Strategic Credit Line Usage and Performance

MISSING IMAGE

Material Information

Title:
Strategic Credit Line Usage and Performance
Physical Description:
1 online resource (98 p.)
Language:
english
Creator:
Manakyan, Ani
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:
James, Christophe M
Committee Members:
Ryngaert, Michael D
Houston, Joel F
Hamersma, Sarah E

Subjects

Subjects / Keywords:
capital -- credit -- iq -- line -- revolving
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

Abstract:
In my study, I use a unique hand-collected dataset of corporate credit line usage. In the first part of the study, I examine whether firms draw precautionary balances from their credit lines in anticipation of a decline in future availability. I find that credit line drawdowns precede declines in cash flow. Further, I model predicted drawdowns and estimate unexpected drawdowns as the residual from the predictive regression. I show that unexpected drawdowns predict future cash flow declines, future net worth declines, future current ratio declines, future covenant violations and concurrent ratings downgrades. Consistent with the use of unexpected drawdowns as a proxy for precautionary balances that are not needed for immediate investment, unexpected drawdowns are associated with increases in cash balances. Firms with unexpected drawdowns do not get better terms in renegotiations, but they are more able to finance future capital expenditures following a covenant violation than those firms without unexpected drawdowns. Overall,these findings support the hypothesis that firms strategically draw on their credit lines to access cash before their performance deteriorates. In the second part of the study, I use my hand-collected dataset to evaluate Capital IQ data on credit line usage. Recent academic work in finance has made use of Capital IQ’s more detailed data on capital structure.I illustrate several potential issues with relying on Capital IQ data to determine if a firm uses or has access to corporate credit lines. These include reporting missing values when there is data available on credit line usage and access in the company’s 10-K filings, misreporting fiscal year end dates for firms with a fiscal year end month prior to June, underrepresenting the number of firm-year observations in which the sampled companies have access to and availability on a corporate credit line, and using data from tables that may misrepresent credit line activity. I suggest some means for improvement when making use of the Capital IQ database.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Ani Manakyan.
Thesis:
Thesis (Ph.D.)--University of Florida, 2012.
Local:
Adviser: James, Christophe M.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2014-08-31

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2012
System ID:
UFE0044420:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 STRATEGIC CREDIT LINE USAGE AND PERFORMANCE By K. ANI MANAKYAN A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 201 2

PAGE 2

2 201 2 K. Ani Manakyan

PAGE 3

3 To my parents, colleagues and friends, for all of your help and support.

PAGE 4

4 ACKNOWLEDGMENTS I thank my parents for instilling in me a love of learning and for always supporting me on this difficult road. I also want to thank all of my friends, my PhD colleagues, and Mike especially, for keeping me sane during the countless ups and downs. Finally I want to sincerely thank my committee, Christopher James, Joel Houston, Michael Ryngaert and Sarah Hamersma, for all of their guidance and wisdom.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 6 ABSTRACT ................................ ................................ ................................ ..................... 8 CHAPTER 1 INTRODUCTORY REMARKS ................................ ................................ ................ 10 Overview of Chapter 2 ................................ ................................ ............................ 10 Overvi ew of Chapter 3 ................................ ................................ ............................ 13 2 STRATEGIC CREDIT LINE USAGE AND PERFORMANCE ................................ 15 Introduction ................................ ................................ ................................ ............. 15 Hypothesis Devel opment and Literature Review ................................ .................... 19 Data ................................ ................................ ................................ ........................ 22 Sample ................................ ................................ ................................ ............. 22 Covenant Compliance Sample ................................ ................................ ......... 23 DealScan Sample ................................ ................................ ............................. 24 Control Vari ables ................................ ................................ .............................. 24 Results ................................ ................................ ................................ .................... 28 Unexpected Drawdowns ................................ ................................ ................... 30 Lender Response ................................ ................................ ............................. 39 Conclusion ................................ ................................ ................................ .............. 44 3 .................... 70 Introduction ................................ ................................ ................................ ............. 70 Data ................................ ................................ ................................ ........................ 72 Results ................................ ................................ ................................ .................... 73 Conclusion ................................ ................................ ................................ .............. 77 4 CONCLUDING REMARKS ................................ ................................ ..................... 89 A PPENDIX : VARIABLE DEFINITIONS ................................ ................................ ...... 93 LIST OF REFERENCES ................................ ................................ ............................... 96 BIOGRAPHIC AL SKETCH ................................ ................................ ............................ 98

PAGE 6

6 LIST OF TABLES Table page 2 1 Summary statis tics ................................ ................................ ............................. 46 2 2 Sum mary statistics by draw dummy ................................ ................................ ... 48 2 3 Effect of drawdowns on future cash flows ................................ .......................... 50 2 4 Effect of change in drawdowns on th e change in future cash flows .................... 51 2 5 Tobit regression to p redict credit line drawdowns ................................ ............... 52 2 6 Effect of unexpect ed drawdow ns on cash holdings ................................ ............ 53 2 7 Effect of unexpected d rawdowns on future cash flows ................................ ....... 54 2 8 Effect of unexpected d rawdowns on future cash flows ................................ ....... 56 2 9 Effect of unexpected drawdown s on future covenant variables .......................... 57 2 1 0 Effect of unexpected drawdown on the probability of a new covenant violation ................................ ................................ ................................ ............. 58 2 11 Effect of unexpected drawdown on probabili ty of credit rating downgrade ......... 60 2 12 Effect of covenant violations on credi t line availability and usage ....................... 62 2 13 E ffect of unexpected drawdowns on change in future capital expenditures ....... 6 3 2 14 Summary statistics on changes in contract terms ................................ ............... 65 2 15 Summary statistics for changes in contract terms for f irms with covenant v iolations ................................ ................................ ................................ ............ 66 2 16 Summary statistics for changes in contract terms for fi rms with unexpected drawdowns ................................ ................................ ................................ ......... 67 2 17 Summary statistics on changes in contract terms around unexpected drawdowns when t here is no covenant violation ................................ ................. 68 2 18 Summary statistics on changes in contract terms around unexpected drawdowns when there is a covenant violation ................................ .................. 69 3 1 Summary statistics for hand collected and Capita l IQ data on credit line usage ................................ ................................ ................................ .................. 79

PAGE 7

7 3 2 S ummary statistics for Capital IQ data on credit line usage relative to hand collected data ................................ ................................ ................................ ..... 80 3 3 Summary statistics for Capital IQ data on credit line usage relative to hand collected data condition al on Capital IQ availability ................................ ............ 81 3 4 Accuracy of Capital IQ data on credit line access and availability re lative to hand collected data ................................ ................................ ............................ 82 3 5 Accuracy of Capital IQ data on credit line access and availability relative to hand collected data by year ................................ ................................ ................ 83 3 6 Accuracy of Capital IQ data on credit line usage relative to hand coll ected data with exact matches ................................ ................................ ..................... 85 3 7 Accuracy of Capital IQ data on credit line usage relative to hand collecte d data with matches within 10% ................................ ................................ ............ 86 3 8 Probability that Capital IQ credit line data matche s hand collected data exactly ................................ ................................ ................................ ................ 87 3 9 Probability that Capital IQ credit line data matches h and collected data within 10% ................................ ................................ ................................ .................... 88

PAGE 8

8 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy STRATEGIC CREDIT LINE USAGE AND PERFORMANCE By K. Ani Manakyan August 201 2 Chair: Christopher James Major: Business Administration In my study, I use a unique hand collected dataset of corporate credit line usage. In the first part of the study, I examine whether firms draw precautionary balances from their credit lines in anticipation of a decline in future availability. I find that credit line drawdowns precede declines in cash flow. Further, I model predicted drawdowns and estimate unexpected dr awdowns as the residual from the predictive regression. I show that unexpected drawdowns predict future cash flow declines, future net worth declines, future current ratio declines, future covenant violations and concurrent ratings downgrades. Consistent w ith the use of unexpected drawdowns as a proxy for precautionary balances that are not needed for immediate investment, unexpected drawdowns are associated with increases in cash balances. Firms with unexpected drawdowns do not get better terms in renegoti ations, but they are more able to finance future capital expenditures following a covenant violation than those firms without unexpected drawdowns. Overall, these findings support the hypothesis that firms strategically draw on their credit lines to access cash before their performance deteriorates.

PAGE 9

9 In the second part of the study, I use my hand collected dataset to evaluate Capital IQ data on credit line usage. Recent academic work in finance has made use of ture. I illustrate several potential issues with relying on Capital IQ data to determine if a firm uses or has access to corporate credit lines These include reporting missing values when there is data available on credit line usage and access in the comp K filings, underrepresenting the number of firm year observations in which the sampled companies have access to and availability on a corporate credit line and using data from tables that may misrepresent credit line activity I suggest some means for improvement when making use of the Capital IQ database

PAGE 10

10 CHAPTER 1 INTRODUCTORY REMARKS In both of the two chapters that constitute my study I investigate corporate credit line usage. In the first part, Chapter 2, I utilize a unique dataset to examine whether firms draw precautionary balances from their credit lines in anticipation of a decline in future availability. To measure the pre emptive balances drawn by firms, I model predicted drawdowns and measure unexpected drawdowns as the residual from the predictive regression. I find evidence that firms draw unexpectedly on their credit lines in advance of poor performance, covenant violat ions and decreased availability. I also show that these unexpected drawdowns do not put the firm in a better bargaining position with their bank but also do not result in harsh retaliation by the bank. I find that unexpected drawdowns allow the company to maintain future capital expenditures. In the second part of the study, Chapter 3, I use my hand collected dataset to evaluate Capital IQ data on credit line usage. I illustrate several potential issues with relying on Capital IQ data to determine if a firm uses or has access to corporate credit lines These include inconsistent reporting of data for a given firm over time, a consistent mislabeling of fiscal years and an overreliance on tables reported in filings instead of a more in depth reading of the filing. I suggest some means for improving Capital IQ data for academic purposes Overview of Chapter 2 Most of the theoretical models of credit lines are based upon t he assumption that access to a committed line is not contingent on the financial condition of the borrower at the time a draw is requested. However, recent empirical studies indicate that this assumption may not be realistic. For example, Sufi (2009) finds evidence that access to

PAGE 11

11 operating cash flows of the borrower. Sufi argues that this positive relationship arises from cash flow and other financial covenants associ ated with bank lines of credit. These restrictions would appear to limit the liquidity insurance that lines provide. The objective of C hapter 2 is to better understand what role lines of credit play in corporate liquidity management. I examine whether fir ms strategically time their credit line drawdowns. I hypothesize that if firms have private information concerning their future operating performance and if banks condition line access on operating performance at the time of a draw request, then firms may have an incentive to preemptively draw on their lines as a way of accumulating precautionary cash balances. a large strategic draw. If, for example, lenders react more aggressively to covenant violations conditional on strategic line use, then borrowers may limit line borrowing as a way of acquiring precautionary balances. Using my hand collected sample on cred it line usage, I model drawdowns and estimate unexpected drawdowns as the residual from the predictive regression. Consistent with their use as a proxy for preemptive drawdowns, I show that unexpected drawdowns increase cash holdings. I find that unexpecte d drawdowns predict future cash flow declines, future covenant violations, and concurrent credit rating downgrades. These findings support my hypothesis regarding strategic credit line drawdowns. I also address the bank response to unexpected drawdowns. S ufi (2009) shows that covenant violations are associated with reductions in the size, or overall availability,

PAGE 12

12 of the credit line. I extend his results to show that covenant violations are not associated with subsequent declines in the amount of borrowing, or actual credit line usage, although they are associated with declines in the unused amount of the line available to the borrower in the future. I f credit line availability is likely to be cut after a default but drawn amounts are not required to be imme diately fully repaid, then corporations have an incentive to draw on their lines in anticipation of poor future cash flows and potential covenant violations. Furthermore, Nini, Smith and Sufi (2011) argue that covenant violations lead to a reduction in ca pital expenditures. They contend that this is due to new restrictions and control rights imposed by the lender. I find that unexpected drawdowns mitigate future capital expenditure declines following a covenant violation. Thus unexpected drawdowns seem to put the firm in a better position. I further explore the change in credit line contract terms following a covenant violation using DealScan data on credit line contracts to see if firms enhance their bargaining power with their lender I find no evidence t hat contract terms improve for borrowers who draw preemptively on their lines of credit. If anything, I tend to see a worsening of contract terms. 1 I interpret this evidence on the bank reaction to unexpected drawdowns as follows: banks gain leverage in re negotiating contract terms following a covenant violation and tend to tighten loan terms, but they do not immediately require repayment of the outstanding balances or impose any other strong punishment on firms with unexpected drawdowns. 1 Chen, Hu and Mao (2011) find that all drawdowns lead to worse future contract terms on a new credit line agreement. I do not find significant differences in contract terms for those firms with and without unexpected drawdowns conditional on a covenant vio lation.

PAGE 13

13 Therefore, the cas h available from a preemptive draw on the credit line can be used to maintain future capital expenditures Overview of Chapter 3 structure. Capital IQ provides data from 2002 2010 on drawn and available amounts on corporate credit line contracts. Paolo, Ippolito, and Li (2011) and Ippolito and Perez (2011) use Capital IQ to identify firms with access to credit lines. They also make use of drawn amounts outstanding as reported in Capital IQ in their analysis I compare the drawn and available amounts in Capital IQ to hand collected data on credit line usage collection method relative to hand collecting. Of course, my hand collecti on is subject to possible errors or subjective judgments, as well. I do not claim to have a perfect database, but I believe that I am able to document some specific features of Capital IQ data that should be of use for individuals who wish to use this data set for studies involving credit line data. I identify several concerns with the use of Capital IQ data on credit line usage and give examples of the relevant issues First, Capital IQ often reports blanks when there is data available on credit line usage K filings. Worse, this lack of reporting is not consistent over time In addition, I show that using Capita l underrepresents the number of firm year observations in which the sampled companies have access to and availability on a corporate credit line. Only 27% of the firm year observations in the hand collected sample match Capital IQ data in establishing whether or not there are amounts available to borrow on a credit line. If I foll ow existing work by Ippolito and Perez (2011)

PAGE 14

14 and use reported Capital IQ credit line availability to determine whether or not a firm has a line of credit, then Capital IQ captures only 28% of the firm year observations where there is a corporate credit li ne. I suggest an improvement for the purposes of determining credit line access. variables, the accuracy of using Capital IQ to determine if a company has a credit line increases to 71%. Finally, Capita l IQ often uses data from tables. This may eliminate some important granularity that can be obtained by reading the filing. For instance, the reporting of outstanding amounts on a credit line agreement may aggregate the revolving and term loan po rtion of t he agreement. This study is the first to systematically examine the Capital IQ credit line database, so it may prove useful for future academic research.

PAGE 15

15 CHAPTER 2 STRATEGIC CREDIT LIN E USAGE AND PERFORMA NCE Introduction The objective of my paper is to better understand what role lines of credit play in corporate liquidity management. Specifically, a prominent theme in the theoretical literature is that corporate credit lines are a form of insurance against liquidity shocks (Holmstrom and Ti role (1998) and Boot et al (1987) ) Holmstrom and Tirole (1998) argue that banks are able to aggregate liquidity across firms and redistribute excess liquidity to those that need funds. Boot et al (1987) note that committed lines of credit allow companie s to pay a fee at initiation in order to have access to a low cost source of financing if they face a liquidity shock in the future. Most of the theoretical models of credit lines are based upon the assumption that access to a committed line is not conting ent on the financial condition of the borrower at the time a draw is requested. However, recent empirical studies indicate that this assumption is not realistic. For example, Sufi (2009) finds evidence that access to credit credit lines relative to cash is increasing in the operating cash flows of the borrower. Sufi argues that this positive relationship arises from cash flow and other financial covenants associated with bank lines of credit. The existence of tight financial covenants and other conditions placed on the use of lines, such as Material Adverse Condition Clauses (MACs), make credit lines a contingent source of liquidity. 1 These restrictions would appear to limit the liquidity insurance that lines provide. 1 Shockley and Thakor (1997) find that every credit line in their sample contains a MAC. However, other studies, such as Ivashina and Scharfstein (2010) note that MACs are rarely invoked. Covenant violations,

PAGE 16

16 Anecdo tal evidence suggests that firms strategically draw down their credit lines in anticipation of future declines in operating performance that may impair their access to credi t line in 2008 raised red flags for turnaround specialist Wilbur Ross who remarked, [It] makes you wonder if there is some danger that when the September results come 2 Additionally rating agencies consider drawdown December of 2008 in which they document an increase in the use of secured lines of dra wdown activity prior to default because it affects the loss given default, especially for unsecured creditors (Emery et al (2008)). In my paper, I examine whether firms strategically time their credit line drawdowns. I hypothesize that if firms have priva te information concerning their future operating performance and if banks condition line access on operating performance at the time of a draw request, then firms may have an incentive to preemptively draw on their lines as a way of accumulating precaution performance declines given a large strategic draw. If, for example, lenders react more aggressively to covenant vio lations conditional on strategic line use or change loan terms on subsequent deals in an adverse fashion then borrowers may limit line borrowing as a way of acquiring precautionary balances. on the other hand, occur with some freq uency (e.g. Sufi (2009), Roberts and Sufi (2009), Nini, Smith and Sufi (2009) and Chava and Roberts (2008)). 2 Humer and Erman (2008)

PAGE 17

17 I investigate the strategic use of lines of credit using a uniqu e hand collected data base. I randomly sample 900 firms and collect data on credit line usage, availability and total size for the 763 of those companies with a credit line in at least one year from 2000 to 2010 I use this annual credit line data to exami ne whether future cash flow performance is related to credit line drawdowns. I model predicted drawdowns and estimate unexpected drawdowns as the residual from the predictive regression. Consistent with their use as a proxy for preemptive drawdowns, I sho w that unexpected drawdowns result in an increase in cash holdings. I find that unexpected drawdowns predict future cash flow declines, future covenant violations, and concurrent credit rating downgrades. I also address the bank response to unexpected draw downs. Sufi (2009) shows that covenant violations are associated with reductions in the size, or overall availability, of the credit line. I extend his results to show that covenant violations are not associated with subsequent declines in the amount of bo rrowing, or actual credit line usage, although they are associated with declines in the unused amount of the line available to the borrower in the future. I posit that i f credit line availability is likely to be cut after a default but drawn amounts are no t required to be immediately fully repaid, then corporations have an incentive to draw on their lines in anticipation of poor future cash flows and potential covenant violations. In deed it may be the case that preemptive credit line drawdowns actually im prove decision to draw down the entirety of its $250 million line of credit in late 2008, even though it had in excess of $300 million in cash on its balance sheet Goldm an Sachs downgraded Idearc stating [W e ] believe that [Idearc] may have decided to [draw down

PAGE 18

18 on its revolver] as a way to gain greater leverage with its bank group as it gets closer to Gorton and Kahn (2000) argue that a larger impairing their strategic bargaining position. I further investigate the reaction of lenders to unexpected drawdowns. Nini, Smith and Sufi (2011) argue that covenant violations lead to a reduction in capital expenditures. They contend that this is due to new restrictions and control rights imposed by the lender. I find that unexpected drawdowns mitigate future capital expenditu re declines following a covenant violation. Thus unexpected drawdowns seem to put the firm in a better position. I then further explore the change in credit line contract terms following a covenant violation. Gorton and Kahn (2000) argue that a greater amo I find no evidence that contract terms improve for borrowers who draw preemptively on their lines of credit. If anything, I tend to see a worsening of contract terms. 3 I interpret this ev idence on bank reaction to unexpected drawdowns as follows: banks gain leverage in renegotiating contract terms following a covenant violation and tend to tighten loan terms, but they do not immediately require repayment of the outstanding balances or impo se any other strong punishment on firms with unexpected drawdowns. Therefore, the cash available from a preemptive draw on the credit line can be used to maintain future capital expenditures. 3 Chen, Hu and Mao (2011) find that all drawdowns lead to worse future contract terms on a new credit line agreement. I d o not find significant differences in contract terms for those firms with and without unexpected drawdowns conditional on a covenant violation.

