is the return on the momentum portfolio (Carhart, 1997), and LIQ t factor. Equation (4 1) is estimated using a five year (60 mon th) rolling window to obtain for each stock in the CRSP database. I start with monthly return data from January 1993, since the first observation in the voluntary disclosure database is from January 1998, and roll forw ard one month at a time until I have estim ates for all months in the sample period. For each month in the sample period, I then construct a value weighted portfolio consisting of those firms that exhibit a positive and statistical ly significant (at the 5 % level) sentiment beta over the five year estimation period. This allows for a rebalancing of the sentiment portfolio each month based on which firms are most sensitive to changes in investor sentiment. The median number of firms included in the sentiment portfolio is approximately 170. The t ime series average sentiment beta for firms included in the sentiment portfolio is 0.07 Return and Control Variable Data Sources. I Security Prices (CRSP). For the vol untary d isclosure sample, I require firms have 12 months of daily return data available in the CRSP database prior to the event date. As a m easure of market fundamentals I utilize the excess return on the value weighted CRSP market portfolio (MKT). To control for return comovement that may be related to firm characteristics other than their exposure to sentiment and the market, I construct value weighted indices related to size and industry classifications. For size portfolios, I categorize firms into size deciles based on market capitalization breakpoints at the end of each month. For industry portfolios, I utilize the Fama

PAGE 119

119 and French 49 industry classifications. Size breakpoints and industry classifications are obtained Descriptive S tatistics Table 4 1 report s descriptive statistics regarding the frequency of long horizon disclosures by year for all firms in the CIG database, the LD sample, and the ID samples. Focusing on the first long horizon disclosure in a fiscal year for all firm s that provide voluntary disclosure, I begin with a sample of 1 1,024 firm observations over the eleven year sample period. The top panel of Table 4 1 indicates that the median number of long horizon forecasts for all disclosing firms was a single disclosur e prior to 2001. However, after 2001 there was a considerable increase in the frequency of disclosure for each firm within a particular year as well as the number of firms choosing to voluntarily disclose. This is consistent with Regulation Fair Disclosure (Reg. FD) encouraging firms to be more forthcoming with their disclosure policy beginning in late 2000. After 2000, the number of firms disclosing appears to remain fairly consistent, which may indicate that once firms begin their disclosure policy, they continue to provide guidance in the next year. When I focus on firms that have not disclosed in the previous y ear (LD), as reported in the middle panel of Table 4 1 disclosure stickiness becomes less of an issue. This removes firms that are consistently p roviding earnings guidance from year to year and therefore places the focus on those firms that are more likely to be strategically varying their disclosure of earnings guidance. After applying this filter to the data, I am left with 3,420 firm observation s. The average frequency of disclosure within year for these firms is less than the full disclosure sample, which eases concerns about isolating the first disclo sure in a particular year as the event date for this subset of firm s. When I compare the number of firms in the limited disclosure sample to all firms in the CIG database the prevalence of repeat disclosure in the full sample b ecomes clear.

PAGE 120

120 For example, if I compare the number of firms disclosing in year 2000 in the LD sample to those in all firms in the CIG database, 289 of the 617 firms that provided long horizon earnings guidance did so in the previous year as well. The bottom panel of Table 4 1 report s descriptive statistics for the ID sample. By including just those firms that disc lose during o nly one year of the sample period and do not disclose in future years, I am left with 1,003 firm observations over the eleven year sampl e period. Throughout most of the sample period, the median number of long horizon forecasts for these firms is a single disclosure within a year. Focusing on empirical test as they appear to be the most likely to strategically select the timeliness of their n the decision to disclose may be linked to taking advantage of when it is most advantageous for them to do so. Prior research indicates that the level of investor sentiment is a key determinant of the timeliness of these disclosures. Therefore, it is imp ortant to understand how these disclosures are concentrated in high sentiment versus low sentiment periods. Panels A and B of Figure 4 1 providing vol untary earnings guidance for the LD and ID samples, respectively. In both Panel A and Panel B, the number of firms that are disclosing is considerably higher when sentiment is high and decreases as investors become less optimistic. The contemporaneous correlation between the number of firms disclosing and movements in the sentiment index over the sample period are positive and significant for both the limited disclosure (0.17) and isolated disclosure (0.20) samples. This suggests that firms may provide voluntary earnings g uidance when sentiment is high in order to take advantage of misguided investor beliefs. It is interesting to note that during the most significant decrease in sentiment, which occurred with the bursting off the

PAGE 121

121 technology bubble in mid 2001, there are zer o firms initiating new earnings guidance for their fiscal year. This suggests that firms may be more likely to strategically time their first voluntary disclosures to take advantage of increasing investor optimism. Tables 4 2, 4 3, and 4 4 report descrip tive statistics regarding industry, size, and return characteristics respectively, for each set of disclosure firms as well as the sentiment portfolio. Table 4 2 documents the percentage of firms in each industry classificati on. For ease of presentation, I only report the top ten industry distributions from the Fama French 49 industry categories. Baker and Wurgler (2006, 2007) suggest that small growth oriented firms are highly susceptible to sentiment induced mispricing. Consistent wi th this hypothesis, f irms in the sentiment portfolio are concentrated in growth oriented industries such as Drugs (Pharmaceuticals), Chips, and Telecom. Firms that provided earnings guidance, on the other hand, are generally concentrated in value oriented industries such as Ut ilities, Retail, and Business Services. This suggests that ex ante these firms would be less likely to be influenced by investor sentiment. On e exception is the overlap ob serve d in the software sector, where the divide between sentiment and disclosure firm s is less prominent. This suggests that controlling for industry comovement will also be important in the multivariate specifications. Table 4 3 reports the percentage of firms in each size decile. Consistent with Baker and Wurgle is, the sentiment portfolio consists of a significant proporti on of small stocks. In fact, 49 % of the stocks in the sentiment portfolio are from the smallest size decile. In general, firms providing voluntary guidance appear to be more evenly distributed a cross size deciles. However, it is worth noting that the ID sample exhibits some similarities to the composition of the sentiment portfolio as it too includes a high proportion of firms from the

PAGE 122

122 smallest size decile. Therefore, it will also be necessary to control for size based comovement in multivariate regressions in order to isolate the sentiment effect. While my primary regression specifications utilize indi vidual firm stock returns as dependent variables, I construct aggregat e return indices for each of the disclosure groups a s well as the sentiment por tfolio using daily data over the full sample period in order to provide some insight into the general characteristic s of these portfolio returns. I report summary statistics for each of set of daily ret urns in Table 4 4 On average, the returns on the sentiment portfolio are higher and more volatile than those of firms that voluntarily disclose. This is consistent with be most sensitive to changes in sentiment. This is also consistent with the premise that disclosing firms are more transparent and therefore have lower price volatility. Empirical Methodology Bivariate Regression Approach In an attempt to distinguish betw een traditional and sentiment based return comovement, Barberis, Shleifer and Wurgler (2005) are the first to utilize a bivariate regression approach to test whether return comovement shifts as firms are added to or removed from the S&P 500. In support of the sentiment based view of comovement, the authors find that firms will covary more with S&P index stocks and less with non S&P stocks after inclusion in the index, even if no new fundamental information about the firm has been revealed. Green and Hwang ( 2009) utilize a similar bivariate regression approach that includes high and low price portfolios to test whether firms experience a shift in return comovement around a stock split event. If investors categorize stocks based on price, the stock will covary more with the low price index and less with the high price index following the split.

PAGE 123

123 I utilize a similar approach to identify whether firms experience a shift in return comovement around a volun tary disclosure event, except I focus on whether firms incre ase their comovemen t with the sentiment portfolio following the disclosure. If firm managers strategically issue voluntary earnings guidance that is consistent with misguided investor beliefs, following the disclosure event one would expect these stocks to covary more with stocks that are most sensitive to investor sentiment and less with the market portfolio, particularly during periods of high sentiment. This return comovement approach has the ability to not only capture the short run momentum created aro und the disclosure event, but also the long run reversal that these stocks may exp erience in subsequent months. I estimate the following bivariate regression separately before and after the disc losure for each firm in both the LD and ID samples: (4 2) where R i,t is the excess return of stock i at time t, SENT i,t is the excess return on the sentiment portfolio and MKT i,t is the excess return on the value weighted market portfolio. To avoid spurious effects, I remove the contribution of t he disclosure stock from the independent variables where appropriate. I estimate the regression using daily returns. The pre event regression is run over the 12 month period ending one month prior to the disclosure and the post event regression is run over the 12 month period staring one month after the disclosure. I report results for disclosures made over the entire sample period and then disaggregate disclosures into periods in which sentiment is positive (high sentiment) and periods in which sentiment i s negative (low sentiment) to isolate the sentiment effect. Multivariate Regressions To control for other fundamental firm characteristics that may be driving return comovement around the disclosure event, I modify Equation 4 2 to include size and industry

PAGE 124

124 portfolios. I estimate the following multivariate regression separately before and after the disclosure for both subsets of firms in the sample: (4 3) where R i,t is the excess retu rn of stock i at time t, SENT i, t is the excess return on t he sentiment portfolio, MKT i,t is the excess return on the value weighted market portfolio, SIZE i,t is the value weighted size index return (using size deciles), and IND i,t is the value weighted industry return (using the Fama and French 49 industry classi fication). The inclusion of industry and size portfolios controls for other potential causes of fundamental s induced return comovement. Barberis, Shleifer and Wurgler (2005) suggest that investors may categorize assets according to size or industry charac teristics, thereby causing firm valuations within these categories to move together. Furthermore, firms in certain industries may share common fundamental information concerning cash flows. Therefore, comovement may be driven by industry level news outside of the disclosure. Similarly, firm size will be highly correlated with analyst following, liquidity, and institutional ownership, all of which may have an impact on how information is incorporated into prices. Finally, Baker and Wurgler (2006, 2007) hypot since these firms are generally more difficult to value and more likely to be speculative stocks. To ensure that I am not just capturing small size and growth oriented comove ment with the role around the disclosure event. Empirical Results Bivariate Regressions Table 4 5 provides estimates of the bivariate regression m odel for th e LD sample. T he cross SENT MKT and their

