Short-term econometric forecasting analysis for Latin America

Material Information

Short-term econometric forecasting analysis for Latin America
Fullerton, Thomas M
Publication Date:
Physical Description:
vi, 91 leaves : ; 29 cm.


Subjects / Keywords:
Debt ( jstor )
Economic inflation ( jstor )
Economic models ( jstor )
Exchange rates ( jstor )
Forecasting models ( jstor )
Market prices ( jstor )
Modeling ( jstor )
Prices ( jstor )
Secondary markets ( jstor )
Time series forecasting ( jstor )
Dissertations, Academic -- Economics -- UF
Economic Forecasting -- Latin America ( lcsh )
Economic conditions -- Latin America -- 1982- ( lcsh )
Economic policy -- Latin America ( lcsh )
Economics thesis, Ph. D
bibliography ( marcgt )
non-fiction ( marcgt )


Thesis (Ph. D.)--University of Florida, 1996.
Includes bibliographical references (leaves 84-90).
General Note:
General Note:
Statement of Responsibility:
by Thomas M. Fullerton, Jr.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
023809897 ( ALEPH )
35730871 ( OCLC )

Full Text









Professor Carol Taylor West is probably the best known

and most widely published regional economic forecaster in the

world. I was acquainted with her research and professional

reputation long before applying for the senior economist

position on her staff at the Bureau of Economic and Business

Research. Over the past five years, we have faced computer

crashes, parameter instability, hurricane shocks, business

cycle inflection points, submission deadlines, bureaucratic

regulations, jet lag, and mind boggling statistical revisions.

Forecasting Program work was always accomplished in a

responsible manner befitting the tradition of econometric

excellence with which the University of Florida is so well

known. It has been a tremendous experience to forecast the

Florida economy and conduct research with Carol.

David Denslow has been a friend, colleague, and coauthor.

We have spent many hours discussing different policy issues

and business outlooks that one or the other was called upon to

discuss in a public forum. Similarly, Terry McCoy and I have

participated in a number of conferences on topics relating to

policy problems facing Latin America. Terry definitely has

keen insights with respect to the entire region.

Chunrong Ai helped clarify many of the mysteries of
modern econometric theory. He also saved me a lot of time and

effort by indicating which estimation strategies were viable

with respect to Chapter 4 and subsequent research developed

with that data set. Bill Bomberger spent generous amounts of

time going over Chapter 3 and making helpful suggestions.

Prior to his retirement, Henri Theil was on my dissertation

committee. Subsequent to his retirement, Professor Theil has

dropped by my office at least once a week to monitor my

progress and talk about the post-war history of econometrics.

Many other individuals in the College of Business

Administration helped me navigate exams, computer systems, and

Graduate School requirements. In no particular order, they

include Stan Smith, Larry Kenny, Janet Galvez, Tony Tracy, Min

Zhu, Yikang Li, Jennifer Cobb, Jon Hamilton, Richard Romano,

Janet Fletcher, Pam Middleton, Tony Daniele, Janet Rose, Cindy

Houser, Dot Evans, Dian Studstill, David Lenze, and Richard

Lutz. My parents, relatives, and friends provided welcome

relief on numerous occasions. It takes a lot of friendly

advice to get through a doctoral program.




ACKNOWLEDGEMENTS .........................................

ABSTRACT .................................................




2.1 Introduction ...............................
2.2 Previous Research ..........................
2.3 Methodology ................................
2.4 Estimation Results .........................
2.5 Policy Simulation Results..................
2.6 Conclusion .................................


3.1 Introduction ...............................
3.2 Literature Review..........................
3.3 Theoretical Model ..........................
3.4 Estimation Results .........................
3.5 Policy Simulation Results..................
3.6 Conclusion .................................


4.1 Introduction ...............................
4.2 Earlier Studies ............................
4.3 Empirical Analysis .........................
4.4 Conclusion .................................


REFERENCES ................................................

BIOGRAPHICALSKETCH .......................................


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



Thomas M. Fullerton, Jr.

August 1996

Chair: Professor Carol Taylor West
Major Department: Economics

Given the short-term nature, one year or less, of many

policy goals espoused by decision makers in Latin America, the

objective of this study is to examine the applicability of

different forecasting techniques to a subset of short-run

economic questions. Because quarterly national income and

product series are typically not available, most commercial

econometric forecasting analysis for the region is conducted

utilizing annual data. Monthly time series do, however, exist

for a large number of key macroeconomic and financial

variables. It is from the latter group of publicly available

data sets that the econometric modeling and simulation

exercises are drawn.

Chapter 2 examines potential impacts associated with

Colombian monetary authority efforts to cut the rate of

inflation by 10 over a 12-month period. Principal tools

relied upon include slower rates of nominal exchange rate

devaluation and money supply growth. Empirical analysis in

Chapter 2 is carried out utilizing transfer function

autoregressive moving average modeling. Model simulations

indicate that adherence to such a program can lead to

noticeable disinflation over a 24-month period.

Similar to other Latin American economies, Ecuador has

faced persistently high rates of inflation in recent years.

In May 1994, the government signed a stand-by loan agreement

with the International Monetary Fund that established a goal

of reducing the inflation rate to 15 percent over a 19-month

period. Chapter 3 develops an exchange rate augmented

monetary model to assess viability of the price stabilization

program. In contrast to the time series approach of Chapter

2, short-run inflationary dynamics are modeled using an

econometric framework.

Economic debates regarding Latin America in recent years

have been dominated by the debt crisis. In response to debtor

country defaults, many lenders reduced or reshuffled risk

exposures by selling debt certificates at discounts from face

value. Chapter 4 analyzes the predictability of secondary-

market debt prices for Colombia, Ecuador, and Venezuela.

Estimation is accomplished via generalized least squares over

24 separate historical periods utilizing monthly data. Model

simulations indicate that forecasting these prices is a

difficult task.


Econometric forecasting analysis began developing as a

field of research with the initial endeavors of Jan Tinbergen

and others in the 1930s (Dhane and Barten, 1989). These

relatively small macroeconometric models of the Dutch economy

were developed using annual data. Research on business cycles

in the United States began replicating and extending the Dutch

modeling examples in the 1940s. Increasing interest in short-

term economic fluctuations eventually led to the development

of quarterly forecasting models (Barger and Klein, 1954).

Along with the expansion of structural econometric

modeling and research with respect to forecasting and policy

analysis, time series statistics also became increasingly

sophisticated in the realm of predictive modeling. Most of

these efforts occurred with respect to high frequency monthly

data, especially univariate autoregressive moving average

models (Box and Jenkins, 1976). While often regarded as

competitors, time series models are frequently utilized as

complements to econometric equation sets and can also be

imbedded within a variety of forecasting systems (Clemen,

1989; Fullerton, 1989a; Zellner, 1994; West, 1996).


Latin American macroeconomic forecasting models began to

appear in the 1960s (Beltrdn del Rio, 1991). Similar to the

first models for the Netherlands and the United States, the

early Latin American models utilized systems of simultaneous

equations designed around national income and product account

(NIPA) data. Unlike most industrialized economies, however,

NIPA data in Latin America during this period tended to be

published only at an annual frequency. This constraint

precluded the development of Latin American quarterly

forecasting models in a manner analogous to what occurred in

many industrialized economies.

Although quarterly forecasting models have not been

widely disseminated in Latin America, large-scale forecasting

models built using annual NIPA data abound. Representative

examples include the CIEMEX-WEFA model for Mexico (Beltran del

Rio, 1991), the CIEPLAN model for Chile (Vial, 1988), the

Metroecon6mica model for Venezuela (Palma and Fontiveros,

1988), and the WEFA models for Colombia and Ecuador

(Fullerton, 1993a, 1993b). Among the most distinguishing

characteristics shared by these models are continuous

maintenance and enhancement over sustained periods of time.

All of the aforementioned studies provide detailed references

to the history of macroeconometric modeling in the region.

Econometric forecasting analysis using annual data in Latin

America has a fairly distinguished history and track record

that is well-documented.


Given the volatile behaviors of the majority of the

economies in Latin America in the 1980s, the absence of

quarterly forecasting tools hampered business planning

efforts. Forecasting conferences sponsored by commercial

entities such as Wharton Econometrics during the late 1980s

utilized simulation output from annual structural models for

Latin American countries of interest. Client questions at

these meetings were generally directed toward the first year

of the forecast period, largely ignoring outer period

extrapolation results (for example, see Fullerton, 1991).

This is not to imply that the traditional

macroeconometric Latin American models are regarded as

useless. Most analyses of international indebtedness

typically rely on annual data in order to examine the

consequences of changes in regional balance of payment

factors. Not surprisingly, significant effort was put forth

in recent years to enhance the current account-capital account

linkages and simulation performances in Latin American

macroeconomic models (Fullerton, 1993a). The relative lack of

forecasting models estimated with higher frequency data,

nevertheless, continues to pose an obstacle to corporate,

public sector, and multilateral agency planners and


Although quarterly national income and product account

data are still not widely available to researchers in Latin

America, monthly time series for many economic variables do


exist. Examples of the latter include inflation indices,

money supply measures, currency exchange rates, commodity

export prices, and international reserves. A logical step,

therefore, is to investigate the econometric properties and

predictability of the series presently available at the higher

frequency. The centerpiece chapters of this dissertation

examine three empirical forecasting questions using monthly

data sets from Colombia, Ecuador, and Venezuela. Distinct

estimation techniques are employed in each chapter and

simulation accuracy is summarized for all of the models

presented therein.

Chapter 2 utilizes a transfer function autoregressive

integrated moving average (transfer ARIMA) modeling framework

to analyze movements in consumer prices in Colombia.

Forecasting experiments are conducted with the resulting model

to shed light on potential impacts associated with an anti-

inflationary program enacted by the central bank. Results

indicate that this class of time series modeling can provide

useful insights with respect to macroeconomic trends in Latin

American countries. A revised version of this chapter was

published in the Journal of Policy Modeling (Fullerton,


Chapter 3 also examines the question of forecasting

short-term price movements. Consumer prices in Ecuador are

modeled using an econometric framework that incorporates

domestic macroeconomic factors and international input costs.


Parameter estimation is accomplished with a nonlinear

procedure that provides sufficient flexibility to handle even

mixed error structures. Simulation exercises are also

utilized to examine possible consequences associated with a

price stabilization program implemented by the central bank.

An abridged version of this study was published in Proceedings

of the American Statistical Association, Business and Economic

Statistics Section (Fullerton, 1995a).

Secondary market trades involving sovereign debt

certificates became widespread following the outbreak of the

international debt crisis in 1982. Chapter 4 employs a

generalized least squares modeling strategy to study the

predictability of short-run movements in secondary market debt

prices in Colombia, Ecuador, and Venezuela. As discussed

below, forecast period lengths and information inputs are

selected to reflect considerations facing financial market

participants. A revised version of this chapter appeared in

Applied Economics (Fullerton, 1993d). Additional research

extending the initial results on this topic was presented at

the 65th annual Southern Economic Association conference

(Fullerton, 1995b).


2.1 Introduction

Inflation has long been one of the most serious economic

problems facing policymakers in Latin America. Although

Colombia has traditionally enjoyed lower rates of inflation

than neighboring South American countries, in 1990, a

presidential election year, consumer prices rose in excess of

32 percent. In response, Colombian monetary authorities

introduced a series of policy innovations designed to lower

the inflation rate. Measures adopted include a progressive

opening of the economy to greater volumes of international

trade, fiscal austerity, tighter credit controls, and a slower

rate of currency devaluation (for details, see Fullerton,


This chapter examines some of the potential results

associated with the principal tools of the policy adjustment

effort. Time series characteristics of the consumer price

index (CPI), the money supply (Ml, defined as currency in

circulation plus demand deposits), and the peso/dollar

exchange rate (REX) are investigated. To measure the short-

run relationships among the CPI, Ml, and REX, transfer


function autoregressive-moving average (ARIMA) analysis is

applied (see Box and Jenkins, 1976, ch. 11).

Selection of the nonlinear model estimation methodology

was motivated by two factors. First was the usefulness of

transfer function ARIMA analysis for short-term forecasting

applications. Second was interest expressed in a previous

study which utilized this technique to analyze price dynamics

in the United States (Fullerton, Hirth, and Smith, 1991).

Specifically, economists within the Research Department of the

Central Bank in Bogota wished to see whether the same approach

would prove useful with respect to the Colombian economy.

Subsequent sections of the paper present a brief review

of the literature, methodology, and empirical results. Policy

simulation exercises designed to reflect exchange rate and

monetary policies in Colombia follow. A summary and

conclusion form the final section of the chapter.