PAGE 19

19 Hypothesis Development and Literature Review My hypothesis, term ed the availability hypothesis, states that firms may access capital while it is still available, even if they have no immediate use for it, because they are concerned about future limitations on availability. The alternative hypothesis is that drawdowns a re based on current cash needs and are not predictive of future performance. If this is the case, then I would expect to find no relationship between drawdowns and future cash flow performance after controlling for concurrent cash flows and cash needs. My paper is related to a number of papers examining the use of credit lines in liquidity management. Lins, Servaes and Tufano (2011) use evidence from a survey conducted in 2005 to show that the contingent nature of credit lines lead s a majority of companies to view them differently from cash. Based on their survey evidence, only 41% of surveyed CFOs view cash and credit lines as substitutes. Lins et al find that firms with high cash flows are more likely to view cash and credit lines as substitutes They co nclude that companies recognize that credit lines are not a guaranteed source of liquidity. This is c onsistent with the results in Sufi (2009) on the importance of cash flows in maintaining a credit line b ehavior in converting contingent lines of credit into cash. Campello, Giambona, Graham and Harvey (2011) conducted a survey in the first quarter of 2009 about current and previous year financial figures, including credit line availability and usage. They model credit line availability and usage and show that private, small and junk rated companies drew more on their credit lines during the financial crisis than their financially unconstrained counterparts. I build on their model of drawdown activity to pre dict drawdowns in my sample. In addition Campello et al

PAGE 20

20 conduct a supplementary survey during the second quarter of 2010 regarding covenant violations. They report that only 10% of the firms who violate a covenant lose their credit lines due to the viola tion. Instead, more than half of the covenant violations lead to the renegotiation of the credit line and one third lead to no action at all. This is an important empirical fact that influences my hypothesis. If banks ended the credit line agreement and re quired immediate repayment, then there would be no value to drawing on a line in anticipation of a violation. Interest in the determinants of credit line usage has heightened since the financial crisis of 2008 because of the increased drawdown activity fo llowing the failure of Lehman Brothers in September of 2008. The increase in drawdowns contributed to increased pressure on banks (Ivashina and Scharfstein (2010)). One possible explanation for the increase in drawdown activity is that firms drew precauti o nary balances on their lines in response to concerns about bank solvency and future access to credit. This is supported by small sample evidence from drawdown announcements during the financial crisis that companies drawing on their lines state a desire to increase financial flexibility and liquidity during a time of market uncertainty. I propose that, in addition to this possible explanation, firm specific problems alone may cause accelerated credit line drawdowns. I examine and find support for this proposition using a large sample of credit line usage data. Sufi (2009) demonstrates that cash flow declines predict covenant violations on credit lines, lead ing to decreased line availability as banks reevaluate the risk iness of the borrower More generally, negative cash flow changes predict unfavorable renegotiations of debt contracts (Roberts and Sufi (2009)). If firms anticipate that a

PAGE 21

21 decline in operating performance will trigger a renegotiation of their line of cred it agreement and a subsequent limitation of availability, then they are likely to respond by converting their lines into cash. That is, precautionary drawdowns are motivated by e of credit availability. The motivation for drawdowns in this context differs from previous studies, such as Ivashina and Scharfstein (2010), wherein precautionary draws are motivated by concerns about the financial performance and solvency of the lender, though the two are not mutually exclusive. be influenced by the reaction of the lender. I posit that firms are able to enhance their strategic bargaining position w ith the lender by drawing precautionary balances. Since negotiate with a borrower with a larger amount outstanding. In fact, Gorton and Kahn (2000) argue that the outcome of bargaining following renegotiation depends on the liquidate. The availability hypothesis predicts that firms will strategically draw their credit lines so long as the b enefits of preemptive draws exceed any lender imposed costs associated with drawing down lines in advance of deteriorating performance. T o test the availability hypothesis I need a proxy for preemptive drawdowns. I use balance sheet and income sheet data to estimate expected credit line drawdowns and use the residuals from the estimation, or unexpected drawdowns, as a measure of precautionary drawdowns. The availability hypothesis predicts that precautionary

PAGE 22

22 drawdowns occur prior to cash flow declines and covenant violations. I also expect to find that precautionary drawdowns increase cash balances as firms have unused capital following a preemptive drawdown Upon the violation of a financial covenant, lenders may restrict corporate flexibility in decision making. Existing evidence shows that covenant violations lead to binding new restrictions on capital expenditures, as evidenced by decreases in future capital expenditure spending (Nini, Smith and Sufi (2009 and 2011)). However, I expect that the negative effect of a covenant violation on future capital expenditures is mitigated when the company has high unexpected drawdowns. This could be due to fewer lender imposed restrictions or simply due to the increased availability of cash to invest. To disentangle the two and assess whether lenders treat borrowers favorably in renegotiations following unexpected drawdowns, I examine the change in contract terms around unexpected drawdowns. Data Sample I first obtain data from Capital IQ on firms with either positive bank debt or a reported credit line in at least one year between 200 1 and 2010. I then keep only those companies that also appear in the Compustat industrial database, are not in the financial or utility sector, and have non missing asset data during the sample period. From the 5,619 companies that met these criteria, I take a random sample of 900. For the random sample of 900 firms, I use a program similar to the one described in Sufi (2009) to run a web crawl on 10 K fillings for fiscal years 2000 to 20 10. 4 I search 4 I modify a web crawler code for SAS provided in Engelberg and Sankaraguruswamy (2007).

PAGE 23

23 terms related to credit lines 5 I then manually examine the filings to establish the total size of the credit line, the drawn amount, and the unused amount available for draw. 6 As in Sufi (2009), if a line of credit backs up outstanding commercial paper or letters of credit, those amounts are subtracted from the available portion of the line of credit but not included as part of the amount drawn on the line. My collection method results in a sample of 76 3 companies with a credit line in at least one year from 2000 to 2010 for a total of 5, 311 firm year observations. Covenant Compliance Sample Data from 10 Q and 10 K statements on quarterly covenant compliance status is provided by Amir Sufi on his website. The dataset includes all U.S. nonfinancial firms in the Compustat universe during the years 1996 to the first quarter of 2008. The sample is limited to companies with average book assets greater than $10 million in 2000 dollar s and available data on assets, sales, common shares outstanding, closing share price, and the calendar quarter of the filing. Nini, Smith and Sufi (2011) describe the data collection process in detail. The variable of interest is an indicator variable for whether or not a company reports a violation of a financial covenant on any debt contract in the given quarter. 7 I annualize their violation data to match it with my data on credit line usage, so that I consider a firm to be in violation of a covenant in a given fiscal year if it was in violation of a covenant in at least one quarter during the fiscal 5 My 6 Corporations report detailed information related to their credit lines in annua l 10 K filings. Item 303 of SEC Regulation S K requires companies to report information related to their liquidity, capital resources, and results of operations for the fiscal year. While some companies choose to disclose credit line information more frequ ently, only annual reporting is required. 7 Note that reported violations for which the company has obtained a waiver are also recorded as a violation in the current quarter.

PAGE 24

2 4 year. I also distinguish a new violation following Nini, Smith and Sufi (2011) as a violation that occurs when there has been no violation in the previous th ree quarters. Merging my random sample with the covenant compliance data leaves me with 689 unique firms with both credit line and covenant violation information, for a total of 3,963 firm years. DealScan Sample I match both the overall sample and the cove nant compliance DealScan database on loan terms. I keep only those deals classified as completed and focus only on DealScan loans that are classified as credit lines. I augment the link file used by Chava and Roberts (2008) to match Deal Scan to Compustat by company name to extend the sample period through 2009. To compare how contract terms change, I first match observations for which there is an active pre existing credit line contract in Dealscan. I then keep only those observations wit h a new credit line contract reported within the next year. 8 I rely on a feature of Dealscan reporting that others (e.g. Nini, Smith and Sufi (2011)) have previously noted, which is that Dealscan reports renegotiations as new loans. This results in 1,087 firm year observations. Control Variables I define a number of control variables using Compustat annual data items. Since firms are likely to determine line of credit usage and cash holdings jointly, there is a mechanical negative correlation between any variable scaled by total assets and credit l ine usage. Following Sufi (2009), I scale Compustat variables by lagged non cash assets. However, the use of non cash assets creates outliers in the scaled variables for 8 I do not match to new loans within the current fiscal year because I cannot ide ntify the relative timing of the drawdown and the new loan contract.

PAGE 25

25 firms that hold most of their assets in cash. T o minimize the impact of outliers, I wi nsorize ratios and firm size at the 5 th and 95 th percentiles. 9 The specifics of the variable definitions, including the Compustat data items used, can be found in the Appendix I include industry fixed effects at the one digit SIC code level (as in Sufi (2009)), as the use of credit lines varies substantially across industries. 10 My results are qualitatively similar if I include two digit industry fixed effects. In addition, I incl ude year fixed effects to control for unobserved macroeconomic conditions, such as changes in interest rate expectations, which may cause firms to have differing incentives to draw on credit lines across time. Finally, I clu ster standard errors by firm to correct for the within firm correlation of residuals across years. This is important since companies may be serial users or non users of credit lines. Table 2 1 provides summary statistics for my random sample of firms. It is important to keep in mind that unlike the Compustat universe, my sample consists entirely of companies with credit line access. I partition the variables of interest into hand collected credit line variables and firm characteristics. The average drawn amount outstanding is $56.6 milli on, or 7.2% of the lagged non cash assets. There is a non zero amount drawn on the credit line in 56.7% of firm year observations. The prevalence of zero drawn amounts outstanding accounts for the substantially smaller median amount drawn. Of the firm year observations that I am able to match to the covenant compliance data from Nini, Smith and Sufi (2011), 16.6% have a covenant violation and 9 My results are qualitatively similar if I winsorize at the 1 st or 2.5 th percentiles. 10 I use the one digit SIC code because my sample is relatively small. Using two digit SIC codes I wo uld have 54 industry groups, but 28 (36) of those industries would contain fewer than five (ten) companies.

PAGE 26

26 10% report a new covenant violation, where new covenant violations are defined as the first violation in a four quar ter period. Recall that the covenant compliance sample contains information on whether or not a firm is in violation of a covenant at a given time. Of the 689 unique firms in the covenant compliance sample, 299 are in violation of a covenant at some point in the sample and 390 are never in violation of a covenant during the sample period. Summary statistics of firm characteristics are reported for use in computing marginal effects and for comparison to other studies. The average firm in my random sample ha s $1.6 billion in non cash assets, cash flows of 13% of non cash assets, and leverage of 31% of non cash assets. I present several measures of financial constraints because I suspect that they will be related to the likelihood that a firm uses its credit l ine. Just under one third of my sample companie s have credit ratings, and 13.5% of the sample is rated investment grade. Thirty percent of the firm year observations have a non zero dividend payment. Additionally, the credit market tightness variable tells me that, on average, banks were tightening lending standards during my time period. The average of credit market tightness indicates that more bank CFOs reported that they were tightening lending standards for commercial and industrial loans to large and medium firms than those that reported loosening standards. The negative number in the 10 th percentile, however, indicates that there are some periods of loosening standards, as well. Table 2 2 partitions firm years into those with drawn amounts outstanding and those with no drawn amounts outstanding. Almeida, Campello, and Weisbach (2004) discuss several features of financially constrained firms. Based on their findings, I

PAGE 27

27 examine several key characteristics of firms that do and do not have drawn amounts ou tstanding. I find that companies that draw on their credit lines are smaller, less likely to be rated, and less likely to be rated investment grade. Firms that draw on their credit lines are also less likely to pay a dividend. These findings are consistent with the notion that firms that use their credit lines are more financially constrained than those that do not make use of their line of credit. In addition, the firms that draw on their credit lines are more likely to be in v iolation of a debt covenant This is another indication that their financial situation may be less stable. I also find that f irms that draw on their credit lines have lower market to book ratios than those that do not. This could be indicative of credit li ne users having more stable operations made up of fewer growth opportunities, which is potentially supported by the lower investment in R&D spending for firms with credit line drawdowns. However, a lower market to book ratio is also consistent with firms h aving a lower valuation due to concerns about financial constraint. I also examine characteristics that I expect to be important based on the findings of Sufi (2009). Sufi (2009) argues that firms with higher and more stable cash flows are more able to obt ain and maintain a line of credit. Therefore, I should expect credit lines to be a less contingent (and more utilized) source of liquidity for those firms with higher and more stable cash flows. In addition, I expect firms that use credit line drawdowns as a substitute source of liquidity to hold less cash. I find that companies that draw on their credit lines hold less cash than those that do not draw This could be due to the credit line being utilized because there is a cash shortage or because the firms credit lines as liquidity substitutes Consistent with the findings in Sufi (2009) firms that draw on their lines have lower cash flow volatility than those that do n o t draw. However,

PAGE 28

28 companies that draw on their credit lines have lower ca sh flows than those with no drawn amount outstanding So while firms need high cash flows to gain access to a credit line, I find that those firms that utilize their credit lines tend to have lower cash flows. I also see that firms with drawn amounts outst anding have lower current ratios, higher leverage, and lower net worth than those that do not have any amount outstanding. While I find that the higher leverage and lower net worth are due solely to the difference in the amount outstanding on the credit li ne when I adjust for drawn amounts, the difference in current ratios hold when I examine non revolving debt, too. Results Table 2 3 provides initial insights into the relationship between credit line dra wdowns and future cash flows. I employ an OLS regres sion in which the dependent variable is cash flow in fiscal year t+1. I investigate the effect of drawdowns on future cash flows after controlling for firm size, cash holdings, and market to book ratio. 11 In this table and all others, I do not argue that dr awdowns cause a decline in cash flows, but rather that firms draw down on their credit lines in anticipation of negative future expectations about future cash flow performance and based on the results of Sufi (2009), future credit line availability. 12 I measure cash flow as EBITDA (oibdp in Compustat) scaled by lagged non cash assets because EBITDA it is the most common measure of cash flow used in covenants on lines of credit (Sufi (2009)). Since cash 11 I am unable to pinpoint the exact timing of the drawdown because I observe only a snapshot of fiscal year end drawn amounts. However, using annual data adds noise and biases against finding results since I may miss the period in which performance changes or violations occur. 12 or contracted upon by t he lender. Therefore, credit line access is not limited before future cash flows are realized.

PAGE 29

29 flows are highly serially correlated, I include prio r cash flows as control variables in Models 2 and 3. As expected, past cash flows are a strong predictor of future cash flows. Even after controlling for the cash f low path, the results of Table 2 3 show that credit line drawdowns are negatively related to future cash flows. As I would expect, the inclusion of lagged cash flows leads to an increase in the explanatory power of the model from an adjusted R 2 of 13.5% in Model 1 to an adjusted R 2 of approximately 61.5% in Models 2 and 3. Using the coefficients from Model 2, a two standard deviation increase in the drawn amount outstanding is associated with a decline in future cash flows of 0.012, or 9.5% at the mean of Cash Flow t+1 These regressions also show that cash flows as a fraction of lagged non cash as sets are increasing in firm size and market to book. I investigate whether drawdowns increase prior to declines in cash flow by modeling an OLS regression in which the dependent variable is the change in cash flow from fiscal year t to fiscal year t+1. Th e results are presented in Table 2 4 Note that this is not a first difference regression, as I am interested in the coefficient on the change, not the level of the outstanding credit line balance I include size and market to book as control variables be ca use I expect that small growth companies are more likely to have volatile cash flows and experience large cash flow changes. I find that market to book behaves as I would expect, but size does not. In Model 1 of Table 2 4 I find that increased drawdown s are associated with negative changes in cash flows from fiscal year t to t+1. I then segment observations by whether the firm increases or decreases the drawn amount outstanding. In the overall sample in Model 1, a change in drawn amounts outstanding tha t is one standard

PAGE 30

30 deviation above its mean is associated with a change in future cash flows of 0.003. As the mean change in future cash flows is 0.00199, this is 150% larger than the mean change in future cash flows. I include the current cash flows and the change in cash flows from year t 1 to year t in order to control for prior performance, as cash flow measures exhibit mean reversion (Nini, Smith and Sufi (2009)). 13 The availability hypothesis predicts that firms increase their drawn amounts prior to c ash flow declines, but it does not predict the opposite relationship between repayments and cash flows. In Model 2, I find that increases in drawn amounts outstanding are predictive of declines in future cash flows. Consistent with the explanatory power co ming from increases in drawn amounts outstanding, both the adjusted R 2 and the size of the coefficient on Draw are larger in Model 2 than in Models 1 and 3. Models 3 shows that decreases in drawn amounts are statistically unrelated to the change in future cash flows. Unexpected Drawdowns So far I have only examined the relationship between drawdowns and future operating performance. However, the relationship between realized drawdowns and future cash flows may arise from either a precautionary motive for drawdowns or an actual need for the cash from drawdowns. I would like to separate cash needs from a precautionary motive for credit line drawdowns. In Table 2 5 I examine the determinants of drawdown activity to predict the component that is based on immediate cash needs. Because drawn amounts are non negative and 43.4% of the firm year observations have no drawn amount outstanding, I employ a Tobit model when 13 The inclusion of Cash Flow t and Cash Flow t creates an endogeneity problem, as Cash Flow t appears in both the dependent and independent variables; however, I a m not interested in examining these coefficients, so I am not concerned that this is a problem. The results are also robust to their exclusion.