PAGE 125

125 p values over the full sample period, during high sentiment periods, and during low sentiment periods are reported. Focusing first on the fu ll sampl e period results, I provide preliminary, although weak, evidence that firms experience a shift in comovement with the sentiment SENT is positive (0.0126) and marginally significant (p value =0.094). This is consistent with firms increasing their comovement with the sentiment portfolio after the disclosure. If firms are covarying more with stocks that are highly sensitive to changes in sentiment following the disclosure, then one would expect these firms to be moving less with market fundamental s after the event as well. While the MKT is negative ( 0.0176), as expected, the change is not statistically significant (p value = 0.185). By aggregating results across the full sample period, it is possible to miss the diffe rential effect that voluntary guidance may have on asset price movements in high versus l ow sentiment periods. Recall, I hypothesize that firm managers are likely to strategically issue voluntary guidance during high sentiment periods in such a manner that caters to misguided investor beliefs. Therefore, one would expect a shift in return comovement with the sentiment portfolio to be concentrated around disclosures that occur in high sentiment periods. When results are disaggregate d into hi gh and low sentim ent periods, I find this to be the case. Focusing on the h igh sentiment period results, I provide stronger evidence that firms increase their return comovement with the sentiment portfolio after voluntarily issuing earnings guidance. Although not reported it is important to note that in each high sen timent regression specification SENT for the 12 months prior to the event is either negative or statistically indistinguishable from zero. This is consistent with the belief that in general, firms participating in voluntary disclosure practices are more transparent and therefore ex ante should

PAGE 126

126 be less susceptible to sentiment induced mispricing. However, if firms are using guidance to strategically cater to misguided investor beliefs d uring high sentiment periods, one would expect these firms to covary more with the sentiment portfolio after the disclosure. The coefficient on SENT for the high sentiment specification is positive (0.0188) and strongly significant (p value=0.037), supporting the hypothesis that strategic disclosure caters towards misguided investo r be liefs. Equally important, I MKT = 0.0530) and statistically significant shift (p value = 0.001) in comovement away from the market portfolio after the event. The corresponding decrease in the coefficient on the market portfolio indi cates that the shift in comovement is less likely to be driven by a change in fundamentals following the disclosure. sentiment based comovement. Moving to the low sentim ent period results, I do not find evidence of a significant shift in comovement with the sentiment por SENT = 0.0035, p value = 0.797) after the disclosure, suggesting that the effect is in fact concentrated in high sentiment periods. It is interesting to note that in the low sentiment state, however, I observe a significant increase in comovemen t with the market portfolio, suggesting that disclosures communicate more about fundamentals during market declines. This is consistent with prior literature documenting the use of earnings guidance to provide warnings to investors, particularly during eco nomic downturns (Kasznik and Lev, 1995; Soffer, Thiagarajan and Walther, 2000). Table 4 6 provides estimates of the bi variate regression model for the ID sample. Recall, firms in the ID sampl e only appear in one year of the sample period, so these firms m ay be more likely to cater their earnings guidance to misguided investor beliefs during high sentiment periods. In the full sample period specifications, I find results similar to those using the LD

PAGE 127

127 sample. I document a marginally significant (p value = 0. 094) positive ( SENT = 0.0341) shift in return comovement with the sentiment portfolio around the disclosure event. However, even in the full sample period, I MKT = 0.0901) and statistically significant shift (p value = 0.012) in co movement away from the market portfolio after the event. Moving to the hig h sentiment period results, coefficient estimates grow stronger in both magnitude and statistical significance. Following the issuance of earnings guidance during a high sentiment pe riod, these firms experience an increase in their comovement with the sentiment portfolio ( SENT = 0.0438, p value = 0.068) and a decrease in comovement with the market MKT = 0.1321, p value = 0.002). However, in low sentiment periods, I do not observe any significant shifts in comovement around the disclosure event. Taken togethe r, these results suggest that providing voluntary earnings guidance during periods of high sentiment causes firms to increase their return comovement with sentiment sensitive stocks and delink their returns from market fundamentals. Multivariate Regression s To ease concerns that the observed shift in comovement with the sentiment portfolio subsequent to the disclosure event is due to a change in market fundamentals, I estimate multivariate regressions that include size and industry portfolios in addition to the sentiment and market portfolios. Table 4 7 presents results for the LD sample. What I am most intere sted in observing is whether the high sentiment period results remai n robust to the inclusion of size and industry controls. Even after controlling for other potential sources of fundamental based comovement, I continue to document a significant positive shift ( SENT = 0.0165, p value = 0.065) in return comovement with the sentiment portfolio following the issuance of voluntary earnings guidance in high sentiment periods. I also continue to find firms delinking from the market portfolio following the disclosure, as the shift in return comovement is negative and

PAGE 128

128 MKT = 0.0491, p value = 0.083). R esults for the low sen timent period are similar to those from the bivariate regressions, though the previously documented shift in return comovem ent with the market portfolio has weakened with the inclusion of industry and size con trols. This may suggest that market, industry, and size portfolios are capturing some common sources of fundamental based comovement. Estimates from multivariate regres si ons using the ID sample are reported in Table 4 8 After controlling for indust ry and size based comovement, I continue to find that firm returns covary more with the sentiment portfolio ( SENT = 0.0558, p value = 0.023) and less with fundamentals following the issuance of earnings guidance in high sentiment periods. The magnitude of the coefficients and the precision of the estimates on the sentiment portfolio in the ID sample are conside rably larger than in the case of the LD firms. This pro vides additional support for my initial hypothesis that sentiment based return comovement increases following a voluntary disclo sure event, particularly when focus ing on firms that are more likely to s trategically provide earnings guidance during high s entiment periods. Similar to the LD sample, results from the full sample and low sentiment period estimations have weakened with the inclusion of industry and size controls. Matched Sample Regressions Ber gman and Roychowdhury (2008) suggest that firms reduce their frequency of induced be equally as likely to increase their comovement with the sentiment portfolio during high sentiment periods. If this is the case, then the observed shift in comovement may not be related to the disclosure event, but rather to the period of hi gh sentiment in which the event occurs. In

PAGE 129

129 order to address this alternative explanation, I construct a matched sample o f non disclosing firms using the ID firms as the base case. Using the ID sample as the base case for this match is appealing for severa l reasons. Fi rst of all, in comparison to the LD sample, firms in the ID sample exhibit size and industry characteristics that are more closely related to the sentiment portfolio than to the full sample of disclosure firms (Tables 4 2 and 4 3 ). Second, som e may argue that creating a matched sample of non disclosure firms suffers from an inherent endogeneity problem in that firms who disclose may be fundamentally different from those that never provide guidance (Houston, Lev and Tucker, 2010). Since the ID f irms are those firms that only disclose during one year of the sample period and never disclose again, I believe that using thes e firms as the base case for the matched sample minimizes the potential endogeneity issue. 5 Finally, based on previous results, ID firms appear to exhibit the most significant shift in return comovement around disclosure events. Therefore, if the comovement was being driven purely by the level of sentiment in these months rather than the disclosure event, matching based on these po ints in time would be most appealing. For each firm that voluntarily issues earnings guidance in the ID sample, I select a control firm from the same industry and same size decile that does not disclose at any point during the sample period. F or each match ed stock, I run bivariate and multivariate regressions over the 12 12 5 An alternative approach for creating a matched sample would be to identify firms that have disclosed in the past, but have ceased to continue their disclosure policy in high sentiment periods. However, Houston, Lev, and Tucker (2010) argue tha analyst expectations, not to market timing.

PAGE 130

130 Table 4 9 and 4 10 report bivariate and multivariat e results, respectively, for the matched sample. For the purpose of this test, I focus only on high senti ment period results. In both the bivariate and multivariate specifications, matched firms do not exhibit the same significant positive s hift in return comovement with the sentiment portfolio nor a significant delinking from fundamentals as documented previously for those firms that do issue voluntary earnings guidance in these same high sentiment periods. This further supports the hypothes is that firms may provide voluntary earnings guidance to take advantage of misguided investor beliefs during periods of high investor sentiment. Multivariate Regressions by Disclosure Type One of the contributions of this study is new evidence I provide in support of sentiment based theories of return comove ment. Some may argue that I am unable to distinguish between traditional (fundamental) and s entiment based comovement in these empirical tests due to the nature of voluntary earnings guidance. In other w ords, voluntary earnings guidance may comovement b etween disclosing stocks and the sentiment portfolio may be related to some underlying change in the fundamentals of these stocks that is currently unac counted for. In other words, the specifications may suffer from an omit ted variable bias. Therefore, I design an additional test based on the type of disclosure that will provide further evidence supporting the sentim ent based theory. One of the key distinctions between fundamental a nd sentiment based comovement depends upon how the revelation of new information, or lack thereof, impacts asset price movements. According to the fundamental view of return comovement, ev ents that do not signal

PAGE 131

131 comovement following the event. However, in the sentiment based view, shifts in return comovement may occur even if no new information regarding fundamentals is disseminated. The CIG database includes a field (CIG Description Code) that categorizes company issued earnings guidance based on how it compares to the consensus analyst forecast at the time the guidance is made. If the earnings guidance is greater than the consensus analyst estimate, then this disclosure is classified as a positive earnings surprise. If the earnings guidance is less than the consensus analyst estimate, then this disclosure is classified as a negative earnings surprise. If the earnings guidance meets consensus analyst expectations, then it is classified as neutral or as meeting ana lyst expectations. While I am aware of the limitations in using these classifications, such as the significant loss of observations due to missin g codes (NA) or erroneous classifications, 6 they provide sufficient information for an additional test of the underlying hypotheses. Since I am interested in understanding the impact of voluntary earnings guidanc e in times of high sentiment, I focus on ne utral earnings guidance and positive earnings surprises during these periods. If managers wish to cater to optimistic investor beliefs duri ng periods of high sentiment, one would expect them to issue either guidance that is consistent with these misguided beliefs (neutral) or that expands upon investor optimism (positive surprises). The distinguishing feature between these two types of guidance is the information, or lack thereof, being conveyed. Managers who strategically provide neutral guidance that mere ly meets analyst manager issues a positive earnings surprise, then new information is being released. Therefore, if one observe s a significant positive shift in r eturn comovement with the sentiment portfolio 6 Rees and Wynalda (2008) provide a thorough analysis of the main issues regarding CIG Description Code classi fications.