2.2 Previous Research

In one of the earliest time series studies developed for

an economy in Latin America, Cabrera and Montes (1978) utilize

univariate ARIMA techniques to model the CPI in Colombia.

Logarithmic transformation, regular and seasonal differencing

of the monthly CPI series are used to induce stationarity. An

equation containing an autoregressive term at lag 1 and a

seasonal moving average term at lag 12 yields statistically

significant parameter estimates. While simple in structure,


the model exhibits good statistical traits and is found to

simulate historical movements of the CPI successfully.

Empirical evidence is provided that Colombian inflation,

although high relative to many industrial economies, is stable

enough to be modeled and predicted with a fair degree of


Montes and Candelo (1982) propose a simultaneous system

of equations for money, prices, international reserves, and

the exchange rate. Full information maximum likelihood

estimation is used to calculate model parameters which reflect

the hypothesized coefficient restrictions. Although quarterly

data are used, the lag structure of the model is fairly

simple. Domestic monetary conditions and the rate of

devaluation are both found to positively affect consumer

prices in a statistically significant manner. The magnitudes

of the exchange rate coefficients exceed those of the monetary

variables in each of the different sample periods.

Leiderman (1984) utilizes vector autoregressions to

analyze inflation in Colombia and Mexico. Changes in the

rates of growth in the economy and the money supply are

included in each model. In the case of Colombia, the results

indicate that variations in the rate of change of Ml

systematically affect the CPI, but not the converse. From a

policy perspective, this implies that monetary authorities do

not engage in "accommodative" measures in response to

production and inflationary shocks. This result may stem from


the usage of monetary policy to achieve goals other than price

stability. These include economic growth and export

diversification. From an econometric standpoint, this result

is also important because it implies that unidirectional

Granger causality exists between M1 and the CPI in Colombia.

A number of recent studies have examined inflationary

dynamics in the United States. Koch, Rosensweig, and Whitt

(1988) investigate the relationship between the exchange rate

and consumer prices. Cross correlation functions are used to

suggest the number of lags to be included in the regression

equations. Granger causality tests imply a unidirectional

channel of influence from the exchange rate to prices. The

inflationary impacts associated with a change in the

international value of the dollar are found to extend more

than 12 months.

Fullerton, Hirth, and Smith (1991) consider the effects

of exchange rate and other financial and commodity price

variable movements on the CPI in the United States. Transfer

function ARIMA analysis indicates that inflationary impacts

resulting from variations in the exchange rate generally take

more than a year. Credit conditions, as proxied by interest

rate spreads, are found to influence consumer prices

relatively quickly. The slow speed of adjustment from the

exchange rate compared to domestic financial conditions may

reflect incomplete pass-through effects which characterize

large industrial economies.


This is in contrast to what might be expected for a

smaller economy such as Colombia's, where pass-through

effects, statistically significant relationships between

international currency values and inflation, are often strong

and relatively quick (see Leith, 1991). Empirical evidence

reported elsewhere (Edwards, 1985) indicates that movements in

the rate of devaluation are helpful in modeling nominal

interest rates in Colombia. Unfortunately, the latter study

does not directly test the relationship between prices and the

peso/dollar exchange rate.

2.3 Methodology

The methodology utilized in this paper is similar to the

multiple ARIMA approach applied to the United States by

Fullerton, Hirth, and Smith (1991). This technique does not

apply any a priori modeling restrictions on the equations

estimated. It is a useful procedure for investigating high

frequency time series data because it can accommodate

different lag structures in a flexible and efficient manner.

Equations developed using this approach are also easy to

simulate and can help analyze policy proposals.

Univariate ARIMA equations are estimated for the

stationary components of the CPI, Ml, and REX series. The

results are then used to specify and estimate an ARMA transfer

function. A weakly stationary series is defined as one whose

mean and variance do not change over time. Stationarity in


the means of the series is attained through regular and

seasonal differencing. Logarithmic transformations are used

to induce homoscedasticity (see Pankratz, 1983).

The general form of a univariate ARIMA model estimated

for the CPI can be written as follows:

(2.1) Pt = [Q(B)QS(B)Ut] / [H(B)Hs(B)]

where Pt is a stationary series derived from the original CPI

series, B is a backshift or lag operator, Q(B) is a moving

average polynomial of order q, QS(B) is a seasonal moving

average polynomial of order qS, Ut is the disturbance term,

H(B) is an autoregressive polynomial of order p, and Hs(B) is

a seasonal autoregressive polynomial of order pS. The ARMA

model for the CPI is also used in the estimation of the

transfer function equations.

To examine the effects of other variables on the CPI,

transfer function ARIMA models are estimated. These models

are used to determine if the input series are incrementally

useful in explaining the variation of the CPI beyond the

information obtained within the price index itself (see Box

and Jenkins, 1976, ch. 11). Prior to estimating a transfer

function equation, cross correlation functions (CCF) are used

to indicate which lags of an input series may contribute

incremental information to the univariate ARMA model already

estimated for the CPI. Residuals from the three univariate


equations are used to calculate CCFs between the CPI and the

other variables (see Chatfield, 1984, p. 173).

Statistical, or Granger, causality is assumed to be

unidirectional in transfer ARIMA models. If this assumption

is reasonable, movements in REX and M1 will chronologically

precede statistically significant changes in CPI. The

converse will not hold if causality is unidirectional. To

examine the possibility that feedback or endogeneity exists

between the CPI and the other series, ordinary least squares

regression is used to construct Granger causality F-tests.

The general form of the transfer function can be written

in the following manner:

(2.2) Pt = [W(B)Rt + G(B)Mt + Q(B)Qs(B)Ut] / [H(B)Hs(B)]

where the univariate model presented in Equation 2.1 is

augmented by incorporating general order polynomials, W(B) and

G(B), for the respective input variables. Rt and M. represent

stationary input series derived from the exchange rate and the

money supply data discussed above. Because the analysis is

conducted within a dynamic framework, coefficient restrictions

are not hypothesized, but the sums of the coefficients

associated with each individual input series are expected to

be positive.


2.4 Estimation Results

Data used in this chapter are from the central bank in

Bogota. Monthly observations for all three series are

published in Revista del Banco de la Rep~blica (for example,

see Cabrera and Montes, 1978). The sample period, January

1967 through December 1990, corresponds to the crawling peg

era of exchange rate determination in Colombia.

Similar to Cabrera and Montes (1978), a logarithmic

transformation of the CPI is taken prior to regular and

seasonal differencing to obtain stationarity. The same steps

are taken with respect to REX and Ml prior to modeling.

Results of the stationarity tests for all three series appear

in Table 2.1. The unit root tests are performed with both

constant and trend terms.

Autocorrelation and partial autocorrelation functions

suggested the forms of the univariate ARIMA equations reported

in Table 2.2. Despite the presence of 11 years of additional

data, the results confirm the AR(l), SMA(12) specification

employed by Cabrera and Montes (1978, p. 1137) to model the

CPI. Parameter estimates reported in Model 2.3 carry the same

signs and are similar in magnitude to those estimated in the

previous study (0.450 versus 0.645, and -0.924 versus -0.858).

Equation 2.3 is incorporated in the estimation of the transfer

function equations.

Table 2.1: Unit Root Tests for Stationarity

Series Aug Dickey-Fuller t-stat MacKinnon crit value

P 9.081 (with const, trend, 1 lag) -4.001 (1% lvl)
R 5.252 (with const, trend, 1 lag)
M -13.195 (with const, trend, 1 lag)

Table 2.2: Univariate ARIMA Models


+ 0.450*Pti

- 0.924*Ut.12

Q(38) = 52.222

+ 0. 795*Rt1

- 0.784*Ut12

Q(38) = 31.893

- 0.229*Ut.i

- 0. 634*Ut12

Q(38) = 59.339










The sample period is January 1967 December 1990.
Numbers in parentheses are computed t-statistics.
Ljung-Box Q-statistics calculated for 38 lags are reported.


Ordinary least squares F-tests are used to determine if

unidirectional Granger causality exists between consumer

prices and the two input series. Results for the F-tests,

calculated for 12 and 18 lags, are reported in Table 2.3. The

tests constructed to examine the relationship between the CPI

and M1 utilize seven years of additional data, but support the

conclusions reported by Leiderman (1984). Because monetary

policy in Colombia appears to be conducted independently of

inflationary shocks, the transfer function methodology can be

used to measure the effect of the money supply on inflation.

Similar to results reported for the United States (Koch,

Rosensweig, and Whitt, 1988), the causality tests for the CPI

and REX series indicate that changes in the inflation rate do

not precede systematic variations in the exchange rate. From

an historical perspective, the result is not surprising.

There have been several episodes during the crawling peg era

in Colombia when authorities have permitted the exchange rate

to become overvalued by failing to devalue the local currency

quickly enough to account for inflationary differentials with

major trading partners. Typically, this has tended to take

place following "coffee bonanzas" when Colombian international

reserves are high due to strongly positive merchandise trade

surpluses (Kamas, 1986; Ocampo, 1983).

Table 2.3: Granger Causality Tests

Causality Direction Number of Lags Computed F-stat

CPI => REX 12 0.834

CPI => REX 18 0.799

CPI => M1 12 0.009

CPI => M1 18 0.278

Degrees of freedom for the regressions with 12 lags:
12 for the numerator and 263 for the denominator.

Degrees of freedom for the regressions with 18 lags:
18 for the numerator and 251 for the denominator.


There have also been periods when local price changes

have moved in concert with those of Colombia's trading

partners, rendering unnecessary any modification in the

crawling peg. As a result, the rate of devaluation has not

always been adjusted proportionately to variations in the

domestic rate of inflation. From an economic policy

perspective, the results in Table 2.3 indicate that Colombian

monetary authorities take into account goals and variables

other than domestic inflation when determining rate of

devaluation embodied in the crawling peg for the peso.

Econometrically, this implies that transfer function ARIMA

analysis can be used to model the influence of the

international value of the peso on domestic prices.

Similar to the CPI, stationarity in the REX and M1 series

is attained after logarithmic transformation, and seasonal and

regular differencing. Unit root tests for both variables are

reported in Table 2.1. As noted above, univariate equations

for the exchange rate and money supply series appear in Table

2.2. These equations also exhibit good statistical traits

such as high computed t-statistics and relativley low Q-

statistics. Residuals from the three univariate models were

used to estimate CCFs containing 18 lags. Both CCFs indicated

that the principal effects resulting from a change in either

input variable are incorporated in the CPI within a relatively

short period.


Transfer function equations are reported in Table 2.4.

Models 2.6 and 2.7 include only one independent variable, REX

and Ml, respectively. Model 2.8 includes both input series.

The exchange rate is included with lags of 2 and 10 months.

The money supply is included with a 9-month lag.

Autoregressive terms at lag 1 and seasonal moving average

terms at lag 12 are included in all three equations. Similar

to Montes and Candelo (1982), the exchange rate input

coefficients are larger than that of Ml.

Inclusion of the input series improves the Q-statistic

estimated from the residuals associated with each equation.

Virtually all of the improvement in the white noise test

results from the introduction of the lagged stationary

exchange rate data. Coefficients estimated for these series

are statistically significant in Model 2.6 and Model 2.8. The

lagged stationary component of the money supply exhibits the

hypothesized sign, but is not significant at the 5-percent


Because money is a useful predictor of inflation in

Colombian macroeconometric models using annual data

(Fullerton, 1993a), the results encountered in this chapter

are unexpected and further research is warranted. Another

possibility is that M1 may not be the correct money stock

measure with respect to Colombian price dynamics. Studies

completed for other economies indicate that broader liquidity

aggregates such as M2 may be useful (Hallman, Porter, and

Table 2.4: ARIMA Transfer Functions


+ 0.459Pt.i

+ 0.201*Rt2

- 0.890Ut.12

+ 0.076Rt.10

Q(38) = 39.826

+ 0.447*Pti

+ 0.030*Mt.9

- 0.929*Ut12

Q(38) = 51.777

+ 0.457*Pt.I

+ 0.216*Rt2

+ 0.034*Mt9

- 0.895Ut.12

+ 0.077*RtI0

Q(38) = 39.389











The sample period is January 1967 December 1990.
Numbers in parentheses are computed t-statistics.
Ljung-Box Q-statistics calculated for 38 lags are reported.