PAGE 31

31 estim ating the effect of explanatory variables on the drawn amount. Unlike a traditional ordinary least squares model, the Tobit model accounts for left censoring of the dependent variable at zero. 14 Following Campello, Giambona, Graham and Harvey (2011), I cons ider firm size, cash holdings, cash flows, investment opportunities and credit ratings in determining draw down activity. I also include capital expenditures and the change in net working capital. I hypothesize that companies with greater investment activi ty will have a greater need for cash from credit lines. In addition, firms with a greater need for cash because of a decline in non cash working capital may need to use their credit line to make up the difference. In Model 3, I drop the year fixed effects and explicitly control for credit market conditions and the US annual GDP. I expect that firms will draw more on their lines when credit markets are tight and they have less access to outside capital. Worse general market conditions, as proxied by US GDP, may also increase the need for cash from lines of credit. Consistent with the results in the existing literature and t he summary statistics in Table 2 2 Table 2 5 shows that smaller firms draw more, those with more cash draw less, and firms with high cas h flows draw less. 15 In addition, companies with greater capital expenditures draw more on their credit lines and rated companies tend to draw less on their lines. This is consistent with rated firms having less expensive and more 14 While I believe that the Tobit model is theoretically justified, I acknowledge that it imposes strict assumpti ons about the normality of the dependent variable and the independence of error terms in order to obtain consistency of the estimators. In unreported tests, I confirm that my results are statistically and economically robust to the use of an OLS model to p redict credit line drawdowns. 15 As I noted in discussing the summary statistics, several of these variables have been associated with financial constraints in other papers. In unreported results, I directly include the Kaplan and Zingales (1997) KZ index and the Altman (1968) Z score as a measure of financial constraint. The KZ index is not statistically significant in my model. Consistent with the idea that more financially constrained firms are more likely to utilize their credit lines, I find that a dec line in the Z score predicts larger drawdowns. All of my other results are robust to including the Z score in the first stage.

PAGE 32

32 available sources of credi t. I also find that time effects are important. In most specifications, time effects are accounted for in a year fixed effect. When I drop year fixed effects to examine the source of annual variations, I specifically see that firms tend to draw more on the ir credit lines when credit markets are tight and when the economy is worse, as measured by lower GDP. Year fixed effects may also include current lending rates and expectations about future rates as these could influence the relative attractiveness of dra wing on credit lines today. In Model 4, I also include the lagged value of Drawn to account for possible stickiness in drawn amounts outstanding over time. This stickiness could be due to a lack of cash for repayments or a company specific sentiment regard ing credit line usage, such as an aversion to debt financing. While I do see that prior year drawn amounts are predictive of current year drawn amounts, my other findings remain unchanged. I use the estimates from my model of predicted credit line drawdown s to construct a measure of unexpected draws, i.e. the difference between the predicted and realized drawn amounts. 16 This forms the basis for all of the analysis to follow. I call the residual from the predictive regressions in Table 2 5 Unexpected Draw. 17 As shown in Table 2 1 the mean of the change in drawn amount is approximately zero. On average, firms hold approximately 24% of their total line as an outstanding balance in a given year, but do not change the outstanding amount from year to year. Based o n my predictive model, a 16 The Tobit model supposes that there is an unobservable variable y* that is linear in x but is unobserved for certain values. In my case, the observed y = max(0, y*). The predicted values are computed for the observed variable, E(y|x) as All of the regression results using Unexpected Draw are statistically and economically robust to using the estimation of the latent variable E(y*|x) for the predicted values. 17 I acknowledge the problem of estimating standard errors for gene rated regressors and have concluded that it is not a concern in my situation. According to Pagan (1984), regression results are consistent when including only the residual from a predictive regression as an independent variable.

PAGE 33

33 positive value for Unexpected Draw could mean that a firm actively draws more than expected, or that they fail to repay their lines when I would otherwise expect them to. This should result in a cash hoard. I utilize unexpected dr awn amounts because I believe that they are a reasonable proxy for precautionary balances drawn by firms. Since holding drawn amounts is costly as the firm must begin paying interest upon drawing on the line and firms can draw on their credit line at will, I would expect to see that firms draw for immediate cash deployment. However, I would expect that companies who draw preemptively on their credit line do not utilize all of the cash immediately. Therefore, I examine the effect of Unexpected Draw on cash h oldings to confirm its use as a proxy for preemptive drawdowns. In Table 2 6 I follow the model of Opler, Pinkowitz, Stulz and Williamson (1999), hereafter OPSW, and construct a model of cash holdings. The model numbers correspond to the model numbers in Table 2 5 such that Unexpected Draw in Model 1 of this table is the residual from Model 1 in Table 2 5 The coefficients on the explanatory variables in my model are consistent with the results in OPSW in directionality; however, several variables that ar e statistically significant in OPSW are insignificant in my model. These include cash flow, net working capital, leverage and industry cash flow volatility. My results indicate that small, growth, and non dividend paying companies tend to hold more cash. I n addition, companies with greater capital expenditures and those with greater research and development expenditures tend to hold more cash. In addition to the explanatory variables used in OPSW, I also include unexpected drawdowns as a predictor of cash h oldings. It is important to remember that

PAGE 34

34 the general relationship between cash and drawdowns is negative, since holding cash and utilizing lines of credit are generally substitute sources of liquidity. However, unexpected drawdowns are positively and sign ificantly related to cash holdings. In other words, firms with unexpected drawdowns hold larger cash balances than I would otherwise expect. This is consistent with the firm holding some of the drawn amounts in the form of cash. This evidence lends further support to the use of unexpected drawdowns as a proxy for precautionary, or preemptive, drawdowns. In Table 2 7 I examine the relationship between unexpected drawdowns and future cash flows through the use of an OLS model. As before, the model numbers in Table 2 7 correspond with the model numbers of the predictive regressions in Table 2 5 That is, the unexp ected drawdowns used in Table 2 7 are from the respective Table 2 5 predictive regression. In examining future cash flows, I control for size, market to book, cash holdings, and previous cash flows. I also include the variables used to predict drawdown activity in the Tobit model in Table 2 5 This ensures that the unexpected draws that I measure are truly orthogonal to the vari ables used in the predictive regression. 18 I find that unexpected credit line drawdowns are associated with declines in future cash flows. Table 2 7 demonstrates this relationship both for my sample and for the publicly available Sufi (2009) sample of credi t line drawdowns. The Sufi (2009) sample is collected using the same method that I employ but on a different random sample of 300 companies from 1996 2003. These results demonstrate that my results are not time period or sample specific. In unreported re sults, I find that the negative and significant relationship between unexpected drawdowns and cash flows is 18 My results in this and fu ture tables are quantitatively and qualitatively similar when I exclude the first stage variables.

PAGE 35

35 limited to the positive component of unexpected drawdowns. That is, consistent with my hypothesis that companies draw precautionary balances on thei r line of credit prior to poor future cash flow performance, unexpected drawdowns are statistically significantly related to cash flows only when the company holds larger drawn amounts than would be expected. 19 In Table 2 8 I rerun the regressions from Ta ble 2 7, but I drop the oth er control variable from Table 2 5. I include the predicted drawdown from Table 2 5 instead I find that the predicted, or expected, portion of drawdowns has no predictive power on future cash flows, but I continue to see a negat ive and significant relationship between unexpected drawdowns and future cash flows. As I have previously stated, I consider the future realizations of cash flows to be a r elated to credit line availability, an expectation of declines in future operating performance would be indicative of a possible decline in future availability. The concern about future availability may then trigger management to preemptively draw on their credit line. However, a decline in cash flow is just one predictor of a covenant violation. Other common covenant restrictions include specifying a maximum debt ratio or a minimum net worth or current ratio. In Table 2 9 I also examine the relationship b etween unexpected drawdowns and these other covenant terms. The use of these covenant terms presents a problem, however, as they are all mechanically related to debt levels. Since drawdowns increase debt outstanding, I look at each of these variables net o f drawn amounts outstanding. Exact variable definitions can be found in 19 I also partition the sample into pre and post 2008 and find that these results hold both during and after the financial crisis.

PAGE 36

36 the Appendix Table 2 9 presents results using Unexpected Draw based on the residuals from Model 1 of Table 2 5 As before, I include the first stage variables from Table 2 5 and also co ntrol for two lags of the dependent variable. The results are quantitatively and qualit atively similar using Models 2 through 4 from Table 2 5 to define Unexpected Draw. The results in Table 2 9 show that an increase in unexpected drawdowns during year t p redicts a statistically significant decline in net worth in year t+1. This is consistent with my hypothesis that unexpected drawdowns occur in advance of deterioration in covenant related variables. I also find that an increase in unexpected drawdowns pred icts a decline in future current ratios. 20 In the final column, I model future leverage. Once again, I net drawn amounts outstanding from my measure of leverage and scale by lagged non cash assets. Unlike my previous results, I find no evidence that an incr ease in unexpected drawdowns is predictive of an increase in future leverage. This is not surprising if I view unexpected drawdowns as preemptive actions. A new debt issue would require additional scrutiny by a new lender. If the company anticipates deteri orating performance, then they are unlikely to obtain new debt financing at favorable terms, if at all. Thus far I have provided evidence consistent with the hypothesis that companies draw on their credit lines prior to a decline in cash flow, net worth an d current ratio. My initial analysis focuses on cash flow performance because it is the most strongly related 20 The current ratio is mea sured as the ratio of current assets to current liabilities where current liabilities are reduced by the total drawn amount outstanding. This assumes that all drawdowns are classified as current debt due to mature within one year. While I know that this is not the case, the adjustment yields a conservative estimate of the current ratio, as an increase in drawdowns absent any other change would result in a current ratio that is too high.

PAGE 37

37 (Sufi (2009)). All of this analysis is based on t he idea that managers have some prediction about a worsening of these covenant triggers and are concerned about the future availability of credit lines. In Table 2 10 instead of looking at covenant triggers, I directly examine whether unexpected drawdown s predict covenant violations after controlling for the typical variables on which covenants are written. I utilize a probit model in which the dependent variable is a binary variable equal to one if there is a new covenant violation during the fiscal year, and zero otherwise. I control for predictors of covenant violations, which include cash flows, the current ratio, leverage and net worth. Market to book and size are also included because I expect that large and value firms tend to generate more stable cash flows and have less need for credit lines. I want to know the effect of unexpected credit line drawdowns on the probability of a future covenant violation; however, drawdowns are mechanically related to most of the variables that I would expect to predict covenant violations because an increase in credit line drawdowns g. To account for this, I remove the total drawn amount outstanding from the debt amounts in the leverage and current ratio. As detailed earlier, while I do not have information of the portion of the drawn amount that is due in less than one year, I assume that the total drawn amount is included in current liabilities. This may frequently be the case due to the short term nature of many credit line agreements; however, I expect that this assumption will result in an overstatement of the current ratio. The r esults are quantitatively and qualitatively similar if I exclude the current ratio from th e model. The results of Table 2 10 confirm the strong relationship

PAGE 38

38 between cash flows and covenant violations and also show that an increase in unexpected drawdowns i ncreases the probability of a covenant violation. 21 If credit line drawdowns are important predictors of future cash flow performance and the probability of future covenant violations, then creditors should care about corporate credit line usage. In Table 2 11 I examine whether corporate credit ratings are related to unexpected drawdowns. I look at the long term issuer credit rating issued downgrade as a decline in the corpor ate credit rating over any month in the fiscal year. Recall from Table 2 1 that approximately one third of my sample has a credit rating. Of the observations that have a credit rating, approximately 42% of the ratings are investment grade (i.e. above BBB ) Table 2 11 presents probit models predicting the probability of a downgrade in credit rating during the fiscal year of the drawdown for each of the definitions of Unexpected Draw found in Table 2 5 22 Following the literature on the determinants of rating s (i.e. Pogue and Soldofsky (1969)), I control for cash flow, industry cash flow volatility, size, market to book and leverage. As before, I measure leverage net of drawdowns. I also include the variables from the predictive models in Table 2 5 The relati onships between the control variables and the probability of downgrade are consistent with the literature. Higher cash flows decrease the probability of a downgrade, while increases in leverage increase the probability of a downgrade. 21 Table 2 10 reports both coefficients and marginal effects for the pro bit model. The inclusion of year and industry fixed effects make the interpretation of marginal effects for the probit model difficult. I also run a linear probability model to aid interpretation and find that the marginal effect of Unexpected Draw is rema rkably similar (0.101, 0.091, 0.091 and 0.149 in Models 1 4, respectively). 22 In unreported results, I also fit an OLS model where I examine the impact of Unexpected Draw on the number of notches that a firm is downgraded in a given year. The results are similar in statistical importance.

PAGE 39

39 Cash and CapEx factor significantly into the model, as well. Consistent with intuition, an increase in cash holdings decreases the probability of a downgrade. Increased capital expenditures may decrease the probability of a downgrade since the company should only invest in cap ital improvements with positive expected future returns. Using the marginal effects from Model 3, a change in unexpected draw from one standard deviation below to one standard deviation above the mean implies an increased probability of a downgrade during the year of 6.4%. Compared to the mean probability of downgrade of 17%, that is an increase in the probability of downgrade of 38%. 23 Lender Response The evidence thus far is consistent with the argument that firms draw on their lines in order to access cas h before a decline in availability. I am also interested in the engage in strategic credit line usage. The first two columns o f Table 2 12 present results that replicate Ta ble 6 in Sufi (2009). Consistent with the findings in Sufi (2009), I find that a covenant violation decreases the total size of the credit line and the available (unused) portion of the credit line. In addition to examining the effect of a covenant violati on on the total and unused portion of the line, as in Sufi (2009), I also test the effect of a covenant violation on drawn amounts outstanding. The results indicate that a covenant violation does not result in a statistically significant decrease in the dr awn amount outstanding, implying that outstanding amounts are not required to be immediately repaid following a violation. Additionally, there is empirical evidence that 23 Table 2 11 reports both coefficients and marginal effects for the probit model predicting credit line downgrades. The inclusion of year and industry fixed effects make the interpretation of marginal effects for the probit model difficult. I also run a lin ear probability model to ease the interpretation issue and find that the comparable marginal effect of a one unit change in Unexpected Draw are 0.291, 0.423, 0.421 and 0.547 in Models 1 4, respectively.