PAGE 132

132 following the issuance of neutral guidance, which by nature contains no new information, this would further support the sentiment based theory of comovement. Table s 4 11 and 4 12 report estimates from multivar iate regressions disaggregated by CIG Description Code for the LD and ID samples, respectively. I report results only for high sentiment period speci fications. Focusing first on the LD sample in Table 4 11 consistent with the sentiment based theory, I fin d a shift in return comovement with the sentiment portfolio concentrated in firm disclosures that provide no new information about firm fundamentals. In particular, I SENT = 0.02 24, p value = 0.076) following the issuance of neutral earnings guidance. There is a MKT = 0.0652, p value = 0.076) following the event as well. In contrast, I do not observe any change in comovement wit h the sentiment portfolio or market portfolio subsequent to a positive earnings surprise. This indicates that when information is communicated, these firms do not tend to comove more with those stocks that are highly sensitive to sentiment. In Table 4 12, I observe even stronger results when considering return comovement around the issuance of neutral ea rnings guidance for firms in the ID sample. Consistent with the hypothesis that these firms are more likely to use disclosure to take advantage of optimist ic SENT is over three times as large as the change reported in Table 4 11 and is sta tistically significant at the 5 % SENT = 0.0785, p value = 0.043). Again, I find a corresponding decrease in comovement with the ma MKT = 0.2084, p value = 0.035) following the disclosure. Just as in the LD sample, I do not find any evidence of a shift in comovement when information is communicated through a positive earnings surprise.

PAGE 133

133 Additional Robustness Checks Tw o remaining issues that I address in further robustness checks involve the manner in which I construct the two main samples of firms that voluntarily provide earnings guidance The LD sample consists of firms that have not provided voluntary earnings guida nce in the 12 month s prior to the event. T his particular criterion was chosen in order to mitigate the well consistent with Christensen et trict the sample of firms in a similar way to their study and include only those firms that have not provided guidance over the 2 years prior to the event and then test whether results remain consistent using this alternate classification. Table 4 13 repor ts estimates from mu ltivariate regressions using the restricted LD sample (LD2). Utilizing the LD2 sample, I find even stronger results than originally reported in Table 4 7 First of all, the increase in comovement with the sentiment portfolio and decreas e in comovement with the market portfolio following the disclosure is not only strongly significant in the high sentiment period estimations, but also in the regressions that span the entire sample period. Furthermore, in the high sentiment results, the ch SENT is approximately one and a half times that of the original estimate using the LD sample and is significant at the 5% SENT = 0.0245, p value = 0.018). Overall, results using the LD2 sample are stronger than estimates originally report ed in Table 4 7. The second issue that is address ed ce nters on the construction of the ID sample. Recall, the ID sample consists of firms that disclose during only one year of the sample period and do not disclose in future years. The main concern with th is group of f irms is the possibility that this sample suffers from a look ahead bias. Therefore, I utilize an alternative approach in which I identify only the first time a firm voluntarily pro vides earnings guidance over the entire sample period, regardle ss of whether they disclose again in the future. Table 4 14 reports estimates from

PAGE 134

134 multivariate regressions using the first disclosure sample. I observe similar results to those reported in Table 4 8 using the ID sample. Therefore, I conclude that the issu e of look ahead bias is of secon dary importance in affecting the empirical results of this study Summary and Conclusion Whether the use of voluntary earnings guidance is in fact beneficial to firm shareholders is a debate that is far from over. While the initial motivation for providing guidance may have been to increase firm transparency and provide shareholders with a better understanding of a facilitate hig her asset valuations, particularly at times when investors are overly optimistic. If firms issue earnings guidance that is consistent with misguided investor beliefs, voluntary disclosure may contribute to sentiment induced mispricing, rather than aid in r esolving sentiment driven overvaluation. This study adds to the growing literature that examines whether the strategic disclosure of earnings guidance caters to changes in investor expectations. The main contribution of this analysis focuses on determining whether firm management strategically issues voluntary earnings guidance in order to take advantage of misguided investor beliefs, particularly during periods of high investor optimism, and whether such disclosures have a significant impact on future asse t price movements. In particular, I test whether firms that strategically disclose during high sentiment periods increase their comovement with stocks that are highly sensitive to shifts in investor sentiment (i.e. those firms that experience significant i ncreases in their stock prices when investor sentiment rises) following the disclosure announcement. In so doing, this analysis is the first to study sentiment based return comovement using voluntary disclosure events as a testing ground. Overall, I bridge the gap between the voluntary disclosure, investor sentiment and return comovement literatures.

PAGE 135

135 Using bivariate and mu ltivariate regression models, I provide new evidence, contrary to Seybert and Yang (2011), supporting the hypothesis that strategically i ssued earnings guidance contributes to sentiment induc ed mispricing. In particular, I find that firms issuing voluntary earnings guidance during high sentiment periods experience a significant increa se in return comovement with the sentiment portfolio and a significant decrease with market fundamentals following the d isclosure event. Furthermore, I provide evidence that this effect is concentrated around the issuance of neutral earnings guidance, which by nature contains no new information about undamental value. This result supports the sentiment based theory of comovement as originally examined in Barberis, Shleifer, and Wurgler (2005).

PAGE 136

136 A B Figure 4 1. Investor sentiment and voluntary disclosure. A) This figure plots Baker and 006, 2007) investor sentiment index and the number of firms providing sample consists of those firms that have not provided voluntary earnings guidance in the 12 months prior to t (2006,2007) investor sentiment index and the number of firms providing voluntary of those firms that disclose during only one year of the sample period. The sample period spans from January 1998 to December 2008.

PAGE 137

137 Table 4 1. Descriptive statistics : voluntary disclosure Year Mean Median St Dev Min Max N All Firms 1998 1.583 1 1.157 1 15 489 1999 1.788 1 1.51 1 1 24 676 2000 1.893 1 1.452 1 12 617 2001 2.501 2 2.017 1 25 1147 2002 3.137 2 2.620 1 31 1150 2003 3.209 3 2.579 1 33 1162 2004 3.369 3 2.637 1 36 1274 2005 3.523 3 2.469 1 24 1132 2006 3.545 3 2.500 1 25 1195 2007 3.415 3 2.276 1 24 116 4 2008 3.677 3 2.281 1 20 1018 Limited Disclosure (LD) 1998 1.621 1 1.017 1 7 449 1999 1.726 1 1.387 1 16 442 2000 1.668 1 1.076 1 8 328 2001 2.718 2 2.238 1 25 365 2002 2.900 2 2.238 1 17 680 2003 2.327 2 1.990 1 16 260 2004 2.118 2 1.603 1 13 289 2005 2.122 2 1.432 1 9 139 2006 2.163 2 1.488 1 9 208 2007 2.149 2 1.372 1 7 148 2008 2.596 2 2.022 1 18 104 Isolated Disclosure (ID) 1998 1.461 1 0.913 1 7 152 1999 1.518 1 0.947 1 7 199 2000 1.390 1 0.709 1 5 100 2001 2.071 2 1.438 1 8 84 2002 1.970 1 1.609 1 12 101 2003 1.517 1 1.000 1 6 60 2004 1.553 1 1.006 1 6 85 2005 1.550 1 0.986 1 4 40 2006 2.070 2 1.498 1 9 57 2007 1.678 1 0.937 1 4 59 2008 2.528 2 1.289 1 6 72 This table reports descriptive statistics regarding the frequency of disclosure for all firms in the disclosure (ID) sample for each year of the sample period. The focus is on long horizon earnings estimates defined as forecasts that were issued more than 90 days before the estimate period end date. Mean, median, standard deviation, minimum, maximum, and the number of firm observations for each year are reported. The sample period spans from January 1998 to December 2008.