Small, 1991, and Ikhide and Fullerton, 1995). Unfortunately,

money supply estimates other than M1 currently do not exist

for Colombia. This problem is fairly widespread in South

America and has long posed difficulties in the analysis of

monetary economics in the region (see Fullerton and Kapur,


2.5 Policy Simulation Results

The Gaviria administration announced in December 1990

that it would attempt to lower the inflation rate to 22

percent in only twelve months (for discussion, see Fullerton,

1991). To attain this goal, two principal tools were to be

employed. After an extended period of aggressive devaluation,

the nominal rate of devaluation for the peso/dollar exchange

rate was to be reduced. Policymakers also announced that

growth in the money supply would be limited to 22 percent.

To examine the potential effects of these policy

innovations on the CPI, simulations are created using Equation

2.8. In the first exercise, the 12-month growth rate for Ml

is assumed to be immediately limited to 22 percent for an

initial 12-month period, and then lowered to 19 percent the

following year. Similarly, the 12-month crawling peg rate of

devaluation is assumed to be held at 22 percent throughout the

first year, and later be revised downward to 19 percent during

the following year. The 19 percent rates of change are chosen

to reflect longer-term policy goals discussed by the

government, including eventual attainment of a 15 percent

annual rate of change for consumer prices (Fullerton, 1991).

Simulating Model 2.8 under these policy assumptions

yields several interesting results that are reproduced in

Table 2.5. During the first six months, the inflation rate

oscillates near 29 percent. Subsequently, the 12-month rate

of change in the CPI declines sharply to 24 percent by year-

end. During the next 12 months, disinflation subsides as the

severity of the cutbacks in the rates of change of both input

variables is moderated. Year-end inflation declines to 22

percent under this scenario. As will be illustrated below,

not all of the 10-point improvement can be attributed to the

government policy package.

Not surprisingly, Colombian monetary authorities did not

introduce the exchange rate and credit policy changes in the

abrupt manner depicted above. Accordingly, Equation 2.8 is

also simulated under an alternative set of assumptions whereby

intermediate policy steps are implemented more gradually.

Additionally, actual exchange rate and money supply data for

the first six months of 1991 are employed. These data reflect

slower attainment of the new policy goals espoused by the

Finance Ministry and the Central Bank. Subsequent movements

in the 12-month rates of change for the exchange rate and the

money supply are assumed to steadily decline to 19 percent by

December 1992 for both variables.

Table 2.5: Immediate Implementation Policy Simulation Results


1 32.1 22.0 22.0
2 30.7 22.0 22.0
3 28.9 22.0 22.0
4 29.1 22.0 22.0
5 29.6 22.0 22.0
6 29.2 22.0 22.0
7 28.9 22.0 22.0
8 27.9 22.0 22.0
9 26.8 22.0 22.0
10 26.2 22.0 22.0
11 25.2 22.0 22.0
12 24.1 22.0 22.0

13 23.5 19.0 19.0
14 23.3 19.0 19.0
15 22.5 19.0 19.0
16 22.5 19.0 19.0
17 22.4 19.0 19.0
18 22.4 19.0 19.0
19 22.4 19.0 19.0
20 22.4 19.0 19.0
21 22.4 19.0 19.0
22 22.3 19.0 19.0
23 22.1 19.0 19.0
24 22.1 19.0 19.0


Results of the second simulation exercise are reported in

Table 2.6. The more rapid rate of depreciation allows the 12-

month inflation rates to remain above 30 percent throughout

the first semester of the test period. Steady declines in the

rates of change for both REX and M1 force inflation to decline

noticeably during the next six months, eventually reaching 25

percent. Year-end inflation slows to 22 percent during the

subsequent 12-months of the simulation period, when growth

rates for the input variables gradually decline to 19 percent.

It is useful to note that the second set of medium-range

simulation results reported in Table 2.6 do not vary

significantly from those obtained in the first exercise.

Despite imperfect implementation efforts, these results

indicate that the government can still attain its stated

policy goals. Because a 24-month period is still needed to

lower the rate of change in the CPI by 10 percentage points,

the government policy claims are again shown to be slightly

optimistic. Furthermore, as shown in the final simulation

exercise presented in Table 2.7, much of the improvement in

the inflation rate cannot be attributed to policy design


Of course, policy indecision can also result in no

progress being made toward achieving either intermediate

target. To examine the potential consequences associated with

such an eventuality, model simulations were also tested in

Table 2.6: Gradual Implementation Policy Simulation Results








Table 2.7: No Implementation Policy Simulation Results


1 32.2 30.0 30.0
2 31.9 30.0 30.0
3 31.6 30.0 30.0
4 31.2 30.0 30.0
5 30.9 30.0 30.0
6 30.6 30.0 30.0
7 30.3 30.0 30.0
8 29.9 30.0 30.0
9 29.6 30.0 30.0
10 29.4 30.0 30.0
11 29.2 30.0 30.0
12 28.8 30.0 30.0

13 28.5 30.0 30.0
14 28.3 30.0 30.0
15 28.0 30.0 30.0
16 27.7 30.0 30.0
17 27.4 30.0 30.0
18 27.1 30.0 30.0
19 26.8 30.0 30.0
20 26.6 30.0 30.0
21 26.4 30.0 30.0
22 26.2 30.0 30.0
23 26.3 30.0 30.0
24 26.2 30.0 30.0


which the growth rates of the input series were held constant

at 30 percent per year. In the absence of any change in the

conduct of monetary management and depreciation rate

determination, the annual rate of inflation stabilizes at 26

percent by the end of the 24-month simulation period. That

rate is well above the announced government target range.

More importantly, the 6-point improvement which results

in a simulation exercise in which no progress is made with

respect to the intermediate policy targets. This result

indicates that only 40 percent of policy attainment embodied

in Tables 2.5 and 2.6 is by government design. The remaining

six-tenths of the 10-point reduction in the annual rate of

change in consumer prices would, on the basis of the above

modeling and simulation framework, have resulted anyway.

2.6 Conclusion

In response to growing inflationary pressures, economic

policymakers in Colombia announced that the inflation rate

would be slashed by 10 percentage points to 22 percent over a

12-month period. Two principal tools were selected to foster

disinflation. The nominal rate of devaluation for the

peso/dollar exchange rate was cut and the rate of growth of

the money stock was reduced.

This chapter examines the empirical relationship between

those variables and the consumer price index. Econometric

results are similar to those reported in previous studies for


Colombia and the United States. Transfer ARIMA functions are

estimated and simulated to determine the potential impacts of

the new policies. These exercises indicate that substantial

progress in the anti-inflationary program may be attained

following the implementation of said policy efforts, although

not as quickly as stated by government officials. More

importantly, over half of the 10-point gain results even if

the rates of nominal currency devaluation and money supply

expansion are held constant.

Because of structural economic and administrative policy

changes taking place in Colombia, additional research will

eventually be necessary. Future studies may find it useful to

consider the effects of other variables such as wage rates,

industrial capacity utilization, and commodity prices on the

CPI. Given the insignificant, at the 5-percent level,

coefficient associated with M1, model estimation utilizing

alternative series designed to reflect monetary conditions may

also prove helpful. Introduction of new variables such as

wage rates may necessitate usage of an estimation technique

different from the transfer function methodology described

above. This is due to the possibility that feedback effects,

or simultaneity, may exist between the CPI and other potential

input series.

At present, the Colombian economy is being opened to

international trade and a free-market exchange rate system is

slated to be implemented. These policy innovations could


potentially render the above parameter estimates obsolete.

ARIMA intervention analysis (Box and Tiao, 1975) may prove

beneficial in subsequent empirical research designed to

examine this possibility. It is interesting to note, however,

that the similarities between the empirical results reported

in this paper and those analyzed in earlier studies indicate

that agent responses have generally been relatively inelastic

with respect to changing monetary and exchange rate policies

in Colombia.


3.1 Introduction

Similar to other Latin American economies, Ecuador has

faced persistently high rates of inflation in recent years.

Although inflation was substantially lower in 1993 and 1994,

excessive money supply growth in early 1995 clouded prospects

for additional short-term improvements. Prior to the first-

quarter 1995 border skirmish with Peru, the government had

signed a stand-by loan agreement with the International

Monetary Fund that established a goal of reducing the

inflation rate to 15 percent over a 19-month time frame (Banco

del Pacifico, 1994a). Not withstanding government assurances

that inflation would decelerate to its target rate by year-end

1995, very little econometric analysis using short-term

forecasting methods appears to have been relied upon in

developing the new policy targets.

In its attempt to slow price movements, the Duran Ballen-

Dahik Garzozi administration introduced a variety of new

policy measures. They include import liberalization, fiscal

austerity, and a slower rate of currency depreciation. By

reducing price pressures, the government hopes to improve

economic welfare by enabling the Ecuadorian economy to operate


more efficiently. This argument is very similar to those

aired in advanced economies such as the United States (Motley,

1993) and analyzed in other developing nations (Zind, 1993).

What is unique, however, is the magnitude of the

disinflationary goals set by the Ecuadorian policymakers. As

a result, short-run price stabilization has become the center

piece of government policy efforts in Ecuador.

This chapter examines potential results associated with

the two principal adjustment tools, money supply growth and

exchange rate movements. Despite ongoing difficulties with

inflationary uncertainty and the ambitious nature of current

policy goals, careful econometric analysis of short-run price

movements in Ecuador has not previously been conducted. To

bridge this gap, a modeling framework is proposed, tested, and

used to develop policy simulation exercises for monthly

Ecuadorian price data.

In contrast to the time series methodology utilized for

Colombia, an econometric approach is followed in the analysis

conducted for Ecuadorian price movements. Selection of this

alternative approach was motivated by two factors. First was

a discussion on quantitative analysis of developing country

inflation held at the 32nd International Atlantic Economic

Conference. Second was interest expressed by economists at

the Research Department of the Central Bank in Quito with

respect to attempting to develop a short-run inflationary

model similar to the long-run monetary-import cost approach


incorporated in Fullerton (1993b). Subsequent sections of

this chapter offer a review of the literature, theoretical

model, and empirical results. Suggestions for future research

are summarized in the conclusion.

3.2 Literature Review

The seminal research on inflationary dynamics in

developing countries was conducted on Chilean data by

Harberger (1963). That early paper interestingly points out

that analyzing nominal data in level form could result in

spurious correlations in equations estimated for highly

inflationary economies. To circumvent this problem,

percentage rates of change are utilized in a linear regression

framework based on the quantity theory of money. What became

known as the "Harberger" framework incorporates real income,

current and lagged values of the money supply, and the

opportunity cost of holding cash balances.

The success of this initial effort conducted on Chilean

data spurred a series of replicated studies for other

developing countries. Vogel (1974) estimates an inflation

equation for several Latin American economies, including

Ecuador, using annual data. Results confirm the overall

usefulness of the Harberger model. Unlike the study at hand,

Vogel utilizes a sample period during which inflation averaged

less than 4 percent per year in Ecuador and the exchange rate

was fixed.


Following numerous applied econometric studies utilizing

this approach, it became apparent that its reliance on

domestic variables alone often provided unsatisfactory

results. Bomberger and Makinen (1979) provide a thorough

examination of the Harberger model using quarterly data for

Korea, Taiwan, and Vietnam. Extensive testing is conducted

using quarterly data in order to establish whether a suitable

characterization of inflation is provided. Encouragingly, the

parameter estimates do not appear sensitive to the time period

selected. However, the elasticities with respect to money and

real income are not always unitary as hypothesized. Also, the

coefficient signs for the cost of holding money variables are

sometimes negative.

Hanson (1985) extends the Harberger framework in a

systematic fashion to incorporate an important missing

component, import costs. An implicit cost function is

utilized to derive an aggregate supply curve which includes

local prices of imported inputs. When the underlying

production function is homogeneous of degree one, inflation

becomes a weighted sum of money supply changes and import

prices. This is important for studies using higher frequency

data if the problem of measurement bias engendered by

interpolated values of real output, generally published on

either a quarterly or annual basis in developing countries, is

to be avoided (see Bomberger and Makinen, 1979). The model

also implies the elasticity of inflation with respect to money


growth is less than one. Empirical results in the Hanson

article strongly support the inclusion of import prices or the

rate of devaluation in models of inflation.

Subsequent research has provided additional evidence in

favor of the augmented Harberger-Hanson approach wherein the

effect of import prices on inflation is considered. Koch,

Rosensweig, and Witt (1988) and Fullerton, Hirth, and Smith

(1991) both report positive linkages between the trade-

weighted exchange value of the dollar and consumer prices in

the United States. These empirical studies indicate a

unidirectional channel of influence from the exchange rate to

domestic prices exists in the United States economy. As will

be discussed below, causality direction has important

implications for both model form and estimation technique.