PAGE 40

40 covenant violations rarely lead to a demand for immediate repayment or a complete canc ellation of the line (e.g. Campello et al (2010)). In unreported tests, I also find that unexpected drawdowns are not related to the probability that a line is cancelled or fully repaid. This leads me to conclude that creditors do not respond harshly to d rawdowns. I also want to investigate whether companies are able to enhance their strategic bargaining position by increasing the drawn amount outstanding prior to violating a covenant. A loan renegotiation may change the covenant structure of the loan. Nini, Smith and Sufi (2009) examine the prevalence of covenants that restrict capital expenditures and find that 46% of firms in their sample have a capital expenditure restriction at some point. Moreover, the probability of a restrictive covenant on capital expenditures is larger after a covenant violation. Nini, Smith and Sufi (2011) look at the changes in several operating variables after covenant violations and show evidence that capital expenditures decline following a covenant violation. In Table 2 13 I utilize the ir regression specification to examine whether the decrease in capital expenditures subsequent to a covenant violation is mitigated when the firm preemptively draws on its credit line. Consistent with this explanation, I find that the interaction of a new covenant violation and unexpected drawdowns is positive and significant. This result is fairly consistent across the first three models, which correspond with the predicted drawdown regressions in Table 2 5 but are less statistically significant in Model 4. The results for the control variables are consistent in sign and magnitude with the results in Nini, Smith and Sufi (2011). Nini, Smith and Sufi (2011) interpret the decline in capital expenditures following covenant violations as evidence that lenders gain control rights following contract violations. However, the decline in capital expenditure following a covenant

PAGE 41

41 violation could also be the result of the company experiencing a cash shortage. In that case, I would expect to see that unexpected drawdown s mitigate capital expenditure declines because they provide the company with cash prior to the violation. To further explore the relationship between covenant violations, unexpected drawdowns and creditor reactions, I directly examine changes in loan terms using the Deal S can database. I keep fiscal year observations for which there is an existing credit line contract and there is a new contract in the next year. Table 2 14 provides summary statistics on the change in select loan terms when a new credit line contract is reported in Deal S can. Variable definitions can be found in the Appendix. On average, fees on new credit line contracts change little (All In Drawn), probability of being secured, probability of a borrowing base and number of lenders compa red to the existing contract terms. New credit line contacts do see a statistically significant increase in maturity and the size of the total line, on average. Table 2 15 presents summary statistics and mean difference tests for changes in credit line con tract terms by whether or not the firm violated a loan covenant in the current year. Consistent with expectations, I find that terms generally worsen for covenant violators as compared to non violators. Specifically, I find that covenant violators see a la rge and statistically significant increase in All In Drawn fees of 29 basis points, which is 34 basis points larger than the change in fee facing non violators. Covenant violators also see a statistically significantly smaller increase in the total size of the credit line upon the introduction of a new contract compared to non violators. In addition, there is an increased probability that the new credit line becomes secured and contains a new borrowing base agreement when the firm has violated a covenant,

PAGE 42

42 c ompared to those firms who have not violated a covenant. Finally, I find that there are, on average, 5 fewer lenders in the loan syndicate following a violation while there is an insignificant change in the number of lenders for non violators. Table 2 16 p rovides the results of splitting the sample by Unexpected Draw. I see that those firms with positive unexpected drawdowns experience a 13 basis point higher fee upon a new loan than those who do not have positive unexpected drawdowns. I also find that thos e firms with positive unexpected drawdowns experience a $60 million lesser increase in the size of their total credit line and are more likely to have a new borrowing base compared to firms with non positive unexpected drawdowns. There is no statistically significant difference in the proportion of credit lines set to expire in the current fiscal year between the two groups. The evidence does not indicate that firms who draw unexpectedly on their credit lines are in a better bargaining position than those t hat do not. If anything, those firms with unexpected drawdowns see worse terms. Table s 2 17 and 2 18 take the analysis one step further to see how credit line terms change when I look at both covenant violations and unexpected drawdowns. Table 2 17 presen ts the mean difference tests for changes in credit line terms between firm observations with positive and non positive unexpected drawdowns when the firm does not experience a covenant violation in the current year. Absent a covenant violation, new credit lines for firms with positive unexpected drawdowns see a larger increase in All In Drawn fees, a lesser increase in the size of the total line and a higher incidence of new borrowing base restrictions. Absent a covenant violation, firms who take out new cr edit lines following unexpected drawdowns tend to see relatively worse changes in their credit line terms when compared to those without unexpected

PAGE 43

43 drawdowns. This is consistent with the findings of Chen, Hu and Mao (2011) that borrowers are penalized with higher loan spreads and annual fee on new credit lines issued after drawdowns. Table 2 18 examines the relationship between unexpected drawdowns and contract changes when there is a covenant violation in the current year. While I found in Table 2 15 that the changes in contract terms are generally worse when a firm violates a covenant, I do not find extensive evidence that the terms for covenant violators differ substantially for those firms who have high unexpected drawdowns. The only statistically signi ficant difference that I find is in the probability that the new credit line contract will be newly secured. The takeaway from these results is that firms with large unexpected drawdowns are not treated more favorably by their lender. If anything, they ar e subject to worse terms than their counterparts without unexpected drawdowns, though most of the differences in term changes disappear when I control for the effect of covenant violations on term changes. In addition to the actual fees paid on drawn amoun ts outstanding, the potential for a tightening of contract terms adds a layer of potential costs for drawing unexpectedly on an existing credit line. However, firms that draw unexpectedly on their credit lines do not experience as large of a decline in cap ital expenditures following a covenant violation. This ability to continue funding capital expenditures highlights the potential benefits of drawing on the line in advance of a covenant violation. I interpret the mitigating effect of unexpected drawdowns o n future capital expenditure declines not as a more lax stance by creditors but as a byproduct of

PAGE 44

44 the early drawdown. Following an unexpected drawdown, the firm has a larger cash reserve and is not forced to cut capital expenditures by as much. Conclusion The existing literature has established that cash flows are an important predictor of covenant violations and that covenant violations are an important determinant of credit line availability (Sufi (2009)). Furthermore, Roberts and Sufi (2009) show that ca sh flows are an important predictor of debt renegotiations outside of covenant violations. If the ability to draw on a line of credit is conditional on cash flow performance, then I expect corporations to manage their liquidity according to their future ca sh flow expectations. I have shown evidence that credit line drawdowns are predictive of declining future cash flows. The availability hypothesis argues that these drawdowns are motivated by precautionary reasons as firms recognize that credit lines are a contingent source of liquidity and may not be available in poor cash flow states In order to identify balances drawn for precautionary motives, I construct a predictive regression of drawdown amounts and use the residuals from the model as a measure of u nexpected drawdowns. To test whether my measure of unexpected drawdowns provides information about preemptive balances withdrawn by the firm, I test their impact on cash holdings. I find an increase in unexpected drawdowns implies an increase in cash. This is consistent with the drawdown being held for precautionary purposes as opposed to meeting immediate cash needs. Consistent with the availability hypothesis, unexpected drawdowns are predictive of decreased future cash flows, decreased future net worth, a lower current ratio and an increased probability of a covenant violation. Moreover, I find that the probability of a

PAGE 45

45 credit rating downgrade is increasing in unexpected draws This is consistent with anecdotal evidence that ratings agencies care about drawdown activity My findings also show that lenders do not cancel credit line agreements and firms are not required to immediately repay outstanding amounts upon the violation of a covenant. I find evidence that borrowers who draw preemptively from thei r line generally see the terms of their credit agreement change against them. However, if there is a covenant violation, then terms worsen equally for those firms who did and did not draw preemptively on their credit line. Additionally, firms with preempti ve drawdowns tend to see a lesser decline in future capital expenditures following a covenant violation. Thus the preemptive access to cash seems to provide a net benefit for the average firm.

PAGE 46

46 Table 2 1. S ummary statistics N Median Mean St d Dev 10th Pctl 90th Pctl Credit Line Variables Draw Dummy 5311 1.000 0.567 0.496 0.000 1.000 Drawn Outstanding ($MM) 5311 1.350 56.572 263.683 0.000 115.000 Drawn 5269 0.011 0.072 0.104 0.000 0.246 4545 0.000 0.000 0.054 0.072 0.069 Undrawn 5311 32.600 257.650 958.300 0.800 510.025 Total 5311 60.000 352.255 1172.673 3.000 758.300 Drawn/Total 5311 0.075 0.241 0.303 0.000 0.727 Covenant Violation 3963 0.000 0.166 0.372 0.000 1.000 New Covenant Violation 3963 0.000 0.100 0.300 0.000 0.000 Firm Characteristics Non Cash Assets ($MM) 5309 310.599 1625.583 3222.265 16.452 5368.110 Size 5274 5.668 5.583 2.070 2.717 8.535 Cash Flow 5262 0.134 0.127 0.171 0.081 0.334 5190 0.000 0.002 0.122 0.116 0.096 Industry Cash Flow Volatility 5309 0.104 0.112 0.054 0.052 0.210 Cash 5270 0.084 0.210 0.295 0.007 0.650 CapEx 5246 0.042 0.069 0.072 0.011 0.179 Net Working Capital 5163 0.088 0.096 0.229 0.192 0.406 Leverage 5269 0.253 0.311 0.273 0.000 0.722 Adj Leverage 5269 0.160 0.232 0.252 0.000 0.620 Current Ratio 5201 1.831 2.156 1.261 0.818 4.025 Adj Current Ratio 5197 2.090 2.471 2.126 0.419 5.473 Net Worth 5269 0.789 0.777 0.326 0.315 1.189 Adj Net Worth 5269 0.871 0.855 0.332 0.399 1.27 Market to Book 5075 1.569 2.130 1.550 0.902 4.431 This table presents summary statistics for a random sample of firms with credit lines from 2000 2010. Draw Dummy is a binary variable equal to one if there is some non zero drawn amount outstanding. Drawn Outstanding ($MM) is the drawn amount reported at the end of the fiscal year, in millions of dollars. Drawn is Drawn Outstanding as a fraction of lagged non cash assets. Undrawn is the amount available for future drawdowns as a fraction of lagged non cash assets. Total is the con tracted size of the credit line as a fraction of lagged non cash assets. Cash Flow, Cash, CapEx, Net Working Capital, Leverage, Net Worth, Adj Leverage, Adj Net Worth, Acquisitions and R&D are all measured as a proportion of lagged non cash assets. Adj Lev erage, Adj Net Worth, and Adj Current Ratio are measured net of drawdowns. Industry Cash Flow Volatility is the median of the standard deviation of the cash flows of all of the firms in the same two digit industry over a ten year period. Dividend Payer, Ra ted and Investment Grade are all binary variables equal to one when the firm is a dividend payer, rated by S&P and rated investment grade, respectively. Credit Market Tightness is measured as the average over the four quarters of the fiscal year of the net fraction of domestic banks who respond to the Federal Reserve Bank study by saying that they are tightening standards for C&I loans to large and medium sized firms. Specifics of variable definitions can be found in the Appendix variable of interest from year t 1 to t, unless otherwise specified. All ratios are winsorized at the 5th and 95th percentiles to reduce the effect of outliers. 10th Pctl and 90th Pctl are the 10th and 90th percentiles of observations, respectively.

PAGE 47

47 Tab le 2 1. Continued N Median Mean Std Dev 10th Pctl 90th Pctl Dividend Payer 5299 0.000 0.298 0.457 0.000 1.000 Rated 5311 0.000 0.321 0.467 0.000 1.000 Investment Grade 5311 0.000 0.135 0.341 0.000 0.000 Credit Market Tightness 5311 0.082 0.147 0.262 0.204 0.503 Acquisitions 5018 0.000 0.030 0.070 0.000 0.116 R&D 3127 0.028 0.088 0.138 0.000 0.289 This table presents summary statistics for a random sample of firms with credit lines from 2000 2010. Draw Dummy is a binary variable equal to one if there is some non zero drawn amount outstanding. Drawn Outstanding ($MM) is the drawn amount reported at the end of the fiscal year, in millions of dollars. Drawn is Drawn Outstanding as a fraction of lagged non cash assets. Undrawn is the amount available for future drawdowns as a fraction of lagged non cash assets. Total is the contracted size of the credit line as a fraction of lagged non cash assets. Cash Flow, Cash, CapEx, Net Working Capital, Leverage, Net Worth, Adj Leverage, Adj Net Worth, Acquisiti ons and R&D are all measured as a proportion of lagged non cash assets. Adj Leverage, Adj Net Worth, and Adj Current Ratio are measured net of drawdowns. Industry Cash Flow Volatility is the median of the standard deviation of the cash flows of all of the firms in the same two digit industry over a ten year period. Dividend Payer, Rated and Investment Grade are all binary variables equal to one when the firm is a dividend payer, rated by S&P and rated investment grade, respectively. Credit Market Tightness is measured as the average over the four quarters of the fiscal year of the net fraction of domestic banks who respond to the Federal Reserve Bank study by saying that they are tightening standards for C&I loans to large and medium sized firms. Specifics o f variable definitions can be found in the Appendix interest from year t 1 to t, unless otherwise specified. All ratios are winsorized at the 5th and 95th percentiles to reduce the effect of outliers. 10th Pctl and 90 th Pctl are the 10th and 90th percentiles of observations, respectively.

PAGE 48

48 Table 2 2. Summary statistics by draw dummy. Drawn > 0 Drawn = 0 N Mean N Mean Difference Credit Line Variables Drawn 2983 0.127 2286 0.000 0.127*** Drawn Outstanding ($MM) 3009 99.851 2302 0.000 99.851*** Drawn/Total 3009 0.426 2302 0.000 0.426*** 2532 0.009 2013 0.012 0.021*** Covenant Violation 2287 0.121 1676 0.072 0.049*** New Covenant Violation 2287 0.206 1676 0.111 0.095*** Firm Characteristics Non Cash Assets ($MM) 3008 1426.082 2300 1887.141 461.059*** Cash 2983 0.129 2286 0.315 0.186*** 2932 0.002 2257 0.002 0.000 Cash Flow t+1 2774 0.104 2091 0.156 0.052*** Cash Flow 2980 0.102 2281 0.159 0.057*** Cash Flow t 1 2934 0.105 2259 0.161 0.056*** Industry Cash Flow Volatility 3008 0.110 2300 0.115 0.005*** CapEx 2968 0.070 2277 0.068 0.003 2451 0.006 1946 0.005 0.001 Market to Book 2868 1.897 2206 2.432 0.535*** Net Working Capital 2906 0.086 2256 0.107 0.021*** Leverage 2983 0.371 2286 0.234 0.137*** Adj Leverage 2983 0.235 2286 0.229 0.006 This table presents a comparison of the mean characteristics of firms when Drawn is positive or zero. Draw Dummy is a binary variable equal to one if there is some non zero drawn amount outstanding. Drawn Outstanding ($MM) is the drawn amount reported at t he end of the fiscal year, in millions of dollars. Drawn is Drawn Outstanding as a fraction of lagged non cash assets. Undrawn is the amount available for future drawdowns as a fraction of lagged non cash assets. Total is the contracted size of the credit line as a fraction of lagged non cash assets. Cash Flow, Cash, CapEx, Net Working Capital, Leverage, Net Worth, Adj Leverage, Adj Net Worth, Acquisitions and R&D are all measured as a proportion of lagged non cash assets. Adj Leverage, Adj Net Worth, and A dj Current Ratio are measured net of drawdowns. Industry Cash Flow Volatility is the median of the standard deviation of the cash flows of all of the firms in the same two digit industry over a ten year period. Dividend Payer, Rated and Investment Grade ar e all binary variables equal to one when the firm is a dividend payer, rated by S&P and rated investment grade, respectively. Credit Market Tightness is measured as the average over the four quarters of the fiscal year of the net fraction of domestic banks who respond to the Federal Reserve Bank study by saying that they are tightening standards for C&I loans to large and medium sized firms. Specifics of variable definitions can be found in the Appendix interest from y ear t 1 to t, unless otherwise specified. All ratios are winsorized at the 5th and 95th percentiles to reduce the effect of outliers. I test the null hypothesis that the means for the two groups are equal using t tests. I assume unequal variances for t tes ts. ***, **, and denote significance in differences between the sample means at the 1%, 5%, and 10% levels, respectively.

PAGE 49

49 Table 2 2. Continued Drawn > 0 Drawn = 0 N Mean N Mean Difference Current Ratio 2931 1.854 2270 2.545 0.691*** Adj Current Ratio 2928 2.323 2269 2.662 0.339*** Net Worth 2984 0.725 2286 0.847 0.122*** Adj Net Worth 2983 0.854 2286 0.857 0.011 Dividend Payer 3002 0.267 2296 0.338 0.072*** Rated 3009 0.292 2302 0.359 0.066*** Investment Grade 3009 0.108 2302 0.169 0.061*** Credit Market Tightness 3009 0.147 2302 0.147 0.000 R&D 1646 0.077 1481 0.100 0.023*** Acquisitions 2840 0.032 2177 0.029 0.003 This table presents a comparison of the mean characteristics of firms when Drawn is positive or zero. Draw Dummy is a binary variable equal to one if there is some non zero drawn amount outstanding. Drawn Outstanding ($MM) is the drawn amount reported at t he end of the fiscal year, in millions of dollars. Drawn is Drawn Outstanding as a fraction of lagged non cash assets. Undrawn is the amount available for future drawdowns as a fraction of lagged non cash assets. Total is the contracted size of the credit line as a fraction of lagged non cash assets. Cash Flow, Cash, CapEx, Net Working Capital, Leverage, Net Worth, Adj Leverage, Adj Net Worth, Acquisitions and R&D are all measured as a proportion of lagged non cash assets. Adj Leverage, Adj Net Worth, and A dj Current Ratio are measured net of drawdowns. Industry Cash Flow Volatility is the median of the standard deviation of the cash flows of all of the firms in the same two digit industry over a ten year period. Dividend Payer, Rated and Investment Grade ar e all binary variables equal to one when the firm is a dividend payer, rated by S&P and rated investment grade, respectively. Credit Market Tightness is measured as the average over the four quarters of the fiscal year of the net fraction of domestic banks who respond to the Federal Reserve Bank study by saying that they are tightening standards for C&I loans to large and medium sized firms. Specifics of variable definitions can be found in t he Appendix interest from y ear t 1 to t, unless otherwise specified. All ratios are winsorized at the 5th and 95th percentiles to reduce the effect of outliers. I test the null hypothesis that the means for the two groups are equal using t tests. I assume unequal variances for t tes ts. ***, **, and denote significance in differences between the sample means at the 1%, 5%, and 10% levels, respectively.