PAGE 138

138 Table 4 2. I ndustry classifications : voluntary disclosure and sentiment portfolios Drugs Oil Software Telecom Chips Util. Bus. Services Retail Wholesale Machine Sent. Port 9.921 5.186 10.665 2.701 8.972 1.517 5.732 4.198 4.268 3.050 LD Firms 3.197 1.896 10.884 1.245 7.048 2.193 7.987 5.889 6.293 2.980 ID Firms 3.514 3.268 11.717 3.107 9.039 1.818 8.306 4.721 5.687 3.359 All CIG 3.251 1.489 9.554 1.104 5.885 3.083 7.434 8.311 5.802 3.939 This table reports descriptive statist ics pertaining to the industry characteristics of firms in the sentiment portfolio, limited disclosure sample, isolated disclosure sample, and all firms in the Thomson First Call CIG database. A sentiment beta approach similar to Glushkov (2006) is used to identify those firms that are most sensitive to changes in investor sentiment and a value weighted portfolio is constructed for each month in the sample period. T he specification is as follows: where R i,t is the exces s return of stock i at time t, BW t is Baker and Wurgler s (2006, 2007) stock market sentiment index MKT t is the excess return on the value weighted market portfolio, SMB t is the return on the small minus big size portfolio, HML t is the ret urn on the high min us low book to market portfolio MOM t is the return on the momentum portfolio and LIQ t is the liquidity factor The limited disclosure sample consists of those firms that have not provided voluntary earning s guidance in the 12 months prior to the disclosure event. The isolated disclosure sample consists of those firms that disclose during only one year of the sample period. Industry classifications are based upon Fama and French 49 industry classifications. Only the top ten industry distributions are reported. Industry classifications are obtained from Kenneth January 1998 to December 2008.

PAGE 139

139 Table 4 3. Size classif ications : voluntary disclosure and sentiment portfolios Decile 1 Decile 2 Decile 3 Decile 4 Decile 5 Decile 6 Decile 7 Decile 8 Decile 9 Decile 10 Sent. Port 49.060 14.509 7.881 5.802 4.345 3.632 3.437 3.285 3.210 4.839 LD Firms 27.812 14.628 11.952 9. 199 7.712 7.864 5.433 5.286 5.534 4.580 ID Firms 43.548 15.340 8.552 9.108 7.170 5.079 3.638 2.809 2.646 2.111 All CIG 21.410 13.524 11.533 9.465 8.825 8.519 6.454 6.654 7.013 6.601 This table reports descriptive statistics pertaining to the size charac teristics of firms in the sentiment portfolio, limited disclosure sample, isolated disclosure sample, and all firms in the Thomson First Call CIG database. A sentiment beta approach similar to Glushkov (2006) is used to identify those firms that are most s ensitive to changes in investor sentiment and a value weighted portfolio is constructed for each month in the sample period T he specification is as follows: where R i,t is the exces s return of stock i at time t, BW t is Baker and Wurgler s (2006, 2007) stock market sentiment index MKT t is the excess return on the value weighted market portfolio, SMB t is the return on the small minus big size portfolio, HML t is the ret urn on the high min us low book to market portfolio MOM t is the return on the momentum portfolio and LIQ t is the liquidity factor The limited disclosure sample consists of those firms that have not provided voluntary earnings guidance in the 12 months prior to the disclosure event. The isolated disclosure sample consists of those firms that disclose during only one year of the sample period. Size categories are deciles (from smallest to largest) based upon market capitalization breakpoints at the end of each mon website. All statistics are reported in percentage form. The sample period spans from January 1998 to December 2008.

PAGE 140

140 Table 4 4. Portfolio daily returns : voluntary disclosure and sentiment portf olios Mean Median St Dev Min Max Serial Correlation Sent. Port 0.092 0.132 2.381 11.289 21.885 0.03 LD Firms 0.065 0.057 1.652 12.651 10.884 0.01 ID Firms 0.053 0.008 2.125 15.891 20.769 0.04 All CIG 0.062 0.049 1.630 17 .892 9.697 0.01 This table reports descriptive statistics pertaining to the daily returns of value weighted portfolios constructed for firms in the sentiment portfolio, limited disclosure sample, isolated disclosure sample, and all firms in the Th omson First Call CIG database for each month in the sample period. A sentiment beta approach similar to Glushkov (2006) is used to identify those firms that are most sensitive to changes in investor sentiment and a value weighted portfolio is constructed f or e ach month in the sample period. T he specification is as follows: where R i,t is the exces s return of stock i at time t, BW t is Baker and Wurgler s (2006, 2007) stock market sentiment index MKT t is the excess return on the value weighted market portfolio, SMB t is the return on the small minus big size portfolio, HML t is the ret urn on the high min us low book to market portfolio MOM t is the return on the momentum portfolio and LIQ t is the liquidity factor The limited disclosure sample consists of those firms that have not provided voluntary earnings guidance in the 12 months prior to the disclosure event. The isolated disclosure sample consists of those firms that disc los e during only one year of the sample period. All statistics are reported in percentage form. Mean, median, standard deviation, minimum, maximum, and serial correlation are reported. The sample period spans from January 1998 to December 2008.

PAGE 141

141 Table 4 5 B ivariate regression results: limited disclosure. Period SENT MKT R 2 N Full Sample Period 0.013 0.018 0.018 3420 (0.094) (0.185) (0.000) High Sentiment 0.019 ** 0.053 *** 0.026 2461 (0.037) (0.001) (0.000) Low Sentiment 0.003 0.073 *** 0.002 959 (0.797) (0.003) (0.526) This table pr esents results obtained from estimating bivariate regression models for the limited disclosure (LD) sample. The LD sample consists of those firms that have not provided voluntary earnings guidance in the 12 months prior to the disclosure event. The bivaria te specification can be represented as follows: where R i,t is the excess return of stock i at time t, SENT i,t is the excess return on the value weighted sentiment portfolio and MKT i,t is the excess return on the value weighted market portfolio. Bivaria te regressions are estimated for each firm using daily returns over the 12 month period ending one month prior to the disclosure and the 12 month period staring one month after the disclosure. Cross sectional means of the changes in the slope coefficients, SENT MKT and their p values over the full sample period, during high sentiment periods, and during low sentiment periods are reported. The sample period spans from January 1998 to December 2008 ***, **, and represent 1%, 5%, and 10% significan ce levels respectively.

PAGE 142

142 Table 4 6. Bivariate regression results: isolated disclosure. Period SENT MKT R 2 N Full Sample Period 0.034 0.090 ** 0.006 1003 (0.094) (0.012) (0 .082) High Sentiment 0.044 0.132 *** 0.006 752 (0.068) (0.002) (0.121) Low Sentiment 0.005 0.036 0.005 251 (0.898) (0.573) (0.431) This table presents results obtained from estimating bivariate regression models for the isolated disclosure (ID) sample. The ID sample consists of those firms that disclose during only one year of the sample period. The bivariate specificatio n can be represented as follows: where R i,t is the excess return of stock i at time t, SENT i,t is the excess return on the value weighted sentiment portfolio and MKT i,t is the excess return on the value weighted market portfolio. Bivariate regressions are estimated for each firm using daily returns over the 12 month period ending one month prior to the disclosure and the 12 month period staring one month afte r the disclosure. Cross SENT MKT and their p values over the full sample period, during high sentiment periods, and during low sentiment periods are reported. The sample period spans from J anuary 1998 to December 2008 ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 143

143 Table 4 7. Multivariate regression results: limited disclosure. Period SENT MKT SIZE IND R 2 N Full Sample Period 0.012 0.052 ** 0.031 0.023 0.020 3420 (0.114) (0.028) (0.433) (0.138) (0.000) High Sentiment 0.016 0.049 0.056 0.014 0.028 2461 (0.065) (0.083) (0.303) (0.454) (0.000) Low Sentiment 0.001 0.059 0.032 0.049 0.001 959 (0.961) (0.170) (0.251) (0.129) (0.812) This table presents results obtained from estimating multivariate regression mode ls for the limited disclosure (LD) sample. The LD sample consists of those firms that have not provided voluntary earnings guidance in the 12 months prior to the disclosure event. The multivariate specification can be represented as follows: where Ri,t is the excess return of stock i at time t, SENTi,t is the excess return on the value weighted sentiment portfolio, MKTi,t is the excess return on the value weighted market portfolio, SIZEi,t is the value weighted size index return (using size deciles), an d INDi,t is the value weighted industry return (using the Fama and French 49 industry classification). Multivariate regressions are estimated for each firm using daily returns over the 12 month period ending one month prior to the disclosure and the 12 mon th period staring one month af ter the disclosure The cross sectional means of t values over the full sample period, during high sentiment periods, and during low sentiment periods are reported. The sample period spans from January 1998 to December 2008 ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 144

144 Table 4 8. Multivariate regression results: isolated disclosure. Period SENT MKT SIZE IND R 2 N Full Sample Period 0.048 ** 0.046 0.00 4 0.059 0.003 1003 (0.029) (0.487) (0.948) (0.186) (0.505) High Sentiment 0.056 ** 0.062 0.005 0.056 0.003 752 (0.023) (0.441) (0.944) (0.247) (0.528) Low S entiment 0.025 0.002 0.000 0.068 0.002 251 (0.599) (0.982) (0.998) (0.515) (0.821) This table presents results obtained from estimating multivariate regression models for the isolated disclosure (I D) sample. The ID sample consists of those firms that disclose during only one year of the sample period. The multivariate specification can be represented as follows: where R i,t is the excess return of stock i at time t, SENT i,t is the excess return o n the value weighted sentiment portfolio, MKT i,t is the excess return on the value weighted market portfolio, SIZE i,t is the value weighted size index return (using size deciles), and IND i,t is the value weighted industry return (using the Fama and French 49 industry classification). Multivariate regressions are estimated for each firm using daily returns over the 12 month period ending one month prior to the disclosure and the 12 month period staring one month after the disclosure. The cross sectional mean SENT MKT SIZE and IND and their p values over the full sample period, during high sentiment periods, and during low sentiment periods are reported. The sample period spans from January 1998 to Decembe r 2008. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 145