Developing country studies have also confirmed the

usefulness of an augmented modeling treatment of inflationary

dynamics. Sheehey (1976) reports some of the early

econometric work along these lines. Sheehey (1980) reaches

additional conclusions on the basis of empirical tests that

indicate that accurate assessment of austerity policy efforts

will likely require explanatory variables representing cost

push factors. More recently, Brajer (1992) provides evidence

that the latter category of models may offer better

specifications than those which rely solely on domestic

economic factors. Conclusions in that article are reached on

the basis of F-tests for different regressor sets. Similarly,


Fullerton (1993b) successfully imbeds a variant of this

approach in a large-scale macroeconometric forecasting model

for Ecuador using annual data.

There have been very few dynamic models estimated on the

basis of monthly data for developing economies. Given that

most business decisions in highly inflationary countries are

reached within a short-range context, this is an area which

needs to be addressed. As detailed in Chapter 2, Fullerton

(1993c) empirically examines Colombian anti-inflationary

efforts utilizing monthly data with an ARIMA transfer

function. The estimated model is found to generate realistic

simulation scenarios for policy analysis. The results also

support the hypothesis of inflation rate inelasticity with

respect to monetary growth. As in the Chilean equation

reported by Hanson (1985), and the Argentine model presented

in Sheehey (1976), exchange rate price effects are found to

outweigh the monetary coefficient. The latter is somewhat

surprising given that imported goods and services comprise

less than 20 percent of Colombian gross domestic product.

3.3 Theoretical Model

Harberger's (1963) model is based on the traditional

quantity theory of money equation:

(3.1) MV = PQ,


where M represents some measure of the money stock, V is the

velocity of circulation, P is the price level, and Q is real

output. Velocity is not assumed to be constant. Instead, it

is hypothesized to be a predictable function of other

macroeconomic variables such as the cost of holding cash

balances. Given the typical variability of velocity in many

Latin American economies, this aspect of the theoretical model

is potentially important (see Clavijo, 1987).

To utilize percentage changes, the variables can be

transformed by natural logarithms and first differenced.

Introduction of a time subscript, and rearrangement of the

terms, yields the basic Harberger equation:

(3.2) DPt = DMt DQt + D(DPt.1),

where the last term results from substituting for velocity and

D represents a difference or backshift lag operator. Usage of

the lagged change in the inflation rate to proxy for the

implicit cost of holding money is motivated by the fact that

developing countries such as Ecuador have frequently imposed

government regulations on interest rates. The latter have

occasionally caused savings and loan rates to become negative

in real terms. Unadjusted interest rates from these periods

in Ecuadorian economic history do not, therefore, provide

accurate estimates for the cost of holding idle cash balances.


Equation 3.2 implies that inflation will vary positively

with the money supply and inversely with respect to real

output. A statistically significant intercept term will enter

the estimated equation if there is a trend in the velocity of

circulation. If only contemporaneous lags of M and Q enter in

the equation, the parameters for both variables are

hypothesized to be unitary. This can be tested empirically

with the following specification:

(3.3) DPt = ao + a1DMt a2DQt + a3D(DPt.1) + u3,

where al and a3 are hypothesized to be positive, and the

absolute values of al and a2 should both be statistically

indistinguishable from one. The last argument in the

expression represents the disturbance term.

Hanson (1985) proposes an implicit cost function dual of

an aggregate production function which is homogeneous of

degree one. Derived output supply functions from this

framework will be homogeneous of degree zero in input and

output prices. Equation 3.4 expresses this relationship using

logarithmic first differences:

(3.4) DQt = b0 + bIDPt b2DPIt + u4,

where PI represents imported input prices. When the relative

prices of imported inputs increase, output is assumed to


decline. The standard homogeneity assumptions for production

and derived supply relations imply that bI b2 = 0.

Equation 3.4 can be substituted into Equation 3.3 to

eliminate the output term from the expression to be estimated.

As noted in the literature review, this step is useful for

avoiding interpolation bias in empirical studies of monthly

inflation for countries such as Ecuador where GDP is published

at quarterly and/or annual frequencies. The resulting

equation can be written as follows:

(3.5) (1 + a2bl)DPt = a0 a2b0 + a1DMt + a2b2DPIt +

a3D(DPt.1) + u5.

Equation 3.5 can be further simplified prior to

estimation. Dividing through by the left-hand side constant

term and rearranging terms such that the price series remains

as the dependent variable yields the following relation:

(3.6) DPt = co + cDMt + c2DPIt + c3D(DPt.1) + u6,

which also has testable properties. Importantly, the

coefficient on the monetary variable, cl, is now hypothesized

to be significantly less than one. Also important, with the

possible exception of the intercept, all of the regression

parameters in Equation 6 are expected to be positive.


Several other properties of this model are worth noting.

In particular, the theoretical coefficient restrictions

described earlier have interesting implications. Namely, a,

and a2 are hypothesized as equal to one, and bI and b2 are

equal in absolute value in the version of the model developed

thus far. Substitution into Equation 3.5 implies cI + c2 = ,

which can also be tested.

As indicated in the literature review, Equation 3.6 has

provided a useful framework for analyzing quarterly and annual

inflation rates. But because the lag structure in this model

is fairly short, it may require additional modification prior

to estimation. This possibility does not reflect any

deficiencies in the theoretical model as such, but arises due

to the fact that short-term models rely upon monthly data. As

a result, if the inflationary impact of a change in the money

supply is felt over the course of one calendar year, the

implied lag structure for a model estimated with data

published at a monthly frequency would potentially range up to

12-months in length. Equation 3.7 takes into account this

empirical issue which has confronted and confounded

researchers for many years (see Laidler, 1993):

(3.7) DPt = co + c1DMti + c2DPIt.j + c3D(DPt.l.k) + u7,

where lag subscripts i, j, k = 0, ..., n, respectively.


The above model provides an attractive starting point for

examining inflationary trends in an economy. It is not,

however, without potential problems for analyzing price

movements in a relatively high inflation country such as

Ecuador. A principal concern arises from the fact that

Equation 3.7 treats all of the regressors as exogenous or pre-

determined. In doing so, it does not allow for the

possibility of statistical feedback or endogeneity between the

left-hand and right-hand side variables.

If a central bank yields to political pressures and

engages in accommodative monetary policy in the face of

inflation shocks, this assumption would be violated. As noted

in Chapter 2, research conducted using higher frequency data

for Colombia indicates that the causality paths in that

economy are unidirectional as implied by Equation 3.7 (see

Fullerton, 1993c, and Leiderman, 1984). While short-term

forecasting models for Ecuadorian inflation have not been

previously developed, domestic prices, monetary aggregates,

and import prices are modeled simultaneously in the

macroeconometric model estimated using annual data by

Fullerton (1993b). It would not be surprising if the feedback

relations encountered in that paper also emerge in the monthly

time series utilized below. As noted elsewhere, monetary

authorities in Ecuador have occasionally been forced to yield

to political pressures (Garlow, 1993). Granger causality


tests will be used to test the severity of this potential


A second possible concern arises from utilizing first

differenced, log-transformed time series data in the equation

to be estimated. If the resulting series are stationary, the

equation can be estimated without risk of obtaining spurious

correlations in the results. As shown in many studies of

hyperinflationary economies, however, higher order

differencing may be required to induce stationarity during

periods in which prices increase rapidly (Engsted, 1993).

Because Ecuador has not undergone any hyperinflationary

episodes, first differencing should remove nonstationary

trends from the variables in question but this assumption must

be tested. The latter tests are accomplished below via a

battery of unit root tests, not all of which are reported.

A third concern arises from the fact that monthly import

price deflators do not exist for Ecuador. To circumvent this

problem and also avoid interpolation bias, a trade-weighted

exchange rate index is used as a proxy for imported input

prices. The index utilized was developed econometrically and

takes into account export and import volume changes with

Ecuador's major trading partners. It also offers a single

monthly index for periods when the government has instituted

multiple exchange rate systems (Fullerton, 1989b).

To construct the currency index, individual currency

weights are calculated as the sum of imports and exports with


each of ten major trading partners and divided by total

international trade in each year. Over the sample period,

annual trade with Ecuador's top ten import sources and export

destinations accounts for more than 70 percent of its total

trade volume in any given year. Products of the bilateral

trade coefficients and the respective currencies are then used

to construct the exchange rate index using a geometric mean.

The latter method is selected to avoid problems which can

potentially result for indexes constructed using arithmetic

means during periods of inflationary variability (see Batten

and Belongia, 1986, Dutton and Grennes, 1987, and Kercheval,


For periods when multiple exchange rates were instituted,

a blended index is calculated. Weights for the free-market

and official government intervention exchange rates are

obtained from the Central Bank publication, Inforuacift

Estadistica Mensual (various issues). Econometric results for

total and disaggregated imports, and non-petroleum exports

indicate that the blended rate provides a more accurate

measure of the appropriate currency basket for Ecuador

(Fullerton, 1989b). Diagnostic tests were also conducted

using currency baskets with different numbers of trading


Introduction of the trade-weighted exchange rate index to

Equation 3.7 causes the model to be estimated to take the

following form:


(3.8) DPt = go + g1DMt-i + g2DTWXt- + g3D(DPt-j-) + u8,

where TWX stands for logarithmic first differences of the

nominal version of the monthly trade-weighted exchange rate

index calculated for the sucre by The WEFA Group (formerly

Wharton Econometrics). Although incorporation of the monthly

exchange rate index avoids the problems associated with

implicit price deflator interpolation, it may introduce other

problems due to the fact that import prices in Ecuador are

affected by global supply and international demand conditions

in addition to exchange rate movements (see Fullerton 1993b).

A fourth observation regarding potential pitfalls

associated with the theoretical specification of the model is

worth noting. Only one class of factor input prices, that for

imports, is included. While this represents an improvement

over the original Harberger framework, it may overlook

additional important inputs such as labor. If the version of

the model developed herein omits relevant variables to the

inflationary process in Ecuador, it is likely that estimated

residuals associated with the empirical version will not be

randomly distributed. If this is the case, then correction

for serial correlation will be necessary. Extension of the

model to overcome this problem cannot currently be

accomplished due to data availability.


3.4 Estimation Results

In order to examine whether the working series included

in Equation 3.8 are stationary, unit root tests are conducted

for each series utilized in the model. Estimation is

conducted for the 1964-1994 sample period for which data are

available. Applying unit-root tests to what could be

considered a relatively short time span may be risky due to

the fact that these tests typically have low power unless

long-run data sets are used (Hakkio and Rush, 1991). Because

time series data in Latin America generally date back to 1957

at most, there is little that can be done to circumvent this

potential problem.

Augmented Dickey-Fuller t-statistics are estimated for

equations with both intercepts and trends. These results,

compared against the corresponding MacKinnon critical value,

appear in Table 3.1. In all cases, tests for unit roots in

the first differenced log transformed series for consumer

prices, money, and the trade-weighted exchange rate index are

rejected at the 1-percent level. Based on this evidence, the

first-order differenced series used to estimate Equation 3.8

are assumed to be stationary.

As specified, the model is explicitly built around a set

of unidirectional causality relations from movements in the

regressors to consumer prices. To examine whether the absence

of simultaneity in the model is plausible, a set of Granger

causality tests are calculated for the stationary components


of the series of interest. These results are reported in

Table 3.2 for lags of 6, 12, 18, and 24 months. Tests are

conducted for prices and the money supply, as well as prices

and the trade-weighted exchange rate.

Similar to the Colombian results reported in Chapter 2,

movements in M1 do not appear to be systematically preceded by

changes in the CPI in a statistically significant manner at

the 5-percent level. Essentially, this implies that monetary

policy in Ecuador is conducted in a manner that is not

accommodative of price shocks. Although central bank linkages

to the executive branch of the government are relatively

strong, steps have been taken in recent years to increase both

monetary policy autonomy and currency stability (Banco del

Pacifico, 1994b).