PAGE 50

50 Table 2 3. Effect of drawdowns on future cash flows. (1) (2) (3) Drawn 0.150*** 0.058*** 0.054*** (0.047) (0.018) (0.018) Cash 0.087*** 0.046*** 0.046*** (0.026) (0.010) (0.010) Market to Book 0.026*** 0.011*** 0.012*** (0.005) (0.002) (0.002) Size 0.020*** 0.002** 0.001 (0.003) (0.001) (0.001) Cash Flow 0.709*** 0.652*** (0.018) (0.027) Cash Flow t 1 0.065*** (0.023) Constant 0.113*** 0.042*** 0.041*** (0.014) (0.008) (0.008) Industry Fixed Effect s Y Y Y Year Fixed Effect s Y Y Y Number of observations 4,637 4,634 4,596 Adjusted R2 0.135 0.617 0.616 This table presents the results of OLS regressions where the dependent variable is Cash Flow in fiscal year t+1. Drawn is the drawn amount outstanding on the credit line as a fraction of lagged non cash assets. Cash and Cash Flow are both measured as a fra ction on lagged non cash assets. Specifics of variable definitions are available in t he Appendix Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 51

51 T able 2 4. Effect of change in drawdowns on the change in future cash flows. (1) (2) (3) Sample All 0.056* 0.143** 0.014 (0.033) (0.072) (0.071) 0.033** 0.064** 0.016 (0.014) (0.028) (0.024) Market to Book 0.009*** 0.009** 0.005** (0.002) (0.004) (0.002) Size 0.004*** 0.000 0.004*** (0.001) (0.002) (0.001) Cash Flow 0.262*** 0.290*** 0.219*** (0.019) (0.032) (0.035) 0.075*** 0.122*** 0.002 (0.025) (0.045) (0.041) Constant 0.051*** 0.039** 0.020 (0.007) (0.019) (0.015) Industry Fixed Effect s Y Y Y Year Fixed Effect s Y Y Y Number of observations 3,948 1,165 1,332 Adjusted R2 0.189 0.234 0.138 This table presents the results of the OLS regressions where the dependent variable is the change in Cash Flow from year t to year t+1. All independent variables are measured at time t. Drawn is the drawn amount outstanding on the credit line as a fraction of lagged non cash assets. Cash and Cash Flow are both measured as a fraction on lagged non from year t 1 to t, unless otherwise specified. Specifics of variable definitions are available in the Ap pendix Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 52

52 Table 2 5. Tobit regression to predict credit line drawdowns. (1) (2) (3) (4) Cash Flow 0.142*** 0.173*** 0.172*** 0.089*** (0.029) (0.030) (0.030) (0.017) Size 0.020*** 0.010*** 0.011*** 0.004* (0.003) (0.004) (0.004) (0.002) Market to Book 0.002 0.004 0.004 0.000 (0.004) (0.004) (0.004) (0.002) Cash 0.199*** 0.213*** 0.217*** 0.123*** (0.023) (0.024) (0.024) (0.015) CapEx 0.216*** 0.250*** 0.189*** (0.066) (0.065) (0.040) 0.046** 0.049** 0.060*** (0.019) (0.019) (0.022) Rated 0.046*** 0.045*** 0.012* (0.013) (0.013) (0.007) Investment Grade 0.021 0.020 0.007 (0.016) (0.016) (0.008) Drawn t 1 0.854*** (0.024) Credit Market Tightness 0.015* (0.009) Annual GDP 0.009** (0.004) Constant 0.047*** 0.010 0.048 0.022 (0.014) (0.020) (0.050) (0.013) Industry Fixed Effects Y Y Y Y Year Fixed Effects Y Y N Y Number of observations 5,035 4,872 4,872 4,230 Pseduo R2 0.969 1.077 1.035 3.758 p value of F statistic 0.000 0.000 0.000 0.000 This table presents the coefficients of Tobit models in which the dependent variable is Drawn. Drawn is the drawn amount outstanding on the credit line as a fraction of lagged non cash assets. Cash Flow, Cash, CapEx and Net Working Capital are measured as a fraction on lagged non cash assets. Net Working Capital is working capital net of cash. Rated and Investment Grade are both binary variables equal to one when the firm is rated by S&P and rated investment grade, respectively. All independent variables ar interest from year t 1 to t, unless otherwise specified. Specifics of variable defini tions are available in the Appendix Standard errors are clustered by firm and repor ted in parentheses. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively

PAGE 53

53 Table 2 6. Effect of unexpected drawdowns on cash holdings. (1) (2) (3) (4) Unexpected Draw 0.152* 0.154** 0.165** 0.185** (0.078) (0.078) (0.078) (0.083) Market to Book 0.054*** 0.054*** 0.054*** 0.057*** (0.007) (0.007) (0.007) (0.008) Size 0.012** 0.008 0.008 0.006 (0.005) (0.006) (0.006) (0.007) Cash Flow 0.169** 0.168** 0.168** 0.165** (0.068) (0.068) (0.068) (0.075) Net Working Capital 0.068 0.074 0.073 0.104* (0.050) (0.050) (0.050) (0.055) CapEx 0.406*** 0.406*** 0.409*** 0.234 (0.143) (0.142) (0.143) (0.155) Adj Leverage 0.082* 0.072 0.072 0.081 (0.045) (0.047) (0.047) (0.054) Industry Cash Flow Volatility 0.163 0.161 0.161 0.145 (0.160) (0.161) (0.161) (0.168) Dividend Payer 0.044*** 0.036** 0.036** 0.039** (0.017) (0.018) (0.018) (0.020) Acquisitions 0.060 0.064 0.062 0.020 (0.080) (0.081) (0.081) (0.091) R&D 1.205*** 1.222*** 1.221*** 1.175*** (0.125) (0.124) (0.124) (0.142) Rated 0.019 0.018 0.026 (0.024) (0.024) (0.025) Investment Grade 0.027 0.027 0.030 (0.023) (0.023) (0.024) Industry Fixed Effect s Y Y Y Y Year Fixed Effect s Y Y Y Y Number of observations 2,830 2,816 2,816 2,440 Adjusted R2 0.569 0.573 0.574 0.548 This table presents the coefficients of OLS models where the dependent variable is Cash. Unexpected Draw is the residual from the predictive regression on drawdowns in Table 2 5. The model numbers in this table correspond to the predictive regression model numbers in Table 2 5. Cash, Cash Flow, CapEx, Net Working Capital, Leverage, Acquisitions and R&D are all measured as a proportion of lagged non cash assets. Industry Cash Flow Volatility is the median of the standard deviation of the cash flows of all of the firms in the same two digit industry over a ten year period. Dividend Payer, Rated and Investment Grade are all binary variables equal to one when the firm is a dividend payer, rated by S&P and rated investment grade, respectively. Specifics of variab le definitions can be found in the Appendix All independent variables are measured at time t unless otherwise specified. All models include an unreported constant term. Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 54

54 Table 2 7. Effect of unexpected drawdowns on future cash flows. 2000 2010 Credit Line Sample Sufi (2009) Credit Line Sample (1996 2003) (1) (2) (3) (4) (1) (2) (3) (4) Unexpected Draw 0.043** 0.048** 0.061*** 0.067** 0.118*** 0.120*** 0.122*** 0.148*** (0.018) (0.019) (0.020) (0.027) (0.030) (0.032) (0.032) (0.054) Cash 0.038*** 0.036*** 0.032*** 0.026** 0.073*** 0.081*** 0.079*** 0.065*** (0.010) (0.010) (0.010) (0.010) (0.021) (0.021) (0.022) (0.024) Cash Flow 0.655*** 0.658*** 0.658*** 0.685*** 0.570*** 0.595*** 0.603*** 0.585*** (0.027) (0.030) (0.029) (0.034) (0.044) (0.044) (0.045) (0.044) Cash Flow t 1 0.062*** 0.060** 0.057** 0.056** 0.002 0.013 0.022 0.040 (0.023) (0.025) (0.025) (0.028) (0.030) (0.031) (0.032) (0.032) Market to Book 0.012*** 0.012*** 0.012*** 0.012*** 0.023*** 0.023*** 0.023*** 0.023*** (0.002) (0.002) (0.002) (0.002) (0.005) (0.005) (0.005) (0.006) Size 0.003*** 0.004*** 0.004*** 0.003** 0.002 0.002 0.003* 0.002 (0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002) CapEx 0.073** 0.112*** 0.072** 0.017 0.014 0.065 (0.033) (0.033) (0.034) (0.056) (0.057) (0.065) Capital 0.023 0.022 0.026 0.063** 0.057** 0.066** (0.022) (0.022) (0.028) (0.027) (0.026) (0.030) Rated 0.003 0.004 0.001 0.018** 0.022*** 0.020** (0.004) (0.004) (0.004) (0.007) (0.007) (0.008) Investment Grade 0.001 0.002 0.000 0.023*** 0.016** 0.023*** (0.004) (0.004) (0.004) (0.008) (0.008) (0.008) This table presents the coefficients of OLS models in which the dependent variable is Cash Flow in year t+1. This table looks at two different time periods: my sample, which includes 2000 2010, and the publicly available sample from Sufi (2009), which in cludes the years 1996 2003. Unexpected Draw is the residual from the predictive regression on drawdowns in Table 2 5. The model numbers in this table correspond to the predictive regression model numbers in Table 2 5. Cash, Cash Flow, CapEx and Net Worki ng Capital are all measured as a proportion of lagged non e of interest from year t 1 to t. Credit Market Tightness is measured as the average over the four quarters of the fiscal year of the net fraction of domestic banks who respond to the Federal Reserve Bank study by saying that they are tightening standards for C&I loans to large and medium size d firms. Annual GDP is the annual U nited States gross domestic product for fiscal year t, measured in billions of 2005 dollars. Specifics of variable definition s can be found in the Appendix All models include an unreported constant term. Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 55

55 Table 2 7. Continued 2000 2010 Credit Line Sample Sufi (2009) Credit Line Sample (1996 2003) (1) (2) (3) (4) (1) (2) (3) (4) Credit Market Tightness 0.006 0.051** (0.006) (0.023) Annual GDP 0.004** 0.007 (0.002) (0.005) Industry Fixed Effects Y Y Y Y Y Y Y Y Year Fixed Effects Y Y N Y Y Y N Y Number of observations 4,596 4,476 4,476 3,850 1,317 1,265 1,265 1,033 Adjusted R2 0.617 0.613 0.604 0.621 0.482 0.498 0.488 0.481 This table presents the coefficients of OLS models in which the dependent variable is Cash Flow in year t+1. This table looks at two different time periods: my sample, which includes 2000 2010, and the publicly available sample from Sufi (2009), which in cludes the years 1996 2003. Unexpected Draw is the residual from the predictive regression on drawdowns in Table 2 5. The model numbers in this table correspond to the predictive regression model numbers in Table 2 5. Cash, Cash Flow, CapEx and Net Worki ng Capital are all measured as a proportion of lagged non e of interest from year t 1 to t. Credit Market Tightness is measured as the average over the four quarters of the fiscal year of the net fraction of domestic banks who respond to the Federal Reserve Bank study by saying that they are tightening standards for C&I loans to large and medium size d firms. Annual GDP is the annual U nited States gross domestic product for fiscal year t, measured in billions of 2005 dollars. Specifics of variable definition s can be found in the Appendix All models include an unreported constant term. Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 56

56 Table 2 8 Effect of unexpected drawdowns on future cash flows 2000 2010 Credit Line Sample Sufi (2009) Credit Line Sample (1996 2003) (1) (2) (3) (4) (1) (2) (3) (4) Unexpected Draw 0.044** 0.048** 0.046** 0.067** 0.120*** 0.119*** 0.119*** 0.146*** (0.018) (0.019) (0.019) (0.027) (0.030) (0.032) (0.032) (0.054) Predicted Draw 0.345 0.167 0.238* 0.032 0.705 0.005 0.049 0.085* (0.275) (0.129) (0.128) (0.028) (0.443) (0.152) (0.191) (0.048) Cash 0.072*** 0.054*** 0.061*** 0.031*** 0.011 0.081*** 0.076** 0.074*** (0.028) (0.015) (0.015) (0.011) (0.060) (0.028) (0.031) (0.025) Cash Flow 0.628*** 0.632*** 0.626*** 0.673*** 0.589*** 0.575*** 0.577*** 0.569*** (0.034) (0.031) (0.031) (0.031) (0.045) (0.045) (0.045) (0.043) Cash Flow t 1 0.063*** 0.066*** 0.067*** 0.058** 0.000 0.002 0.002 0.014 (0.023) (0.024) (0.024) (0.026) (0.030) (0.031) (0.031) (0.032) Market to Book 0.013*** 0.012*** 0.013*** 0.012*** 0.024*** 0.024*** 0.024*** 0.023*** (0.002) (0.002) (0.002) (0.002) (0.005) (0.005) (0.005) (0.006) Size 0.001 0.002 0.001 0.003*** 0.011* 0.002 0.002 0.000 (0.003) (0.002) (0.002) (0.001) (0.006) (0.003) (0.003) (0.002) Constant 0.017 0.037** 0.034*** 0.051*** 0.198 0.027 0.012 0.027 (0.028) (0.015) (0.013) (0.007) (0.145) (0.055) (0.067) (0.030) Industry Fixed Effects Y Y Y Y Y Y Y Y Year Fixed Effects Y Y Y Y Y Y Y Y Number of observations 4,596 4,476 4,476 3,850 1,317 1,265 1,265 1,033 Adjusted R2 0.617 0.612 0.612 0.620 0.483 0.494 0.494 0.478 This table presents the coefficients of OLS models in which the dependent variable is Cash Flow in year t+1. This table prese nts a different set of independent variables from Table 2 7. Unexpected Draw is the residual from the predictive regression on draw downs in Table 2 5. Predicted Draw is the predicted drawn amount from the predictive regression in Table 2 5. The model numbers in this table correspond to the predictive regression model numbers in Table 2 5. Cash, Cash Flow, CapEx and Net Working Capita l are all measured as a proportion of lagged non cash interest from year t 1 to t. Specifics of variable definitions can be found i n the Appendix All models include an unreported constant term. Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 57

57 Table 2 9. Effect of unexpect ed drawdowns on future covenant variables. Dependent Variable Adj Net Worth t+1 Adj Current Ratio t+1 Adj Leverage t+1 Unexpected Draw 0.192*** 0.704* 0.025 (0.057) (0.402) (0.025) Cash 0.062** 0.471*** 0.029*** (0.024) (0.121) (0.011) Adj Net Worth 0.430*** (0.024) Adj Net Worth t 1 0.187*** (0.020) Adj Current Ratio 0.410*** (0.027) Adj Current Ratio t 1 0.167*** (0.025) Adj Leverage 0.628*** (0.024) Adj Leverage t 1 0.190*** (0.022) Market to Book 0.018*** 0.008 0.010*** (0.005) (0.023) (0.002) Size 0.009*** 0.066*** 0.005*** (0.002) (0.014) (0.001) Constant 0.260*** 0.532*** 0.007 (0.028) (0.177) (0.010) Industry Fixed Effects Y Y Y Year Fixed Effects Y Y Y Number of Observations 3,534 3,465 3,534 Adjusted R2 0.409 0.372 0.677 The following table presents the coefficients of OLS models in which the dependent variable is measured in year t+1 and specified in the first row. Unexpected Draw is the residual from the predictive regression on drawdowns in Model 1 of Table 2 5. Cash, Adj Net Worth and Adj Leverage are measured as a fraction of lagged non cash assets. All independent variables are measured at time t unless otherwise specified. Specifics of variable definitions can b e found in the Appendix Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 58

58 Table 2 10 Effect of unexpected drawdown on the probability of a new covenant violation. (1) (2) (3) (4) Coefficient Marginal Effect Coefficient Marginal Effect Coefficient Marginal Effect Coefficient Marginal Effect Unexpected Draw 0.622** 0.097** 0.584* 0.090* 0.583* 0.090* 0.844 0.124 (0.299) (0.047) (0.309) (0.047) (0.309) (0.047) (0.536) (0.078) Cash Flow 0.390** 0.061** 0.396** 0.061** 0.396** 0.061** 0.767*** 0.113*** (0.188) (0.029) (0.198) (0.030) (0.198) (0.030) (0.248) (0.036) Adj Current Ratio 0.010 0.002 0.016 0.002 0.015 0.002 0.002 0.000 (0.016) (0.003) (0.016) (0.003) (0.016) (0.003) (0.019) (0.003) Net Worth 0.262** 0.041** 0.303** 0.047** 0.303** 0.047** 0.354** 0.052** (0.126) (0.020) (0.136) (0.021) (0.136) (0.021) (0.160) (0.023) Adj Leverage 0.322** 0.050** 0.385** 0.059** 0.385** 0.059** 0.271 0.040 (0.152) (0.024) (0.165) (0.025) (0.165) (0.025) (0.195) (0.029) Market to Book 0.072*** 0.011*** 0.061** 0.009** 0.061** 0.009** 0.046 0.007 (0.025) (0.004) (0.026) (0.004) (0.026) (0.004) (0.030) (0.004) Size 0.087*** 0.014*** 0.044* 0.007* 0.044* 0.007* 0.032 0.005 (0.018) (0.003) (0.025) (0.004) (0.025) (0.004) (0.028) (0.004) Cash 0.065 0.010 0.068 0.010 0.069 0.011 0.158 0.023 (0.132) (0.021) (0.134) (0.021) (0.134) (0.021) (0.158) (0.023) CapEx 0.356 0.055 0.347 0.053 0.037 0.006 (0.540) (0.083) (0.539) (0.083) (0.609) (0.089) This table presents the results of probit regressions of the probability of a new covenant violation in year t+1. The depende nt variable is New Covenant Violation t+1 New Covenant Violation is a binary variable equal to one if the firm violates a covenant during the fiscal year and has not violated a covenant in any of the previous three quarters. Unexpected Draw is the residual from the predictive regression on drawdowns in Table 2 5. The model numbers in this table correspond to the predictive regression model numbers in Table 2 5. Cash Flow, Net Worth, Adj Leverage, Cash, CapEx and Net Working Capital are all measured as a fraction of lagged non cash assets. Net Working Capital is working capital net of cash. Adj Leverage and Adj Current Ratio are measure d net of credit line drawdowns. Rated and Investment Grade are binary variables equal to 1 icates a change in the variable of inter est from year t 1 to t. Specifics of variable definitions can be found in the Appendix All models include an unreported constant term. Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 1 0% levels, respectively.

PAGE 59

59 Table 2 10. Continued (1) (2) (3) (4) Coefficient Marginal Effect Coefficient Marginal Effect Coefficient Marginal Effect Coefficient Marginal Effect 0.112 0.017 0.112 0.017 0.362 0.053 (0.249) (0.038) (0.249) (0.038) (0.320) (0.047) Rated 0.123 0.018 0.123 0.018 0.144 0.020 (0.124) (0.018) (0.124) (0.018) (0.135) (0.018) Investment Grade 0.340** 0.044** 0.340** 0.044** 0.354* 0.044** (0.168) (0.018) (0.168) (0.018) (0.187) (0.019) Industry Fixed Effects Y Y Y Y Y Y Y Y Year Fixed Effects Y Y Y Y Y Y Y Y Number of observations 3,174 3,174 3,131 3,131 3,131 3,131 2,557 2,557 Adjusted R2 0.038 0.038 0.042 0.042 0.042 0.042 0.051 0.051 This table presents the results of probit regressions of the probability of a new covenant violation in year t+1. The depende nt variable is New Covenant Violation t+1 New Covenant Violation is a binary variable equal to one if the firm violates a covenant during the fiscal year and has not violated a covenant in any of the previous three quarters. Unexpected Draw is the residual from the predictive regression on drawdowns in Table 2 5. The model numbers in this table correspond to the predictive regression model numbers in Table 2 5. Cash Flow, Net Worth, Adj Leverage, Cash, CapEx and Net Working Capital are all measured as a fraction of lagged non cash assets. Net Working Capital is working capital net of cash. Adj Leverage and Adj Current Ratio are measure d net of credit line drawdowns. Rated and Investment Grade are binary variables equal to 1 icates a change in the variable of inter est from year t 1 to t. Specifics of variable definitions can be found in the Appendix All models include an unreported constant term. Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 1 0% levels, respectively.