145 Table 4 9. Bivariate regression results: matched sample. Period SENT MKT R 2 N Full Sample Period 0.002 0.069 0.011 *** 1003 (0.950) (0.452) (0.003) High Sentiment 0.018 0.074 0.014 *** 752 (0.647) (0.217) (0.001) Low Sentiment 0.049 0.170 0.002 251 (0.451) (0.082) (0.783) This table presents results obtained from estimating a bivariate regression model for a matched sample for firms. For each firm that voluntarily issues earnings guid ance in the isolated disclosure (ID) sample, I select a control firm from the same industry and same size decile that does not disclose at any point during the sample period. The bivariate specification can be represented as follows: where R i,t is the excess return of stock i at time t, SENT i,t is the excess return on the value weighted sentiment portfolio and MKT i,t is the excess return on the value weighted market portfolio. Regressions are estimated for each firm using daily returns over the 12 month period month period staring one month sectional means of the changes in the slope coefficients, SENT and MKT and their p values over the full sample p eriod, during high sentiment periods, and during low sentiment periods are reported. The sample period spans from January 1998 to December 2008. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 146

146 Table 4 10. Mul tivariate regression results: matched sample. Period SENT MKT SIZE IND R 2 N Full Sample Period 0.017 0.039 0.036 0.017 0.014 1003 (0.441) (0.649) (0.354) (0.810) (0.002) High Sentiment 0.023 0.130 0.03 3 0.073 0.016 752 (0.375) (0.171) (0.487) (0.310) (0.002) Low Sentiment 0.002 0.255 0.049 0.167 0.006 251 (0.952) (0.200) (0.474) ( 0.366) (0.492) This table presents results obtained from estimating a multivariate regression model for a matched sample of firms. For each firm that voluntarily issues earnings guidance in the isolated disclosure (ID) sample, I select a control fir m from the same industry and same size decile that does not disclose at any point during the sample period. The multivariate specification can be represented as follows: where R i,t is the excess return of stock i at time t, SENT i, t is the excess return on the value weighted sentiment portfolio, MKT i,t is the excess return on the value weighted market portfolio, SIZE i,t is the value weighted size index return (using size deciles), and IND i,t is the value weighted industry return (using the Fama and Frenc h 49 industry classification). Regressions are estimated for each firm using daily returns over the 12 month period ending one month prior to disclosure. Cross sectional means of t SENT MKT SIZE IND and their p values over the full sample period, during high sentiment periods, and during low sentiment periods are reported. The sample period spans from January 1998 to December 2008 ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 147

147 Table 4 11. Multivariate regression results by disclosure type: limited disclosure. Disclosure Type SENT MKT SIZE IND R 2 N Neutral 0.022 0.065 0.041 0.025 0.034 1169 (0.076) (0.076) (0.318) (0.269) (0.000) Positive Surprise 0.006 0.066 0.044 0.028 0.070 348 (0.782) (0.324) (0.377) (0.539) (0.000) This table presents results obtained from estimating multivariate regre ssion models by disclosure type for the limited disclosure (LD) sample. The LD sample consists of those firms that have not provided voluntary earn ings guidance in the 12 months prior to the disclosure event. Disclosures guidance is greater than the consensus analyst estimate, then this disclosure is classified a s a where R i,t is the excess return of stock i at time t, SENT i ,t is t he excess return on the value weighted sentiment portfolio, MKT i,t is the excess return on the value weighted market portfolio, SIZE i,t is the value weighted size index return (using size deciles), and IND i,t is the value weighted industry return (using th e Fama and French 49 industry classification). Regressions are estimated for each firm using daily returns over the 12 month period ending one month prior to disclosure. Cross SENT MKT SIZE IND and their p values only for high sentiment periods are reported. The sample period spans from January 1998 to December 2008. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 148

148 Table 4 12. Multivariate regression results by disclosure type: isolated disclosure. Disclosure Type SENT MKT SIZE IND R 2 N Neutral 0.078 ** 0.208 ** 0.099 0 .072 0.011 295 (0.043) (0.035) (0.477) (0.260) (0.142) Positive Surprise 0.005 0.082 0.096 0.174 0.047 109 (0.943) (0.775) (0.503) (0.206) (0. 010) This table presents results obtained from estimating multivariate regression models by disclosure type for the isolated disclosure (ID) sample. The ID sample consists of those firms that disclose during only one year of the sample period. Disclosur CIG Description Code variable. If the earnings guidance is greater than the consensus analyst consensus analy can be represented as follows: where R i,t is the excess return of stock i at time t, SENT i, t is the excess return on the value weighted sentiment portfolio, MKT i,t is t he excess return on the value weighted market portfolio, SIZE i,t is the value weighted size index return (using size deciles), and IND i,t is the value weighted industry return (using the Fama and French 49 industry classification). Regressions are estimate d for each firm using daily returns over the 12 month period ending one month prior to disclosure. Cross sectional means of the changes in the slope coefficient SENT MKT SIZE IND and their p values only for high sentiment periods are reported. The sample period spans from January 1998 to December 2008. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 149

149 Table 4 1 3 Multivariate regression results for robustness checks: restricted limited disclosure. Period SENT MKT SIZE IND R 2 N Full Sample Period 0.023 ** 0.059 ** 0.024 0.016 0.020 2646 (0.012) (0.034) (0.625) (0.389) (0.000) High Sentiment 0.024 ** 0.074 0.044 0.014 0.026 1977 (0.018) (0.057) (0.492) (0.507) (0.000) Low Sentiment 0.017 0.050 0.037 0.023 0.002 669 (0.353) (0.350) (0.299) (0.577) (0.644) This table presents results obtained from estimating multivariate regression models for the restricted limited disclosure sample (LD2). The LD2 sample consists of those firms that have not provided guidance over the two years prior to the disclosure event. The multivariate specification can be represented as follows: where R i,t is the excess return of stock i at time t, SENT i, t is the excess return on the value weighted sentiment portfolio, MKT i,t is the excess return on the value weighted market portfolio, SIZE i,t is the value weighted size index return (using size deciles), and IND i,t is the value weighted industry retu rn (using the Fama and French 49 industry classification). Regressions are estimated for each firm using daily returns over the 12 month period ending one month prior to osure. Cross SENT MKT SIZE IND and their p values over the full sample period, during high sentiment periods, and during low sentiment periods are reported. The sample period spans from Jan uary 1998 to December 2008. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 150

150 Table 4 14. Multivariate regression results for robustness checks: first disclosure. Period SENT MKT SIZE IND R 2 N Full Sa mple Period 0.022 ** 0.051 0.007 0.001 0.012 2415 (0.021) (0.085) (0.790) (0.947) (0.000) High Sentiment 0.027 ** 0.048 0.002 0.010 0.018 1877 (0.014) (0.154) (0.944) (0.632) (0.000) Low Sentiment 0.007 0.061 0.038 0.031 0.008 538 (0.735) (0.323) (0.345) (0.528) (0.085) This table presents results ob tained from estimating multivariate regression models for the first disclosure sample. The first disclosure sample keeps only the first time a firm voluntarily provides earnings guidance over the entire sample period, regardless of whether they disclose ag ain in the future. The multivariate specification can be represented as follows: where R i,t is the excess return of stock i at time t, SENT i, t is the excess return on the value weighted sentiment portfolio, MKT i,t is the excess return on the value weig hted market portfolio, SIZE i,t is the value weighted size index return (using size deciles), and IND i,t is the value weighted industry return (using the Fama and French 49 industry classification). Regressions are estimated for each firm using daily return s over t he 12 month period ending one month prior to Cross SENT MKT SIZE IND and their p values o ver the full sample period, during high sentiment periods, and during low sentiment periods are reported. The sample period spans from January 1998 to December 2008. ***, **, and represent 1%, 5%, and 10% significance levels respectively.

PAGE 151

151 CHAPTER 5 CONCLUSION Efficient financial markets assume that security prices represent fair valuations by rational investors, explicitly incorporate the perceived debt and equity financing costs of the marginal investor, and reflect all available public informat ion. However, over the past decade, the rapid rise of asset prices in the technology sector and real estate market, liquidity crises such as those of Long Term Capital Management and Lehman Brothers, and the use of misleading information at the heart of ac counting scandals such as those of Enron and WorldCom, have weakened support among economists for asset pricing models based on the assumptions of efficient markets, rational expectations, and costless arbitrage. In each scenario, factors outside of tradit ional fundamental determinants of value had a significant impact on asset price movements. During periods of asset pricing booms, funds flow freely to speculative investors, thereby providing them with the means to bid asset prices above their fundamental value. As financial markets become less liquid in times of economic distress, investors are limited in their ability to enter the market and may even be forced to sell assets in order to meet outstanding funding requirements, thereby causing asset prices to decline below what market fundamentals dictate. When managers utilize information to mislead investors, this information may be incorporated into asset values even if other fundamental determinants of value remain unchanged. Put simply, asset prices ar e sensitive to market frictions and behavioral influences that may, at times, lead asset prices away from their fundamental value. Testing the impact of these factors and identifying the market conditions in which they matter most provides a significant co ntribution to the field of financial research. In three studies, I examine the impact that behavioral factors, market frictions, and information dissemination have on asset price movements in both private and public markets. In

PAGE 152

152 particular, I study the rol es that investor sentiment, credit availability and voluntary disclosure of earnings guidance play in driving asset prices away from their fundamental values. In my first study, I examine the differential impact of investor sentiment on both short and lo ng run returns in public and private commercial real estate markets. Theory predicts that t he illiquidity, information asymmetries, more limited price revelation and inability to short sell inherent in private markets may lead to prolonged periods of sent iment induced mispricing I find that investor sentiment plays a more persistent role in pushing asset prices away from their fundamental values in private markets because of the increased ill iquidity and limits to arbitrage This study is the first to dir ectly investigate the relative importance of sentiment in public and private asset markets. In my second study, I examine the relation between credit availability market liquidity and asset price movements in both private and public commercial real estat e markets. Given the relative illiquidity and significant use of leverage in acquisitions within commercial real estate markets, theory predicts that funding constraints are likely to play a significant role in asset price determination. I find asset value s of highly leveraged assets, such as commercial real estate, to b e sensitive to the availability of credit, even after controlling for other fundamental determinants of value. In addition, I provide evidence that assets trading in illiquid segments of the commercial real estate market are highly susceptible to a spiral effect, in which changes in asset prices lead to further changes in the availability of credit, thus magnifying the overall price impact of a funding shock. These results suggest that while leverage plays a significant role in determining credit market availability pricing effects, the underlying liquidity with which these assets trade is a key factor in determining the likelihoo d of a liquidity spiral, with lower liquidity creating the marke t setting for a spiral effect.