Implications based on the Granger causality tests

estimated for consumer prices and the exchange rate series are

less clear. At shorter lag lengths, the null hypothesis that

changes in consumer prices do not lead to subsequent changes

in the international value of the sucre is rejected at the 5-

percent level. This conclusion, similar to the Colombian

results obtained by Fullerton (1993c) and Kamas (1995), is not

upheld at longer lag lengths and makes it difficult to reach

concrete conclusions regarding potential feedback effects

between the two series. However, truncation of the longer lag

lengths can bias the hypothesis toward incorrect rejection of

the null hypothesis (Feige and Pearce, 1979). Consequently,

Table 3.1: Unit Root Tests for Stationarity

Series Aug Dickey-Fuller t-stat MacKinnon crit value

P 9.778 (with const, trend, 1 lag) -3.995 (1% ivl)
M -16.626 (with const, trend, 1 lag)
TWX -13.652 (with const, trend, 1 lag)

Table 3.2: Granger Causality Tests

Causality Direction Number of Lags Computed F-stat

CPI => M1 6 1.659

CPI => Ml 12 1.309

CPI => M1 18 1.420

CPI => M1 24 1.509

CPI => TWX 6 2.900

CPI => TWX 12 2.214

CPI => TWX 18 1.664

CPI => TWX 24 1.029


a unidirectional channel of influence from exchange rate

movements to prices appears to be a reasonable assumption.

Given the latter, model estimation is conducted without

resorting to instrumental variables, or developing a system of

simultaneous equations, and the resulting coefficients are

assumed to be unbiased and consistent.

Regression results for Equation 3.8 are summarized here:

(3.9) DPt = 0.015 + 0.013*DMt +
(10.224) (0.797)

0.028*DMt.4 + 0.039*DMt.9 +
(1.735) (2.333)

0.060*DTWXt + 0.007*DTWXt.2 +
(5.932) (0.642)

0.063*D(DPt.6) + 0.405*Ut2 +
(1.638) (8.439)

0. 369*Vt.,

R2 0.379 S.E.R. 0.015 Log likelihood 1013.715
DW 2.087 F-stat 26.798 Prob(F-stat) 0.001,

where the numbers in parentheses are computed t-statistics and

Vt is the error term associated with the ARMAX model for Ut.

Lag lengths of 24 months were used in the initial estimates

for Equation 3.9. While a large number of the resulting

coefficients were significant at the 5-percent level, serial

correlation was present in the residuals. To avoid

potentially spurious estimation results (see Hamilton, 1994),

a nonlinear ARMAX procedure is utilized to correct for

autocorrelation. This estimator (Pagan, 1974) is useful


because of its flexibility in handling a variety of different

error generating processes.

Correcting for serially correlated disturbances caused

the computed t-statistics for many coefficients to become

insignificant at the 5-percent level. Because inclusion and

exclusion of numerous different lags did not yield clear

results, the model structure reported in Equation 3.9 was

selected on the basis of likelihood ratio tests. Although the

relatively short lag components may seem unexpected, they were

confirmed by cross correlation function analysis. To handle

autocorrelation, an ARMA(2,1) specification is used to

characterize the data generating process for the residuals.

As would be expected in an inflationary economy, the

algebraic sign of the intercept in Equation 3.9 is positive.

Because the model is estimated using differenced data, this

result indicates that a systematic upward trend exists in the

Ecuadorian consumer price index. As hypothesized by Hanson

(1985), the sum of the coefficients for the lagged monetary

series is significantly less than one. Similar to Fullerton

(1993a, 1993c), but unlike Kamas (1995), the exchange rate

appears to play an important role in determining price

movements. The coefficient for the velocity of money supply

circulation proxy was not, however, significant at the 5-

percent level.

Although the sum of the lagged monetary aggregate

parameters is less than one, the sum of those coefficients


with the exchange rate variable coefficients does not equal

one as implied by the reduced form of the theoretical model

specification. These results cast doubt upon the relatively

simple version of the Harberger-Hanson framework developed

above. As noted previously, this is not completely unexpected

and is a potential root cause underlying the presence of

autocorrelated residuals in the initial empirical results. To

see if the theoretical model can be improved, an alternative

version of the approach is currently under development for an

economy where data shortages are less severe and a richer

input structure can be handled (Kim and Fullerton, 1996). The

latter effort will also benefit from the incorporation of an

interest rate variable to measure the cost of holding idle

cash balances, as well as an import price index rather than

the exchange rate proxy utilized herein.

In 1983, the Central Bank introduced new exchange rate

policies that allowed the sucre to fluctuate more freely. To

allow for potential parameter heterogeneity caused by

structural change associated with periodic currency

devaluations, a shorter sample was also used for estimation

purposes. Results from these exercises, not reported here,

generally support the empirical estimates presented above.

Experimentation with the lag structure over the shorter sample

period does not yield strong evidence of parameter instability

or any other major shortcomings with Equation 3.9.


3.5 Policy Simulation Results

From the alternative specification and sample size

results, Equation 3.9 does not seem overly fragile.

Alternative lag structures and ARMAX processes were compared

to the model using likelihood ratio tests to evaluate the

results. While it was not possible to reject the

specification shown above, it does seem likely, however, that

additional time series analysis will be required in order

reach firm conclusions regarding inflationary dynamics in

Ecuador. For that reason, the modeling and simulation results

associated with Equation 3.9 should be regarded as preliminary

in nature. They do, however, provide a good starting point

for understanding short-term Ecuadorian price movements and

assessing government policy objectives.

Cognizant of the paucity of comparative research results

in this segment of the literature, a variety of simulation

experiments are conducted using Equation 3.9. The goal of the

simulation tests is to shed light on the feasibility of

attaining the inflationary stabilization targets announced by

the Durdn Ballen-Dahik Garzozi administration in 1995. Sample

data used to estimate Equation 3.9 only includes information

available to government policy makers at the time the

inflation target was announced. The simulation analyses thus

satisfy the Klein (1984) and Christ (1993) criteria for

forecast evaluation. Of course, additional policy analysis

with different model specifications may also be useful.


To examine the feasibility of the government's inflation

goals, four simulation exercises are conducted using Equation

3.9. The first exercise assumes immediate implementation of

price stabilization plan wherein the annualized rate of growth

in the money supply and the rate of devaluation are reduced to

20 percent within one month of the policy announcement.

Scenario two examines rapid implementation of the price

stabilization whereby growth in the money supply and the rate

of devaluation are reduced to 20 percent over a 6-month phase-

in period. The third simulation test analyzes the effects of

gradual introduction of the anti-inflationary program with

money supply growth and the rate of devaluation lowered to 20

percent during a 12-month period. The rates of money supply

growth and currency depreciation reflect the revised policy

targets adopted following the border conflict with Peru and

the aftermath of the "Tequila effect" associated with Mexico's

December 1994 peso devaluation (Banco del Pacifico, 1995).

Policy simulation results are reported in Table 3.3. The

impacts associated with all three implementation scenarios are

striking. In each, reducing the rate of money supply growth

and slowing the rate of nominal currency depreciation to 20

percent over a one-to-twelve month implementation period

causes inflation to decline to less than 25 percent. This is

somewhat close to the policy target established by the

government in late 1994. If implementation of the

stabilization program is immediate, annual consumer prices


increases fall to slightly more than 23 percent. In a more

likely scenario under which intermediate policy targets are

attained more gradually, disinflation is still fairly rapid

and overall policy credibility would not appear to be at risk.

To actually achieve the announced inflation target, however,

does not appear likely.

The final column of Table 3.3 contains the results

associated with a scenario in which no progress is made with

respect to lowering the rate of growth in the money stock.

Similarly, nominal depreciation of the sucre is not lowered

under this simulation. In contrast to the "no implementation"

policy simulation results reported in Table 2.7, the inflation

rate remains practically unchanged if no intermediate steps

are taken by Ecuadorian monetary authorities. This result is

due in large part to the fact that consumer prices in Ecuador

rose on average by approximately 26.7 percent per year during

the past three decades. As a result, the policy experiments

illustrated in Table 3.3 have a common starting point that is

almost identical to the sample period mean.

On the basis of the empirical evidence obtained in this

chapter, it appears that Ecuador's anti-inflationary program

is fairly credible. This conclusion is predicated upon

eventual deceleration in the rates of liquidity growth and

currency depreciation brought about by the central bank. In

light of previous monetary policy analysis conducted for

Table 3.3: Policy Simulation Results

Month Immediate Rapid Gradual None

1 26.1 26.2 26.2 26.2

3 24.9 25.1 25.3 26.0

6 24.3 24.8 25.1 25.8

9 24.0 24.5 25.1 25.9

12 23.2 23.7 24.2 26.0


Ecuador, the results obtained are pleasantly surprising

(Fullerton, 1990a, offers a negative assessment of an earlier

price stabilization effort). That the government's announced

targets are not completely attained is in line with other

studies of Latin American inflation policies (Fullerton,

1993c, and Kamas, 1995).

3.6 Conclusion

An empirical model of Ecuadorian consumer price inflation

is developed and estimated in this chapter by incorporating

both monetary and import cost effects in a theoretically

plausible manner. Specification and simulation of the model

are relatively easy to accomplish. Experimentation with the

estimated equation indicates that the current anti-

inflationary goals of Ecuadorian monetary authorities are in

large part attainable. Because the model does not pose

stringent data requirements, it may be applicable to other

Latin American economies where inflation remains a problem.

Examples include Brazil, Colombia, Mexico, and Venezuela where

authorities continue to grapple with short-term price

stabilization goals.

Additional econometric testing should prove useful.

Initial results reported above indicate this framework will

likely benefit from incorporating a more realistic conceptual

model. Because ARMAX treatment of nonrandom movement in the

initial model residuals was necessary, expansion of the scope


of the model to include additional factors such as labor costs

may be required to more completely specify the inflationary

process in Latin America. Doing so, however, may necessitate

the usage of instrumental variables or the introduction of

multiple equations which allow for potential endogeneity

between prices, money, import costs, and wage rates. Even if

the latter are not required for parameter estimation

consistency, they could enrich subsequent policy simulation


These suggested changes represent avenues for refinement

to the basic model outlined above. They are not likely to

result in wholesale alterations to the general framework.

Similarly, it is not clear that policy simulation impacts and

conclusions will change markedly due to expanding the scope of

the empirical techniques presented above. But given the

breadth of economic conditions prevailing across Latin

America, steps in these directions may prove helpful to

subsequent econometric research of this nature. Given the

divergence between the theoretical model parameters and the

estimated coefficients, additional empirical testing is

certainly warranted.


4.1 Introduction

While inflation has undoubtedly been one of the most

hotly debated items in Latin American policy debates, issues

related to external obligations have also been important

topics of discussion in the region. Following the outbreak of

the international debt crisis in September 1982, a secondary

market for sovereign debt instruments became active in the

major world financial centers. As the payments crisis spread

from Mexico to the rest of Latin America, much of Africa, and

parts of Asia, secondary market trades in developing country

sovereign debt paper increased. Key issues in the emerging

debate regarding potential solutions to the payments problem

often involved the treatment of discounts from face value

implied by secondary market debt prices.

Accordingly, researchers began investigating different

aspects of the behavior of secondary market developing country

debt prices. Much of this work investigates the applicability

of theoretical valuation models to assessing implied discounts

from face. Other research has used secondary market prices

and other financial data to evaluate the probability of

payment rescheduling requests. Authors have also attempted to

assess the sensitivity of the secondary market to

macroeconomic fundamentals, but these efforts have not

investigated whether or not the discounts from face value on

these obligations are predictable.

Several basic questions are investigated in the chapter

at hand. An important issue to be considered is the nature of

the time series behavior of these prices. Under certain

conditions in an efficient market, each individual series

might be expected to follow a random walk. Of course, in a

relatively thin secondary market such as that for developing

country debt, the perfectly competitive hypothesis may not

always be satisfied. If movements in the price series are

nonrandom, it may be possible to relate these variations to

changes in other domestic and international economic

indicators which are generally included in commercial

forecasts of the region. Candidate series which help assess

creditworthiness include variables such as interest rates,

export prices, and international reserves (see Fullerton,

1991, 1993a).

Anecdotal evidence indicates that countries with good

payments records such as Colombia have seen their access to

international commercial credit diminish over the last fifteen

years. If the "good debtor in a bad neighborhood" contagion

effect is present, this will be manifested in the movements of

the discounted prices. Debt prices for individual countries

would consequently be affected by developments in neighboring


countries and correlated with the prices for the instruments

of those economies and the market in general. The generalized

least squares estimation procedure used below implicitly takes

this possibility into account by allowing for

contemporaneously correlated residuals across equations.

To examine the predictability of these series, forecasts

are generated using jointly estimated individual country

models. For comparison purposes, baseline projections are

developed using the random walk assumption that the best

forecast is one of no deviation from the last observation

available in each data set. Modified Theil U-coefficients are

then calculated on the basis of root mean squared error (RMSE)

ratios estimated for each set of model based and random walk

forecasts. The U-statistic for an individual forecast step-

length is equal to the ratio of the model based RMSE to that

of a no change RMSE. When the resulting coefficient is less

than one, the equation forecast has outperformed the random

walk prediction.