PAGE 60

60 Table 2 11. Effect of unexpected drawdown on probability of credit rating downgrade. (1) (2) (3) (4) Coefficient Marginal Effect Coefficient Marginal Effect Coefficient Marginal Effect Coefficient Marginal Effect Unexpected Draw 1.242 0.254 2.111*** 0.406*** 2.101*** 0.404*** 2.824*** 0.539*** (0.765) (0.156) (0.764) (0.148) (0.765) (0.148) (1.072) (0.205) Industry Cash Flow Volatility 0.015 0.003 0.173 0.033 0.172 0.033 0.321 0.061 (1.007) (0.206) (1.087) (0.209) (1.087) (0.209) (1.065) (0.204) Cash Flow 5.211*** 1.065*** 5.930*** 1.142*** 5.926*** 1.141*** 5.955*** 1.137*** (0.745) (0.143) (0.886) (0.164) (0.886) (0.164) (0.941) (0.168) Adj Leverage 1.193*** 0.244*** 1.604*** 0.309*** 1.604*** 0.309*** 1.475*** 0.282*** (0.242) (0.052) (0.279) (0.056) (0.279) (0.056) (0.281) (0.055) Size 0.061 0.012 0.028 0.005 0.029 0.006 0.043 0.008 (0.041) (0.008) (0.054) (0.010) (0.054) (0.010) (0.054) (0.010) Market to Book 0.091 0.019 0.132 0.025 0.132 0.025 0.137 0.026 (0.096) (0.019) (0.105) (0.020) (0.106) (0.020) (0.112) (0.021) Cash 0.644** 0.132** 0.276 0.053 0.283 0.054 0.225 0.043 (0.316) (0.064) (0.341) (0.066) (0.341) (0.065) (0.361) (0.069) CapEx 3.217*** 0.619*** 3.186*** 0.613*** 2.985** 0.570** (1.232) (0.231) (1.233) (0.231) (1.314) (0.245) 0.476 0.092 0.479 0.092 0.912* 0.174* (0.478) (0.092) (0.478) (0.092) (0.526) (0.100) This table presents the results of probit regressions of the probability of a downgrade in any month of the fiscal year. The dependent variable is a binary variable equal to one if the firm experiences a downgrade in their S&P long term credit rating (Comp ustat variable splticrm) during the year. Unexpected Draw is the residual from the predictive regression on drawdowns in Table 2 5. The model numbers in this table correspond to the predictive regression model numbers in Table 2 5 Industry Cash Flow Volat ility is the median of the standard deviation of the cash flows of all of the firms in the same two digit industry over a ten year period. Cash Flow, Adj Leverage, Cash, CapEx and Net Working Capital are all measured as a fraction of lagged non cash assets Net Working Capital is working capital net of cash. Adj Leverage is measured net of credit line drawdowns. Investment Grade is a binary variable equal to 1 if the firm is rated investment grade. All independent variables are measured at time t. Specifics of variable definitions can be found in the Appendix All models include an unreported constant term. Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 61

61 Table 2 11. Continued (1) (2) (3) (4) Coefficient Marginal Effect Coefficient Marginal Effect Coefficient Marginal Effect Coefficient Marginal Effect Investment Grade 0.374*** 0.074*** 0.375*** 0.074*** 0.365** 0.071** (0.138) (0.028) (0.138) (0.028) (0.143) (0.029) Industry Fixed Effects Y Y Y Y Y Y Y Y Year Fixed Effects Y Y Y Y Y Y Y Y Number of observations 1,475 1,475 1,444 1,444 1,444 1,444 1,303 1,303 Pseudo R2 0.166 0.166 0.185 0.185 0.185 0.185 0.188 0.188 This table presents the results of probit regressions of the probability of a downgrade in any month of the fiscal year. The dependent variable is a binary variable equal to one if the firm experiences a downgrade in their S&P long term credit rating (Compustat variable spl ticrm) during the year. Unexpected Draw is the residual from the predictive regression on drawdowns in Table 2 5. The model numbers in this table correspond to the predictive reg ression model numbers in Table 2 5. Industry Cash Flow Volatility is the median of the stand ard deviation of the cash flows of all of the firms in the same two digit industry over a ten year period. Cash Flow, Adj Leverage, Cash, CapEx and Net Working Capital are all measured as a fraction of lagged non cash assets. Net Working Capital is working capital net of cash. Adj Leverage is measured net of credit line drawdowns. Investment Grade is a binary variable equal to 1 if the firm is rated investment grade. All independent variables are measured at time t. Specifics of variable definitions can be found in the Appendix All models include an unreported constant term. Standard errors are clustered by firm and reported in parentheses. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 62

62 Table 2 12. Effect of covenant violat ions on credit line availability and usage. Total line t+1 / (Non Cash Assets) t Unused line t+1 / (Non Cash Assets) t Drawn line t+1 / (Non Cash Assets) t Covenant Violation 0.017** 0.020* 0.005 (0.007) (0.011) (0.006) Cash Flow 0.129*** 0.091** 0.023 (0.022) (0.035) (0.019) Leverage 0.078*** 0.018 0.065*** (0.012) (0.021) (0.012) Net Worth 0.038*** 0.079*** 0.028*** (0.009) (0.015) (0.008) Current Ratio 0.002 0.012** 0.008*** (0.004) (0.006) (0.003) Market to Book 0.006*** 0.007*** 0.002* (0.002) (0.003) (0.001) Size 0.011*** 0.031*** 0.015*** (0.002) (0.003) (0.002) Constant 0.107*** 0.279*** 0.129*** (0.020) (0.034) (0.017) Industry Fixed Effect s Y Y Y Year Fixed Effect s Y Y Y Number of observations 3,622 3,622 3,622 Adjusted R2 0.110 0.140 0.159 This table presents the results of OLS regressions where the dependent variable is specified in the column heading. All indep endent variables are measured at time t. Specifics of variable definitions can be found in the Appendix All Standard errors are clustered by firm and reported in parentheses ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 63

63 Table 2 13. Effect of unexpected drawdowns on change in future capital expenditures. Predicted Draw Model 2 Predicted Draw Model 3 Predicted Draw Model 4 (1) (2) (3) (1) (2) (3) (1) (2) (3) New Covenant Violation 0.004* 0.001 0.001 0.006 0.004 0.003 0.004* 0.001 0.001 (0.002) (0.002) (0.002) (0.005) (0.005) (0.005) (0.002) (0.002) (0.002) Unexpected Draw 0.009 0.014 0.016 0.034 0.036 0.038* 0.006 0.010 0.019 (0.011) (0.011) (0.011) (0.021) (0.022) (0.022) (0.014) (0.013) (0.014) Unexpected Draw*Violation 0.047** 0.048** 0.044** 0.082** 0.081** 0.074* 0.045 0.052* 0.057* (0.019) (0.019) (0.020) (0.037) (0.039) (0.039) (0.031) (0.030) (0.031) Cash Flow 0.037*** 0.083*** 0.075*** 0.032*** 0.085*** 0.098*** 0.037*** 0.082*** 0.075*** (0.008) (0.012) (0.015) (0.011) (0.023) (0.029) (0.008) (0.012) (0.015) Leverage 0.039*** 0.055* 0.031 0.032*** 0.049 0.020 0.039*** 0.058* 0.030 (0.008) (0.029) (0.029) (0.012) (0.070) (0.067) (0.008) (0.030) (0.030) Interest 0.141 0.049 0.335 0.083 0.153 0.076 0.149 0.027 0.326 (0.096) (0.390) (0.397) (0.113) (0.698) (0.688) (0.095) (0.391) (0.397) Net Worth 0.024*** 0.016 0.008 0.043*** 0.011 0.018 0.023*** 0.010 0.001 (0.009) (0.086) (0.086) (0.013) (0.129) (0.131) (0.009) (0.086) (0.086) Current Ratio 0.001* 0.005 0.003 0.002 0.010 0.005 0.001* 0.005 0.004 (0.001) (0.009) (0.009) (0.001) (0.017) (0.017) (0.001) (0.009) (0.009) Market to Book 0.004*** 0.021*** 0.025*** 0.005*** 0.031*** 0.033*** 0.004*** 0.022*** 0.026*** (0.001) (0.006) (0.006) (0.001) (0.010) (0.010) (0.001) (0.006) (0.006) NSS Controls Y Y Y Y Y Y Y Y Y This table presents the results of OLS regressions where the dependent variable is the change in CapEx from year t to year t+ 1 following the specifications in Table VI of Nini, Smith and Sufi (2011). New Covenant Violation is a binary variable equal to one if the firm violates a covenant during the fiscal year and has not violated a covenant in any of the previous three quarters. Unexpected Draw is the residual from the predictive regression on drawdowns in Table 2 5. Unexpected Draw*Violation is the product of New Covenant Violation and Unexpected Draw. NSS Controls are the level and first difference of Size and the level and first difference of PPE. The six covenant control variables are Cash Flow, Leverage, Inter est, Net Worth, Current Ratio, and Market to Book. Higher Covenant Controls are the squared and cubed values of the six covenant control variables. Lagged Covenant Controls are the six covenant variables measured at t 1. CapEx, Cash Flow, Leverage, Interes t, Net Worth, and PPE are measured as a fraction of lagged non cash assets. All independent variables are measured at time t. Specifics of variable definitions can be found in the Appendix Standard errors are clustered by firm and reported in parentheses. ** *, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 64

64 Table 2 13. Continued Predicted Draw Model 2 Predicted Draw Model 3 Predicted Draw Model 4 (1) (2) (3) (1) (2) (3) (1) (2) (3) Industry Fixed Effects Y Y Y Y Y Y Y Y Y Year Fixed Effects Y Y Y Y Y Y Y Y Y Higher Order Covenant Controls N Y Y N Y Y N Y Y Lagged Covenant Controls N N Y N N Y N N Y Number of observations 2,844 2,844 2,805 849 849 839 2,843 2,843 2,805 Adjusted R2 0.356 0.371 0.381 0.398 0.408 0.410 0.356 0.371 0.381 This table presents the results of OLS regressions where the dependent variable is the change in CapEx from year t to year t+ 1 following the specifications in Table VI of Nini, Smith and Sufi (2011). New Covenant Violation is a binary va riable equal to one if the firm violates a covenant during the fiscal year and has not violated a covenant in any of the previous three quarters. Unexpected Draw is the residual from the predictive regression on drawdowns in Table 2 5. Unexpected Draw*Viol ation is the product of New Covenant Violation and Unexpected Draw. NSS Controls are the level and first difference of Size and the level and first difference of PPE. The six covenant control variables are Cash Flow, Leverage, Interest, Net Worth, Current Ratio, and Market to Book. Higher Covenant Controls are the squared and cubed values of the six covenant control variables. Lagged Covenant Controls are the six covenant variables measured at t 1. CapEx, Cash Flow, Leverage, Interest, Net Worth, and PPE ar e measured as a fraction of lagged non cash assets. All independent variables are measured at time t. Specifics of variable definitions can be found in the Appendix Standard errors are clustered by firm and reported in parentheses. ***, **, and denote sign ificance at the 1%, 5%, and 10% levels, respectively.

PAGE 65

65 Table 2 14. Summary statistics on changes in contract terms N Mean Std Dev 10th Pctl 90th Pctl T stat 909 0.106 83.165 100.000 100.000 0.03829 965 3.684 19.933 20.000 26.000 5.74115 1017 54.096 520.750 101.000 280.000 3.312823 995 0.015 0.365 0.000 0.000 1.30112 1017 0.010 0.476 1.000 1.000 0.659194 1017 0.382 19.732 13.000 12.000 0.61819 This table displays summary statistics on changes in credit line contract terms when a new contract is reported in DealScan within the following year. Covenant terms are measured using DealScan data. The sample is restricted to companies with ex isting credit line agreements where a new agreement is reported 1 to t, unless otherwise specified. Unexpected Draw is the residual from the predictive regression on drawdowns in Table 2 5. T stat reports the t statistic for the significance of the reported mean from zero.

PAGE 66

66 Table 2 15. Summary statistics for changes in contract terms for f irms with covenant violations. Covenant Violation = 1 Covenant Violation = 0 N Mean N Mean Difference 117 29.192 792 4.434 33.626*** 128 2.719 837 3.832 1.113 132 3.629 885 61.624 57.994*** 126 0.119 869 0.000 0.119** 132 0.083 885 0.001 0.084* 132 5.394 885 0.365 5.759** This table presents summary statistics and mean difference tests for changes in credit line contract terms before and after a covenant violation. Covenant terms are measured using DealScan data. The sample is restricted to companies with existing credit li ne agreements where a new agreement is reported in 1 to t, unless otherwise specified. Unexpected Draw is the residual from the predictive regression on drawdowns in Mod el 2 of Table 2 5. I test the null hypothesis that the means for the two groups are equal using t tests. I assume unequal variances for t tests. ***, **, and denote significance in differences between the sample means at the 1%, 5%, and 10% levels, respe ctively.

PAGE 67

67 Table 2 1 6 Summary statistics for changes in contract terms for f irms with unexpected drawdowns. Unexpected Draw > 0 Unexpected Draw < =0 N Mean N Mean Difference 228 9.523 681 3.329 12.852** 240 2.671 725 4.019 1.348 251 8.994 766 68.875 59.882* 245 0.073 750 0.004 0.077** 251 0.036 766 0.001 0.035 251 0.036 766 0.520 0.555 Expire 244 0.053 736 0.054 0.001 This table divides the sample into firms with positive and negative unexpected drawdowns and provides summary statistics and mean difference tests for changes in credit line contract terms.. Covenant terms are measured using DealScan data. The sample is re stricted to companies with existing credit line variable of interest from year t 1 to t, unless otherwise specified. Unexpected Draw is the residual from the predictive regression on drawdowns in Model 2 of Table 2 5. I test the null hypothesis that the means for the two groups are equal using t tests. I assume unequal variances for t tests. ***, **, and denote significance in differences between the samp le means at the 1%, 5%, and 10% levels, respectively.

PAGE 68

68 Table 2 17. Summary s tatistics on changes in c ontract t erms a round u nexpected d rawdowns when there is no covenant violation Unexpected Draw > 0 Unexpected Draw < =0 N Mean N Mean Difference 190 4.599 602 7.284 11.883* 200 3.020 637 4.086 1.066 209 10.032 676 77.574 67.542* 205 0.044 664 0.014 0.057* 209 0.005 676 0.000 0.005 209 2.091 676 0.169 2.260 This table presents summary statistics and mean difference tests on changes in credit line contract terms for firms with positive or negative unexpected drawdowns when there is no covenant violation in the current year. Covenant terms are measured using De alScan data. The sample is restricted to companies indicates a change in the variable of interest from year t 1 to t, unless otherwise specified. Unexpect ed Draw is the residual from the predictive regression on drawdowns in Model 2 of Table 2 5. I test the null hypothesis that the means for the two groups are equal using t tests. I assume unequal variances for t tests. ***, **, and denote significance in differences between the sample means at the 1%, 5%, and 10% levels, respectively.

PAGE 69

69 Table 2 1 8 Summary s tatistics on changes in c ontract t erms a round u nexpected d rawdowns when there is a covenant violation Unexpected Draw > 0 Unexpected Draw < =0 N Mean N Mean Difference 38 34.145 79 26.810 7.335 40 0.925 88 3.534 2.609 42 3.826 90 3.538 0.288 40 0.225 86 0.070 0.155 42 0.238 90 0.011 0.227** 42 10.190 90 3.156 7.035 This table presents summary statistics and mean difference tests on changes in credit line contract terms for firms with positive or negative unexpected drawdowns when there is a covenant violation in the current year. Covenant terms are measured using Dea lScan data. The sample is restricted to companies indicates a change in the variable of interest from year t 1 to t, unless otherwise specified. Unexpecte d Draw is the residual from the predictive regression on drawdowns in Model 2 of Table 2 5. I test the null hypothesis that the means for the two groups are equal using t tests. I assume unequal variances for t tests. ***, **, and denote significance in differences between the sample means at the 1%, 5%, and 10% levels, respectively.