PAGE 153

153 In my third study, I examine whether firm management strategically issues voluntary earnings guidance in order to take advantage of misguided investor beliefs, particularly during periods of high investor optimism, and wheth er such disclosures have a significant impact on future asset price movements. While the initial motivation for providing voluntary earnings guidance may have been to increase firm transparency and provide shareholders with a better understanding of a comp this opportunity to facilitate higher asset valuations, particularly at times when investors are overly optimistic. I find that firms issuing voluntary earnings guidance during perio ds of investor confidence covary more with firms that are most sensitive to changes in investor sentiment and less with market fundamentals following the d isclosure event. Furthermore, I provide evidence that this effect is concentrated around the issuance of neutral earnings guidance, which by nature contains no new information about result directly supports the sentiment based theory of comovement Taken together, these three studies provide evidence that behavioral facto rs, such as investor sentiment, market frictions, such as a la ck of liquidity or changes in the credit supply, and the strategic issuance of value relevant information, such as the voluntary disclosure of earnings guidance, can have significant impacts on asset price movements. Furtherm ore, my results indicate that certain market conditions, such as significant limits to arbitrage, funding constraints in relatively illiquid markets, or periods of investor optimism may act as a catalyst for non fundamental b ased pricing effects.

PAGE 154

154 APPENDIX BOOTSTRAPPING WITH LONG HORIZON REGRESSIONS Bootstrap Simulation Procedure Inference based on asymptotic theory can be extremely misleading with small sample sizes, where many asymptotic tests may over reject or under rejec t the null hypothesis depending on the particular sample size. Several recent studies have utilized bootstrap simulation procedures to produce more accurate coefficient estimates and more reliable measures of inference within small sample settings. Instead of imposing a restrictive shape on the sampling distribution of desired parameter estimates, the bootstrap empirically estimates the sampling distribution from the original sample by sampling repeatedly with replacement from the actual data. I follow the framework of the bootstrap simulation procedure described in Brown and Cliff (2005), but also incorporate several additional adjustments proposed in the more recent literature. 1 First, I begin by running long horiz on regressions as specified in Equation 2 2 The original OLS coefficient estimates on measures of investor sentiment are saved, as they will be utilized to calculate bias adjusted coefficient estimates. The second step in the simulation procedure is to generate a pseudo return series under the n ull hypothesis that sentiment does not matter. I utilize the following vector autoregressive (VAR) model as the underlying data generating process of the pseudo return series: VAR(1) for y t = [r t S t z t ], ( A 1 ) where r t is the contemporaneous log re turn (i.e., the set of returns that was the basis for calculating the original future k period return series), z t is the set of contemporaneous control 1 See also Menkhoff and Rebitzky (2008), Schmeling (2007), and Schmeling and Schrimpf (2008).

PAGE 155

155 variables, and S t is the measure of investor sentiment. 2 The beta coefficient on investor sentiment is s et to zero in this case to ensure that the pseudo return series is generated under the null hypothesis that sentiment does not affect returns and the constant in the constrained model is adjusted to restore the original mean. I save the predicted y values and the residuals from the VAR specification. Before proceeding, the residuals must first be adjusted to correct for a downward bias that results from the use of Least Squares in the VAR framework. Thus, each residual is multiplied by (n/n v) 1/2 where n is the number of observations and v refers to the degrees of freedom of the VAR (MacKinnon, 2002). I then sample with replacement from these residuals to generate a new set of bootstrapped residuals. This sampling process is repeated 10,100 times, with t he first 100 samples of bootstrapped residuals being discarded to avoid any startup effects. I use the remaining residual datasets to create 10,000 bootstrapped dependent variables by adding the predicted y value from the VAR to the new residual. From this pseudo return series, I create 10,000 new sets of k period future returns. I then run long horizon retu rn regressions as specified in Equation 2 2, except in this case I use the new k period future return s as the dependent variables. I save the beta coeff icients from each of the regressions, yielding 10,000 betas for each of the return horizons. I then calculate the mean beta coefficient and the standard deviation of the estimated values of beta (bootstrap standard error) across the 10,000 simulations. The se will be used to create a bias adjusted beta coefficient and a new empirical distribution of t statistics for inference. MacKinnon and Smith (1998) demonstrate how bootstrap simulations can be utilized to generate more accurate beta coefficient estimate s within finite sample settings. When the bias 2 SBIC model selection criteria indicates that a VAR (1) specification is appropriate.

PAGE 156

156 function is not known analytically, the authors hypothesize that it can be estimated through the bootstrap simulation procedure as previously discussed. Assuming the bias is constant for a particular return ho rizon, the estimated bias can be calculated as follows: Estimated Bias = ( A 2 ) where is the original OLS beta coefficient and is the sample mean of the 10,000 simulated beta coefficients. Because the si mulated samples are assumed to be drawn from the same model as the original data, the estimated bias function should converge to the actual bias function as the number of simulations approaches infinity. The adjusted beta coefficient can therefore be speci fied as: ( A 3 ) This is the mathematical equivalent of subtracting the estimated bias from the original OLS coefficient estimate. This bias adjusted estimator has been used extensively in the bootstrap literature and has been shown to provide more reliable coefficient estimates in numerous simulation based studies. measures for hypothesis testing and more appropriate confidence intervals for inference. I n fact, the bootstrap procedure was originally proposed as an alternative method to compute standard errors (Efron, 1979). Prior literature has documented that Newey West standard errors perform poorly in finite samples with overlapping observations. In pa rticular, research has shown that standard errors calculated in this fashion suffer from a significant downward bias in small samples (Andrews, 1991). This implies that test statistics based on these estimated error terms are likely to result in false infe rences. When asymptotic standard errors are unreliable, the

PAGE 157

157 simplest alternative is to utilize the bootstrap standard error, which is calculated using the following specification: (A 4 ) where is the estimated beta coeffi cient from iteration i of the simulation process, is the mean beta of the 10,000 simulated beta coefficients, and B is the number of simulations. In other words, se( ) is simply the standard deviation of the estimated values of be ta across the 10,000 simulations. The new bias adjusted t statistic can be calculated as follows: ( A 5 ) where the numerator is the bias adjusted beta coefficient and the denominator is the bootstrap standard error. Because t stat istics constructed in this fashion do not always follow the t distribution in finite samples, standard critical values may not be appropriate for inference. Therefore, I utilize two alternative approaches to compute p values based on a bootstrap distributi on of t statistics to yield a more accurate test of the null hypothesis. The first step in both approaches is to calculate an adjusted t statistic for each of the 10,000 simulations, thus creating a new empirical distribution of t statistics. This test s tatistic is calculated as follows: ( A 6 ) where is the estimated beta coefficient for iteration i of the simulation procedure, is the mean beta coefficient across the 10,000 simulations, and se( ) is the bootstrap standard error. An appealing aspect of this technique is that I am able to construct an empirical distribution of test statistics that adheres closely to the normal distribution, yet provides new critical values for inference.

PAGE 158

158 The fir st approach for calculating a bootstrap p value can be depicted as follows: ( A 7 ) where is the absolute value of the simulated t statistic of iteration i t hat is specified in Equation A 6 is the absolute value of the bias ad justed t statistic detailed in Equation A 5 and B is the number of simulations. This approach implicitly assumes the distribution of the test statistic is symmetric around zero. Because the empirical distribution of test statistics m ay not always be entirely symmetric around zero, I compute an alternative p value as follows: ( A 8 ) where is the simul ated t statistic calculated in Equation A 6 is the bias adju sted t statistic calculated in Equation A 5 and B is the number of simulations. Unlike the previous specification, this technique does not assume the empirical distribution of test statistics is symmetric around zero. MacKinnon (2006) points out that these two approaches could very well lead to significantly different results if the mean value of the simulated t statistics is vastly different from zero. However, I find that these two approaches yield essentially identical p values, thus implying that the empirical distributions of t est statistics in the present study are virtually symmetric around zero. Reported p values are calculated based on the f irst specification detailed in Equation A 7 Impact of Bootstrap Simulation Adjustments Figure A 1 provides a graphical illustration of the bootstrap simulation adjustment. Panel A displays the empirical distributions of unadjusted and adjusted t statistics. The solid line represents the bootstrap adjusted t statist ics, as depicted in Equation A 6 while the dashed line represents the unad justed t statistic, which is calculated using the OLS beta coefficient and

PAGE 159

159 Newey West standard error. In this case, the unadjusted test statistic distribution has thick tails. Thus, standard inference would result in a rejection of the null hypothesis that sentiment does not affect long horizon returns more often than is actually the case. For example, the empirical distribution of the unadjusted t statistics falls below the critical value of 1.96 19.62% of the time. The adjusted t statistics, on the other hand, more closely resemble a normal distribution. In fact, 2.45% of the empirical distribution of simulation adjusted t statistics falls below the critical value of 1.96 in the particular example shown. Panel B displays the empirical distribution of un adjusted and adjusted coefficient estimates from the bootstrap simulation. The solid line represents bootstrap adjusted coefficients, while the dashed line represents OLS coefficient estimates. The distribution of unbiased coefficient estimates, under the null hypothesis that investor sentiment does not affect long horizon returns, should be centered on zero. However, the unadjusted coefficients appear to be biased, as the empirical distribution is centered slightly to the right of zero, with a median value of 0.0005. Therefore, bias adjusted coefficient estimates are reported and are used to calculate the adjusted test statistics.