4.2 Earlier Studies

There have been a growing number of studies regarding

developing economy external indebtedness and the secondary

market in recent years. Gennotte, Kharas, and Sadeq (1987)

develop a numerical debt valuation method based on a financial

options pricing technique. Using liquid international

reserves plus the estimated values of the capital stocks in


mining and manufacturing as proxies for collateral in each

country, theoretical values for developing country foreign

debts are simulated under different scenarios involving

interest rate changes, principal due, and front-end fees.

Simulation results are found to be positively correlated with

secondary market prices reported for 1985.

Stone (1991) examines the behavior of implied returns on

secondary market sovereign debt instruments. Using an

arbitrage pricing model approach, he examines the empirical

relationships between secondary market returns and various

macroeconomic variables. To control for cross equation

disturbance simultaneity, a seemingly unrelated regression

estimator is utilized. Poor equation fits and weak t-

statistics indicate that movements in implied sovereign debt

returns are not readily explained by arbitrage pricing


Other empirical studies of sovereign debt problems have

been more successful. Rahnama-Moghadam and Samavati (1991)

employ probit models to examine the propensity to default.

Ten different macroeconomic and international financial ratios

are used to quantify the probability of rescheduling. Among

the ratios found most useful in predicting debt moratoria,

formal or informal, are the following: international reserves

relative to imports of goods and services; international

reserves to disbursed debt; disbursed debt relative to

exports; disbursed debt to gross domestic product; and


interest payments relative to exports of goods and services,

also known as the interest service ratio. Parameter estimates

are based on annual cross country sample data from around the


The probit results discussed in the preceding paragraph

were later confirmed by subsequent research which utilized

data from Latin American economies (Rahnama-Moghadam,

Samavati, and Haber, 1991). Reasons offered for segmenting

the data in this fashion include geographic, structural, and

institutional similarities among countries in Latin America.

The authors also point out that economies in this smaller

sample are all middle-income countries which share similar

characteristics in terms of overall development. The models

exhibit relatively high goodness-of-fit statistics, individual

coefficients with expected algebraic signs, statistically

significant parameter estimates, and coefficient stability

across different specifications. As debt problems arise from

a number of different determinants, it would appear that a

further refinement from a regional focus to that of an

individual economy may be useful.

The roles of fundamental economic and financial factors

in the evolution of secondary market sovereign debt prices

have been directly examined in recent research. As shown by

Anayiotos and de Pinies (1990), these types of

characterizations are straightforward and intuitively

appealing. In a statistical framework, variables selected to


capture market fundamentals can also be combined with

regressors designed to represent exogenous risks. Using

pooled observations and annual data, the econometric results

of these authors show that even simple specifications can

represent secondary market developing debt prices with a high

degree of accuracy.

Perasso (1989) also emphasizes economic factors in his

study of secondary market prices. To reflect the importance

of debt-equity conversions, his pricing model is derived from

a profit maximizing framework that includes real interest rate

measures, real costs of capital, international wage

differentials (assumed to induce manufacturers to invest

abroad), and individual economy performance variables. Time

series and cross country annual data are pooled prior to

estimation. Some coefficients are statistically insignificant

or of the wrong sign, but overall empirical results for the

estimated equations for secondary market prices are fairly


Empirical models developed by investment bankers indicate

that secondary market debt prices can be modeled and

forecasted (see Purcell and Orlanski, 1988, 1989). Similar to

the studies mentioned above, these models also rely upon

pooled cross section time series data for different countries.

Regressors used to estimate equation parameters include debt

to export ratios and per capita incomes. Dummy variables for

payment rescheduling programs, principal payment moratoria,


and debt retirement agreements are also entered as right-hand

side variables. Model simulations using individual country

data are used to calculate specific secondary price forecasts.

As mentioned previously, not all of the statistical

evidence with respect to the behavior of secondary market

prices leads to the same conclusions. Laney (1987) concludes

that economic factors are more important as explanatory

variables than political and structural risk factors. Sachs

and Huizinga (1987) report regression results that indicate

both economic and political variables have key roles to play

in modeling developing country debt discounts. Similar to

other studies, the latter also utilize pooled cross country

data in calculating equation parameters.

4.3 Empirical Analysis

As the literature review indicates, there have been a

variety of studies with interesting econometric results

published in recent years. However, none of the initial

efforts attempted to model secondary market debt prices for

individual countries using time series sample observations.

Given the differing sources of debt servicing difficulties, a

country by country examination of secondary market price

movements appears warranted. Similarly, individual

forecasting models or equations have not been systematically

tested to examine whether debt prices can be predicted with

any degree of accuracy. Despite the fact that financial


market participants focus almost exclusively on short-term

movements in sovereign debt prices, earlier research efforts

also failed to study high-frequency data from this market.

In this section of the chapter, simple forecasting models

are proposed, estimated, and simulated for three individual

debtor countries. These equations are modeled jointly using

monthly data series. As stated in the introduction, modified

Theil inequality coefficients are calculated using random walk

forecasts as the benchmarks to which the model projections are

compared. Although the individual equation specifications

employed are straightforward, recent financial market research

underscores the potential success of simple forecasting models

(see Granger, 1992, and Christ, 1993). This initial attempt

provides a useful starting point for addressing questions

regarding the predictability of secondary market developing

country debt prices.

Debt price data used in the empirical estimates are

collected by Salomon Brothers in New York, with monthly

averages published by The WEFA Group in Philadelphia. The

sample period is March 1986 December 1991. Countries

included in the sample are Colombia, Ecuador, and Venezuela.

These countries have interesting and differing histories with

respect to their external debt management practices and their

individual approaches to economic policymaking in general.

Colombian debt management practices have traditionally

been more conservative than those of either of its neighbors.


Government economists have consistently treated international

credit markets as sources of financing to bridge domestic

savings gaps (Fullerton, 1990b). Colombian debt negotiators

have never sought a Brady initiative write-off, arguing that

to do so would only impair the nation's creditworthiness.

Over the period from 1945 forward, Colombia has never declared

an interest payment moratorium and has one of the best debt

service records among developing countries which have utilized

external financing sources.

Ecuador declared principal and interest moratoria in 1987

as a result of the financial aftermath following the

earthquake which shattered the country's transAndean oil

pipeline. The latter event interrupted Ecuador's principal

source of export earnings and destroyed much of its physical

infrastructure. In 1989, the Borja administration resumed

negotiations with commercial creditors and eventually began

honoring 30 percent of the interest coming due on commercial

loans (Fullerton, 1989c). Progress regarding the treatment of

growing amortization and interest arrears remained elusive

through the balance of the Borja government which stepped down

in 1992. A new round of discussions with Ecuador's bank

advisory committee began after the Duran Ballen-Dahik Garzozi

government took office.

Venezuela also encountered debt service problems

following negative oil price shocks in 1986 and 1988.

Eventually, the Perez administration rescheduled commercial


credits under a Brady initiative agreement with Venezuela's

bank advisory committee (Fullerton, 1990c). The agreement

offered bank loan syndicate members five menu options designed

to relieve balance of payment pressures faced by this economy.

Several Euromoney bond issues were successfully floated in

subsequent periods and Venezuela temporarily regained access

to international credit markets.

Debt instruments for all three countries have traded at

substantial discounts from face value in recent years. Given

its superior service record, it is not surprising that

Colombian paper generally carries a higher price than that of

its two Andean neighbors. Similarly, given its higher level

of payment arrears and worse economic performance, Ecuadorian

paper tends to trade at sharper discounts than those of the

other countries in the sample.

Table 4.1 presents summary statistics for each secondary

market sovereign debt price series. Discounts from face value

on Colombian debt certificates were less variable than those

of Ecuador and Venezuela during the March 1986 December 1991

sample period. The range and standard deviation for Colombian

debt quotes are smaller than the others, while those for

Ecuador are the largest of the three. The arithmetic means

for each series follow the opposite pattern in terms of

ranking. Data in Table 4.1 are in cents per dollar, or

percent of face value of the loans, the units in which

transactions are conducted at money center trading desks.

Table 4.1: Secondary Market

Country Average Maximum

Colombia 69.990 86.00

Ecuador 30.332 68.00

Venezuela 57.635 78.50

The sample period is March

Debt Price Summary Statistics

Minimum Standard Deviation

50.00 10.415

10.67 18.728

31.50 14.859

L986 December 1991.


Each series is modeled as a function of key economic and

financial variables which are easily observed as well as

likely to be used by secondary market participants as

indicators of the creditworthiness of the individual country.

Possible regressors include lagged debt quotes, world interest

rates, commodity export prices, international reserves, and

domestic price indexes (for discussion, see Wakeman-Linn,

1991). As mentioned above, series such as these are typically

included in macroeconometric forecasting models of Latin

America due to their usefulness in predicting balance of

payment movements and general economic performance (see

Fullerton, 1993a, 1993b).

A three-stage generalized least squares (3SLS) regression

technique is used to jointly estimate model parameters

(Zellner and Theil, 1962). Doing so permits incorporating

potential cross-equation simultaneity effects of events such

as Citibank's decision to unilaterally increase loan loss

reserves in 1987. The latter is believed to have reduced

overall secondary market liquidity and also reduced the

attractiveness sovereign debt paper to most creditors,

irrespective of their individual payment records (Gajdeczka

and Stone, 1990, and Snowden, 1989). This estimator also

allows for potential simultaneity between the dependent

variables and the independent variables. In the case of

Ecuador, this is important because feedback exists between the

secondary market debt price and the 180-day LIBOR.


Individual models may be written conceptually as:

4.1 Pt = b0 + blxlt + ... + bnxnt + et,

where Pt is the secondary market sovereign debt price series

for an individual country at time period t, x1t, ..., Xnt are

predetermined domestic and international variables for each

economy, et is a random disturbance term, and bl, ..., bn are

regression coefficients. From a theoretical modeling

perspective, this approach may seem informal. Two points are


First, alternative methodologies were originally

considered but deemed inappropriate due to data requirements

and difficulties in applying them to forecasting problems

(Fullerton, 1990d). Second, the complete absence of other

studies of this nature increases the value of an initial

attempt to establish whether any regularities at all are

present in the data (Christ, 1994). Both points were

repeatedly raised by financial economists who participate in

the secondary market and attended the Sovereign Debt

Conference sponsored by The WEFA Group in New York in 1990.

Similar observations are also made by Friscia (1993).

As mentioned above, parameter estimation is accomplished

using the 3SLS methodology developed by Zellner and Theil

(1962). Other procedures were considered, but 70 monthly

observations constitutes a fairly small sample for many time


series estimators. The series are not differenced prior to

estimation, but are logarithmically transformed. Because the

data are in levels, it is important to assess whether the

series are cointegrated. The latter assumption is tested via

unit-root tests on the individual model residuals (for

discussion, see Hamilton, 1994).

Because monthly data are used, it is not possible to

utilize the same regressor variables as have been used in

earlier studies incorporating quarterly or annual series from

national income and product accounts. There are still a

number of potential candidate series which the financial

community may use as indicators of a country's

creditworthiness and will potentially influence secondary

market price quotes. It should be noted that sets of

indicators will vary for different countries according to

individual economic endowments and performance records. This

argument is similar to that previously made for equations used

to estimate developing country borrowing levels under

different regimes (Eaton and Gersovitz, 1981).

For Colombia, Equation 4.2 in Table 4.2, the regressors

include a one-month lag of the debt price, the effective

annual rate for the 180-day London interbank offer rate

(LIBOR), and the one month change in the national consumer

price index (CPI). The one-month lag on the debt price is

included to provide information on how the market has valued

Colombian paper in the most recent period. International


loans to sovereign nations typically carry variable interest

rates defined in terms of a fixed spread over the 6-month

LIBOR. Upward changes in the variable interest rate assessed

on such loans will reduce Colombia's current account balance

and exert downward pressure on secondary market prices. The

rate of inflation is used as a proxy for overall economic

conditions. When inflation rises, Colombian monetary

authorities generally attempt to tighten credit conditions

(Fullerton, Fainboim, and Agudelo, 1992). Parameter estimates

for each variable have the expected arithmetic signs, but the

t-statistic for the inflation term is not significant at the

5-percent level.

In the case of Ecuador, Equation 4.3 in Table 4.2, only

two predetermined variables are included in the three-staged

least squares equation. The first is a one-period lag of the

secondary market debt price series. The second is the 6-month

LIBOR rate. These series are included for the same reasons as

they were used in the Colombian equation. Both coefficients

have the expected signs and are statistically significant.