PAGE 70

70 CHAPTER 3 Introduction s tructure. Capital IQ provides data from 2002 2010 on drawn and available amounts on corporate credit line contracts. Paolo, Ippolito, and Li (2011) and Ippolito and Perez (2011) use Capital IQ to identify firms with access to credit lines. They also make use of drawn amounts outstanding as reported in Capital IQ in their analysis I compare the drawn and available amounts in Capital IQ to hand collected data on credit line usage for a random sample of firms. The goal is to document the accuracy of Capital collection method relative to hand collecting. Of course, my hand collection is subject to possible errors or subjective judgments, as well. I do not claim t o have a perfect database, but I believe that I am able to document some specific features of Capital IQ data that should be of use for individuals who wish to use this dataset for studies involving credit line data. I identify several concerns with the use of Capital IQ data on credit line usage. First, Capital IQ often reports blanks when there is data available on credit line usage K filings. Worse, this lack of reporting is not consistent over time. For example, in the case of Kimberley Clark Corp., Capital IQ reports non missing data only in 2006 and 2010. In al l other years from 2002 2010, I find varying amounts available reported in 10 K filings with the SEC, but Capital IQ reports only blanks. Similarly with Alaska Air Group, Capital IQ reports a non missing value of the drawn amount outstanding for 2005 2007 and 2010. In the other years from 2002

PAGE 71

71 2010, Capital IQ reports a missing for the drawn amount outstanding despite the line being fully drawn at the end of fiscal year 2002. In addition, I rrepresents the number of firm year observations in which the sampled companies have access to and availability on a corporate credit line. Only 27% of the firm year observations in the hand collected sample match Capital IQ data in establishing whether or not there are amounts available to borrow on a credit line. If I follow existing work by Ippolito and Perez (2011) and use reported Capital IQ credit line availability to determine whether or not a firm has a line of credit, then Capital IQ captures only 28% of the firm year drawn and available variables, the accuracy of using Capital IQ to determine if a company has a credit line increases to 71%. Finally, Capital IQ of ten uses data from tables. This may eliminate some important granularity that can be obtained by reading the filing. For example, Capital IQ reports $12.381 billion outstanding on a credit line for Time Warner Inc. in 2006 based on the line item reporting outstanding debt on the bank credit agreement and commercial paper programs. Upon closer reading, $8 billion of the credit agreement is in the form of two $4 billion term loans. I do not include term loans in computing credit line use and access, so I report $4.381 billion outstanding. This same chart used to obtain the $12.381 billion outstanding reported in Capital IQ also reports unused committed capacity on the bank credit agreement of $8.382 billion. However, Capital IQ reports a missing value for the unused capacity of Time Warner Inc. in 2006.

PAGE 72

72 Data I first obtain data from Capital IQ on firms with either positive bank debt or a reported credit line in at least one year between 200 1 and 2010. I then keep only those companies that also appear in the Compustat industrial database, are not in the financial or utility sector, and have non missing asset data during the sample period. From the 5,619 companies that met these criteria, I take a random sa mple of 900. For the random sample of 900 firms, I use a program similar to the one described in Sufi (2009) to run a web crawl on 10 K fillings for fiscal years 2000 to 2010. 1 I search terms related to credit lines 2 I then manually examine the filings t o establish the total size of the credit line, the drawn amount, and the unused amount available for draw. 3 As in Sufi (2009), if a line of credit backs up outstanding commercial paper or letters of credit, those amounts are subtracted from the available p ortion of the line of credit but not included as part of the amount drawn on the line. This collection method results in a sample of 76 3 companies with a credit line in at least one year from 2000 to 2010 for a total of 5, 311 firm year observations. I ma tch these observations to Capital IQ data. Capital IQ reports information from 2002 2010 on two variables of interest. The first, Total Revolving Credit reports the amount drawn on the credit line. The second, Undrawn Revolving Credit states the amount available for future drawdowns. All 763 companies from my sample match 1 I modify a web crawler code for SAS provided in E ngelberg and Sankaraguruswamy (2007). 2 My 3 Corporations report detailed informat ion related to their credit lines in annual 10 K filings. Item 303 of SEC Regulation S K requires companies to report information related to their liquidity, capital resources, and results of operations for the fiscal year. While some companies choose to d isclose credit line information more frequently, only annual reporting is required.

PAGE 73

73 Capital IQ due to my initial data screening process. The matching results in 4,217 firm year observations from 2002 2010. Results Table 3 1 reports summary statistics for the hand co llected and Capital IQ data on credit line usage. Capital IQ reports which I refer to as Drawn Outstanding in my hand collected sample. Similarly, t he Capital IQ equivalent of my Undrawn variable directly report a value for the total size of the line, which I dub Total. The best available estimate is the sum of the two reported items, Total Revolving Credit and Undrawn Revolving Credit, which I For my hand collected sample, Total is not equal to the sum of Drawn Outstanding and Undrawn if the line of credit supports outstanding letters of credit or commercial paper, as these reduce availability but are not included in drawn amounts outstanding. T here is a significant difference in the number of observations between the hand collected and Capital IQ reported data due to the prevalence of missing observations in Capital IQ. Table s 3 2 and 3 3 present mean difference tests between the hand collected and Capital IQ data. Table 3 2 uses the entire sample of each, while Table 3 3 restricts the sample to only those observations where Capital IQ has non missing data. Perhaps surprisingly, both tables show similar results. Capital IQ tends to report larger drawn amounts outstanding, but there is no statistically significant difference between the hand collected and Capital IQ data with regards to the average credit line availability. Using my nave definition of the total size of the line based on Capital IQ data, I find that the hand collected data shows a larger total line, on average.

PAGE 74

74 In Table s 3 4 and 3 5 I fairly represent the number of firm year observations in which the sampled companies hav e either access to or availability on a corporate credit line compared to what is reported in 10 K filings 4 Ippolito and Perez (201 2 ) use s Capital IQ Revolving Credit to determine whether or not a firm has a line of credit. Table 3 4 sh ows that Capital IQ captures only 28% of the firm year observations where there is a corporate credit line. This discrepancy occurs because Capital IQ often reports blanks when there is in fact data available in SEC filings. I investigate whether I can imp rove the accuracy of using Capital IQ data to determine credit line access. I find that using the implied total (the s drawn and available variables) increases the accuracy to 71%. While both Total Revolving Credit and Undrawn Revolving C redit are often reported as missing, they do not always overlap. Therefore, including both variables in the determination of credit line access substantially increases the usefulness of the Capital IQ database. Callo, Ippolito, and Li (2011) use Capital I determine whether or not there is credit line availability for future drawdowns Table 3 4 also shows that only 27% of the firm year observations in the Capital IQ database match the hand collected data from SEC filings with regards to whether or not there is borrowing availability on a credit line. 5 4 By access, I simply mean the existence of a corporate credit line. The term availability is used to describe both the existence of a credit line and the amount available for drawdowns. Therefore, a firm can only have availability if it has access, but access does not imply availability. 5 An availability observation is counted as a match if both Capital IQ and hand collected data report zero availability, or if both report positive availability.

PAGE 75

75 In Table 3 5 I break the data into fiscal years to see if there are time series credit line access and availability. Interestingly, I find that there is not much time variance until 2010. In fiscal year 2010, the Capital IQ data becomes substantially more accurate at establishing both credit line access and availability, with both having an approximately 75% success rate. 6 Again, using Total (Implied) improves upon using Undrawn Revolving Credit when establish ing credit line access. The percentage matches using Total (Implied) improves to nearly 90% in 2010. Ippolito and Perez ( 2012 ) also use Capital IQ to determine the amount of credit line access that a company has available for future drawdowns. In Table s 3 6 and 3 7 I show the number of matches between the Capital IQ provided and hand collected datasets. Table 3 6 show the overall number of matches, the percentage matches to the entire hand collected sample, and the percentage matches when the sample is con strained to those observations with non missing data in Capital IQ. As I would expect, the percentage matches improve when the dataset is restricted, but the amount of credit line availability still matches less than 50% of the time. When I include the vas t number of times that Undrawn Revolving Credit is missing in Capital IQ, the percentage matches fall drastically to only 12%. To account for differences in rounding, or possible small judgment differences, I investigate the number of Capital IQ observati ons that are within 10% of the hand collected data in Table 3 7 While the percentage matched increases, there is still only 6 In unreported tests, I also explored the results from Table 3 2 by year. I do not find any time series collected data throughout the sa mple period.

PAGE 76

76 an 18% match rate for the full sample. This increases to 63% when I condition on Capital IQ having non missing data. Both Table 3 6 and Table 3 7 also shows that the data on drawn amounts outstanding is better matched between the two sources than are total amounts. However, the percentage of exact matches is still quite low (34% in the full sample and 52% in the non mi ssing Capital I Q sample). When I both limit the sample to observations with non missing Capital IQ data and relax the match to within 10%, I find substantial improvement. Capital IQ matches the hand collected data 72.5% of the time. However, I should note that restrictin g the sample in this way is only possible for some study designs. Any study interested in both firms with and without credit line access would be unable to limit the data to only non missing Capital IQ observations. In Table 3 4 I showed that using Total (Implied) gives a more accurate representation o f credit line access. However, I do not expect Total (Implied) to be a particularly accurate representation of the exact size of the credit line due to various mechanisms that decrease credit line availabilit y but are not included in outstanding drawn amounts. Consistent with this expectation, in Table s 3 6 and 3 7 I find that Total is the dimension with the lowest percentage matches between Capital IQ and hand collected data. Even after conditioning on non missing Capital IQ data and widening the interval for a match to within 10%, there is still only a 25% match between Capital IQ and the hand collected sample. Thus, while using Total (Implied) is useful for creating a binomial variable that indicates c redit line access, it is not particularly useful for liquidity management strategy.

PAGE 77

77 Finally, in Table s 3 8 and 3 9 I address possible determinants of Capital I accuracy in a multivariate probit model. In Table 3 8 the dependent variable is a binary variable equal to 1 when there are exact matches between Capital IQ and the hand collected data on Drawn Outstanding, Undrawn and Total in Columns 1, 2 and 3, re spectively. In Table 3 9 I repeat the analysis where the binary variable equals one when the Capital IQ observations are within 10% of the hand collected value. In the majority of the regressions, I find that Capital IQ is more likely to match the hand c ollected data for smaller firms. In addition, several of the year dummies are statistically significant. Notably, the fiscal year 2010 dummy variable is statistically significantly positive in all specifications. This confirms the ye ar by year results from Table 3 5 that showed a m uch greater accuracy in 2010. I include an additional variable, Unavailable, in the probit regressions estimating the probability of a match on Total. Since the Capital IQ equivalent of Total is estimated by summing the drawn and available amounts, it will overstate the total contractual size if there are letters of credit outstanding that do not count towards drawn amounts but do reduce availability. On the other hand, it may understate the total contractual size of the credit lin e if there are borrowing base restrictions that reduce availability below the contracted level. Unavailable measures the gap between the contracted size of the credit line and the sum of the drawn and available components in the han d collected data. As exp ected, I find that greater unavailable amounts reduce the probability of a match on Total. Conclusion I show, for a random sample of firms, that there exist significant differences collecte d data from annu al reports filed with the SEC. I find that Capital IQ reports missing data even when

PAGE 78

78 The same firm may have data reported for some years but not fo r others, even while operating under th e same credit line agreement. The bigg est caution, and advice, that I am able to give Capital IQ users regards the reliance on Undrawn Revolving Credit to determine credit line access and availability. This measure do es a poor job of establishing whether or not a firm has access to a credit line, as it is correct less than 30% of the time. However, I show that by utilizing the sum of both credit line variables provided by Capital IQ (Undrawn Revolving Credit and Total Revolving Credit), one can successfully identify 71% of the sample firm year observations with credit line access.

PAGE 79

79 Table 3 1. Summary statistics for hand collected and Capital IQ data on credit line usage. N Mean Median Std Dev Minimum Maximum 10 th Pctl 90 th Pctl Drawn Outstanding 4217 53.470 0.929 235.617 0.000 6500.000 0.000 108.000 Undrawn 4217 271.664 35.000 985.743 0.000 18264.000 0.913 560.200 Total 4217 364.375 60.000 1198.600 0.025 21031.000 3.000 800.000 CIQ: Total Revolving Credit 2735 89.427 10.500 417.347 0.000 12381.000 0.000 188.000 CIQ: Undrawn Revolving Credit 1164 291.067 29.850 1463.520 0.000 18264.000 0.553 400.000 CIQ: Total (Implied) 2994 194.851 21.100 1005.740 0.000 18264.000 0.000 331.400 This table presents summary statistics on credit line usage information provided by hand collecting and by Capital IQ. Drawn Outstanding is the amount outstanding on all lines of credit according to hand collected data. The Capital IQ equivalent of Drawn Outstanding is repor ted under the collected data. The Capital IQ equivalent of edit based on the hand collected data. Capital IQ does not directly report an equivalent to Total. The best available estimate is the sum of the two reported items, Total Revo lving Credit and tl and 90th Pctl are the 10th and 90th percentiles of observations, respectively.

PAGE 80

80 Table 3 2. Summary statistics for Capital IQ data on credit line usage relative to hand collected data. Hand collected Capital IQ N Mean N Mean Difference Drawn Outstanding 4217 53.470 2735 89.427 35.958*** Undrawn 4217 271.664 1164 291.067 19.403 Total 4217 364.375 2994 194.851 169.523*** This table presents summary statistics and mean difference tests on credit line usage information provided by hand collecting and by Capital IQ. Drawn Outstanding is the outstanding amount drawn on the credit line at the end of fiscal year t from hand coll ected data. Capital IQ records Drawn Outstanding available for future drawdowns as of the end of fiscal year t from hand collected data. Capital IQ records U end of fiscal year t from hand collected data. Capital IQ does not directly report an equivalent to Total. I use the best available estimate: the sum of the two reported items, Total Revolving Credit and Undrawn are equal using t tests. I assume unequal variances for t tests. ***, **, and deno te significance in differences between the sample means at the 1%, 5%, and 10% levels, respectively.

PAGE 81

81 Table 3 3 Summary statistics for Capital IQ data on credit line usage relative to hand collected data conditional on Capital IQ availab ility. Hand collected Capital IQ N Mean N Mean Difference Drawn Outstanding 2735 67.881 2735 89.427 21.547** Undrawn 1164 350.657 1164 291.067 59.590 Total 2994 343.060 2994 194.851 148.209*** This table presents summary statistics and mean difference tests on credit line usage information provided by hand collecting and by Capital IQ conditional on data being reported in Capital IQ. Drawn Outstanding is the outstanding amount drawn on the credi t line at the end of fiscal year t from hand of the credit line that is available for future drawdowns as of the end of fiscal year t from hand collected data. Capital IQ does not directly report an equivalent to T otal. I use the best available estimate: the sum of the two reported items, Total Revolving Credit and Undrawn t tests. I assume u nequal variances for t tests. ***, **, and denote significance in differences between the sample means at the 1%, 5%, and 10% levels, respectively.

PAGE 82

82 Table 3 4 Accuracy of Capital IQ da ta on c redit line a ccess and availability r elative to h and c ollecte d d ata Percent Agreement with Hand Collected Data Use CIQ: Undrawn Revolving Credit to base if company has a credit line 0.276 Use CIQ: Undrawn Revolving Credit to base if company has any credit line availability 0.269 Use CIQ: Total (Implied) to base if company has a credit line 0.710 This table assesses the accuracy of using Capital IQ data to identify firm credit line access and availability when compared to hand collected data. Drawn Outstanding is the outstanding amount drawn on the credit li ne at the end of fiscal year t from hand collected data. Capital IQ records e for future drawdowns as of the end of fiscal year t f rom hand contractual size of the reported credit line as of the end of fiscal year t from hand collected data. Capital IQ does not directly report an equivalent to Total I use the best available estimate: the sum of the two reported items, Total Revolving Credit and Undrawn Revolving Credit, which I dub

PAGE 83

83 Table 3 5. Accuracy of Capital IQ da ta on c redit line a ccess and availability r elative to h and c ollected d ata by year Percent Agreement with Hand Collected Data Fiscal Year: 2002 Use CIQ: Undrawn Revolving Credit to base if company has a credit line 0.254 Use CIQ: Undrawn Revolving Credit to base if company has any credit line availability 0.245 Use CIQ: Total (Implied) to base if company has a credit line 0.679 Fiscal Year: 2003 Use CIQ: Undrawn Revolving Credit to base if company has a credit line 0.305 Use CIQ: Undrawn Revolving Credit to base if company has any credit line availability 0.299 Use CIQ: Total (Implied) to base if company has a credit line 0.676 Fiscal Year: 2004 Use CIQ: Undrawn Revolving Credit to base if company has a credit line 0.246 Use CIQ: Undrawn Revolving Credit to base if company has any credit line availability 0.240 Use CIQ: Total (Implied) to base if company has a credit line 0.673 Fiscal Year: 2005 Use CIQ: Undrawn Revolving Credit to base if company has a credit line 0.325 Use CIQ: Undrawn Revolving Credit to base if company has any credit line availability 0.320 Use CIQ: Total (Implied) to base if company has a credit line 0.714 Fiscal Year: 2006 Use CIQ: Undrawn Revolving Credit to base if company has a credit line 0.290 Use CIQ: Undrawn Revolving Credit to base if company has any credit line availability 0.278 Use CIQ: Total (Implied) to base if company has a credit line 0.712 This table assesses the accuracy of using Capital IQ data to identify firm credit line access and availability when compared to hand collected data. Drawn Outstanding is the outstanding amount drawn on the credit line at the end of fiscal year t from hand collected data. Capital IQ records is available for future drawdowns as of the end of fiscal year t from hand contractual size of the reported credit line as of the end of fiscal year t from hand collecte d data. Capital IQ does not directly report an equivalent to Total. I use the best available estimate: the sum of the two reported items, Total Revolving Credit and Undrawn Revolving Credit, which I dub

PAGE 84

84 Table 3 5 Continued Percent Agreement with Hand Collected Data Fiscal Year: 2007 Use CIQ: Undrawn Revolving Credit to base if company has a credit line 0.200 Use CIQ: Undrawn Revolving Credit to base if company has any credit line availability 0.198 Use CIQ: Total (Implied) to base if company has a credit line 0.709 Fiscal Year: 2008 Use CIQ: Undrawn Revolving Credit to base if company has a credit line 0.189 Use CIQ: Undrawn Revolving Credit to base if company has any credit line availability 0.183 Use CIQ: Total (Implied) to base if company has a credit line 0.719 Fiscal Year: 2009 Use CIQ: Undrawn Revolving Credit to base if company has a credit line 0.184 Use CIQ: Undrawn Revolving Credit to base if company has any credit line availability 0.177 Use CIQ: Total (Implied) to base if company has a credit line 0.725 Fiscal Year: 2010 Use CIQ: Undrawn Revolving Credit to base if company has a credit line 0.751 Use CIQ: Undrawn Revolving Credit to base if company has any credit line availability 0.742 Use CIQ: Total (Implied) to base if company has a credit line 0.897 This table assesses the accuracy of using Capital IQ data to identify firm credit line access and availability when compared to hand collected data. Drawn Outstanding is the outstanding amount drawn on the credit line at the end of fiscal year t from hand collected data. Capital IQ records e for future drawdowns as of the end of fiscal year t from hand contractual size of the reported credit line as of the end of fiscal year t from hand collected data. Capital IQ does not directly report an equivalent to Total. I use the best available estimate: the sum of the two reported items, Total Revolving Credit and Undrawn Revolving Credit, which I dub