PAGE 160

160 A B Figure A 1. Empirical distribution of bootstrap test statistics and coefficient estimates. A) This figure provides a graphical illustration of the impact of the bootstrap simulation adjustment on the test statistics used for inference. In particular, this figure displays the distributions of unadjusted and adjusted t statistics. The solid line represents bootstrap adjust ed t statistics, while the dashed line represents unadjusted t statistics, which are calculated using the unadjusted OLS coefficient estimate and Newey West standard error. B) This figure provides a graphical illustration of the impact of the bootstrap sim ulation adjustment on the coefficient estimates from the long horizon regressions. In particular, this figure displays the empirical distribution of unadjusted and adjusted coefficient estimates for the measure of investor sentiment. The solid line represe nts the bootstrap adjusted coefficients, while the dashed line represents the unadjusted OLS coefficient estimates. This example utilizes returns in public commercial real estate markets, REITRET, at the four year horizon as the dependent variable and the direct measure of investor sentiment, DRES, as the independent variable.

PAGE 161

161 LIST OF REFERENCES Adrian, T and H Shin. 2010. Liquidity and Leverage. Journal of Financial Intermediation 19 (3): 418 437. Amihud, Y 2002. Illiquidity and Stock Returns: Cross Section and Time Series Effects. Journal of Financial Markets 5(1): 31 56. Andrews, D W. 1991. Heteroskedasticity and Autocorrelation Consistent Covariance Matrix Estimation. Econometrica 59(3): 817 858. Anilowski, C., M. Feng, and D. Skinner. 2007. Do es Earnings Guidance Affect Market Returns? The Nature and Information Content of Aggregate Earnings Guidance. Journal of Accounting and Economics 44: 36 63. Baker, M. and J. Stein. 2004. Market Liquidity as a Sentiment Indicator. Journal of Financial Ma rkets 7: 271 299. Baker, M. and J. Wurgler. 2000. The Equity Share in New Issues and Aggregate Stock Returns. Journal of Finance 55(5): 2219 2257. Baker, M. and J. Wurgler. 2004. A Catering Theory of Dividends. Journal of Finance 59(3): 1125 1165. B aker, M. and J. Wurgler. 2006. Investor Sentiment and the Cross Section of Stock Returns. Journal of Finance 61(4): 1645 1680. Baker, M. and J. Wurgler. 2007. Investor Sentiment in the Stock Market. The Journal of Economic Perspectives 21 (2): 129 151. Baker, M., J. Wurgler, and Y. Yuan. 2009. Global, Local, and Contagious Investor Sentiment. Working Paper, Harvard University. Barinov, A 2010. Turnover: Liquidity or Uncertainty? Working Paper, University of Georgia. Barber, B. and T. Odean. 2008. A ll That Glitters: The Effect of Attention and News on the Buying Behavior of Individual and Institutional Investors. The Review of Financial Studies 21(2): 785 818. Barberis, N., A. Shleifer and J. Wurgler. 2005. Comovement. Journal of Financial Economics 75: 283 317. Bergman, N. K. and S. Roychowdhury. 2008. Investor Sentiment and Corporate Disclosure. Journal of Accounting Research 46(5): 1057 1083. Bernanke, B., and M. Gertler. 1989. Agency Costs, Net Worth, and Business Fluctuations. American Econom ic Review 79: 14 31.

PAGE 162

162 Bernstein, P 2007. Capital Ideas Evolving John Wiley & Sons, Inc.: Hoboken, New Jersey. Boyson, N., C. Stahel, and R. Stulz. 2010. Hedge Fund Contagion and Liquidity Shocks, forthcoming Journal of Finance. Brown, G. W. and M. T. C liff. 2004. Investor Sentiment and the Near Term Stock Market. Journal of Empirical Finance 11: 1 27. Brown, G. W. and M. T. Cliff. 2005. Investor Sentiment and Asset Valuation. Journal of Business 78(2): 405 440. Brown, N. C., T. E. Christensen, W. B. Elliot and R. Mergenthaler Jr. 2011. Investor Sentiment and Pro Forma Earnings Disclosures. Working Paper, Georgia State University. Brown S. J., W. N. Goetzmann, T. Hiraki, N. Shiraishi, and M. Watanabe. 2002. Investor Sentiment in Japanese and U.S. Dai ly Mutual Fund Flows. NBER Working Paper. Brunnermeier, M 2009. Deciphering the Liquidity and Credit Crunch 2007 2008. Journal of Economic Perspectives 23(1): 77 100. Brunnermeier, M. and L. Pedersen. 2009. Market Liquidity and Funding Liquidity. The R eview of Financial Studies 22(6): 2201 2238. Carhart, M 1997. On Persistence in Mutual Fund Performance. Journal of Finance 52(1): 57 82. Chan, J., D. Hong, and M. Subrahmanyam. 2008. A Tale of Two Prices: Liquidity and Asset Prices in Multiple Markets. Journal of Banking and Finance 32: 947 960. Chan, S., J. Erickson and K. Wang. 2003. Real Estate Investment Trusts: Structure, Performance, and Investment Opportunities. Oxford University Press: New York. Chen, N., R. Roll, and S. Ross. 1986. Economic Forces and the Stock Market. Journal of Business 59: 383 403. Christensen, T. E., K. J. Merkley, J. W. Tucker and S. V enkataraman. 2011. Do Managers U se Earnings Guidance to Influence Street Earnings Exclusions? Review of Accounting Studies forthcoming. Chuk, E., D Matsumoto and G. Miller. 2009. Assessing Methods of Identifying Management Forecasts: CIG vs. Research Collected. Working Paper University of Washington Clayton, J. D C. Ling, and A Naranjo. 2009. Commercial Real Estate Valuation: Fun damentals Versus Investor Sentiment. Journal of Real Estate Finance and Economics 38: 5 37. Daniel, K D Hirshleifer, and A Subrahmanyam. 1998. Investor Psychology and Security Market Under and Overreactions. Journal of Finance 53: 1839 1886.

PAGE 163

163 De Long, J. B A Shleifer, L H. Summers, and R J. Waldmann. 1990. Noise Trader Risk in Financial Markets. Journal of Political Economy 98: 703 738. Dudley, E. and M. Nimalendran. 2010. Margins and Hedge Fund Contagion. Journal of Financial and Quantitative A nalysis forthcoming Dye, R 1985. Disclosure of Nonproprietary Information. Journal of Accounting Research 23(1): 123 145. Dye, R 1998. Investor Sophistication and Voluntary Disclosures. Review of Accounting Studies 3: 261 287. Economist 2010. Bubbl e Warning. 7 January 2010. Efron, B 1979. Bootstrap Methods: Another Look at the Jackknife. The Annals of Statistics 7(1): 1 26. Fama, E 1970. Efficient Capital Markets: A Review of Theory and Empirical Work. Journal of Finance 25: 383 417. Fama, E., and K. French. 1988. Dividend Yields and Expected Stock Returns. Journal of Financial Economics 22(1): 3 25. Fama, E ., and K French. 1993. Common Risk Factors in the Returns on Stocks and Bonds. Journal of Financial Economics 33: 3 56. Fama, E and K French. 1996. Multifactor Explanations of Asset Pricing Anomalies. Journal of Finance 51(1): 55 83. Fama, E., and G. W. Schwert. 1977. Asset Returns and Inflation. Journal of Financial Economics 5:115 46. Ferson, W.F. and C.R. Harvey. 1991. The Variati on of Economic Risk Premiums. Journal of Political Economy 99: 385 415. Fisher, J. D Gatzlaff, D Geltner, and D Haurin. 2004. An Analysis of the Determinants of Transaction Frequency of Institutional Commercial Real Estate Investment Property. Real Es tate Economics 32(2): 239 264. Fisher, J. D Geltner, and H Pollakowski. 2007. A Quarterly Transactions Based Index (TBI) of Institutional Real Estate Investment Performance and Movements in Supply and Demand. Journal of Real Estate Finance and Economic s 34: 5 33. Fisher, J and M. Young. 2000 Institutional Property Tenure: Evidence from the NCREIF Database. Journal of Real Estate Portfolio Management 6 (4): 327 338.