The absence of other balance of payment indicator series such

as export prices may seem surprising. Because Ecuador,

similar to Colombia, has a relatively diverse commodity export

basket, no single price series will suffice (Fullerton, 1993a,

1993b). This is not the case for Venezuela, where petroleum

products account for more than 80 percent of total merchandise

exports (Fullerton, 1990c).

Table 4.2: Three-Stage Least Squares Regression Results


+ 0.818*Pt.i

- 0.172*LIBOR6M

0. 916*CPIt

R2 0.951
Q(6) 6.118





+ 0.883*Ptoi

- 0.476*LIBOR6M




+ 0.811*Pt.i

0. 061*IRt

- 0.249*LIBOR6M

+ 0.072*OILt



The sample period is March 1986 December 1991.

Pt =

Pt =





Pt =


Q (6)





Four independent variables are included in the Venezuelan

model. As shown in Equation 4.4 in Table 4.2, they include a

one-period lag of the debt price, the 180-day LIBOR series,

international reserves net of gold, and the average price of

petroleum exports. International reserves and the oil export

price are employed as proxies for expected economic conditions

in Venezuela. Although related, the two series do not always

follow parallel paths. More specifically, the level of

international reserves serves as an indicator of domestic

economic policy success or failure (Fullerton, 1990c). World

oil prices provide a measure of global demand for the nation's

principal export. All of the regressor coefficients have the

expected signs. The t-statistic for international reserves,

however, is not significant at the 5-percent level.

The 3SLS technique utilized to estimate the parameters

reported in Table 4.2 allows for potential cross equation

contemporaneous error correlation, a reasonable assumption for

the secondary market for sovereign debt paper. To examine

whether this assumption is necessary, residuals from the first

stage ordinary least squares (OLS) regression are tested for

contemporaneous correlation using a standard t-test

methodology (see Ostle and Mensing, 1975, or Snedecor, 1956).

As shown in Table 4.3, the OLS residuals for Colombia and

Venezuela are correlated in a statistically significant

manner, as are those for Ecuador and Venezuela.


Table 4.3: OLS Cross Equation Correlation Coefficients

Equation Residuals Correlation Coefficient Computed t-stat

Colombia-Ecuador 0.101 0.772

Colombia-Venezuela 0.340 2.750

Ecuador-Venezuela 0.453 3.872


Unit-root cointegration tests performed on the individual

country 3SLS residuals are reported in Table 4.4. In all

three cases, it appears cointegrating vectors have been

obtained. Note that this procedure may be inappropriate for

the short data set which currently exists for secondary market

developing country debt prices. Previous research has shown

that tests associated with unit-root techniques have low power

when applied to short-run data sets (Hakkio and Rush, 1991).

For purposes of the chapter at hand, the results in Table 4.4

are interpreted as evidence that the 3SLS regression results

are not spurious. Given the diagnostic statistics also

reported in Table 4.2, the latter conclusion is probably


To conduct ex ante dynamic simulation exercises, the

equations were jointly reestimated and simulated for 24

different historical subperiods. Four month ahead forecasts

were produced for each secondary market debt price series from

the last subperiod observation forward. A four-month step

length is sufficient to incorporate quarterly portfolio

accounting considerations faced by international banks. As

pointed out by participants at the 1990 Sovereign Debt

Conference, quarterly corporate income tax filing requirements

often trigger loan certificate swaps between secondary market

participants and make a four-month time frame logical for

simulation experiments.


Table 4.4: Unit Root Cointegration Tests

Country Aug Dickey-Fuller t-stat MacKinnon crit value

Colombia -4.895 (with const, trend, 1 lag) -4.122 (1% ivl)

Ecuador -5.140 (with const, trend, 1 lag)

Venezuela -5.320 (with const, trend, 1 lag)


To further enhance the realism of the extrapolation

scenarios, actual forecast data available to the financial

markets during each of the 24 subperiods are incorporated in

the simulations. By relying on forecast data which are

unconditional upon any information not available prior to the

start of any simulation period, Klein's (1984) forecast

evaluation criterion is met. These estimates for the

independent regressors were compiled from international

outlook reports published by The WEFA Group from 1989 to 1991.

Forecast results for each price series are summarized in

Table 4.5. For Colombia, the results indicate that the

secondary market LDC debt price series is predictable using

econometric methods. Although the 1-month ahead Colombian

projections have a slightly greater than unity U-coefficient,

the remaining estimates are all less than one. Interestingly,

the Colombian inequality coefficients monotonically decrease

as the length of the forecast period increases. This suggests

that inclusion of econometric information, in this particular

instance, grows in importance as the timeframe under

consideration expands.

In the case of Ecuador, the results reported here suggest

that its secondary market debt discount is unpredictable.

Consequently, it appears that financial analysts can do no

better than utilize the last available observation for

Ecuadorian paper in anticipating future quotes. Additional

research with alternative estimators and different

Table 4.5: Modified Theil Inequality Coefficients

Series 1 Step 2 Steps 3 Steps 4 Steps

Colombia 1.070 0.888 0.728 0.611

Ecuador 1.715 2.109 2.408 2.614

Venezuela 1.263 1.207 1.108 1.017

The simulation period is January 1990 December 1991.


specifications might obtain superior forecast accuracy, so it

may be premature to conclude that the Ecuadorian debt price

series is not predictable. As presented above, however, the

results obtained in this chapter provide a fairly striking

example of the fact that a relatively high coefficient of

determination does not guarantee automatic simulation


Forecasting results for the Venezuelan debt price series

are also interesting. Similar to Colombia, the inequality

coefficients decline monotonically as the length of the

projections increases. Unlike the Colombian example, however,

the Venezuelan U-statistics remain at least slightly above

unity at each step length. On the basis of the results in

Table 4.5, it therefore appears that a random walk approach

yields better forecasts. Experimentation with other

estimators is probably warranted. Similar to Ecuador,

however, the Venezuelan modeling and simulation results

provide another example of a case in which a high coefficient

of determination does not guarantee prediction accuracy.

4.4 Conclusion

A simple questioned is asked in this chapter. Are

secondary market debt prices predictable? To shed light on

the possible answer, several steps are taken which have not

previously been investigated. A key aspect that distinguishes

this research from earlier efforts is the modeling of monthly

time series data for individual economies, as opposed to cross

section annual or quarterly samples used elsewhere. Also

developed are forecast exercises designed to meet the needs of

participants in international financial markets where

developing country debt instruments are traded.

Modeling and simulation results reported here provide

only limited evidence that it is possible to forecast

secondary market developing country debt discounts from face

value. All three series in the sample exhibited interesting

model characteristics. In particular, the equations and

forecasts for Colombia are encouraging. The same cannot be

said of the Ecuadorian and Venezuelan prediction tests, as in

both cases simple random walk forecasts of the respective debt

price series prove more accurate.

Because the three economies, and their respective debt

price series, differ substantially from one another, it may be

useful to apply this modeling approach to other debt price

series. Obvious candidates include Argentina, Brazil, Chile,

Mexico, and Peru. Further specification enhancements for

Ecuador and Venezuela may also provide useful information.

Inclusion of institutional variables related to Brady-

initiative debt negotiations might prove beneficial,

especially if it is possible to construct monthly indices for

these factors. As noted by Friscia (1993), however,

confidentiality restrictions may preclude this possibility.


Similarly, it may be worthwhile to test alternative

estimators. The latter will become more feasible as

additional observations and information regarding this market

become available. On balance, however, it appears that

movements in secondary market debt prices are not predictable.


In Latin America, short-term econometric forecasting

analysis is still a largely uncharted area of research.

Material presented above indicates that a variety of

techniques, methodologies, and modeling approaches may yield

interesting insights with respect to both forecasting and

policy issues. Not surprisingly, this research has only

hinted at a few of the potentially beneficial topics which

merit further attention.

As discussed elsewhere, the availability of low cost

computer hardware and software, plus the development of wide

coverage high-frequency data banks, will help encourage

additional research of this nature (Fullerton, 1992). With

respect to the results presented above, new efforts are

already underway in terms of alternative estimators for

secondary market debt price forecasting. The underlying

theoretical model presented in Chapter 3 has also been

extended to include labor costs as part of the price vector on

the output side for studying price dynamics.

From a business forecasting perspective, a promising area

of endeavor is likely to arise from the publication of

quarterly national income and product account data. In


countries such as Ecuador, this will permit the development of

large-scale macroeconometric models such as those pioneered by

Barger and Klein (1954). The latter continue to enjoy a

central role in business and government planning exercises

throughout the world. Given the short-term uncertainties

present in Latin America, quarterly modeling and forecasting

will likely be welcomed with enthusiasm. At this juncture,

much work remains to be done, but initial efforts such as

those presented above point to numerous potential successes to

be gained from future research efforts.


Anayiotos, G., and J. de Pinies, 1990, "The Secondary
Market and the International Debt Problem," World Development,
18, 1655-1669.

Banco del Pacifico, 1994a, "Economic Report," 16-5
(September-October), 1-4.

Banco del Pacifico, 1994b, "Economic Report," 16-6
(November-December), 1-4.

Banco del Pacifico, 1995, "Economic Report," 17-2 (March-
April), 1-4.

Barger, H., and L.R. Klein, 1954, "A Quarterly Model for
the United States Economy," Journal of the American
Statistical Association, 49, 413-437.

Batten, D.S., and M.T. Belongia, 1986, "Monetary Policy,
Exchange Rates, and U.S. Agricultural Exports," American
Journal of Agricultural Economics, 68, 422-427.

Beltr~n del Rio, A., 1991, "Macroeconometric Model-
Building of Latin American Countries, 1965-1985," Chapter 12
in A History of Macroeconometric Model-Building, Edited by
R.G. Bodkin, L.R. Klein, and K. Marwah, Brookfield, VT: Edward
Elgar Publishing Company.

Bomberger, W.A., and G.E. Makinen, 1979, "Some Further
Tests of the Harberger Inflation Model using Quarterly Data,"
Economic Development and Cultural Change, 27, 629-644.

Box, G.E.P., and G.M. Jenkins, 1976, Time Series
Analysis: Forecasting and Control, San Francisco: Holden-Day.

Box, G.E.P., and G.C. Tiao, 1975, "Intervention Analysis
with Applications to Economic and Environmental Problems,"
Journal of the American Statistical Association, 70, 70-79.

Brajer, V., 1992, "Empirical Inflation Models in
Developing Countries," Journal of Development Studies, 28,

Cabrera, M., and F. Montes, 1978, "La Dindmica de las
Series de Tiempo: El Caso del Indice Nacional de Precios al
Consumidor-Empleados," Revista del Banco do la Repfiblica, 51,

Chatfield, C., 1984, The Analysis of Time Series, New
York: Chapman and Hall.

Christ, C.F., 1993, "Assessing Applied Econometric
Results," Federal Reserve Bank of St. Louis Review, 75
(March/April), 71-94.

Christ, C.F., 1994, "The Cowles Commission's
Contributions to Econometrics at Chicago, 1939-1955," Journal
of Economic Literature, 32, 30-59.

Clemen, R.T., 1989, "Combining Forecasts: A Review and
Annotated Bibliography," International Journal of Forecasting,
5, 559-583.

Clavijo, S., 1987, "Hacia una Caracterizaci6n del
Comportamiento de la Velocidad de Circulaci6n del Dinero,"
Ensayos Sobre Politica Econ6mica, 12, 59-79.

Dhane, G., and A.P. Barten, 1989, "When it all Began: The
1936 Tinbergen Model Revisited," Economic Modelling, 6, 203-

Direcci6n General de Estudios, various issues,
Informaci6n Estadistica Mensual, Banco Central del Ecuador:

Dutton, J., and T. Grennes, 1987, "Alternative Measures
of Effective Exchange Rates for Agricultural Trade," European
Review of Agricultural Economics, 14, 427-442.

Eaton, J., and M. Gersovitz, 1981, "Debt with Potential
Repudiation: Theoretical and Empirical Analysis," Review of
Economic Studies, 48, 289-309.

Edwards, S., 1985, "Money, the Rate of Devaluation, and
Interest Rates in a Semiopen Economy: Colombia, 1968-82,"
Journal of Money, Credit, and Banking, 17, 59-68.

Engsted, T., 1993, "Cointegration and Cagan's Model of
Hyperinflation under Rational Expectations," Journal of Money,
Credit and Banking, 25, 350-360.