PAGE 85

85 Table 3 6. Accuracy of Capital IQ data on credit line usage re lative to hand collected data with exact matches Number of Matches Percentage Matched Percentage Matched Conditional on non missing CIQ data Drawn Outstanding 1417 0.336 0.518 Undrawn 498 0.118 0.428 Total 400 0.095 0.134 This table provides summary statistics on the match of Capital IQ data on credit line usage to hand collected data. Capital IQ and hand collected data are reported as a match if they are exactly the same values. Drawn Outstanding is the outstanding amount drawn on the cr edit line at the end of fiscal year t from hand amount of the credit line that is available for future drawdowns as of the end of fiscal year t from hand collected data. Capital IQ records Undrawn hand collected data. Capital IQ does not directly report an equivalent to Tot al. I use the best available estimate: the sum of the two reported items, Total Revolving ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 86

86 Table 3 7. Accur acy of Capital IQ data on credit line usage relative to hand collected data with matches within 10% Number of Matches Percentage Matched Percentage Matched Conditional on non missing CIQ data Drawn Outstanding 1983 0.470 0.725 Undrawn 747 0.177 0.642 Total 741 0.176 0.247 This table provides summary statistics on the match of Capital IQ data on credit line usage to hand collected data. Capital IQ and hand collected data are reported as a match if the value reported in Capital IQ is within 10% of the hand collected value. Drawn Outstanding is the outstanding amount drawn on the credit line at the end of fiscal year t from hand collected data. Capital IQ records Drawn Outstanding under the variable f the credit line that is available for future drawdowns as of the end of fiscal year t from hand t line as of the end of fiscal ye ar t from hand collected data. Capital IQ does not directly report an equivalent to Total. I use the best available estimate: the sum of ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 87

87 Table 3 8. Probability that Capital IQ credit line data matches hand collected data exactly Dependent Variable: Drawn Outstanding Undrawn Total Coefficient Marginal Effect Coefficient Marginal Effect Coefficient Marginal Effect Size 0.018* 0.007* 0.037*** 0.007*** 0.124*** 0.017*** (0.010) (0.004) (0.013) (0.002) (0.015) (0.002) Unavailable 0.002*** 0.000*** (0.001) (0.000) Constant 5.985*** 0.090 0.543 (0.406) (0.902) (0.895) Industry Fixed Effect s Y Y Y Y Y Y Year Fixed Effect s Y Y Y Y Y Y Number of observations 4,176 4,176 4,165 4,165 4,176 4,176 Adjusted R2 0.029 0.029 0.058 0.058 0.086 0.086 This table reports the results of a probit regression estimating the probability of an exact match between Capital IQ and hand collected data. The dependent variable is listed in the second row and is a binary variable equal to 1 when there is a match between the Capital IQ and hand collected vari able. Drawn Outstanding is the outstanding amount drawn on the credit line at the end of fiscal year t from hand collected data. Capital IQ that is available for future drawdowns as of the end of fiscal year t based on hand contractual size of the reported credit line as of the end of fiscal year t from han d collected data. Capital IQ does not directly report an equivalent to Total. I use the best available estimate: the sum of the two reported items, Total Revolving Credit and Undrawn Revolving Credit, which I dub d non cash assets. Unavailable is the difference Total (Drawn Outstanding + Undrawn). This amount may be unavailable to draw because it is committed as a letter of credit or because the total availability has been reduced d ue to a borrowing base or other restriction. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 88

88 Table 3 9 Probability that Capital IQ credit line data matches hand collected data within 10% Dependent Variable: Drawn Outstanding Undrawn Total Coefficient Marginal Effect Coefficient Marginal Effect Coefficient Marginal Effect Size 0.027*** 0.011*** 0.040*** 0.010*** 0.080*** 0.019*** (0.010) (0.004) (0.012) (0.003) (0.012) (0.003) Unavailable 0.001** 0.000** (0.000) (0.000) Constant 6.277*** 0.113 0.304 (0.117) (0.900) (0.899) Industry Fixed Effect s Y Y Y Y Y Y Year Fixed Effect s Y Y Y Y Y Y Number of observations 4,176 4,176 4,176 4,176 4,176 4,176 Pseudo R2 0.020 0.020 0.053 0.053 0.061 0.061 This table reports the results of a probit regression estimating the probability that the Capital IQ data reported is within 10% of the hand collected value. The dependent variable is listed in the second row and is a binary variable equal to 1 when there is a match between the Capital IQ and hand collected variable. Drawn Outstanding is the outstanding amount drawn on the credit line at the end of fiscal year t from han d collected data. for future drawdowns as of the end of fiscal year t based on hand Total is the contractual size of the rep orted credit line as of the end of fiscal year t from hand collected data. Capital IQ does not directly report an equivalent to Total. I use the best available estimate: the sum of the two reported items, Total Revolving Credit and Undrawn Revolving Credit cash assets. Unavailable is the difference Total (Drawn Outstanding + Undrawn). This amount may be unavailable to draw because it is committed as a letter of credit or because the total avail ability has been reduced due to a borrowing base or other restriction. ***, **, and denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 89

89 CHAPTER 4 CONCLUDING REMARKS The two chapters that make up my study focus on a hand collected dataset of corporate credit line information. To create this dataset, I randomly sample 900 companies that appear in both Capital IQ and Compustat and are not in the financial or utility sector. For the random sample of 900 firms, I manually examine 10 K filings to establish the total size of the credit line, the drawn amount, and the unused amount available for draw as of the end of the fiscal year I have a final sample of 76 3 companies with a credit line in at least one year from 2000 to 2 010 for a total of 5, 311 firm year observations. In Chapter 2, I utilize this dataset to study strategic credit line usage. Given existing empirical evidence that credit line access is contingent on financial performance, I hypothesize that firms draw on t heir credit lines in advance of poor future performance. This would allow firms to turn their contingent credit line access into realized liquidity in advance of future cash needs. I note that this strategic motive for credit line drawdowns differs from th e theoretical literature on credit line usage which treats corporate credit lines as liquidity insurance that is available to fund projects when external funding is unavailable. In Chapter 2, I show evidence that credit line drawdowns are predictive of fut ure cash flow declines. I separate credit line drawdowns into a predicted component based on cash needs and an unexpected component. I argue that the unexpected component proxies for preemptive drawdowns. To test whether my measure of unexpected drawdowns provides information about preemptive balances withdrawn by the firm, I test their impact on cash. I find an increase in unexpected drawdowns implies an

PAGE 90

90 increase in cash holdings. This is consistent with the drawdown being held for precautionary purposes a s opposed to meeting immediate cash needs. Consistent with the availability hypothesis, I find that unexpected drawdowns are related to performance metrics that creditors care about. I show that unexpected drawdowns predict declines in future cash flows a nd future net worth, a lower current ratio and an increased probability of a covenant violation. Moreover, I find that the probability of a credit rating downgrade is increasing in unexpected draws This is consistent with anecdotal evidence that ratings a gencies care about drawdown activity My findings also examine the lender reaction to unexpected drawdowns. I show that lenders do not cancel credit line agreements and firms are not required to immediately repay outstanding amounts upon the violation of a covenant. I find evidence that those borrowers who draw preemptively on their credit line generally see the terms of their credit agreement change against them if there is a renegotiation of the agreement. However, if there is a covenant violation, then terms worsen equally for those firms who did and did not draw preemptively on their credit line. Additionally, firms with preemptive drawdowns tend to see a lesser decline in future capital expenditures following a covenant violation. Therefore, I conclude that the preemptive access to cash seems to provide a net benefit for the average firm. In Chapter 3 of my study, I compare my hand collected dataset to data provider While I recognize that my hand collected da ta is not perfect and is subject to collection errors, I believe that it is useful to provide a first look at the accuracy of Capital IQ credit line data. Through my comparison and

PAGE 91

91 direct examination of SEC filings, I am able to identify several problems with relying on Capital IQ data. I show that Capital IQ often reports missing information on one or both of the data items collected even when there is information on credit line usage and availability in corporate 10 K filings. Furthermore, this misreporting is not systematic and results in the same firm having data reported for some years but not others, even while operating under the same credit line. This missing reporting may lead to misidentifying firms as not havin g access to a credit line, or as having lost access to a line, when relying on Capital IQ data. I look to contemporaneous studies to see how Capital IQ credit line data is being utilized by academics. I found no existing studies that use Capital IQ data t o draw conclusions about the level of drawn or available balances. However, I did find new whether or not firms have any credit line access and availability. I show that this m ethod does a poor job of establishing whether or not a firm has access to a credit line, as it is correct less than 30% of the time. I suggest an improvement for determining credit line access. I show that by summing the two variables provided by Capital I Q, Undrawn Revolving Credit and Total Revolving Credit, one can successfully identify 71% of the firm year observations with credit line access in my sample. My study demonstrates that there is ample opportunity for future research into corporate credit l ine usage. My empirical findings support the idea that firms draw preemptively from their credit line in anticipation of reduced future availability. This finding points to the need for a theory of credit line usage that accounts for the

PAGE 92

92 contingent nature of credit line contracts. In addition, I find evidence that the existing Capital IQ database for credit line data may need to be manually cleaned for accuracy. While my results are robust to the use of two separately created samples, the lack of reliable c omputer readable data limits the ability of empirical researchers to perform large scale tests for either a large number of firms or a large number of periods.

PAGE 93

93 APPENDIX VARIABLE DEFINITIONS Compustat Variable names are given in italics Non Cash Assets = Total Assets ( at ) Cash and Cash Equivalents ( che ) Draw Dummy is a binary variable equal to one if the total amount drawn on the credit line at the end of fiscal year t is greater than zero. Drawn Outstanding is the outstanding amount drawn on the cr edit line at the end of fiscal year t, measured in millions of dollars. Drawn = Drawn Outstanding/(Lagged Non Cash Assets) = (Drawn Outstanding t Drawn Outstanding t 1 )/(Lagged Non Cash Assets) Undrawn is the amount of the credit line that is available for future drawdowns as of the end of fiscal year t, measured in millions of dollars. Total is the contractual size of the reported credit line at the end of fiscal year t, measured in millions of dollars. Drawn/Total is the amount drawn on the credit line at the end of fiscal year t as a fraction of the size of the credit line at the end of fiscal year t. Covenant Violation is a binary variable equal to one if the firm violates a covenant during the fiscal year. New Covenant Violation is a bi nary variable equal to one if the firm violates a covenant during the fiscal year and has not violated a covenant in any of the previous three quarters. Size = LN(Lagged Non Cash Assets) Cash Flow = (Operating Income Before Depreciation ( oibdp ) Cash an d Cash Equivalents ( che ))/(Lagged Non Cash Assets) Industry Cash Flow Volatility is the median of the standard deviation of the cash flows of all of the firms in the same two digit industry over a ten year period. Cash = Cash and Cash Equivalents ( che )/ (Lagged Non Cash Assets) CapEx = Capital Expenditures ( capx )/(Lagged Non Cash Assets) Net Working Capital = (wcap che) /(Lagged Non Cash Assets)

PAGE 94

94 Leverage = (Long Term Debt ( dltt ) + Total Debt in Current Liabilities ( dlc ))/(Lagged Non Cash Assets) Adj Leverage = Leverage (Drawn Outstanding/Lagged Non Cash Assets) Current Ratio = Current Assets ( act )/Current Liabilities ( lct ) Adj Current Ratio = Current Assets ( act )/(Current Liabilities ( lct ) Drawn Outstanding) Net Worth = (Total Assets ( at ) Cash and Cash Equivalents ( che ) Total Debt in Current Liabilities ( dlc ) Long Term Debt ( dltt )) /(Lagged Non Cash Assets) Adj Net Worth = Net Worth + Drawn Market to Book = (Total Assets ( at ) Cash and Cash Equivalents ( che ) Book Value of Common Equ ity Outstanding ( ceq ) + Closing Share Price ( prcc_f )*Number of Common Shares Outstanding ( csho )) /( Non Cash Assets) PPE = Property Plant and Equipment Net of Accumulated Depreciation ( ppent )/(Lagged Non Cash Assets) Interest = Interest Expense ( xint ) /(L agged Non Cash Assets) Rated is a binary variable equal to one if the firm has a credit rating and zero otherwise. BBB at the end of the fiscal year and equal to zero if the firm is rated below BBB or is unrated. Credit Market Tightness is measured as the average over the f our quarters of the fiscal year of the net fraction of domestic banks who respond to the Federal Reserve Bank study by saying that they are tightening standards for C&I loans to large and medium sized firms. Annual GDP is the annual United States gross d omestic product for fiscal year t, measured in billions of 2005 dollars. Dividend Payer is a binary variable equal to one if the firm pays dividends and zero otherwise. R&D = Research and Development Expense ( xrd )/(Lagged Non Cash Assets) Acquisitions = Cash Used in Acquisitions ( aqc )/(Lagged Non Cash Assets) All In Drawn is the fee over LIBOR paid on each dollar drawn. Maturity is the number of months until the credit line contract matures.

PAGE 95

95 Borrowing Base is a dummy variable equal to one if the credit line contract includes a borrowing base and zero otherwise. Secured is a dummy variable equal to one if the credit line contract is secured and zero otherwise. Number of Lenders is the number of lenders in the loan syndicate.

PAGE 96

96 LIST OF REFERENCES Almeida, Heitor, Murillo Campello, and Michael Weisbach, 2004, The cash flow sensitivity of cash, Journal of Finance 59, 1777 1804. Altman, Edward, 1968, Financial ratios, discriminant analysis and th e prediction of corporate bankruptcy, Journal of Finance 23, 589 609. Bates, Thomas, Kathleen Kahle, and Rene Stulz, 2009, Why do firms hold so much more cash than they used to?, Journal of Finance 64, 1985 2021. Boot, Arnoud, Anjan Thakor, and Gregory Ud ell, 1987, Competition, risk neutrality and loan commitments, Journal of Banking and Finance 11, 449 471. Campello, Murillo, Erasmo Giambona, John Graham, and Campbell Harvey, 2011, Liquidity management and corporate investment during a financial crisis, R eview of Financial Studies 24, 1944 1979. Chava, Sudheer, and Michael Roberts, 2008, How does financing impact investment? The role of debt covenants, Journal of Finance 63, 2085 2121. Chen, Zhaohui, Yan Hu, and Connie Mao, 2011, How much liquidity insuran ce can lines of credit provide? The impact of bank reputation and lending relationship, University of Virginia Working Paper. Colla, Paolo, Filippo Ippolito, and Kai Li, 2011, Debt Specialization, Universit Bocconi Working Paper Emery, Kenneth, Daniel Gates, Tom Marshella, and Sharon Ou, 2008, Migration of debt structures and revolver usage as firms approach default, Special Comment 1 20. Engleberg, Joseph, and Srinivasan Sankaraguruswamy, 2007, How to gathe r data using a web crawler: an application using SAS to search EDGAR, National University of Singapore Working Paper. Finerman, Stacey, 2008, IAR reports 3Q2008 results; downgrading to underperform, Goldman Sachs Global Investment Research October 31. Gor ton, Gary, and James Kahn, 2000, The design of bank loan contracts, Review of Financial Studies 13, 331 364. Holmstrom, Bengt and Jean Tirole, 1998 Private and public supply of liquidity, Journal of Political Economy 106, 1 40. Humer, Caroline, and Michael Erman, 2008, Ross says timing of GM drawdown raises questions, Reuters, September 22.

PAGE 97

97 Ippolito, Filippo, and Ande r Perez, 2012, Credit Lines: The Other Side of Corporate Liquidity, Universitat Pompeu Fabra Working Paper. Ivashina, Victoria, and David Scharfstein, 2010, Bank lending during the financial crisi s of 2008, Journal of Financial Economics 97, 319 338. Kaplan, Steven, and Luigi Zingales, 1997, Do investment cash flow sensitivities provide useful measures of financing constraints?, The Quart erly Journal of Economics 112, 169 215. Lins, Karl, Henri Servaes, and Peter Tufano, 2010, What drives corporate liquidity? An international survey of cash holdings and lines of credit, Journal of Financial Economics 98, 160 176. Nini, Greg, David Smith, a nd Amir Sufi, 2009, Creditor control rights and firm investment policy, Journal of Financial Economics 92, 400 420. Nini, Greg, David Smith, and Amir Sufi, 2011, Creditor control rights, corporate governance, and firm value, University of Pennsylvania Work ing Paper. Opler, Tim, Lee Pinkowitz, Rene Stulz, and Rohan Williamson, 1999, The determinants and implications of corporate cash holdings, Journal of Financial Economics 52, 3 46. Pagan Adrian, 1984, Econometric issues in the analysis of regressions with generated regressors, International Economic Review 25, 221 247. Journal of Financial and Quantitative Analysis 4, 201 228. Roberts, Michael, and Amir Sufi, 2009, Renegotiation of financ ial contracts: evidence from private credit agreements, Journal of Financial Economics 93, 159 184. Shockley, Richard, and Anjan Thakor, 1997, Bank loan commitment contracts: data, theory, and tests, Journal of Money, Credit and Banking 29, 517 534. Sufi, Amir, 2009, Bank lines of credit in corporate finance: an empirical analysis, Review of Financial Studies 22, 1057 1088.

PAGE 98

98 BIOGRAPHICAL SKETCH Ani graduated magna cum laude from University of Maryland with college and departmental honors. She earned a B.S. degree in finance and a minor in statistics Ani joined the PhD program in finance at the University of Florida in the f all of 2007. After completing her PhD in the summer of 2012, she will join the finance department at Salisbury University as an assistant professor of finance.