PAGE 164

164 Fishman, M and K. Hagerty. 2003. Mandatory versus Voluntary Disclosure in Mark ets with Informed and Uninformed Customers The Journal of Law, Economics, and Organization 19(1): 45 63. Frazzini, A and O A. Lamont. 2008. Dumb Money: Mutual Fund Flows and the Cross Section of Stock Returns. Journal of Financial Economics 88: 299 32 2. Froot, K., D. Scharfstein, and J. Stein. 1992. Herd on the Street: Informational Inefficiencies in a Market with Short Term Speculation. Journal of Finance 47(4): 1461 1484. Froot, K. and E.M. Dabora. 1999. How are Stock Prices Affected by the Locat ion of Trade? Journal of Financial Economics 53 (2): 189 216. Garleanu, N. and L Pedersen. 2007, Liquidity and risk management. American Economic Review 97: 193 197. Geanakoplos, J 2003. Liquidity, Default, and Crashes: Endogenous Contracts in General Equilibrium. Advances in Economics And Econometrics: Theory and Applications, Eighth World Conference, Econometric Society Monographs (2) : 170 205. Geltner, D. and D. C. Ling. 2007. Indices for Investment Benchmarking and Return Performance Analysis in P rivate Real Estate. International Real Estate Review 10(1): 93 138. Ghysels, E. A Plazzi, and R Valkanov. 2007. Valuation in US Commercial Real Estate. European Financial Management 13(3): 472 497. Glushkov, D 2006. Sentiment Beta. Working Paper, Un iversity of Texas Austin. Goetzmann, W N. and P Jorion. 1993. Testing the Predictive Power of Dividend Yields. Journal of Finance 48(2): 663 679. Goto, S M Watanabe and Y Xu. 2009. Strategic Disclosure and Stock Returns: Theory and Evidence from U S Cross Listing. The Review of Financial Studies 22(4): 1585 1620. Green, T., and B Hwang. 2009. Price Based Return Comovement. Journal of Financial Economics 93: 37 50. Greenwood, R 2008. Excess Comovement: Evidence from Cross Sectional Variation in Nikkei 225 Weights. The Review of Financial Studies 21: 1153 1186. Grossman, S J. 1981. The Informational Role of Warranties and Private Disclosure about Product Quality. Journal of Law and Economics 24: 461 489. Grossman, S J. and O. Hart. 1980. Disc losure Laws and Takeover Bids. Journal of Finance 35(2): 323 334.

PAGE 165

165 Haggard, K. X Martin and R Pereira. 2008. Does Voluntary Disclosure Improve Stock Price Informativeness? Financial Management 37: 747 768. Hameed, A W Kang, and S. Viswanathan. 2010. Stock Market Declines and Liquidity Journal of Finance 65(1): 257 293. Hansen, C and B. Tuypens. 2004. Long Run Regressions: Theory and Application to US Asset Markets. Working Paper, CUNY. Hansen, L. and R. Hodrick. 1980. Forward Exchange Rates as Optimal Predictors of Future Spot Rates: An Empirical Analysis. Journal of Political Economy 88: 829 853. Houston, J., B Lev, and J W Tucker. 2010. To Guide or Not to Guide? Causes and Consequences of Stopping Quarterly Earnings Guidance. Contemporary Accounting Research 27(1): 143 185. Jegadeesh, N., and S. Titman. 1993. Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency. Journal of Finance 48(1): 65 91. Jiang, G., D. Xu and T. Yao. 2009. The Information Content of Idiosyncratic Volatility. Journal of Financial and Quantitative Analysis 44(1): 1 28. Jin, L., and S Myers. 2006. R 2 Around the World: New Theory and New Tests. Journal of Financial Economics 79(2): 257 292. Kaplan, S., and A. Schoar. 2005. Private Equ ity Performance: Returns, Persistence and Capital Flows. Journal of Finance 60: 1791 1823. Kasznik, R., and B. Lev. 1995. To Warn or Not to Warn: Management Disclosures in the Face of an Earnings Surprise. The Accounting Review 70(1): 113 134. Kiyotaki, N., and J. Moore. 1997. Credit Cycles. Journal of Political Economy 105(2): 211 248. Kumar, A., and C Lee. 2006. Retail Investor Sentiment and Return Comovements. Journal of Finance 61(5): 2451 2486. Lamont, O and R. Thaler. 2003a. Can the Market Add and Subtract? Mispricing in Tech Stock Carve outs. Journal of Political Economy 111(2): 227 268. Lamont, O and R. Thaler. 2003b. Anomalies: The Law of One Price in Financial Markets. Journal of Economic Perspectives 17(4): 191 202. Lee, C A Shleifer and R Thaler. 1991. Investor Sentiment and the Closed End Fund Puzzle. Journal of Finance 46: 75 109.

PAGE 166

166 Lettau, M., and S. Ludvigson. 2001. Resurrecting the (C) CAPM: A Cross Sectional Test When Risk Premia are Time Varying. Journal of Political Economy 109(6): 1238 1287 Liew, J. and M. Vassalou. 2000. Can Book to Market, Size and Momentum be Risk Factors that Predict Economic Growth? Journal of Financial Economics 57: 221 245. Longstaff, F. and J Wang. 2008. Asset Pricing and the Credit Market. Workin g Paper, MIT. Lown, C. D Morgan, and S Rohatgi. 2000. Listening to Loan Officers: The Impact of Commercial Credit Standards on Lending and Output. Federal Reserve Bank of New York Policy Review (July): 1 16. MacKinnon, J. 2002. Bootstrap Inference in Econometrics. Canadian Journal of Economics 35(4): 615 645. MacKinnon, J. 2006. Bootstrap Methods in Econometrics. The Economic Record 82: S2 S18. MacKinnon, J. and A. Smith. 1998. Approximate Bias Correction in Econometrics. Journal of Econometrics 85 : 205 230. Menkhoff, L ., and R. Rebitzky. 2008. Investor Sentiment in the US dollar: Longer Term, Non Linear Orientation on PPP. Journal of Empirical Finance 15: 455 467. Nelson, C., and M Kim. 1993. Predictable Stock Returns: The Role of Small Sample B ias. Journal of Finance 48(2): 641 661. Newey, W and K West. 1987. A Simple Positive Semi Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix. Econometrica 55: 703 708. Pagliari, J., K. Scherer and R. Monopoli. 2005. Public Versus Private Real Estate Equities: A More Refined, Long Term Comparison. Real Estate Economics 33(1): 147 187. Pastor, L ., and R. Stambaugh. 2003. Liquidity Risk and Expected Stock Returns. Journal of Political Economy 111(3): 642 685. Pontiff, J. 1996 Costly Arbitrage: Evidence from Closed End Funds. Quarterly Journal of Economics 111: 1135 1151. Pontiff, J. 2006. Costly Arbitrage and the Myth of Idiosyncratic Risk. Journal of Accounting and Economics 42: 35 52. Qiu, L ., and I Welch. 2005. Investor Sentiment Measures. NBER Working Paper. Rees, L ., and R Database. Advances in Accounting 23: 147 177.

PAGE 167

167 Riddiough, T M Moriarty, and P. Yeatman. 2005. Privately Versus Publicly Held Asset Investment Performance. Real Estate Economics 33(1): 121 146. Ritter, J 1991. The Long run Performance of Initial Public Offerings. Journal of Finance 46: 3 27. Scheinkman, J. and W Xiong. 2003. Overconfidence and Speculative Bubbles. Journal of Political Economy 111(6): 1183 1219. Schmeling, M 2007. Institutional and Individual Sentiment: Smart Money and Noise Trader Risk? International Journal of Forecasting 23: 127 145. Schmeling, M., and A Schrimpf. 2008. Expected Inflation, Expected St ock Returns, and Money Illusion: What Can We Learn from Survey Expectations? Working Paper Humboldt University. Seybert, N ., and H Sentiment Driven Overvaluation. Management Scie nce (Special Issue on Behavioral Economics and Finance), forthcoming. Earnings Forecasts. Review of Economics and Statistics 84:632 48. Shleifer, A ., and R Vishny. 1997. The Limits of Arbitrage. Journal of Finance 52: 35 55. Shleifer, A ., and R Vishny. 2010. Unstable Banking. Journal of Financial Economics 97 : 306 318. Soffer, L S. R Thiagarajan and B Walther. 2000. Earnings Preannouncement Strategies. Review of Accounting Studies 5(1): 5 26. Stambaugh, R. 1999. Predictive Regressions. Journal of Financial Economics 54: 375 421. Veldkamp, L 2006. Information Markets and the Comovement of Asset Prices. The Review of Economic Studies 73(3): 823 845. Verrecch ia, R 1983. Discretionary Disclosure. Journal of Accounting and Economics 5: 179 194. Welch, I 1992. Sequential Sales, Learning, and Cascades. Journal of Finance 47(2): 695 732. Wurgler, J., and E. Zhuravskaya. 2002. Does Arbitrage Flatten Demand Curve s for Stocks? Journal of Business 75(4): 583 608.

PAGE 168

168 BIOGRAPHICAL SKETCH Benjamin received a Bachelor of Science degree in e conomics, a Bachelor of Science degree in business a dmi nistration (f inanc e), and a minor certificate in m athematics from Villanova Un iversity in 2003. He was awarded the Bartley Medallion in Economics by the College of Commerce and Finance, the highest distinction the college can award to a graduating student. In 2003, Benjamin joined Navigant Consulting as an Associate Consultant. Over a three year period, Benjamin was involved in several projects that included working within the private wealth management division of a major investment bank, designing and implementing a reporting system for the account reconciliation team of a hedge fun d administrator, and performing an audit of the real estate investment division of a prominent insurance company. Benjamin left Navigant Consulting as a Senior Consultant to purs ue his Ph.D. in f inance at the University of Florida. Benjamin received his Ph .D. from the University of Florida in 2011 and join ed the faculty at the University of Georgia as an Assistant Professor of Real Estate.