Feige, E.L., and D.K. Pearce, 1979, "The Casual Causal
Relationship between Money and Income: Some Caveats for Time
Series Analysis," Review of Economics and Statistics, 61, 521-

Friscia, A.B., 1993, "Comment on Fullerton," Atlantic
Economic Journal, 21 (March), 87.

Fullerton, T.M., Jr., 1989a, "A Composite Approach to
Forecasting State Government Revenues," International Journal
of Forecasting, 5, 373-380.

Fullerton, T.M., Jr., 1989b, "Exchange Rate Indexes for
Ecuador and Venezuela," Latin America Economic Outlook, The
WEFA Group (June), 1.3-1.7.

Fullerton, T.M., Jr., 1989c, "Ecuador," Latin America
Economic Outlook, The WEFA Group (September), 6.1-6.15.

Fullerton, T.M., Jr., 1990a, "Ecuador," Latin America
Economic Outlook, The WEFA Group (June), 6.1-6.15.

Fullerton, T.M., Jr., 1990b, "Colombia," Latin America
Economic Outlook, The WEFA Group (March), 5.1-5.16.

Fullerton, T.M., Jr., 1990c, "Venezuela," Latin America
Economic Outlook, The WEFA Group (September), 10.1-10.14.

Fullerton, T.M., Jr., 1990d, "Methodologies for
Evaluating Secondary Market Debt Prices," paper presented at
The WEFA Group Sovereign Debt Conference, New York, NY.

Fullerton, T.M., Jr., 1991, "Colombia," Latin America
Economic Outlook, The WEFA Group (March), 5.1-5.16.

Fullerton, T.M., Jr., 1992, "Confessions of an
International Forecaster," Business Economics, 27 (April),

Fullerton, T.M., Jr., 1993a, "Un Modelo Macroeconometrico
para Pronosticar la Economia Colombiana," Ensayos Sobre
Politica Econ6mica, 24, 101-136.

Fullerton, T.M., Jr., 1993b, "Un Modelo Macroeconometrico
para Pronosticar la Economia Ecuatoriana," Cuestiones
Econ6micas, 20, 59-100.

Fullerton, T.M., Jr., 1993c, "Inflationary Trends in
Colombia," Journal of Policy Modeling, 15, 463-468.

Fullerton, T.M., Jr., 1993d, "Predictability of
Secondary-Market Developing-Country Debt Prices," Applied
Economics, 25, 1365-1371.

Fullerton, T.M., Jr., 1995a, "Short-Term Price Movements
in Ecuador," Proceedings of the American Statistical
Association, Business and Economic Statistics Section, 280-

Fullerton, T.M., Jr., 1995b, "Empirical Models for
Secondary Market Debt Prices," paper presented at 65th Annual
Meeting of the Southern Economic Association, New Orleans, LA.

Fullerton, T.M., Jr., I. Fainboim, and M.I. Agudelo,
1992, "Exchange Rate, Money Supply, and Inflationary Dynamics
in Colombia," Best Paper Proceedings of the Atlantic Economic
Society, 2 (January), 54-58.

Fullerton, T.M., Jr., R. Hirth, and M. Smith, 1991,
"Inflationary Dynamics and the Angell-Johnson Proposals,"
Atlantic Economic Journal, 19 (March), 1-14.

Fullerton, T.M., Jr., and A. Kapur, 1991, "Prediccion de
Multiplicadores Monetarios en Colombia, Ecuador, y Venezuela,"
Lecturas de Economia, 35, 53-86.

Gajdeczka, P., and M.R. Stone, 1990, "The Secondary
Market for Developing Country Loans," Finance and Development
(December), 22-25.

Garlow, D.C., 1993, "Ecuador," Latin America Economic
Outlook, The WEFA Group (September), 6.1-6.17.

Gennotte, G., H.J. Kharas, and S.Sadeq, 1987, "A
Valuation Model for Developing-Country Debt with Endogenous
Rescheduling," World Bank Economic Review, 1, 237-271.

Granger, C.W.J., 1992, "Forecasting Stock Market Prices,"
International Journal of Forecasting, 8, 3-13.

Hakkio, C., and M. Rush, 1991, "Cointegration: How Short
is the Long-Run?" Journal of International Money and Finance,
10, 571-581.

Hallman, J.J., R.D. Porter, and D.H. Small, 1991, "Is the
Price Level Tied to the M2 Monetary Aggregate in the Long
Run?" American Economic Review, 81, 841-858.

Hamilton, J.D., 1994, Time Series Analysis, Princeton,
NJ: Princeton University Press.

Hanson, J.A., 1985, "Inflation and Imported Input Prices
in some Inflationary Latin American Economies," Journal of
Development Economics, 18, 395-410.

Harberger, A., 1963, "The Dynamics of Inflation in
Chile," in Measurement in Economics: Studies in Mathematical
Economics and Econometrics in Memory of Yehuda Grunfeld,
edited by C.F. Christ, Stanford, CA: Stanford University

Ikhide, S., and T.M. Fullerton, Jr., 1995, "Inflationary
Dynamics in Nigeria," paper presented at Institute for
Developing Economies, Tokyo, Japan.

Kamas, L., 1986, "Dutch Disease Economics and the
Colombian Export Boom," World Development, 14, 1177-1198.

Kamas, L., 1995, "Monetary Policy and Inflation under the
Crawling Peg: Some Evidence from VARs for Colombia," Journal
of Development Economics, 46, 145-161.

Kercheval, M.P., 1987, "Another Look at the Dollar's
Trade-Weighted Value," Business Economics, 22 (April), 43-47.

Kim, C., and T.M. Fullerton, Jr., 1996, "Import Prices,
Wage Rates, and Korean Price Dynamics," working paper,
Department of Economics, Ajou University, Suwon, Korea.

Klein, L.R., 1984, "The Importance of the Forecast,"
Journal of Forecasting, 3, 1-9.

Koch, P., J. Rosensweig, and J. Whitt, 1988, "The Dynamic
Relationship between the Dollar and U.S. Prices," Journal of
International Money and Finance, 7, 181-204.

Laidler, D., 1993, "Commentary on Assessing Applied
Econometric Results," Federal Reserve Bank of St. Louis
Review, 75 (March/April), 101-102.

Laney, L.O., 1987, "The Secondary Market in Developing
Country Debt," Federal Reserve Bank of Dallas Economic Review,
(July), 1-12.

Leiderman, L., 1984, "On the Monetary-Macro Dynamics of
Colombia and Mexico", Journal of Development Economics, 14,

Leith, J.C., 1991, "The Exchange Rate and the Price Level
in a Small Open Economy," Journal of Policy Modeling, 13, 309-

Montes, G., and R. Candelo, 1982, "El Enfoque Monetario
de la Balanza de Pagos: El Caso de Colombia, 1968-1980,"
Revista do Planeaci6n y Desarrollo, 14, 11-40.

Motley, B., 1993, "Inflation and Growth," FRBSF Weekly
Letter, 93-44 (December 31), 1-3.

Ocampo, J.A., 1983, "En Defensa de la Continuidad del
Regimen Cambiario," Coyuntura Econ6mica, 13, 198-214.

Ostle, B., and R.W. Mensing, 1975, Statistics in
Research, Ames: Iowa State University Press.

Pagan, A.R., 1974, "A Generalized Approach to the
Treatment of Autocorrelation," Australian Economic Papers, 13,

Palma, P., and D. Fontiveros, 1988, "A Comparative
Sensitivity Analysis of the MODVEN VII Macroeconomic Model for
Venezuela," Economic Modelling, 5, 286-345.

Pankratz, A., 1983, Forecasting with Univariate Box-
Jenkins Models, New York: John Wiley and Sons.

Perasso, G., 1989, "The Pricing of LDC Debt in the
Secondary Market," Kyklos, 42, 533-555.

Purcell, J.F.H., and D.J. Orlanski, 1988, "Developing
Country Loans: A New Valuation Model for Secondary Market
Trading," International Loan Trading Analysis, Salomon
Brothers, 1-20.

Purcell, J.F.H., and D.J. Orlanski, 1989, "Developing
Country Loans: Refinements to the Relative Value Model for the
Secondary Market," International Loan Trading Analysis,
Salomon Brothers, 1-16.

Rahnama-Moghadam, M., and Hedayeh Samavati, 1991,
"Predicting Debt Rescheduling by Less-Developed Countries,"
Quarterly Review of Economics and Business, 31, 3-14.

Rahnama-Moghadam, M., H. Samavati, and L.J. Haber, 1991,
"The Determinants of Debt Rescheduling," Southern Economic
Journal, 58, 510-517.

Sachs, J., and H. Huizinga, 1987, "US Commercial Banks
and the Developing-Country Debt Crisis," Brookings Papers on
Economic Activity, 2, 555-601.

Sheehey, E.J., 1976, "The Dynamics of Inflation in Latin
America: Comment," American Economic Review, 66, 692-694.

Sheehey, E.J., 1980, "Money, Income, and Prices in Latin
America," Journal of Development Economics, 7, 345-357.

Snedecor, G., 1956, Statistical Methods, Ames: Iowa State
University Press.

Snowden, P.N., 1989, "The Interpretation of Market
Discounts in Assessing Buy-Back Benefits," IMF Staff Papers,
36, 733-735.

Stone, M.R., 1991, "Are Sovereign Debt Secondary Market
Returns Sensitive to Macroeconomic Fundamentals?," Journal of
International Money and Finance, 10, SI00-S122.

Vial, J., 1988, "Comparaci6n de Modelos
Macroeconometricos Latinoamericanos," Colecci6n Estudios
CIEPLAN, 24, 145-181.

Vogel, R.C., 1974, "The Dynamics of Inflation in Latin
America, 1950-1969," American Economic Review, 64, 102-114.

Wakeman-Linn, J., 1991, "The Market for Developing
Country Debt," The World Bank Research Observer, 6, 191-203.

West, C.T., 1996, "System-based Weights vs. Series-
specific Weights in the Combination of Forecasts," paper
presented at the 42nd Regional Science Association
International North American Meetings, Cincinnati, OH.

Zellner, A., 1994, "Time Series Analysis, Forecasting,
and Econometric Modeling: The Structural Econometric Modeling,
Time Series Analysis (SEMSTA) Approach," Journal of
Forecasting, 13, 215-233.

Zellner, A., and H. Theil, 1962, "Three-Stage Least
Squares: Simultaneous Estimation of Simultaneous Relations,"
Econometrica, 30, 54-78.

Zind, R.G., 1993, "On Inflation and Growth in the LDCs,"
Economic Studies Quarterly, 44, 108-116.


Thomas M. Fullerton, Jr. is a senior economist in the

Forecasting Program of the Bureau of Economic and Business

Research at the University of Florida. Fullerton is co-author

of The Florida Outlook, a quarterly forecast of the state and

20 metropolitan economies. He also teaches a course on Latin

American political economy. Fullerton previously worked at

Wharton Econometrics as international economist in charge of

modeling, forecasting, and policy analysis for Colombia,

Ecuador, and Venezuela. He also worked as an economist in the

Executive Office of the Governor of Idaho where he forecast

the state economy and conducted fiscal policy analysis during

legislative sessions. He began his career in the Planning

Department of El Paso Electric Company. His research has been

published in outlets such as Applied Economics, Journal of

Forecasting, Public Budgeting & Finance, Atlantic Economic

Journal, Journal of Policy Modeling, Business Economics,

Applied Economics Letters, and International Journal of

Forecasting. Fullerton holds degrees from the University of

Texas at El Paso, Iowa State University, and the Wharton

School of the University of Pennsylvania. He is a doctoral

candidate in economics at the University of Florida.

I certify that I have read this study and that in my
opinion it conforms to acceptable standards of scholarly
presentation and is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.

Carol Taylor West, Chair
Professor of Economics

I certify that I have read this study and that in my
opinion it conforms to acceptable standards of scholarly
presentation and is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.

David A. Denslow
Distinguished Service
Professor of Economics

I certify that I have read this study and that in my
opinion it conforms to acceptable standards of scholarly
presentation and is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.

Chunrong Ai
Assistant Professor of
I certify that I have read this study and that in my
opinion it conforms to acceptable standards of scholarly
presentation and is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.

Terry '. McCoy
Professor of Latin American

This dissertation was submitted to the Graduate Faculty
of the Department of Economics in the College of Business
Administration and to the Graduate School and was accepted as
partial fulfillment of the requirements for the degree of
Doctor of Philosophy.

August 1996
Karen A. Holbrook
Dean, Graduate School

Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd