Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Two essays on the impact of exchange rate regime changes in Asia: examples from Thailand and Japan
(USC Thesis Other)
Two essays on the impact of exchange rate regime changes in Asia: examples from Thailand and Japan
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
TWO ESSAYS ON THE IMPACT OF
EXCHANGE RATE REGIME CHANGES IN ASIA:
EXAMPLES FROM THAILAND AND JAPAN
by
Suriya Poolvoralaks
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
May 2007
Copyright 2007 Suriya Poolvoralaks
ii
DEDICATION
I would like to dedicate this dissertation to my mother and sisters
iii
ACKNOWLEDGEMENTS
There are so many people to whom I am indebted to for their support in one
way or another to my success in completing this dissertation. First of all, I would
like to thank Professor Robert Dekle, chairperson of my dissertation committee, for
his continuous support and patience. He is a man of great erudition and impeccable
professionalism and he took me under his wing, always with an intention for the
success of my dissertation. I also would like to thank Professor Caroline Betts, my
graduate advisor, who guided me extensively during my study at USC. Professor
Roger Moon, Professor Aris Protopapadakis, and Professor Yong Kim, with their
impressively diverse and deep knowledge, have been very generous in providing
extremely helpful advices and comments on this dissertation.
I would like to express my gratitude to my family, especially my mother
and my sisters, for their unconditional support and encouragement throughout my
study in the U.S., without whom this dissertation would have never been
completed. I thank them with all my heart for the difference they have made on
this endeavor.
I would also like to thank Miss Onnicha Sawangfa, my dear friend, who has
always been dependable during tough times and whose help was also crucial to the
completion of this dissertation. Many helpful suggestions from my colleague,
Albert Echu Liu, and excellent editing service by Miss Sarah Novak, are also
greatly appreciated.
iv
I am responsible, of course, for any shortcomings that still remain in my
dissertation, but through Professor Dekle’s brilliance, kindness and support, I was
able to produce a document that, at least, expressed and articulated what I had in
mind when I began the project.
v
TABLE OF CONTENTS
DEDICATION ii
ACKNOWLEDGEMENTS iii
LIST OF TABLES vi
LIST OF FIGURES vii
ABSTRACT viii
CHAPTER 1: MAIN INTRODUCTION 1
CHAPTER 2: IMPACT OF CORPORATE RESTRUCTURING
ON THAI FIRMS 7
2.1 Introduction 7
2.2 Data Description 11
2.3 Methodology 17
2.4 Empirical Results 43
2.5 Conclusions 60
CHAPTER 3: THE STABILITY OF MACROECONOMIC VOLATILITY
UNDER DIFFERENT EXCHANGE RATE REGIMES 66
3.1 Introduction 66
3.2 Data Description 69
3.3 Methodology 82
3.4 Empirical Results 85
3.5 Conclusions 99
CHAPTER 4: MAIN CONCLUSION 101
BIBLIOGRAPHY 103
APPENDICES 106
APPENDIX A 107
APPENDIX B 108
APPENDIX C 109
vi
LIST OF TABLES
Table 1: Definitions of Firm Performance Measures 12
Table 2: Descriptive Statistics of the Variables of 219 Non-
financial Thai Firms, 1996 – 2004 14
Table 3: List of Explanatory Variables for Each Performance Measure 31
Table 4: Estimated Average Treatment Effects (ATE) Using
Matching Method on: (1) Excess Stock Market Return;
(2) ROA1; (3) ROA2; (4) Capital Cost; (5) Debt-to-Assets
Ratio; and (6) Coverage Ratio 45
Table 5: Estimated Average Treatment Effects (ATE) of Debt
Restructuring Treatment on Firm Performances using
Endogenous Switching Regression Techniques,
219 Non-financial Thai Firms, 1996-2004 56
Table 6: Structural Break Tests for NER-related Variables 86
Table 7: Structural Break Tests for Non-NER-related Variables 90
vii
LIST OF FIGURES
Figure 1: Japanese Series of Interest after First Differencing
Transformation 71
Figure 2: Conditional Standard Deviation of Japanese Series of
Interest Estimated from GARCH(1,1) 74
Figure 3-1: Tradeoff of Real Exchange Rate Volatility and
Volatility of Other Variables of Interest under Fixed
Exchange Rate Regime 77
Figure 3-2: Tradeoff of Real Exchange Rate Volatility and
Volatility of Other Variables of Interest under Floating
Exchange Rate Regime 80
Figure C1: Japanese Series of Interest before First Differencing
Transformation 111
viii
ABSTRACT
This dissertation is about the impact on exchange rate regime changes in
Asia with examination employed through econometric theory and modeling
analysis. The dissertation consists of two parts.
The first part is about the impact of corporate restructuring on the
performance of Thai firms such as return on assets, capital cost, debt-to-asset ratio,
and coverage ratio after the financial crisis and the following floating of the
exchange rate regime in 1997. Based on two approaches from the program
evaluation literature — the “matching” and the “endogenous switching regression”
approaches, the estimated impacts of corporate restructuring on a firm’s
performance are found to be either insignificant or inconclusive, depending on
which of these two approaches is used. The estimated effects of debt-restructuring
are also found to vary between structured and non-restructured firms for some
performance measures.
The second part of the dissertation examines whether the volatility of
Japanese macroeconomic variables are stable during the fixed regime and whether
the stability of these variables change following floating of the exchange rate
regime in the early 1970s. Through usage of the structural break test model, it was
found that, after floating of the Yen in 1971, there was a strong pattern of one-time
break in volatility for most variables that are related to the nominal exchange rates,
such as real exchange rates, exports, imports and unit value of exports. The same
ix
pattern can be found only in volatility of some of the variables that are not related
to the nominal exchange rates, namely the household consumption expenditure. On
the other hand, volatility of most of the variables that are not related to the nominal
exchange rates appears to be instable throughout the periods of our investigation.
1
CHAPTER 1
MAIN INTRODUCTION
This paper is about the impact on exchange rate regime changes in Asia
with examination employed through econometric theory and modeling analysis.
Using Thailand and Japan as the studied countries that both went through
involuntary exchange rate regime changes, we intend to comprehend the impact
resulting through such exchange rate regime alteration.
As we begin our analysis, we find that there is a vast amount of theoretical
and empirical literature on the consequences in changing exchange rate regimes.
Before the early 1970s, the optimal choice of exchange rate regime was never an
issue because nations were not independently free to choose their exchange rate
regimes; instead, they were subject to prevailing norms of the international
monetary system such as the gold standard or the Bretton Woods Agreement.
Following the collapse of the Bretton Woods Agreement in early 1971, nations
became free to choose their own exchange rate regime which allowed many major
currencies to float soon thereafter. With the independence to choose one’s own
exchange rate regime came the issue of the pursuing an optimal regime choice and
invariably also the issue of the economic consequences of the exchange rate regime
changes.
There are two extreme exchange rate regimes — the pure floats and the
hard pegs — each one existing at the other end of the spectrum. Among the two
2
extreme exchange rate regimes lies a variety intermediate regime with varying
degree of intervention or fixity ranging from basket pegs, crawling pegs to target
zones and target bands. The choice of the adopted exchange rate regime is a matter
of each nation’s individual preference depending on the tradeoffs as well as the
pros and cons between these regimes. The most optimal exchange rate regime
considered for a certain nation depends largely on the current government’s
targeted economic policy and the credibility of that nation’s monetary authorities.
Since these determinants for any nation varies over time, the optimal exchange rate
regime considered will also change over time.
There have been a significant number of nations making changes to the
choice of their exchange rate regimes or what is termed exchange rate regime
transition in the past 30 years. During the period of their study between 1975 and
1999, Ghosh et al. (2005) indicated that 1 out of 10 countries decide to change their
exchange rate regime as often as annually. Some studies (Eichengreen, 1994;
Fischer, 2001) found evidence that support the “hollowing out” hypothesis which
states that countries will gravitate from intermediate regimes toward more extreme
regimes at either end of the spectrum. In addition, Bubula and Otker-Robe (2002)
documented that exits from intermediate regimes in developing countries were
mostly in the form of switching to a floating regime or some type of intermediate
regime, rather than a fixed regime. Different transitions imply varying degrees of
complexity and difficulty. Transitions between intermediate and floating regimes
require only minor adjustments and therefore are relatively easy to manage.
3
Changes that involve the extreme exchange rate regimes, such as the hard pegs and
the pure floats, are more demanding and may require substantial legal and
institutional changes in order to guarantee a smooth transition processes.
Exchange rate regime changes can be categorized into two types. The first
category is related to voluntary changes. Voluntary exchange rate regime change
occurs when a government replaces one exchange rate regime with another in order
to benefit from the superior performance of the replacing regime or to avoid a
future economic crisis. Because it is a voluntary change, it is usually well managed
and prepared and allows the market enough time to settle with these changes, thus
resulting in a smooth and orderly transition process. The immediate aftermath of
such voluntary regime change is often characterized by a relatively leveled
depreciation of the market-determined exchange rate. Ideally, a voluntary
exchange rate change should occur when the expected welfare benefit exceeds the
transition costs. In practice, nations with pegged exchange rate regimes
demonstrate a “fear of floating” because of their government’s reluctance to move
away (notwithstanding the benefits to do so) from a relatively adequate working
pegged exchange rate regime in apprehension of the possible risk of failure. Thus,
many exchange rate regime changes will occur only after a nation has faced severe
currency speculative attacks, at which point a crisis is often inevitable, with an exit
that is anything but orderly and most often times takes the form of spectacular
crash. This leads us to the second category of regime change and that is the
involuntary exchange rate regime change.
4
This second category of exchange rate regime change is more common and
can also be termed as forced exchange rate regime change. Involuntary exchange
rate regime changes occur when the governments had no prior intention to make a
change to its exchange rate regime but were forced to do so due to appreciation or
depreciation pressures. Usually, involuntary exchange rate regime change occurs
when the current regime is fragile and vulnerable to speculative attack which may
have been spurred by developments of increasing international capital mobility.
This type of involuntary change is economically, politically and socially disruptive
not to mention damaging to the reputation of the nation’s central bank. Oftentimes,
The consequences of such involuntary change causes volatile asset prices, unstable
expectations, and higher risk premium. It is also noted that changes from
intermediate exchange rate regimes to a floating exchange rate regime has often
occurred in the context of a currency crisis.
This brings us back to the purpose of this paper, which is to view two
aspects (through two sampled nations – Thailand and Japan) of the consequences of
involuntary exchange rate regime change, in particular, the forced exits from the
pegged or “fixed” exchange rate regime to a floating or more flexible exchange rate
regime.
The first of the two aspects will view the consequences through a systemic
banking and corporate sector crisis. This type of crisis as explored in Chapter 2 is
often part of a currency crisis and its resolution involves many policy choices
ranging from macroeconomic (including the tightness of monetary and fiscal
5
policy) to microeconomic (including capital adequacy rules, corporate governance
requirements, and various others in depth of fundamental reforms)
The study will try to illustrate the effectiveness of a certain resolution to the
systemic crisis of the corporate sector, namely the corporate restructuring program.
In particular, we will consider the effectiveness of corporate restructuring of Thai
firms during the periods of the East Asian financial crisis that started in mid-1997.
Prior to the crisis, many of the ASEAN member economies including Thailand had
been enjoying a boom in foreign capital inflows resulting from currency
appreciation in Japan and the Newly Industrialized Countries (NICs) as well as
domestic financial liberalization occurring in the early 1990s. In Thailand’s case,
high profit margins in stocks, high interest rates, and relatively low risk due largely
to a U.S. dollar-pegged currency, attracted foreign capital into the country. A
reversal of these flows occurred when the currency was floated, with massive
amounts of capital fleeing the country in the second half of 1997. This caused a
national economic crisis that soon spread to other countries in the region. As the
originator of the Asian crisis, Thailand became one of the most affected countries,
seeing its currency drop significantly relative to the U.S. dollar. The key factors
that led to the crisis appears to have been many including the prolonged
maintenance of pegged exchange rate regime, which encouraged external
borrowing and led to excessive exposure to foreign exchange risk in both the
financial and corporate sector.
6
The second aspect of our paper is to explore whether there is a change in the
stability of macroeconomic fundamentals following a shift from a pegged to more
flexible or floating exchange rate regime. In Chapter 3, we examine whether there
is a significant change in volatility of macroeconomic fundamentals after the
floating of Yen following the Bretton Woods collapse in 1971. From 1949, the
value of Yen was fixed at ¥360 per U.S. dollar as part of the Bretton Woods
agreement to stabilize the Japanese economy. This was maintained until 1971
when many major currencies became undervalued against the U.S. dollar. Under
the pressures from the supply and demand in the foreign exchange market, the U.S.
dollar devaluation as well as the new fixed rates was difficult to maintain. This led
major nations, including Japan, to abandon the fixed rate and allowed their
currency to float in early 1973.
7
CHAPTER 2
IMPACT OF CORPORATE RESTRUCTURING
ON THAI FIRMS
2.1 Introduction
In developing countries, corporate restructuring measures are often at the
center of economic adjustment programs following a systemic crisis.
1
In such a
crisis, partly as a result of a general economic slowdown and large shocks to
exchange and interest rates, the corporate sector experiences a large number of
defaults and difficulties in paying contracts on time. Non-performing loans
increase and there is a sharp rise in interest rates and a slowdown or reversal in
capital flows. As part of the remedy for an occurrence of a systemic crisis,
governments typically carry out corporate restructuring programs. With a systemic
crisis the fate of the corporate sector and the general economy are typically
intertwined. For example, investors, both domestic and foreign, will await the
actions of the government in restructuring the corporate sector, thus withholding
capital to corporations when it is needed most. The corporations’ inability to
obtain capital then causes a slow down in its operations and invariably defaults on
its debts. The financial sector is consequently affected that causes for a further
1
A systemic crisis is a situation in which an economy faces large-scale financial and corporate
distress within a short period. Recent examples include the crisis in Nordic countries in the early
1990s, in Mexico in 1994-95, and in the East Asian countries after 1997.
8
action to deter lending through interest rate increases and stricter credit approval
policies.
Supporting the first aspect of the paper in analysis on the impact of changes
in exchange rate regimes, the consequences of corporate restructuring programs on
the performance of corporations in Thailand are examined. During Thailand’s
Crisis of 1996, the Thai government carried out several corporate restructuring
programs. There was a belief that the Thailand’s crisis was partly triggered by
problems in the corporate sector, including too much investment and borrowing,
which led to fragility in the corporate sector. There was also a belief that unless the
corporate sector was made healthy, the Thai economy would not recover from the
crisis. The impact of these corporate restructuring measures, however, has never
been systematically examined.
Did the corporations improve their performance after the corporate
restructuring measures? The answer is mixed. Based on several accounting and
stock market performance measures from 1996 to 2004, it cannot be concluded that
the corporate restructuring measures improved Thai corporate performance.
Methods from the program evaluation literature are applied to examine the impact
of corporate restructuring measures on corporate performance. Thai “restructured”
firms are classified as the “treatment” group and the “non-restructured” firms as the
“control” group. It is obvious that the classification of firms into the two groups:
“treated” and “non-treated” is not random. Firms with low or negative profitability
or otherwise weak balance sheets are often forced by the authorities to enter
9
corporate restructuring programs. It is noted that the “treated” weaker firms’
inability to improve their performance when compared to the “non-treated” healthy
firms should not at all be a surprise.
Because there is an occurrence of endogeneity or the “non-random” factor
in which a corporation is to enter or not enter the government’s corporate
restructuring program, there needs to be further econometrical analysis made before
a conclusion can be prepared on the effectiveness of corporate restructuring. To
deal with the endogeneity of “restructuring” versus “non-restructuring” or
“treatment” versus “non-treatment,” two approaches are adopted from the program
evaluation literature — the “matching” and the “endogenous switching regression”
approaches. The matching approach is used when the non-random factors that
determine the firm’s participation in the restructuring program are observable. The
endogenous switching regression approach is more appropriate when the factors
that determine the firm’s participation in the restructuring program are believed to
be unobservable.
Based on both approaches, corporate restructuring is found to have no
significant impact on stock performance and coverage ratio of firms that actually
participated in the program. Based on matching approach alone, we found that
corporate restructuring has no significant impacts on most performance measures
of firms that actually participated in the program except improving their return on
assets and worsening their debt-to-assets ratio in later years. For firms that did not
participate in a restructuring program, matching method suggests that adopting debt
10
restructuring would have no conclusive effects on their stock performance and even
worsen their balance sheet performances such as capital cost, debt-to-assets ratio
and coverage ratio. Return on assets is the only measure that improves after debt
restructuring for non-restructured firms. Corporate restructuring seems to be more
effective based on endogenous switching regression approach because it either
improves or has no significant effect on firm’s balance sheet performance, with
non-restructured firms benefit more than restructured firms. Based on endogenous
switching regression approach alone, the only unfavorable effect of restructuring is
that it worsens the debt-to-assets ratio of restructured firms.
Thus, it can be concluded that the impact of corporate restructuring on the
subsequent performance of the restructured Thai firms is either insignificant or
inconclusive because of the contradicting results between the two approaches. It
can otherwise be stated that firms that are restructured do not perform any better
before and after the restructuring process has occurred.
The previous literature on the impact of corporate restructuring, either
voluntarily or involuntarily, on subsequent firm performance using firm level data,
is scarce. Claessens, Klingsbiel, and Laeven (2003) examined the impact of
restructuring policies at the country level on firm level earnings-sales ratios in eight
countries that have experienced systemic crisis and significant corporate
restructuring. Essentially, they find that corporations in countries that have
undertaken more corporate restructuring measures have not performed any better
than corporations in countries that have undertaken fewer measures. Polsiri and
11
Wiwattanakantang (2004) compared the performance (corporate earnings) of Thai
firms before and after the corporate restructuring. They found that after controlling
for industry effects, restructuring has improved corporate earnings. However, the
authors did not econometrically control the endogeneity of corporate restructuring
decisions as this paper does. It is found that once the endogeneity of corporate
restructuring decisions are controlled, the positive impact of corporate restructuring
on performance disappears.
The rest of Chapter 2 is organized as follows. Section 2.2 describes the
data, variables of interest and their statistics. Section 2.3 describes the estimation
methods and their underlying assumptions as well as limitations. Section 2.4
presents the results of estimation with various specifications and Section 2.5
concludes this chapter.
2.2 Data Description
The impact of debt restructuring on each firm’s performance is examined
using the annual panel data of 219 non-financial Thai firms in 17 industries from
1996 to 2004. The types of industry and the number of firms in each industry are
reported in Appendix A. For each firm, the variables considered in this analysis
include total assets, total liabilities, capital stock, total sales, total cost of sales,
interest expenses, and net profit. The annual observations of these series were
obtained from the corresponding annual financial statements of each firm. The
12
total assets and liabilities were taken from the balance sheet statement. Capital
stock was based on the balance sheet under the ‘net property, plant and equipment’
category. Total sales, total cost of sales, and interest expenses were available in the
income statement. Net profit was obtained from the cash flow table. All firms
considered were listed in the Thai stock market and their financial statements were
available at the library of the Stock Exchange of Thailand (SET). The post-2000
financial statements were also available online (www.setsmart.com).
Based on the above-mentioned series, five variables were created as
performance measures for each firm. Table 1 details the definitions of these
variables.
Table 1. Definitions of Firm Performance Measures
Performance Measure Definition
Excess stock market returns They are also known as abnormal returns in stock trade and
measured as the difference between the annual growth of each
firm’s stock price and the annual growth of stock market
index. The annual stock market index is calculated by
averaging its monthly indices during the corresponding year.
Return on Assets (ROA)
It shows how profitable a firm’s assets are in generating
revenue.
(i) ROA1 is equal to net profit of a firm divided by
its total assets
(ii) ROA2 is similar to ROA1 but net profit is
replaced by operating profit, which is the total of
net profit and interest expenses.
Capital cost It is the ratio of a firm’s interest expenses over its total
liabilities.
Debt-to-assets ratio It is also known as financial leverage and calculated as a
firm’s total liabilities divided by its total assets.
Coverage ratio It is calculated as a firm’s net profit divided by its interest
expenses.
13
Also incorporated in the analysis are two dummy variables — the political
connection dummy variable and the debt restructuring dummy variable. The
political connection dummy variable of a firm is one in every year if that firm has a
connection with politicians in the ruling or opposing parties; otherwise, the variable
is zero. The debt restructuring dummy variable of a firm in a particular year is one
if the firm is undergoing or has undergone any kind of debt restructuring at that
time; otherwise the variable is zero. Once a firm voluntarily renegotiated its debts
or entered the rehabilitation process, it would be considered debt-restructured in all
subsequent years regardless of the firm’s recovery. In this analysis, debt
restructuring refers to either a firm’s voluntary re-negotiation of its debt contract
with the creditors or its participation in the rehabilitation program called
REHABCO,
2
which began 1998. Starting in 1998, firms that performed poorly
were forced to refinance or to enter the rehabilitation program. If a firm is delisted
from the market, all of its observations after the time of delisting are excluded from
the analysis. Information regarding debt restructuring, including the list of
participating firms and the date each firm entered the program, is collected from the
firm’s financial reports to the Stock Exchange of Thailand (SET). The numbers of
2
The REHABCO program was implemented by the Stock Exchange of Thailand (SET) in 1998
following the change of Thailand’s exchange rate system in order to protect the rights and benefits
of small shareholders. The Stock Exchange of Thailand requires that listed companies that show
negative shareholders' equities on their balance sheets, after adjusting for any unrealized losses due
to the exchange rate regime switch and in accordance with the auditor’s comments, be transferred to
the ‘Companies Under Rehabilitation’ category (REHABCO).
In the rehabilitation process, the SET works closely with related parties including debtors,
creditors and shareholders to resolve REHABCO firms’ financial problems and to help reinstate the
firms into their original categories.
14
firm that enter the debt restructuring in each year for each industry are reported in
Appendix A. The information regarding a firm’s political connection can also be
obtained from the SET.
Table 2 reports the descriptive statistics on the variables of interest. In
particular, the mean averages of each aforementioned performance measure across
debt-restructured and non-debt-restructured firms from year 1996 to 2004. For the
debt-restructured firms, the averages are segregated before and after the entering
the restructuring program. The idea is to compare the corporate performance of the
restructured firms in the pre- and post-restructured periods. All of the continuous
variables will be collapsed into two periods, pre and post restructuring.
Table 2. Descriptive Statistics of the Variables of 219 Non-financial
Thai Firms, 1996 – 2004
Debt-Restructured Firms Regular Firms
Pre
Restructuring
Post
Restructuring
(non
restructured)
Net Profit/Total Assets (ROA1) -0.096 0.089 0.025
Operating Profit/Total Assets (ROA2) -0.040 0.134 0.052
Interest Expenses/Total Capital
(Capital Cost) 0.071 0.046 0.049
Debt-to-Assets Ratio 0.829 1.002 0.514
Excess Stock Market Returns -0.741 -0.652 -0.823
Coverage Ratio -1.469 3.485 4.074
Total Assets 13901 10380 8738
Note:
1. Net Profits and Operating Profits are from financial statements of the firms.
2. Excess Stock Market Returns denotes difference between stock price ratio of firms in year t and
the SET index in year t.
Source: SETSMART, 2006
15
For example, for a particular firm restructured in 2002, the pre-restructured
period will be 1996 to 2001, and the post-restructured period will be 2002 to 2004.
Two values of this firm’s performance measures such as coverage ratio will be
constructed — the average of the coverage ratio between 1996 and 2001 and the
average of the coverage ratio between 2002 and 2004. The idea is to compare the
change in the average performance before and after the restructuring. Each
restructured firm may have different pre-restructured and post-restructured time
periods calendar-wise; however, by this method, the variables of all the restructured
firms can be classified into just two periods. As more firms have restructured their
debts over time, the proportion of debt-restructured firm increases and the averages
across debt-restructured firms are calculated from an expanding set of firms over
the years.
For the debt restructured firms, the first two rows of Table 2 indicate that
the average return on assets increases from negative values to positive value after
the firms adopted the debt restructuring program. This is true for both ROA1 and
ROA2, with the ROA1 and ROA2 averages being -0.096 and -0.04 pre-
restructuring and becoming 0.089 and 0.134 respectively post-restructuring.
The return on assets of debt-restructured firms, averaged from the pre-
restructuring periods, are lower than the average ROA of regular firms that did not
debt restructure during our sample period, which are 0.025 and 0.052 for ROA1
and ROA2. The return on assets of debt-restructured firms, averaged from the
periods after debt-restructuring started, became higher than the average ROA of
16
non-restructured firms. In general, the averages of ROA2 were slightly higher than
those of ROA1, consistent with the definition of ROA variables. The patterns and
relations between the average across debt-restructured firms and the average across
all firms were almost identical for ROA1 and ROA2.
The average capital cost of pre-debt restructured firms was 0.071, higher
than that of restructured firms averaged post debt restructuring and regular firms
averaged throughout the sample periods. The difference was 0.025 and 0.022
respectively. The capital cost average of 0.049 of regular firms remained close to
those of post-debt restructured firms capital cost average of 0.046. This indicates
that the debt restructuring program lower the capital costs of firms that participated
in the program to the same level as those of non-participants.
After the restructuring, the average debts-to-assets ratio across debt-
restructured firms increases approximately from 0.83 to 1. The debt-to-assets ratio
of debt-restructured firms, whether averaged pre-debt restructuring or post-debt
restructuring, are still higher than the average debts-to-assets ratio of non-
participants, at 0.51. This indicates that, on average, pre-debt restructured firms
use higher proportions of debts to finance their assets.
The excess stock market returns averaged across the participating firms
become less negative after the restructuring, from -0.74 to -0.65, which are both
lower than -0.82 the excess stock market returns averaged across the non-
participating firms throughout the sample periods.
17
The average coverage ratio of debt-restructured firms shows significantly
improvement after the restructuring, from -1.47 to 3.48. The coverage ratio of
debt-restructured firms, both pre and post debt restructuring average, are lower than
those of non-restructured firms figures at 4.07.
With the exception of debt-to-assets ratio, every other performance
measures show signs that the situations of the debt-restructured firms have
improved after debt restructuring, as the return on assets, excess stock market
return and the coverage ratio became higher and the capital costs became lower, on
average, in the post-restructuring periods. We proceed to the econometric methods
of measuring the actual effects of debt restructuring on these performance measures
in the next section.
2.3 Methodology
This section presents the econometric models used in this study. Partial
effects of debt restructuring on various measures of firm performance are estimated
using two distinct yet closely related techniques, namely “matching estimators” and
“endogenous switching regression estimators.” Both methods are widely used in
program evaluation literature which concerns measuring the impact of interventions
or treatments on outcomes of interest, particularly in the labor market area. The
choice of method that is more appropriate depends on the observability of non-
random factors that determines the treatment selection criteria. The “matching”
18
method is used when said factors are observable and the “endogenous switching
regression” method is used when these factors are assumed to be unobservable. To
consider our posed problem in the context of program evaluation, treatment effects
are estimated from firm-level observational data, with debt restructuring as the
treatment and various firm performances as the outcome. Which method should be
used depends on whether the firm’s characteristics that determines its participation
in the restructuring program can be observed. The common problem in program
evaluation literature is briefly discussed below. Matching and endogenous
switching regression models and their key assumptions as well as applications in
the context of this analysis will also be discussed in the following subsections.
The fundamental problem in treatment evaluation is its non-experimental
setting. The main purpose of program evaluation is to find the causal effect of
some treatment or program on some outcome experienced by units in the
population of interest. The impact of the treatment on the outcome could be
consistently estimated if the treatment were implemented randomly as in traditional
scientific experiments. That is, since the criteria of selection into treatment are
random, the differences between the treated group and the control group are solely
due to the treatment and not to some other specific characteristics. In a single cross
section, the effects of randomly assigned treatment could be measured by
comparing the outcome between units in the treatment and the control groups. On
the other hand, if only the data for the treatment group are available both before and
after the treatment, the treatment effects can be measured by comparing the
19
outcome of the units in the treatment groups across time, assuming that the
treatment is the only cause of changes in the outcome with no external influences.
However, the debt restructuring treatment in this study is non-experimental
in the sense that a firm’s decision to restructure its debt and the criteria that force
firms into the REHABCO program are not random but initially determined on the
basis of their own performances. For example, firms that perform poorly are more
likely to be forced into the program. If this self-selection is not accounted for, the
estimated treatment effects of the debt-restructuring program on firm performance
may be biased. Formally, this self-selection into the program makes estimation and
inference problematic because the treatment decision, which may be determined by
firms’ observable and unobservable characteristics, is correlated with the outcome.
This creates what is commonly referred to as selection bias.
To solve the problem of “selection on observables,” we used the methods of
matching estimation, where counterfactual outcomes for members in the treatment
group are inferred from members of the comparison or control group that have
similar observable characteristics. Unobservable characteristics are assumed to
play no role in determining the treatment decision under the matching methods. If
this assumption holds, the treatment decision is considered to be random once the
observable characteristics are controlled, thus the selection bias is eliminated. This
key assumption of the matching estimator is called the ‘conditional independence’
assumption, which will later be explained in detail.
20
In other cases where the conditional independence assumption doesn’t hold,
the selection bias still exists even after the relevant observable characteristics are
controlled. In order to correct the problem of “selection on unobservables”
(Heckman, 1979), the methods of endogenous switching regression estimation is
employed and the relatively strong distributional assumptions are needed. The
model can be viewed as a variation of “switching regression model with
endogenous switching” (Maddala, 1983) whose essential elements are common to
sample selection models. The success of this method depends on the availability of
at least one purely exogenous component that is correlated to the possibility of
being selected into the treatment but uncorrelated to outcome variables except
through the treatment. This will be explained in details in the following section.
2.3.1 Matching Approach
Matching is a relatively new nonparametric approach that provides a
statistical framework for the estimation of treatment effects. To better understand
the matching approach, the fundamental problem of program evaluation (the non-
experimental setting) was considered in view of the missing data problem. The
treatment effects could be easily estimated if the potential outcome variables for the
treated units — had they not participated in the treatment — were observable.
However, at any point in time, a unit can either be under the treatment or not, but
not both; hence, suitable counterfactuals need to be constructed. In the pure
scientific experimental setting, treatment assignment is random. Thus, the non-
21
treated units in the control group are good representatives of the missing
counterfactuals since treatment is independent of potential outcomes.
In contrast, in most program evaluations, treatment on the units often
depends on pre-treatment characteristics and is not random; therefore, non-treated
units are no longer appropriate counterfactuals. Alternative methods of
constructing suitable counterfactuals are needed in this non-experimental
environment. The basic idea of matching is to pair each unit of interest with its
own comparison group that represents counterfactual of the unit of interest. This
comparison group contains units that are similar to the unit of interest in terms of
their observable characteristics but have the opposite treatment status. The
treatment effects are then evaluated by comparing the outcome variables of the
interested unit to those of units in its comparison group. Because the comparison
group of each unit has similar characteristics, the treatment effects contain no self-
selection bias that stems from differences in observable characteristics.
If the number of observable characteristics is small, that is, the vector of
covariates has a low dimension, implementing this procedure is straightforward.
This procedure becomes impractical when there are many observable
characteristics, since it is difficult to determine as to which dimensions match a
unit. Hence, characteristics must be mapped into a measure of lower dimensions,
such as the Euclidean norm. Simplifying a comparison between two sets of
characteristics and a derivation of distance between them, this measure of lower
dimensions forms a basis for matching.
22
In their seminal article, Rosenbaum and Rubin (1983) replaced the
observable characteristics with the “propensity score” of these characteristics as the
basis for matching to resolve the problem of high dimensionality. The propensity
score is defined as the conditional probability of being treated given the
individual’s observable characteristics. The propensity score can be estimated
using parametric methods such as logit model or probit model. However, this
imposes strong distributional assumptions on the underlying data. In particular, the
dangers of misspecification may be severe if the error terms are not independent or
identically distributed from their known parametric distributions. To overcome this
restriction, non-parametric methods, such as kernel-based estimator (Heckman,
Ichimura & Todd, 1997), can be used. Propensity score matching estimators have
become increasingly popular in the non-experimental program evaluation because
they reduce the dimensionality of matching on the individual characteristics to
matching solely on the basis of estimated propensity scores.
The mechanism of matching method is illustrated as follows. For unit i,
where i =1,….,N, let Y be the outcome variable experienced by units in the
population of interest,
i
X be the set of observable characteristics of unit i and
i
D be
a binary indicator of the treatment actually received by unit i, equal to one if treated
and zero otherwise.
1,i
Y denotes the potential outcome of unit i if i were exposed to
the treatment, and
0,i
Y denotes the potential outcome of unit i if i were not exposed
to the treatment. The actually observed outcome of unit i is
23
( )
0, 1, 0, ii i i i
YY DY Y =+ − (1.1)
and the treatment effect on the outcome for unit i would be
1, 0, ii i
TE Y Y =−. (1.2)
In general, we are interested in the average treatment effects (ATE) expressed as
10
ATE E Y Y =− ⎡⎤
⎣⎦
. (1.3)
When only the units in the treatment groups are considered, the resulting
comparison is called an average treatment effect on the treated (ATET), expressed
as
10
1 ATET E Y Y D ⎡⎤ = −=
⎣⎦
. (1.4)
Similarly, when only the units in the control groups are considered, the resulting
comparison is an average treatment effect on the control (ATEC), expressed as
10
0 ATEC E Y Y D ⎡⎤ = −=
⎣⎦
. (1.5)
As discussed above, only one of the potential outcomes is observed, while the other
is unobserved. The matching method imputes the missing potential outcome of
unit i by using an average of outcomes of units in the comparison group. This
comparison group for unit i contains units whose characteristics are considered to
be similar to characteristics of unit i but with the opposite treatment status. To find
the suitable matches to unit i, let
, ki
d be the distance between the characteristics of
unit i
i
X , and the
th
k nearest unit with the opposite treatment status. The set of
indices for the matches for unit i that is at least as close as the
th
k match can be
expressed as
24
{ } ,,
1,..., 1 ,
ki l i l i ki
Ml ND DX X d == = − − ≤ (1.6)
when observable characteristics are directly used as the basis for matching where
, ki k i
dX X =− .
The propensity score is expressed as () Pr 1 pxD X EDX ⎡ ⎤⎡ ⎤ ≡= =
⎣ ⎦⎣ ⎦
. When a
propensity score is used as the basis for matching instead of the observable
characteristics themselves,
, ki
d can be considered as the distance between the
propensity score of unit i's characteristics ( )
i
pX and the
th
k closest propensity
score of another unit that has the opposite treatment status. The set of indices for
the matches for unit i that have the propensity score at least as close as the
th
k
match can be expressed as
() ( )
{ } ,,
1,..., 1 ,
ki l i l i ki
Ml ND DpX pX d == = − − ≤ (1.6a)
when the propensity score of the characteristics is the basis of matching where
( ) ( )
, ki k i
dpX pX =− .
For unit i, its estimated potential outcome if non-treated is as follows:
0,
ˆ
ii
YY = if the actual status of unit i is non-treated ( 0
i
D = ),
,
0,
,
1
ˆ
#
ki
il
ki
lM
YY
M
∈
=
∑
if the actual status of unit i is treated ( 1
i
D = ).
Similarly, its estimated potential outcome, if treated, is as follows:
1,
ˆ
ii
YY = if the actual status of unit i is treated ( 1
i
D = ),
,
1,
,
1
ˆ
#
ki
il
ki
lM
YY
M
∈
=
∑
if the actual status of unit i is non-treated ( 0
i
D = ).
25
The matching estimator can then be implemented as
() 1, 0,
1
1
ˆˆ
N
kii
i
ATE Y Y
N
=
=−
∑
. (1.7)
This can easily be modified to estimate the ATE on the treated and ATE on the
control as follows:
() 0,
1
:1
1
ˆ
i
kii
iD
ATET Y Y
N
=
=−
∑
(1.8)
() 1,
0
:0
1
ˆ
i
kii
iD
ATEC Y Y
N
=
=−
∑
(1.9)
where
1
N and
0
N are the number of treated and non-treated units respectively.
An important assumption for matching methods is the “conditional
independence” assumption
3
. This assumption requires that, conditional on the
observable characteristics, the outcome and the treatment are independent, and can
be formally written as
10
, YY DX ⊥ .
In other words, conditional on the observable characteristics included in X,
the treatment decision is independent of the potential outcome. In other words,
once the observable characteristics are controlled for, the treatment assignment is
assumed to be random, and there is no self-selection bias with regard to the
observable characteristics. However, the matching method is silent about the self-
selection bias with regard to unobservable characteristics. If the treatment
assignment is believed to be correlated with some unobservable characteristics,
3
Rosenbaum & Rubin (1983) refer to “conditional independence assumption” as “ignorability of
treatment assumption”; Imbens (2005) referred to it as “unconfoundedness assumption”.
26
then it cannot be considered random even after the observable characteristics are
controlled for, and the estimated treatment effects will suffer from the self-selection
bias. One alternative method to deal with this problem is illustrated in the next
section.
The propensity score matching is particularly appropriate in evaluating
cases in which: (a) some observations receive non-random treatment; (b) selection
of observations for treatment is non-random and based only on observable
characteristics of each observation; (c) few observations in the control group have
similar characteristics to observations in the experimental group; and (d) selecting
comparable treated and non-treated observations is difficult due to the high number
of observable characteristics needed to determine comparability (Dehejia & Wahba,
2002). Because a firm’s participation in debt restructuring is non-random and
heavily based on its own characteristics, matching is an appropriate approach for
evaluating debt-restructuring program. In the context of our analysis, the outcome
variableY denotes various performance measure variable experienced by Thai
firms in our study,
i
X denotes the set of observable characteristics of firm i and the
treatment variable
i
D denotes a binary indicator of the actual status of debt
restructuring of firm i, equal to one if restructured and zero otherwise. The
potential outcome
1,i
Y denotes the performance of firm i if restructured, and
potential outcome
0,i
Y denotes the performance of firm i if not restructured.
Because the impacts of the debt restructuring program may not be time-invariant,
27
we show the matching results segregated annually between 1997 through 2004.
We report the matching estimation results based on both normal and propensity
score basis in section 2.4.1.
2.3.2 Endogenous Switching Regression Approach
When the selection is caused by common unobservable factors, the
restrictive assumption of conditional independence in the matching approach as
previously mentioned in Section 2.3.1 breaks down and we need an alternative
strategy in order to eliminate the self-selection bias. To illustrate this alternative
method, we begin with the base model commonly used in regression estimation.
Formally, we consider a basic framework typically in the form of a linear
parametric model with binary treatment. Let
it
Y be the outcome of interest for
individual i at time t and let
it
D be a binary treatment dummy equal to one if
individual i receives treatment in period t; otherwise, it is zero. To better capture
the outcome dynamics, the traditional way to accommodate the observed
characteristics
it
X in the model is to simply add them linearly to the regression
equation. The model can be written as follows:
it it t t it it
u D X Y + + = π β . (2.1)
To capture the effects of the treatment, one typically performs the Ordinary Least
Squares (OLS) regression on equation (2.1) with observations pooled across
treatment and control groups as well as before and after the treatment in order to
estimate ( ) ,
tt
β π The elements in vector
t
β are the effects of the observed
28
characteristics on the outcome for each period and
t
π represents the impact of
treatment on the treated units where the subscript t indicates that the treatment
effects change over time. Under the assumption that the treatment is exogenous
and uncorrelated with the error terms in the outcome equation (2.1), the OLS
estimator of
t
π is consistent and equals to the average treatment effect of interest
(ATE). This model is too restrictive because the treatment may create interaction
effects with observed or unobserved characteristics. As a result, distribution of
it
u
as well as the parameter of explanatory variables
t
β when treated may not be the
same as those when not treated,
01 DD
tt
ββ
= =
≠ . Also in most program evaluation
problem, the decision of individual whether to participate in the program is based
on individual self-selection. This implies that there are some common
unobservable factors that is not included in the explanatory variables
it
X but affect
both treatment
it
D and outcome
it
Y . Under normal OLS regression, these omitted
components will be in the error terms
it
u and make these error terms to be
correlated with the treatment variable
it
D . In the context of our analysis, firm’s
participation in debt restructuring is not exogenous but partly determined by its
own performance; therefore, we need more general models that can overcome this
problem.
As in matching approach, we let
i
Y be performance measure variable
experienced by Thai firms i and
i
D be a binary debt restructuring variable of firm i,
equal to one if restructured and zero otherwise.
1,i
Y denotes the potential
29
performance of firm i if restructured, and
0,i
Y denotes the potential performance of
firm i if not restructured.
i
X and
i
Z denotes the set of observable characteristics of
firm i with a vector of observed variables ) , , , (
i i i i
D Z X Y :
i i i
u X Y
1 1 1
+ = β , (2.2)
i i i
u X Y
0 0 0
+ = β , (2.3)
i i i
Z D ε γ + =
*
. (2.4)
The first equation is the performance equation of firm i given that firm i has
been debt-restructured. The second equation is the performance equation of firm i
given that firm i has not been debt-restructured. The third equation is the sample
selection function
4
where
*
i
D is a latent variable for the actually observed treatment
variable
i
D which is expressed as the binary-response model such that:
1 =
i
D if 0
*
>
i
D ,
(2.5)
0 =
i
D if 0
*
≤
i
D .
The observed performance variable
i
Y takes one of the two forms depending on the
treatment variable
i
D as follows:
i i
Y Y
1
= if 1 =
i
D ,
(2.6)
i i
Y Y
0
= if 0 =
i
D .
Model that consist of equations in the form of (2.2) to (2.6) are known as
“switching regression model” (Goldfeld & Quandt, 1973) in which the behavior of
4
Also known as decision function, criterion function or participation function in the literature.
30
the agent is described by two regression equations and a sample selection function
that determines which of these two equations is applicable. If the error terms of the
sample selection function
i
ε is correlated to the error terms of outcome equations
) , (
0 1 i i
u u , this is simply the “switching regression model with endogenous
switching” (Maddala & Nelson, 1975) or the “Tobit type 5 model” (Amemiya,
1985). It is analogous to Roy model (1951) in the sense that the outcome of the
treated agents and that of non-treated agents have different conditional mean
functions. For the rest of this paper, we will address equation (2.2) to (2.6) as the
“endogenous switching regression model”.
This model fall in the class of switching models in the sense that firm i can
not simultaneously be in state of restructuring and non-restructuring; hence, we can
not simultaneously observe
i
Y
0
and
i
Y
1
. The observed firm performance
i
Y
switches between the two values and takes the form
i i i i i
Y D Y D Y
0 1
) 1 ( ⋅ − + ⋅ = . (2.7)
This equation is equivalent to equation (1.1) in the matching method estimation. If
i
D is one, firm i is in a state of restructuring and the performance equation take the
form of equation (2.2) where
i
Y is
i
Y
1
. If
i
D is zero, firm i is in a state of non-
restructuring and the performance equation take the form of equation (2.3) where
i
Y
is
i
Y
0
.
In practice for this type of model, explanatory variables in equation (2.2)
and (2.3) ) (
i
X are allowed to overlap with explanatory variables in equation (2.4)
31
) (
i
Z but at least one component of
i
Z needs to be a unique and nontrivial
determinant of the treatment. That is, there is at least one source of variation in the
treatment
i
D that is independent of the outcome
i
Y . Formally, this component,
considered as an instrument variable
i
z , needs to be correlated with the
endogenous treatment variable
i
D and uncorrelated with the performance variable
) , (
0 1 i i
Y Y after
i
D is controlled for. This can be expressed as
i i i i
D Y Y z | ) , (
0 1
⊥ .
These identifying instrument variables may be weak instruments if they are
correlated with the errors in the outcome equations. The set of explanatory
variables of firm performance equation (
i
X ) are shown in table 3.
Table 3. List of Explanatory Variables for Each Performance Measure
Firm Performance Measures (
i
Y ) Explanatory Variables (
i
X )
Excess stock market returns
Return on Assets (ROA)
total asset
capital cost
debt-to-asset ratio
coverage ratio
Capital cost
total asset
debt-to-asset ratio
coverage ratio
debt-to-asset ratio
total asset
capital cost
coverage ratio
coverage ratio
total asset
capital cost
debt-to-asset ratio
32
With the exception of total assets, all explanatory variables in performance
equation (2.2) and (2.3) in Table 3 are also explanatory variables of the sample
selection rule (2.4)
i
Z for that particular performance measure. Two additional
instrument variables
i
z , ‘close price over par value’ and ‘market capitalization’, are
included in
i
Z as the independent sources of variation in the treatment variable
i
D .
The basic idea of estimation using endogenous switching regression model
approach is to initially include selection correction terms, derived from the self
selection rule (2.4), into the outcome equations (2.2) and (2.3) in order to account
for the selection bias. We then can consistently estimate parameters in the new
outcome equations that include selection correction terms by a simple OLS
regression. Once we have consistent estimators of performance measure equation
in both states, we can approximate the potential outcome, or in the context of our
analysis, the performance counterfactuals of every firm considered. All average
treatment effect ATE, ATET and ATEC can then be estimated in the same way
they are estimated in matching approach equation (1.7) to (1.9) in section 2.3.1.
In the following section, we will employ strong distribution assumptions
and fully parametric model to estimate the endogenous switching regression model
in equation (2.2) to (2.6). In essence, this parametric approach requires strong
assumption of the joint normal distribution for the error terms, a common feature in
selection model literature. Later, we apply the alternative semiparametric approach
33
to estimate the endogenous switching regression model, in which the assumption of
joint normal distribution is relaxed.
Parametric approach
In the widely used parametric approach, it is assumed that the error terms in
the outcome equation (2.2), (2.3) and selection rule (2.4) are independent of their
own two set of explanatory variables ( Z X , ):
() 0 , |
1
= Z X u E , ( ) 0 , |
0
= Z X u E and ( ) 0 , | = Z X E ε .
The error terms ) , , (
0 1 i i i
u u ε are assumed to be jointly multivariate normal
distributed with mean vector zero and covariance matrix Σ
5
() Σ , 0 ~ )' , , (
0 1
N u u
i i i
ε with ) , , (
0 1 i i i
u u Cov ε = Σ =
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎣
⎡
1
0 1
0 00 10
1 10 11
ε ε
ε
ε
σ σ
σ σ σ
σ σ σ
.
If
i
ε is independent of ) , (
0 1 i i
u u , the covariance parameters
ε
σ
1
and
ε
σ
0
are equal to
zero. This is simply the switching regression model with ‘exogenous switching’
since the selection is not correlated with the outcome. Since
i
ε is assumed to be
correlated to ) , (
0 1 i i
u u , equation (2.2) to (2.6) combined with nonzero covariance
parameters
ε
σ
1
and
ε
σ
0
creates endogenous switching regression model. This
implies endogeneity of the treatment variable, which in turns create the selection
bias as previously stated. The model represented by equation (2.2) to (2.4) can be
5
The covariance parameter
10
σ reflects the covariance between the outcomes in both states which
is unidentified and generally set to zero for empirical application. Because γ is estimable up to a
scale factor, variance of
i
ε is assumed to be 1 for identification purpose.
34
consistently estimated by the method of maximum likelihood function (ML). The
likelihood function of this model is
() [ ] {}() [ ] {}
∏
=
−
≤ ⋅ ≤ > ⋅ > =
N
i
D
i i i
D
i i i
i i
D D Y f D D Y f L
1
1
* *
0
* *
1
0 Pr 0 | 0 Pr 0|(2.8)
where () ⋅ f denotes the density function. The first term is the continuous
contribution when 0
*
>
i
D and the second term is the continuous contribution
when 0
*
≤
i
D . Given the parameter of interest ( )
ε ε
σ σ σ σ β β θ
0 1 00 11 0 1
, , , , , ' = , its ML
estimator θ
ˆ
can be obtained by maximizing the following log-likelihood function:
() θ L ln () [ ] [ ] {}
∑
=
> ⋅ > =
N
i
i i i i
D D Y f D
1
* *
1
0 Pr 0 | ln
() ( ) [ ] [ ] { } 0 Pr 0 | ln 1
* *
0
≤ ⋅ ≤ − +
i i i i
D D Y f D . (2.10)
Maximization of this log-likelihood function is feasible but can be cumbersome. It
is more common to try to obtain consistent estimators of parameters in performance
equation (2.2) and (2.3) by applying the Heckman two-stage method (Heckman,
1979) to the truncated means of these equations. Once we find the appropriate
form of performance equation (2.2) and (2.3) that account for the selection bias, we
can estimate the conditional mean for both states of restructuring or non-
restructuring of each firm. The benefit of debt restructuring can then be evaluated
from averaging the conditional mean difference of the firm performances of each
firm between the two states. This approach is illustrated as follows.
Before we can apply the Heckman-two stage method, we first need to find
selection correction term that account for selection bias in both performance
35
equations. To find selection correction term in equation (2.2), we consider only the
firms that restructure their debts with
i
D =1 since all these firms have the observed
performances of the equation (2.2) form. The conditional mean of performance
equation
i
Y
1
is
[ ] 0 |
*
1
>
i i
D Y E [] 0 |
1 1
> + + =
i i i i
Z u X E ε γ β
[ ] γ ε β
i i i i
Z u E X − > + = |
1 1
. (2.11)
If the error terms of performance equation under debt-restructuring state and
selection equation ) , (
1 i i
u ε are independent then the last term simplifies to
[][] 0 |
1 1
= = − >
i i i i
u E Z u E γ ε and OLS regression of
i
Y
1
on
i
X will give a
consistent estimator of
1
β . This is simply the case of zero covariance
parameters 0
1
=
ε
σ as mentioned earlier. If instead they are correlated, the
parameter
ε
σ
1
is nonzero hence the expectation term of
i
u
1
in equation (2.11) is
nonzero. As a result, a simple OLS regressing
i
Y
1
on
i
X will give an inconsistent
estimator of
1
β . Therefore we need to account for the selection bias in equation
(2.11).
Given the joint normality assumption of the error terms, we can rewrite
i
u
1
as
i i i
u
1 1 1
ξ ε σ
ε
+ = (2.12)
where [ ]
2
1 11 1
, 0 ~
ε
σ σ ξ − N . We explain how to obtain equation (2.12) in Appendix
B. Replace equation (2.12) in (2.11), we get
36
[ ] 0 |
*
1
>
i i
D Y E [ ] γ ε ξ ε σ β
ε i i i i i
Z E X − > + + = |
1 1 1
[ ] [ ] γ ε ξ γ ε ε σ β
ε i i i i i i i
Z E Z E X − > + − > + = | |
1 1 1
[ ] γ ε ε σ β
ε i i i i
Z E X − > + = |
1 1
( )
() γ
γ φ
σ β
ε
i
i
i
Z
Z
X
Φ
+ =
1 1
(2.13)
where the last term
()
() γ
γ φ
σ
ε
i
i
Z
Z
Φ
1
is the selection correction term. We receive the
third line because
i 1
ξ is independent of
i
ε , therefore [ ][ ] 0 |
1 1
= = − >
i i i i
E Z E ξ γ ε ξ .
The deduction of the last line is explained in Appendix C. Equation (2.13) can be
rewritten as
[ ] ( ) γ λ σ β
ε i i i i
Z X D Y E
1 1
*
1
0 | + = > (2.14)
where ()
()
() z
z
z
Φ
=
φ
λ is known as inverse Mills ratio. And we have the truncated
variance as
[ ] ( ) ( ) ( ) γ λ γ γ λ σ σ
ε i i i i i
Z Z Z D Y Var + − = >
1 11
*
1
0 | . (2.15)
Similarly to equation (2.11) of the treated firms, the conditional expectation of
performance equation for the non-treated firms (2.3) is
[ ] 0 |
*
0
≤
i i
D Y E [ ] γ ε β
i i i i
Z u E X − ≤ + = |
0 0
. (2.16)
Under the joint normal distribution assumption of ) , (
0 i i
u ε , the conditional
expectation of
i
Y
0
equivalent to equation (2.13) of the treated firms is
37
[]
( )
() γ
γ φ
σ β
ε
i
i
i i i
Z
Z
X D Y E
Φ −
−
+ = ≤
1
0 |
0 0
*
0
(2.17)
where the last term
( )
() γ
γ φ
σ
ε
i
i
Z
Z
Φ −
−
1
0
is the selection correction term. Its derivation is
explained in Appendix C.
Once the selection correction terms have been identified, we are ready to
obtain consistent estimators of ) , (
0 1
β β . Heckman suggests a two step method
applied to the truncated means in equation (2.13) and (2.17), which can be rewrite
as the following new performance regression equations:
()
()
i
i
i
i i
v
Z
Z
X Y
1 1 1 1
+
Φ
+ =
γ
γ φ
σ β
ε
, (2.18)
( )
()
i
i
i
i i
v
Z
Z
X Y
0 0 0 0
1
+
Φ −
−
+ =
γ
γ φ
σ β
ε
, (2.19)
where (2.18) is applicable to firms with 1 =
i
D and (2.19) is applicable to firms
with 0 =
i
D . This implies that the new error terms are
()
() γ
γ φ
σ
ε
i
i
i i
Z
Z
u v
Φ
− =
1 1 1
(2.20)
()
() γ
γ φ
σ
ε
i
i
u i
Z
Z
u v
Φ −
+ =
1
0 0 0
(2.21)
for debt-restructured firms and non-restructured firms respectively. The new error
terms now satisfy the zero conditional mean; ( ) = = 1 |
1 i i
D v E () 0 0 |
0
= =
i i
D v E .
The first step is to estimate γ in equation (2.4) using parametric methods such as
logit or probit ML estimation or non-parametric methods such as maximum score
38
estimators, root-N consistent semiparametric estimators or estimator used in Klien
and Spady (1993) with observation
i
D . We can then find the estimated selection
correction terms in performance equation (2.18) and (2.19) as
()
() γ
γ φ
σ
ε
ˆ
ˆ
1
i
i
Z
Z
Φ
and
()
() γ
γ φ
σ
ε
ˆ 1
ˆ
0
i
i
Z
Z
Φ −
−
. The second stage is to estimate equation (2.18) and (2.19) using
simple OLS regression. With the estimated selection correction terms as additional
regressors accounting for the selection bias, regression of ( )
i i
Y Y
0 1
, on
i
X in the
performance equation (2.18) and (2.19) will give consistent estimators of
1
β and
0
β
respectively.
Based on consistent estimators of ()
′
ε ε
σ σ β β
0 1 0 1
, , , obtained by Heckman
two step method above, we can estimate the average treatment effect (ATE, ATET
and ATEC) of debt restructuring on various firm performances in the same way
they are estimated in matching approach equation (1.7) to (1.9) in section 2.3.1.
From matching approach, the average treatment effect (ATE) as expressed
in equation (1.3)
10
ATE E Y Y =− ⎡⎤
⎣⎦
can be approximated by equation (1.7) as
() 1, 0,
1
1
ˆˆ
N
ii
i
YY
N
=
−
∑
. As we already know that every firm is either in the restructuring or
non-restructuring state, equation (1.7) can be rewritten as
()( )
0 1
1, 0, 1, 0,
11
1
ˆˆ ˆˆ
|1 | 0
N N
ii i i i i
ii
YY D Y Y D
N
==
⎡⎤
⎢⎥ −=+ − =
⎢⎥
⎣⎦
∑∑
(2.22)
39
where
1
N and
0
N are the number of restructured and non-restructured firms
respectively as in section 2.3.1. The first summation in the bracket is over firms
who actually participated in the debt restructuring and the last summation is over
firms who did not restructure their debts. In the first summation,
1,
ˆ
|1
ii
YD = can be
estimated as in equation (2.13) and its counterfactual
0,
ˆ
|1
ii
YD = can be estimated by
() 1 |
0
=
i i
D Y E
( )
() γ
γ φ
σ β
ε
i
i
i
Z
Z
X
Φ
+ =
0 0
(2.23)
Similarly in the second summation,
0,
ˆ
|0
ii
YD = can be estimated from equation
(2.17) and its counterfactuals
1,
ˆ
|0
ii
YD = can be estimated from
() 0 |
1
=
i i
D Y E
( )
() γ
γ φ
σ β
ε
i
i
i
Z
Z
X
Φ −
−
+ =
1
1 1
(2.24)
The derivation of equation (2.23) and (2.24) are explained in Appendix C.
Equation (2.13), (2.17), (2.23) and (2.24) define all conditional firm performances
in equation (2.22), therefore ATE can be estimated as follows
()()
( )
()
()()
()
()
⎪
⎪
⎭
⎪
⎪
⎬
⎫
⎪
⎪
⎩
⎪
⎪
⎨
⎧
⎥
⎦
⎤
⎢
⎣
⎡
Φ −
−
− + − +
⎥
⎦
⎤
⎢
⎣
⎡
Φ
− + −
=
∑
∑
=
−
0
1
1
0 1 0 1
1
0 1 0 1
ˆ 1
ˆ
ˆ ˆ
ˆ ˆ
ˆ
ˆ
ˆ ˆ
ˆ ˆ
1
N
i i
i
i
N
i i
i
i
Z
Z
X
Z
Z
X
N
ATE
γ
γ φ
σ σ β β
γ
γ φ
σ σ β β
ε ε
ε ε
(2.25)
By ways of obtaining ATE estimate in (2.25), we can approximate average
treatment effect on the treated (ATET) and average treatment effect on the control
(ATEC) by the following formula
40
()()
( )
()
⎥
⎦
⎤
⎢
⎣
⎡
Φ
− + − =
γ
γ φ
σ σ β β
ε ε
ˆ
ˆ
ˆ ˆ
ˆ ˆ
1
0 1 0 1
1 i
i
i
Z
Z
X
N
ATET (2.26)
()()
( )
()
⎥
⎦
⎤
⎢
⎣
⎡
Φ −
−
− + − =
γ
γ φ
σ σ β β
ε ε
ˆ 1
ˆ
ˆ ˆ
ˆ ˆ
1
0 1 0 1
0 i
i
i
Z
Z
X
N
ATEC (2.27)
Unlike in matching method, we assume that the parameters of the treated
and non-treated distributions are time-invariant, that is ()
′
ε ε
σ σ β β
0 1 0 1
, , , are
unchanged from 1997 to 2004. The estimation results of the average treatment
effect by switching regression method with parametric Heckman two-stage
estimations are reported in section 2.4.2.
Semiparametric approach
The key assumption in the previous section is that the error
terms ) , , (
0 1 i i i
u u ε are assumed to be jointly multivariate normal distributed with
mean vector zero and covariance matrix Σ , ( ) Σ , 0 ~ )' , , (
0 1
N u u
i i i
ε . If the joint
normality assumption doesn’t hold, the Heckman’s two-stage estimators are
inconsistent. Newey (1999) suggested a similar two-stage method but without
imposing any restriction on the distribution of the error term in selection equation,
i
ε . Without the normality assumption, we need to approximate the selection
correction terms in firm performance equation (2.11) and (2.16). Let’s reconsider
the selection correction terms as functions of γ
i
Z ,
[] ( ) γ λ γ ε
i i i i
Z Z u E
1 1
| = − > (2.28)
41
[] ( ) γ λ γ ε
i i i i
Z Z u E
0 0
| = − ≤ (2.29)
Using polynomial function as the basis function ( ) ⋅
1
λ and ( ) ⋅
0
λ as proposed by
Newey (1999), the selection correction term can be approximated as follows
() ()
∑
=
−
=
1
1
1
1 1
K
j
j
i j i
Z Z γ η γ λ ( () ()
∑
=
−
=
1
1
1
1 1
ˆ
ˆ
K
j
j
i j i
Z Z γ η γ λ ) (2.30)
() ( )
∑
=
−
=
0
1
1
0 0
K
j
j
i j i
Z Z γ η γ λ ( () ()
∑
=
−
=
0
1
1
0 0
ˆ
ˆ
K
j
j
i j i
Z Z γ η γ λ ) (2.31)
where
1
K and
0
K are arbitrary fixed numbers. The truncate means equation (2.11)
and (2.16) can be rewritten as
[ ] 0 |
*
1
>
i i
D Y E ()
∑
=
−
+ =
1
1
1
1 1
K
j
j
i j i
Z X γ η β (2.32)
[ ] 0 |
*
0
≤
i i
D Y E ()
∑
=
−
+ =
0
1
1
0 0
K
j
j
i j i
Z X γ η β (2.33)
which are equivalent to truncate mean equations (2.13) and (2.17) in the parametric
approach. The new performance equations to be estimated are derived from (2.32)
and (2.33) as
()
i
K
j
j
i j i i
w Z X Y
1
1
1
1 1 1
1
+ + =
∑
=
−
γ η β (2.34)
()
i
K
j
j
i j i i
w Z X Y
0
1
1
0 0 0
0
+ + =
∑
=
−
γ η β (2.35)
42
which are equivalent to new performance equations (2.18) and (2.19) in the
parametric approach. Similar to (2.20) and (2.21), the new error terms()
i i
w w
0 1
,
also satisfy the zero conditional mean and take the form:
()
∑
=
−
− =
1
1
1
1 1 1
K
j
j
i j i i
Z u w γ η , (2.36)
()
∑
=
−
− =
0
1
1
0 0 0
K
j
j
i j u i
Z u w γ η . (2.37)
In the first stage, Newey suggested the distribution-free methods such as maximum
score estimation or root-N consistent semiparametric estimation to estimate γ .
These methods impose no restriction on the distribution of
i
ε to avoid the
inconsistency associated with misspecifying distribution of
i
ε . In the second stage,
i i
Y D
1
are linearly regressed on
i i
X D and ( ) γ λ
i i
Z D
1
ˆ
, and similarly, ()
i i
Y D
0
1 − are
linearly regressed on ()
i i
X D − 1 and ( ) ( ) γ λ
i i
Z D
0
ˆ
1 − to obtain the estimation of
()
′
0 , 1 , 0 1
, , ,
j j
η η β β . These estimators are shown to be consistent and asymptotically
normally distributed. Hence, we can estimate the average treatment effect (ATE,
ATET and ATEC) of debt restructuring on various firm performances in the same
way they are estimated in matching approach and in the endogenous switching
regression approach with parametric method.
Similar to parametric method, we assume that the parameters of the treated
and non-treated distributions are time-invariant; that is, ( )
′
0 , 1 , 0 1
, , ,
j j
η η β β are
constant throughout our sample periods. The estimation results of the average
43
treatment effect by switching regression method with semiparametric Newey two-
stage estimations are reported in section 2.4.2.
2.4 Empirical Results
2.4.1 Estimation Results of Matching Approach
The estimation of the average treatment effects (ATE), the average
treatment effects on the treated (ATET), and the average treatment effects on the
control (ATEC) of debt restructuring on a firm’s performance using two matching
estimators: normal matching and propensity score matching, as illustrated in
section 2.3.1, are reported in Table 4. For each performance measure, the firm
characteristics used for matching each firm with its counterfactuals
i
X are also
shown on the table as matching variables.
Because we made no assumption that the impact of the debt restructuring
program is constant overtime, we estimate the effects annually between 1997
through 2004. Hence, the estimation results in Table 4 are derived from matching
processes that allow potential counterfactuals of each firm to come from other firms
in any industry during the same time period.
We found that the matching estimation results of the average treatment
effect on all firms are very similar to those only on the non-restructured firms, as
ATE estimates and the ATEC estimates are almost identical in terms of their
magnitudes as well as their significance level. This is due to the fact that there are
44
insufficient numbers of treated observations because only very small numbers of
firms are considered debt-restructured each year especially in the early years. For
this reason and also because the effects of debt restructuring may differ depends on
the group of firms considered, we only focus on the estimated effects of debt
restructuring on debt-restructured firms (ATET) and on non-restructured firms
(ATEC) below.
Excess stock market returns
Under normal matching, debt restructuring has no significant impact on the
excess stock market returns of debt-restructured firms in every year except 2003
where debt restructuring would have increased their excess stock market returns by
17 %. Debt restructuring would decrease the excess stock market returns of non-
restructured firms in every year by approximately 3 percent, except in 1997 and
2003 when their excess stock market returns increase by approximately 18 and 30
percent respectively.
The estimated average treatment effects of debt restructuring on the excess
stock market returns under propensity score matching are very similar to the
normal matching results. Only in year 2003 that we found positive effects of debt
restructuring on excess stock market returns regardless of matching basis and firms
considered.
45
Table 4. Estimated Average Treatment Effects (ATE)
1
Using Matching Method on: (1) Excess Stock Market Return;
(2) ROA1; (3) ROA2; (4) Capital Cost; (5) Debt-to-Assets Ratio; and (6) Coverage Ratio
Estimation of annual average effects of debt restructuring treatment on firm performances using two matching techniques
2,3
,
219 non-financial Thai firms, 1997-2004
YEAR 1997 1998 1999 2000 2001 2002 2003 2004
(1) log (Excess Stock Market Returns)
matching variables: Total Assets, Capital costs, Debt-to-Assets ratio, Coverage ratio
Normal Matching
ATE
Effect 0.18 -0.04 -0.03 -0.03 -0.01 0.00 0.29 -0.03
Z 21.98 -3.71 -1.81 -3.95 -1.16 0.01 13.91 -2.23
ATE on the Treated
Effect 0.17 -0.07 0.07 -0.02 -0.05 0.01 0.17 -0.02
Z 0.18 -0.16 0.28 -0.26 -1.29 0.17 2.91 -0.60
ATE on the control
Effect 0.18 -0.04 -0.03 -0.03 -0.01 0.00 0.32 -0.03
Z 44.23 -7.08 -4.19 -7.56 -0.99 -0.14 24.20 -3.83
Propensity Score Matching
ATE
Effect 0.18 -0.04 -0.06 -0.09 0.02 -0.03 0.29 -0.04
Z 17.71 -2.87 -3.53 -11.05 1.37 -1.73 11.39 -3.48
45
46
Table 4 (continued).
YEAR 1997 1998 1999 2000 2001 2002 2003 2004
ATE on the Treated
Effect 0.11 -0.10 0.03 0.01 0.01 0.04 0.17 -0.03
Z 0.10 -0.17 0.15 0.12 0.15 0.91 2.34 -1.07
ATE on the control
Effect 0.18 -0.04 -0.06 -0.09 0.02 -0.04 0.32 -0.05
Z 35.96 -5.88 -7.30 -22.10 2.54 -4.66 20.40 -5.67
(2) log (ROA1)
matching variables: Total Assets, Capital costs, Debt-to-Assets ratio, Coverage ratio
Normal Matching
ATE
Effect 0.05 -0.04 -0.01 0.02 0.03 0.04 0.02 0.02
Z 24.73 -17.07 -4.58 3.10 11.93 7.70 5.84 3.15
ATE on the Treated
Effect 0.05 -0.02 -0.01 0.01 0.02 0.06 0.03 0.00
Z 0.24 -0.20 -0.18 0.17 1.81 4.62 2.91 -0.11
ATE on the control
Effect 0.05 -0.04 -0.01 0.02 0.03 0.04 0.02 0.02
Z 42.97 -33.33 -8.87 4.64 25.13 9.38 8.47 5.76
46
47
Table 4 (continued).
YEAR 1997 1998 1999 2000 2001 2002 2003 2004
Propensity Score Matching
ATE
Effect 0.05 -0.05 -0.01 0.05 0.02 0.11 0.03 0.02
Z 14.87 -17.41 -2.68 8.14 6.27 12.80 10.16 2.43
ATE on the Treated
Effect 0.03 -0.04 -0.01 0.02 0.04 0.11 0.06 0.00
Z 0.07 -0.44 -0.13 0.46 3.40 4.14 6.74 -0.23
ATE on the control
Effect 0.05 -0.05 -0.01 0.05 0.02 0.11 0.03 0.02
Z 29.78 -34.35 -5.22 12.57 9.48 21.61 13.38 4.83
(3) log (ROA2)
matching variables: Total Assets, Capital costs, Debt-to-Assets ratio, Coverage ratio
Normal Matching
ATE
Effect 0.05 -0.04 -0.01 0.02 0.03 0.04 0.02 0.02
Z 25.91 -15.62 -4.64 5.44 13.99 9.00 6.11 3.61
ATE on the Treated
Effect 0.05 -0.02 -0.01 0.01 0.02 0.06 0.03 0.00
Z 0.24 -0.20 -0.16 0.41 2.77 5.09 3.03 0.06
47
48
Table 4 (continued).
YEAR 1997 1998 1999 2000 2001 2002 2003 2004
ATE on the control
Effect 0.05 -0.04 -0.01 0.02 0.04 0.04 0.02 0.02
Z 45.00 -30.57 -9.17 8.27 27.97 11.29 8.84 6.47
Propensity Score Matching
ATE
Effect 0.05 -0.04 -0.01 0.05 0.02 0.11 0.04 0.02
Z 16.34 -16.39 -3.86 12.61 8.33 15.97 10.54 2.97
ATE on the Treated
Effect 0.03 -0.05 -0.01 0.02 0.04 0.10 0.06 0.00
Z 0.08 -0.51 -0.18 0.73 4.73 4.74 6.79 0.02
ATE on the control
Effect 0.05 -0.04 -0.01 0.05 0.02 0.11 0.03 0.03
Z 32.81 -32.23 -7.48 20.45 12.30 27.08 14.09 5.66
(4) log (Capital Cost)
matching variables: Total Assets, Debt-to- Assets ratio, Coverage ratio
Normal Matching
ATE
Effect 0.00 0.01 -0.01 0.02 0.01 0.01 0.01 0.01
Z 0.72 2.81 -2.32 7.82 6.39 2.79 3.46 4.24
48
49
Table 4 (continued).
YEAR 1997 1998 1999 2000 2001 2002 2003 2004
ATE on the Treated
Effect -0.03 -0.01 0.00 0.02 0.01 0.01 0.01 0.00
Z -0.12 -0.08 0.03 1.50 1.68 1.43 1.81 -0.17
ATE on the control
Effect 0.00 0.01 -0.01 0.02 0.01 0.01 0.00 0.01
Z 1.63 5.91 -4.42 12.34 11.10 3.81 4.90 8.50
Propensity Score Matching
ATE
Effect 0.00 0.00 -0.01 0.02 0.01 0.01 0.01 0.00
Z 0.70 1.32 -3.63 4.63 4.67 4.23 2.67 2.45
ATE on the Treated
Effect -0.04 0.02 -0.02 0.01 0.02 0.00 0.00 0.00
Z -0.17 0.21 -0.32 0.36 2.84 0.70 0.92 -1.19
ATE on the control
Effect 0.00 0.00 -0.01 0.02 0.01 0.01 0.01 0.01
Z 1.56 2.48 -6.93 8.90 6.38 7.03 4.33 6.03
49
50
Table 4 (continued).
YEAR 1997 1998 1999 2000 2001 2002 2003 2004
(5) log (Debt-to-Assets Ratio)
matching variables: Total Assets, Capital costs, Coverage ratio
Normal Matching
ATE
Effect 0.04 0.05 0.09 0.11 0.20 0.15 0.16 0.13
Z 6.36 6.84 9.97 12.54 21.05 14.19 17.16 14.56
ATE on the Treated
Effect 0.05 0.03 0.07 0.06 0.17 0.15 0.10 0.11
Z 0.06 0.12 0.59 0.80 5.73 5.09 4.07 4.50
ATE on the control
Effect 0.04 0.05 0.09 0.11 0.20 0.15 0.17 0.14
Z 13.51 13.03 19.22 22.98 34.92 22.60 29.66 25.95
Propensity Score Matching
ATE
Effect 0.05 0.07 0.13 0.11 0.22 0.17 0.11 0.15
Z 5.28 6.91 11.08 8.81 18.12 16.26 9.96 12.44
ATE on the Treated
Effect 0.30 0.12 0.19 0.07 0.18 0.14 0.10 0.12
Z 0.32 0.36 1.21 0.66 4.90 4.64 3.45 3.80
ATE on the control
Effect 0.04 0.07 0.12 0.11 0.23 0.18 0.11 0.16
Z 10.11 13.30 20.50 16.89 30.86 27.55 15.92 21.14
50
51
Table 4 (continued).
YEAR 1997 1998 1999 2000 2001 2002 2003 2004
(6) log (Coverage Ratio)
matching variables: Total Assets, Capital costs, Debt-to-Assets ratio
Normal Matching
ATE
Effect -0.01 -0.04 -0.06 -0.10 -0.09 -0.09 -0.09 -0.12
Z -0.53 -2.37 -4.23 -3.85 -1.94 -3.21 -3.36 -3.00
ATE on the Treated
Effect 0.02 -0.02 0.00 -0.21 0.04 0.02 0.03 -0.09
Z 0.02 -0.03 -0.02 -1.03 0.30 0.25 0.54 -0.91
ATE on the control
Effect -0.01 -0.04 -0.06 -0.09 -0.12 -0.12 -0.12 -0.13
Z -0.90 -4.29 -8.08 -6.10 -4.30 -7.07 -6.40 -4.96
Propensity Score Matching
ATE
Effect -0.01 -0.04 -0.06 -0.08 -0.08 -0.09 -0.10 -0.12
Z -0.38 -1.44 -3.38 -2.31 -1.50 -2.73 -2.77 -4.32
ATE on the Treated
Effect 0.03 -0.01 -0.04 -0.21 0.05 0.04 0.08 -0.01
Z 0.01 -0.01 -0.17 -0.66 0.27 0.39 0.75 -0.22
ATE on the control
Effect -0.01 -0.04 -0.06 -0.08 -0.11 -0.12 -0.14 -0.15
Z -0.78 -3.11 -6.15 -3.97 -3.64 -5.60 -6.45 -7.52
51
52
Table 4 (continued).
Note:
1. ATE indicates Average Treatment Effects of all observations. Treated ATE indicates Average Treatment Effects on the treated observation only.
Control ATE indicates Average Treatment Effects on the control observation only.
2. For each observation, its unobserved counterfactual is imputed using other observations whose covariates are as close as the first nearest match but
with the opposite treatment (m=1 as in Abadie et al., 2001).
3. Normal matching technique uses vector norm between covariates to measure distance between observations. Propensity score matching use
propensity score to measure distance between observations.
52
53
Return on Assets
For non-restructured firms only, debt restructuring would decrease return on
assets only in 1998 and 1999 by approximately 4% and 1% respectively. Debt
restructuring would increase return on assets of non-restructured firms in all other
years. These increases would be strongly significant and range from 1 to 5 percent
on the basis of both matching methods with the exception of 11 percent increase in
2002 on the basis of propensity score matching.
Debt restructuring would have any significant impact on return on assets of
debt-restructured firms only during 2001 to 2003. Under normal matching, debt
restructuring increases ROA1 of debt-restructured firms by 6% and 3% in 2002 and
2003 respectively. Under propensity score matching, debt restructuring would
increase ROA1 of debt-restructured firms by 4%, 11% and 6% from 2001 to 2003
respectively. The estimated treatment effects of debt restructuring on ROA2 are
very similar to those on ROA1. We observed only significant positive treatment
effects on ROA2 of debt-restructured firms only in 2001 to 2003 and insignificant
in all other years. The treatment effects on ROA2 of non-restructured firms are
negative only during 1998 and 1999 and positive in all other years. The
magnitudes of the treatment effects on ROA2 are also similar to effects on ROA1.
Capital cost
Debt restructuring seems to have no impact on the capital cost of debt-
restructured firms, as the Average Treatment Effects on the Treated (ATET) are
54
statistically insignificant under both matching basis in all periods with one
exception in 2001 where debt restructuring would increase capital costs of debt-
restructured firms by 2% when matching based on the propensity score. Both
matching bases suggest that debt restructuring would slightly increase the capital
cost of non-restructured firms, as the Average Treatment Effects on the Control
(ATEC) are 1 to 2 percent in all periods except 1999 where the debt restructuring
would decrease capital cost of non-restructured firms by 1%.
Debt-to-assets ratio
The matching estimation results strongly suggest that debt restructuring
would greatly increase debt-to-assets ratio during later years. For all firms that
didn’t restructure their debts, debt-restructuring would have raised their debt-to-
assets ratio by small percentage less than 10% in earlier year before 2000 and large
percentage greater than 10% in later years. For debt-restructured firm, debt
restructuring would have no significant impact on their debt-to-assets ratio during
early period from 1997 to 2000. However, similar to non-restructured firms, debt
restructuring would significantly increase debt-to-assets ratio of debt-restructured
firms by greater than 10% during later period from 2001 to 2004. These increases
in debt-to-assets ratio of debt-restructured firms are lower than those of non-
restructured firms each year, as ATET estimates are about 5% lower than ATEC
estimates from 2001 to 2004. The magnitudes of estimation results of treatment
effect are similar in both matching methods.
55
Coverage ratio
Matching estimation results show strong evidences that debt restructuring
would not improve coverage ratio of debt-restructured firms but would
significantly lower coverage ratio of non-restructured firms, with large percentage
decreases in later periods. The estimated ATET of coverage ratio are all
insignificant and the estimated ATEC of coverage ratio are all negative and
statistically significant with decreases greater than 10% in later years from 2001 to
2004.
2.4.2 Estimation Results of Endogenous Switching Regression Approach
The parametric and semiparametric estimation of the Average Treatment
Effects (ATE), the Average Treatment Effects on the Treated (ATET) , and the
Average Treatment Effects on the Control (ATEC) of debt restructuring on a firm’s
performance measures using the endogenous switching regression model illustrated
in section 2.3.2 are reported in Table 5. These estimation results of ATE, ATET
and ATEC are derived from equations (2.25), (2.26) and (2.27), which in turn, were
based on the performance estimation results of equations (2.11) and (2.15) for
parametric method and equations (2.34) and (2.35) for semiparametric method.
The p-values of the hypothesis testing, whether the estimated average treatment
effects are significantly different from zero, are also reported.
Similar to matching estimation results, we found that the estimation results
of the Average Treatment Effect (ATE) on all firms are almost identical to those on
56
the non-restructured firms. This implies that the results of endogenous switching
regression approach also suffer from the same problem found in the results of
matching approach, where there are insufficient numbers of debt-restructured firms
in our estimation each year. For this reason, we only focus on the estimated effects
of debt restructuring on debt-restructured firms (ATET) and on non-restructured
firms (ATEC) because ATE estimates is practically the same as ATEC estimates.
Table 5. Estimated Average Treatment Effects (ATE) of Debt Restructuring
Treatment on Firm Performances using Endogenous Switching
Regression Techniques, 219 Non-financial Thai Firms, 1996-
2004
ATE ATE on the treated ATE on the control
Performance measures coefficient
p-
value
coefficient
p-
value
coefficient
p-
value
(1) Parametric Estimation
Excess Stock Market
Returns -0.019 0.361 -0.002 0.491 -0.021 0.340
ROA1 0.077 0.002 0.031 0.000 0.081 0.002
ROA2 0.070 0.001 0.028 0.000 0.074 0.001
Capital Costs -0.018 0.000 -0.014 0.000 -0.018 0.000
Debt-to-Asset Ratio -0.075 0.021 0.000 0.479 -0.082 0.018
Coverage Ratio 0.092 0.262 0.116 0.285 0.089 0.258
(2) Semiparametric estimation
Excess Stock Market
Returns 0.048 0.210 0.035 0.280 0.049 0.202
ROA1 0.072 0.002 0.035 0.000 0.076 0.002
ROA2 0.065 0.001 0.032 0.000 0.068 0.001
Capital Costs -0.001 0.128 -0.001 0.035 -0.001 0.136
Debt-to-Asset Ratio -0.073 0.013 0.017 0.011 -0.082 0.008
Coverage Ratio 0.145 0.200 0.066 0.338 0.154 0.187
57
Excess stock market returns
Endogenous switching regression approach suggests that debt restructuring
has no significant impact on excess stock market returns of both debt-restructured
and non-restructured firms. This is the case for both parametric and
semiparametric estimation methods.
Return on assets
Endogenous switching regression technique shows strong evidence that
debt structuring has positive impacts on the return on assets. The semi-parametric
estimation results of the average treatment effects on return on assets are almost
identical to the parametric estimation results.
Debt restructuring would increase both ROA1 and ROA2 of debt-
restructured firms by approximately 3%. For non-restructured firms, their ROA1
and ROA2 would increase by 8% and 7% respectively had they restructured their
debts. These results suggest that debt restructuring would benefit non-restructured
firms more than it would debt-restructured firms. All of the estimated treatment
effects on the return on assets are statistically significant at one-percent
significance level.
Capital costs
The estimation results by parametric method are different from those by
semi-parametric method. Under parametric estimation, non-restructured firms
would lower their capital costs by 1.8% while the debt-restructured firms would
58
lower their capital costs by 1.4% had they restructured their debts. Parametric
estimation results suggest that both non-restructured and restructured firms would
have both benefit from debt restructuring but the non-restructured would benefit
more. These results are strongly significant at one-percent significance level.
However, semi-parametric estimation results suggest that debt restructuring has no
significant impact on capital cost of non-restructured firms and only marginal
impact on capital cost of restructured firms as it is lowered by 0.1% at five-percent
significance level.
Debt-to-assets ratio
Under parametric method, non-restructured firms would lower their debt-to-
assets ratio by 8% had they restructured their debts but debt restructuring would
have no significant impact on the debt-to-assets ratio of debt-restructured firms.
Under semi-parametric method, non-restructured firms would lower their debt-to-
assets ratio by 8% at one-percent significant level had they restructured their debts
just as observed under parametric method but debt-restructured firms would
increase their debt-to-assets ratio by approximately 2% at five-percent significant
level. Using debt-to-assets ratio as performance measure, both parametric and
semi-parametric methods suggest that non-restructured firms would benefit from
debt restructuring program while restructured firms would not benefit under
parametric method and even worse off under semi-parametric method.
59
Coverage ratio
Debt restructuring has no significant impact on coverage ratio of both debt-
restructured and non-restructured firms. This is the case for both parametric and
semi-parametric estimation methods as the p-values indicate that estimated effects
of debt restructuring on the coverage ratio are insignificant at 10% significance
level.
Overall, both parametric and semi-parametric methods indicate that debt
restructuring improve firm’s ROA performance but has no significant impact on
firm’s stock performance and coverage ratio. Only parametric method shows that
debt restructuring would lower firm’s capital cost while semi-parametric method
see no links between firm’s capital cost and participation in debt restructuring
program. Only non-restructured firms would benefit from debt restructuring in
terms of debt-to-asset ratio performance. We also found that every significant
impact from this approach is small.
In matching approach, debt restructuring has no significant impact on most
of performance measures of debt-restructured firms, with the exception of
increasing their ROA and debt-to-assets ratio in later years during 2001 to 2004.
However, the impacts of debt structuring on all performance measures of non-
restructured firms would be statistically significant. The impacts of debt
restructuring on firm’s stock performance are mixed and varied on each year.
ROA, capital cost, and debt-to-assets ratio of non-restructured firms would increase
had these firms adopted a debt restructuring program, with small increase for
60
capital cost but large increase for debt-to-assets ratio. Coverage ratio of non-
restructured firms would decrease had these firms restructured their debts.
Estimation results of normal matching and propensity score matching are
very similar in terms of magnitude and significance level. Under matching
approach, we also found that the impacts on firm performances in later years during
2001 to 2004 are larger and more statistically significant than those during early
years before 2001. This may be due to the insufficient number of debt-restructured
firms in the early periods.
2.5 Conclusions
This paper has evaluated the success of Thailand’s corporate debt
restructuring program during the period of 1997 to 2004 in improving corporate
performance, as measured by stock performance, return on assets, capital cost,
debt-to-asset ratio and coverage ratio. Based on two approaches from the program
evaluation literature — the “matching” and the “endogenous switching regression”
approach, the estimated impacts of debt-restructuring on a firm’s performance are
found to be either insignificant or inconclusive depend on which of these two
approach is used. The estimated effects of debt-restructuring are found to vary
between structured and non-restructured firms for some performance measures.
The comparison between two approaches is as follows.
61
When using excess stock market return as measure of stock performance,
both matching and endogenous switching regression approaches suggest that debt
restructuring has no impact on stock performance of debt-restructured firms
because the estimated treatment effects on their excess stock market returns are
insignificant. The results of both approaches are quite different for non-
restructured firms. Under endogenous switching regression approach, debt
restructuring would still have no intended impacts on stock performance of non-
restructured firms but stock prices of non-restructured firms would generally
perform poorer than the Thailand stock market index (SET) had they adopted debt
restructuring under matching approach. Matching approach suggests that the only
year stock returns of non-restructured firms outperform SET index is 2003. This is
surprising because, with non-restructured firms’ already healthy standings, we
expected that a debt-restructuring program should have further strengthened their
positions in the stock market.
The endogenous switching regression approach shows strong evidence that
debt restructuring improve return on assets of both firm groups but non-restructured
firms would benefit more, with their increases about twice as large as those of
restructured firms. This result is expected for it is possible that the debt-
restructuring program has enabled these firms to utilize their assets more
efficiently. The magnitude may simply reflect non-restructured firms’ already
healthy standings and that the debt-restructuring program may have benefit more
from the program. The matching approach shows similar results in the sense that
62
debt restructuring improve return on assets of both firm groups, but only in later
years, with restructured firms benefit more than non-restructured firms during those
years. In particular, the effects of debt restructuring on the return on assets of non-
restructured firms was negative before 2000 and became positive after 2000. For
restructured firms, matching approach suggests that debt restructuring has no
impact on their return on assets pre-2000 but improved their return on assets post-
2000. It should be noted that these post-2000 positive impacts, as measured by
ATET, are close to the 3% increase predicted by the endogenous switching
regression approach.
The results with regard to capital cost are inconclusive at best because
matching and endogenous switching regression approaches give contradicting
estimation results of debt restructuring effects on firm’s capital cost. Matching
approach suggests that debt restructuring has no significant impact on the capital
costs of restructured firms while non-restructured firms are negatively affected as
their capital cost would be slightly increased by debt restructuring. These increases
on the capital cost of non-restructured firms are generally small, around 1-2%,
throughout the sample periods. The endogenous switching regression approach
with parametric estimation shows strong evidence that debt restructuring lowers the
capital cost of both firm groups with capital cost of non-restructured firms lowered
more than that of restructured firms. This is expected because firms that have not
been debt-restructured should be healthier than debt-restructured firms, and thus,
should be able to find a lower cost of capital had they participated in a restructuring
63
program. In the semi-parametric estimation results of the endogenous switching
regression approach, we found that debt restructuring has no significant impact on
the capital cost of non-restructured firms and a very small 0.1% decrease on capital
cost of restructured firms.
The estimation results of debt-restructuring impact on firm’s debt-to-assets
ratio are also inconclusive. Similar to capital cost, matching and switching
regression approaches give contradicting results in ATE estimation of debt
restructuring on firm’s debt-to-assets ratio. Under endogenous switching
regression approach, both parametric and semi-parametric methods show that non-
restructured firms would lower their debt-to-assets ratio had they adopted the debt
restructuring program. For debt-restructured firms, parametric method suggests
that debt restructuring has no significant impact on their debt-to-assets ratio but
semi-parametric method shows that their debt-to-assets ratio would be slightly
increased, approximately 2%, by the debt restructuring program. Matching
approach suggests that debt restructuring would increase debt-to-assets ratio of
both firm groups, with increases in debt-to-assets ratio of non-restructured firms
larger than those of non-restructured firms in every year. The increases are
particularly large (more than 10%) and highly significantly in later years. An
increase in debt-to-assets ratio, as a result of debt-restructuring, may reflect the
practice of the program itself, namely, taking over a firm’s assets or expanding the
total amount of debts (while relaxing repayment rules).
64
Endogenous switching regression approach suggests that debt restructuring
has no apparent effects on the firm’s coverage ratio. This is true only for the group
of debt-restructured firms when estimated by the matching method. Matching
method shows that non-restructured firms would lower its coverage ratio had they
decide to restructure its debt, with less than 10% decrease before 2001 and
approximately 13% decrease after. This result is unexpected because generally, a
decrease in coverage ratio is either due to a decrease in the firm’s profit, an increase
in interest expense, or both. A common practice in a debt-restructuring program
such as a reduction in interest expense leads to an increase in coverage ratio.
It is obvious that the endogenous switching regression approach often does
not agree with the two matching approach, which seem to agree with each other.
The disagreement between the two approaches may be rooted in how firm
characteristics are controlled. While the matching approach compute the treatment
effect by comparing each firm with its counterfactuals that are close in
characteristics in the same period, the endogenous switching regression approach
pool data from all firms in every period to estimate the coefficient of performance
equations in two states in order to calculate the treatment effects.
As the effects of firm’s characteristics on firm performance may differ from
period to period, the treatment effect computed by pooling data across all firms in
every period, which assumes that coefficients in performance equations are time-
invariant, may not be appropriate. In this respect, the matching method is more
preferable because it allows for different impacts on different periods. Moreover,
65
the matching methods seem to be more flexible as the criteria of finding
counterfactuals for each firm can be varied depending on the hypothesis or the
variables of interest.
Although the way that the matching estimators control for characteristics is
designed to mimic the experimental setting of a scientific experiment, it is limited
by the availability of observable characteristics that can be thought to affect the
probability of an observation being in the treatment. Beside the limitation on the
availability of observable characteristics, the validity of the matching estimators
also relies on the assumption that the treatment can be considered as random once
observable characteristics are controlled. Since it is possible that there may be
some corporate characteristics that this study has not accounted for, the above
results should be interpreted with caution. In this respect, the endogenous
switching regression approach is more preferable.
Finally, an interesting extension of this study is to replace annual data with
quarterly or monthly data. With annual data, the estimates reflect a long-term
effect of debt-restructuring on firm performance; with quarterly or monthly data, a
medium-term or a short-term effect may be explored. It is very possible that the
medium-term and the short-term treatment effects differ in both magnitudes and
signs from the long-term treatment effects. Such investigation would complete the
evaluation of Thailand’s debt restructuring program.
66
CHAPTER 3
THE STABILITY OF MACROECONOMIC VOLATILITY
UNDER DIFFERENT EXCHANGE RATE REGIMES
3.1 Introduction
It is well known that nominal exchange rate volatility increased
substantially after the collapse of the fixed exchange rate system during the Bretton
Woods era in 1971. Mussa (1986) showed that similar to the nominal exchange
rates, the volatility of real exchange rates is also substantial. A large collection of
literature has examined the consequences of moving from a fixed to a floating
exchange rate system on the behaviors of real and nominal macroeconomic
variables, including their volatility. Baxter and Stockman (1989) found
inconclusive evidence of relationships between exchange rate volatility and the
volatility of various economic variables, including output, consumption, trade
flows, and government consumption spending. Among 49 industrialized and
developing countries, the volatility of real macroeconomic variables, including
output and consumption, did not appear to change systematically under different
exchange rate systems.
Flood and Rose (1995) found that the volatility of macroeconomic variables
(particularly, interest rates, relative prices, money, reserves, and stock returns) did
67
not change significantly across fixed and floating exchange rate regimes for the
OECD countries. They interpreted this finding as evidence that there was no
tradeoff between the exchange rate stability and the stability of other
macroeconomic variables. Basu and Taylor (1999) confirmed these findings by
studying an earlier regime change from the gold standard to the interwar period
during 1914 to 1919 in addition to the abandonment of Bretton Woods in 1973.
Sopraseuth (1999) investigated the impact of the European Exchange Rate
Mechanism on macroeconomic fundamentals, and again, found that the volatility of
macroeconomic fundamentals were the same before and after the mechanism.
These results are surprising since one would expect that fluctuations in the
nominal exchange rate would change international relative prices, which should, in
turn, lead to changes in trade flows and output. For example, a depreciation of the
home currency should make foreign goods more expensive relative to domestic
goods, resulting in a fall in imports and a rise in output. Thus, macroeconomic
variables such as imports and output should fluctuate more under the floating
exchange rate regime. This stylized fact that exchange rate volatility differs
systematically across exchange rate regimes while the volatility of other
macroeconomic aggregates does not, is commonly known as “the exchange rate
disconnect puzzle”, as discussed by Obstfeld and Rogoff (2000).
Bélanger, Gutiérrez, Racette and Raynauld (1992) suggested that the reason
behind the finding of no significant shifts in the volatility of macroeconomic
variables under floating rates may be the use of aggregate data. However, their
68
results using sectoral trade volume data between the United States and Canada still
failed to discover higher volatility in trade volumes under the floating rate regime.
Ghosh, Gulde, Ostry and Wolf (1997) believed that the time series of
exchange rates used in previous studies may have caused the failure to establish a
link between the volatility of nominal exchange rates and that of macroeconomic
variables, especially inflation and output growth. Using a very comprehensive data
set of 140 countries over 30 years, from 1960 to 1990, and distinguishing between
the stated intention de jure and actual de facto regimes, they found higher volatility
of macroeconomic variables in the fixed regime than in the floating regime. In
particular, the results showed no evidence that output growth rates were different
across fixed and floating regimes. Levy-Yeyati and Strurzenegger (2002) also used
the new de facto classification of regimes to examine relationships between
exchange rate regimes and economic growth among 183 countries over the post-
Bretton Woods period from 1974 to 2000. They found that less flexible regimes
are associated with higher output volatility and slower growth only in developing
countries.
The objective of this paper is to examine whether the volatility of Japanese
macroeconomic variables is stable during the fixed regime and whether these
variables are stable during the floating exchange rate regime. Japan is chosen as
the subject of this study because it is one of the countries that have allowed its
exchange rate to float freely after the collapse of the Bretton Woods Agreement.
69
Its key economic aggregates during the period 1957 to 2002, which
encompassed the country’s fixed exchange rate regime and floating exchange rate
regime
6
, are analyzed. The hypothesis that there exists a structural break or
instability in levels and volatility of macroeconomic variables is tested over the
whole sample period, and separately under each exchange rate regime. The test
results are mixed, depending on whether the whole sample period is considered or
whether the regimes are considered separately.
The rest of Chapter 3 is organized as follows. Section 3.2 describes the data
set and the characteristics macroeconomic variables. The relationship between the
exchange rate volatility and the volatility of macroeconomic variables is
highlighted. Section 3.3 presents the econometric methodology used for stability
testing. The empirical results are reported in section 3.4. Section 3.5 concludes
this chapter
3.2 Data Description
Japan’s key macroeconomic aggregates, obtained from International
Financial Statistics (IFS) in the International Monetary Fund's main statistical
publication, are examined. Based on their relations with Yen nominal exchange
rate (¥NER), variables can be categorized into two groups: (i) NER-related
variables, including real exchange rate (RER), exports, imports, volumes of exports
6
Japan is under the Bretton Woods system up to the second quarter of 1971, and adopted
generalized floating exchange rates after 1971.
70
and imports, unit values of exports and imports, and net primary income from
abroad; and (ii) non NER-related variables, including industrial production, wages,
household consumption expenditure, government consumption expenditure, gross
fixed capital formation, consumption of fixed capital, and changes in inventories
All variables, with the exception of indices, are measured in real terms. The
only price index that has been available since the beginning of the data set is the
consumer price index (CPI), so each year’s CPI is used to transform nominal
variables of that particular year into real variables.
7
The real exchange rate is
computed as e ×P
us
/ P
j
where e is the Yen per U.S. Dollar market exchange rate,
and P
us
and P
j
are the consumer price indices for the U.S. and Japan, respectively.
The data are quarterly and seasonally adjusted, with the sample period starting in
the first quarter of 1957 and ending in the last quarter of 2002. According to
Japan’s history of exchange rate system classification, data up to the second quarter
of 1971 are considered to be under the fixed exchange rate regime, and data of
subsequent years to be under the floating exchange rate regime.
Because time series data are often non-stationary and stationary series are
desired for computations of meaningful statistics, filtering is needed for all
variables, except for change in inventories, which is by itself already filtered. To
filter the non-stationary variables, they are first differenced. The time series of
interest after first differencing are shown in Figure 1. The time series of interest
before transformation are shown in Figure C1 in the Appendix C.
7
Using other price indices yields similar results.
71
1960 1965 1970 1975 1980 1985 1990 1995 2000
-40
-30
-20
-10
0
10
20
30
CPI Indexed
Exchange rate(end of period)
1960 1965 1970 1975 1980 1985 1990 1995 2000
-30
-20
-10
0
10
20
CPI Indexed
Exchange rate(period avg.)
1960 1965 1970 1975 1980 1985 1990 1995 2000
-60
-40
-20
0
20
40
60
CPI Indexed
Exports of Goods and Services
1960 1965 1970 1975 1980 1985 1990 1995 2000
-60
-40
-20
0
20
40
60
CPI Indexed
Imports of Goods and Services
1960 1965 1970 1975 1980 1985 1990 1995 2000
-15
-10
-5
0
5
10
Index
Volume of Exports
1960 1965 1970 1975 1980 1985 1990 1995 2000
-10
-5
0
5
10
Index
Volume of Imports
1960 1965 1970 1975 1980 1985 1990 1995 2000
-15
-10
-5
0
5
10
15
Index
Unit Value of Exports
1960 1965 1970 1975 1980 1985 1990 1995 2000
-40
-20
0
20
40
Index
Unit Value of Imports
Figure 1. Japanese Series of Interest Series after First Differencing
Transformation
72
Figure 1 (continued).
1960 1965 1970 1975 1980 1985 1990 1995 2000
-6
-4
-2
0
2
4
Index
Industrial Production
1960 1965 1970 1975 1980 1985 1990 1995 2000
-8
-6
-4
-2
0
2
4
6
Index
Wages
1960 1965 1970 1975 1980 1985 1990 1995 2000
-4
-2
0
2
4
6
8
Index
Manufacturing Employment
1960 1965 1970 1975 1980 1985 1990 1995 2000
-300
-200
-100
0
100
200
CPI Indexed
GDP
1960 1965 1970 1975 1980 1985 1990 1995 2000
-200
-150
-100
-50
0
50
100
150
CPI Indexed
HH Cons.Expend.
1960 1965 1970 1975 1980 1985 1990 1995 2000
-50
0
50
100
150
CPI Indexed
Government Consumption Expend.
1960 1965 1970 1975 1980 1985 1990 1995 2000
-100
-50
0
50
100
CPI Indexed
Gross Fixed Capital Formation
1960 1965 1970 1975 1980 1985 1990 1995 2000
-30
-20
-10
0
10
20
30
CPI Indexed
Consumption of Fixed Capital
73
1960 1965 1970 1975 1980 1985 1990 1995 2000
-50
0
50
100
150
CPI Indexed
Changes in Inventories
1960 1965 1970 1975 1980 1985 1990 1995 2000
-30
-20
-10
0
10
20
CPI Indexed
Net Primary Income from Abroad
Figure 1 (continued).
Throughout this paper, standard deviation is used as a measure of volatility.
The conditional variances of the first differences of the series are examined to see
whether there have been any changes in the nature of volatility. The conditional
standard deviations are estimated using a GARCH (1,1) model and shown in Figure
2. The conditional standard deviations of all NER-related variables displayed a
distinct volatility break at the time of the exchange rate regime switch, with the
exception of unit value of exports and net income from aboard. The fact that the
volatility of NER-related variables was higher for a fairly extended period of time
after the regime change, and did not return to their original levels prior to the
regime change lends some weight to the structural shift interpretation. However, a
similar pattern of volatility break was also found in some of the non NER-related
variables, including industrial production, household consumption, government
spending and fixed capital formation.
74
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
5
10
15
20
25
30
Conditional SD: d(Exchange Rate(end of period))
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
5
10
15
20
25
30
35
Conditional SD: d(Exchange Rate(period average))
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
5
10
15
20
25
30
35
Conditional SD: d(Exports, Good and Services)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
5
10
15
20
25
Conditional SD: d(Imports, Good and Services)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
2
4
6
8
10
Conditional SD: d(Volume of Exports)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
1
2
3
4
5
6
Conditional SD: d(Volume of Imports )
1960 1965 1970 1975 1980 1985 1990 1995 2000
2
4
6
8
10
12
Conditional SD: d(Unit Value of Exports)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
5
10
15
20
25
Conditional SD: d(Unit Value of Imports)
Figure 2. Conditional Standard Deviation of Japanese Series of Interest
Estimated from GARCH(1,1)
75
Figure 2 (continued).
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
2
4
6
8
Conditional SD: d(Industrial Production)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
2
4
6
8
10
Conditional SD: d(Wage)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
2
4
6
8
10
Conditional SD: d(Employment)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
100
200
300
400
Conditional SD: d(GDP)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
50
100
150
200
250
Conditional SD: d(Consumption)
1960 1965 1970 1975 1980 1985 1990 1995 2000
10
20
30
40
50
60
70
Conditional SD: d(Government spending)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
2
4
6
8
10
12
Conditional SD: d(Fixed Capital)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
5
10
15
20
Conditional SD: d(Fixed Capital consumption)
76
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
20
40
60
80
100
Conditional SD: Change in Inventories
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
2
4
6
8
10
12
Conditional SD: d(Income from abroad)
Figure 2 (continued).
Next, the tradeoff between the RER volatility and the volatility of key
macroeconomic variables under two different exchange rate regimes are examined.
As shown in Figure 3, the estimated RER standard deviation is plotted against the
estimated standard deviations of the key macroeconomic variables from the same
period. If there is a tradeoff between the volatility of RER and that of a certain
variable, a negative correlation is expected in the corresponding graph. Figures 3-1
and 3-2 illustrate volatility tradeoffs during the fixed and floating exchange rate
regime respectively.
Figure 3-1 indicates that, under the fixed exchange rate regime, volatility of
all NER-related variables except for imports and their unit value are negatively
correlated to the RER volatility. However, none of their volatility is correlated to
the RER volatility under the floating exchange rate regime. The plots in Figure 3-1
show only weak evidence of negative correlation between the RER volatility and
volatility of some non-NER-related variables, including industrial production,
77
GDP, and wages, during the time of fixed exchange rate. These correlations are not
found under the floating regime.
The volatility measures of fixed capital formation and net income from
abroad are negatively correlated to the RER volatility throughout our sample
period, but their correlations are more pronounced in the fixed regime. The weak
negative correlation between the volatility of fixed capital consumption and the
RER volatility is present in the fixed regime but disappears during periods of
floating exchange rate. The volatility measures of employment, household
consumption, and government spending, are not correlated with the RER volatility
regardless of the exchange rate regime.
3 4 5 6 7 8 9
4.5
5
5.5
6
6.5
Conditional SD: d(Exports)
Conditional SD: d(Exchange Rate*)
Fixed Regime
4 6 8 10 12 14 16
4.5
5
5.5
6
6.5
Conditional SD: d(Imports)
Conditional SD: d(Exchange Rate*)
Fixed Regime
0 0.5 1 1.5 2 2.5
4.5
5
5.5
6
6.5
Conditional SD: d(Volume of Exports)
Conditional SD: d(Exchange Rate*)
Fixed Regime
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
4.5
5
5.5
6
6.5
Conditional SD: d(Volume of Imports )
Conditional SD: d(Exchange Rate*)
Fixed Regime
Figure 3-1. Tradeoff of Real Exchange Rate Volatility and Volatility of Other
Variables of Interest under Fixed Exchange Rate Regime
78
Figure 3-1 (continued).
1 1.5 2 2.5 3 3.5
4.5
5
5.5
6
6.5
Conditional SD: d(Unit Value of Exports)
Conditional SD: d(Exchange Rate*)
Fixed Regime
1 2 3 4 5 6 7 8
4.5
5
5.5
6
6.5
Conditional SD: d(Unit Value of Imports)
Conditional SD: d(Exchange Rate*)
Fixed Regime
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
4.5
5
5.5
6
6.5
Conditional SD: d(Industrial Production)
Conditional SD: d(Exchange Rate*)
Fixed Regime
0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8
4.5
5
5.5
6
6.5
Conditional SD: d(Wage)
Conditional SD: d(Exchange Rate*)
Fixed Regime
0 0.5 1 1.5 2 2.5 3 3.5
4.5
5
5.5
6
6.5
Conditional SD: d(Employment)
Conditional SD: d(Exchange Rate*)
Fixed Regime
10 20 30 40 50 60
4.5
5
5.5
6
6.5
Conditional SD: d(GDP)
Conditional SD: d(Exchange Rate*)
Fixed Regime
15 20 25 30 35 40 45
4.5
5
5.5
6
6.5
Conditional SD: d(Consumption)
Conditional SD: d(Exchange Rate*)
Fixed Regime
5 6 7 8 9 10 11 12
4.5
5
5.5
6
6.5
Conditional SD: d(Government spending)
Conditional SD: d(Exchange Rate*)
Fixed Regime
79
Figure 3-1 (continued).
5 10 15 20 25 30
4.5
5
5.5
6
6.5
Conditional SD: d(Fixed Capital)
Conditional SD: d(Exchange Rate*)
Fixed Regime
0.5 1 1.5 2 2.5 3 3.5
4.5
5
5.5
6
6.5
Conditional SD: d(Fixed Capital consumption)
Conditional SD: d(Exchange Rate*)
Fixed Regime
5 10 15 20 25 30 35
4.5
5
5.5
6
6.5
Conditional SD: Change in Inventories
Conditional SD: d(Exchange Rate*)
Fixed Regime
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3
4.5
5
5.5
6
6.5
Conditional SD: d(Income from abroad)
Conditional SD: d(Exchange Rate*)
Fixed Regime
80
5 10 15 20 25 30
6
8
10
12
14
16
Conditional SD: d(Exports)
Conditional SD: d(Exchange Rate*)
Flexible regime
0 10 20 30 40 50 60
6
8
10
12
14
16
Conditional SD: d(Imports)
Conditional SD: d(Exchange Rate*)
Flexible regime
2 3 4 5 6 7
6
8
10
12
14
16
Conditional SD: d(Volume of Exports)
Conditional SD: d(Exchange Rate*)
Flexible regime
1 1.5 2 2.5 3 3.5 4 4.5 5
6
8
10
12
14
16
Conditional SD: d(Volume of Imports )
Conditional SD: d(Exchange Rate*)
Flexible regime
2 3 4 5 6 7 8
6
8
10
12
14
16
Conditional SD: d(Unit Value of Exports)
Conditional SD: d(Exchange Rate*)
Flexible regime
0 10 20 30 40 50
6
8
10
12
14
16
Conditional SD: d(Unit Value of Imports)
Conditional SD: d(Exchange Rate*)
Flexible regime
0.5 1 1.5 2 2.5 3 3.5 4 4.5
6
8
10
12
14
16
Conditional SD: d(Industrial Production)
Conditional SD: d(Exchange Rate*)
Flexible regime
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3
6
8
10
12
14
16
Conditional SD: d(Wage)
Conditional SD: d(Exchange Rate*)
Flexible regime
Figure 3-2. Tradeoff of Real Exchange Rate Volatility and Volatility of Other
Variables of Interest under Floating Exchange Rate Regime
81
Figure 3-2. (continued).
0 2 4 6 8 10
6
8
10
12
14
16
Conditional SD: d(Employment)
Conditional SD: d(Exchange Rate*)
Flexible regime
20 40 60 80 100 120
6
8
10
12
14
16
Conditional SD: d(GDP)
Conditional SD: d(Exchange Rate*)
Flexible regime
0 50 100 150 200 250
6
8
10
12
14
16
Conditional SD: d(Consumption)
Conditional SD: d(Exchange Rate*)
Flexible regime
0 10 20 30 40 50
6
8
10
12
14
16
Conditional SD: d(Government spending)
Conditional SD: d(Exchange Rate*)
Flexible regime
10 20 30 40 50 60
6
8
10
12
14
16
Conditional SD: d(Fixed Capital)
Conditional SD: d(Exchange Rate*)
Flexible regime
0 5 10 15 20
6
8
10
12
14
16
Conditional SD: d(Fixed Capital consumption)
Conditional SD: d(Exchange Rate*)
Flexible regime
0 20 40 60 80 100
6
8
10
12
14
16
Conditional SD: Change in Inventories
Conditional SD: d(Exchange Rate*)
Flexible regime
0 2 4 6 8 10 12
6
8
10
12
14
16
Conditional SD: d(Income from abroad)
Conditional SD: d(Exchange Rate*)
Flexible regime
82
3.3 Methodology
A test for structural breaks in volatility of the first difference of Japan’s real
aggregate series, denoted by
t
y & , will be conducted. Following McConnell and
Perez-Quiros (2000), each series is modeled with an AR specification:
t t t
y y ε φ μ + + =
−1
& & T t ,...., 1 = (3.1)
where T denotes the number of observation. Assuming that
t
ε is normally
distributed, we model the standard deviation of
t
ε with its unbiased estimator,
t
ε
π
ˆ
2
, as:
t t
u + = α ε
π
ˆ
2
(3.2)
where α is the estimator of the standard deviation.
The test on the whole sample period can proceed as follows. First,
statistical significance of the coefficient in Equation (3.1) α would, to some extent,
validate the AR specification as an appropriate for modeling the corresponding real
aggregate series, so the Likelihood Ratio Test (LR) is used to establish such claim.
Second, we test the hypothesis that the AR parameters in Equation (3.1) vary with
the exchange regime. Let
t
D is the dummy variable that indicates whether a period
is before and after the period of break point
BP
τ :
⎩
⎨
⎧
=
1
0
1t
D
for
for
BP
BP
t
t
τ
τ
>
≤
83
⎩
⎨
⎧
=
0
1
2t
D
for
for
BP
BP
t
t
τ
τ
>
≤
If the break point
BP
τ is known, Equation (3.1) becomes:
( ) ( )
t t t t t t
D y D y y ε φ μ φ μ + + + + =
− − 2 1 2 2 1 1 1 1
& & & . (3.1’)
A simple Chow test for the stability of the AR parameter in Equation (3.1) can then
be performed on this equation under the null hypothesis:
2 1 0
: μ μ = H and
2 1
φ φ = .
By testing for the stability of parameter α in Equation (3.2), whether a volatility
break exists among the real aggregate variables can be tested, using the same
technique as in the previous step. Equation (3.2) becomes:
t t t t
u D D + + =
2 2 1 1
ˆ
2
α α ε
π
. (3.2’)
This test is performed under two separate specifications regarding the nature of the
AR parameters. In the first specification, the AR parameters are assumed to be
time-invariant, as in Equation (3.1); in the second specification, the possibility of
different AR coefficients under the two exchange rate regimes is allowed, as in
Equation (3.1’). Formally, the following two null hypotheses are tested in the first
and the second specifications, respectively.
2 1 0
: α α = H assuming
2 1 2 1
φ φ μ μ = = and
2 1 0
: α α = H (no restriction on
2 1 2 1
φ φ μ μ = = and )
84
However, we have no prior information regarding the break point
BP
τ of
conditional means and volatility of the aggregate variables. To circumvent this
problem, techniques developed by Andrews (1993) and Andrews and Ploberger
(1994) are used to test for the parameter instability and the structural break without
knowing the time of the break point. In these techniques, all period that might
contain possible break point, given that the earliest and the latest periods that the
break can occur are respectively
1
τ and
2
τ , are
2 2
, 1 ,...., 1 , τ τ τ τ τ − + =
v v v
. As in
Andrews (1993) and Andrews and Ploberger (1994), the time periods
1
τ and
2
τ are
generally set as 0.15T and 0.85T, where T denotes the total number of time periods.
All of the statistics used in these techniques are based on the point-wise
statistics ( )
v T
F τ , which can be a Likelihood Ratio (LR), a Lagrange Multiplier
(LM), or a Wald (W) statistic. More specifically, the statistics considered in
Andrews (1993) is the supreme statistics:
()
v T
F SupF
v
τ
τ τ τ
2 1
sup
≤ ≤
= (3.3)
The statistics considered in Andrews and Ploberger (1994) are the average and
exponential statistics given by
()
∑
=
− −
=
2
1
1
1
1 2
τ
τ τ
τ
τ τ
v
v T
F AvgF (3.4)
()
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
⎟
⎠
⎞
⎜
⎝
⎛
− −
=
∑
=
2
1
2
1
exp
1
1
ln
1 2
τ
τ τ
τ
τ τ
v
v T
F ExpF (3.5)
85
3.4 Empirical Results
All three test statistics and their p-values associated with NER-related
variables are reported in Table 6. When the break point is unknown, the simple
Chow structural break test cannot be used, and a new statistical distribution must be
simulated for each of the three statistics. Based on these distributions, a special p-
value, namely p-value*, is reported. Also reported are break points suggested by
supF statistics. Under the assumption that the break point is known, the break
point suggested by the supF statistic provides a basis for the simple Chow
structural break test, from which a conventional p-value are obtained. The test
statistics, the special p-values, and the conventional p-values associated with non-
NER-related variables are reported in Table 7.
For each variable, a structural break test is conducted for the conditional
mean. If a structural break in conditional mean cannot be detected, then it can be
said that the AR parameters do not vary over time, as in Equation (3.1); thus, more
attention should be paid to ‘volatility1’, the measure of volatility that assumes
accordingly. On the other hand, if a structural break can be detected in the
conditional mean, then it can be said that the AR parameters vary over time, as in
Equation (3.1’); thus, more attention should be paid to ‘volatility2’ measures,
which are computed when no restriction is imposed on the AR parameters.
86
Table 6. Structural Break Tests for NER-related Variables
Structural Break Test: AR1 Coefficient Structural Break Test: Volatility1 Structural Break Test: Volatility2
1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2
fixed ER Floated ER fixed ER Floated ER fixed ER Floated ER
(1) Exchange rate(end of period)
Break point 1986.5 1965.25 1978.5 1971.25 1963 1977.5 1971.25 1969.25 1978.5
supF 7.38 8.78 7.68 52.86 2.59 3.27 52.70 7.84 5.12
p-value 0.00 0.00 0.00 0.00 0.08 0.04 0.00 0.00 0.01
p-value* 0.42 0.26 0.38 0.00 1.00 1.00 0.00 0.36 0.73
expF 2.00 2.96 1.94 22.04 0.34 0.45 21.98 1.36 0.80
p-value* 0.40 0.16 0.42 0.00 1.00 1.00 0.00 0.66 0.90
avgF 1.14 1.59 1.14 2.89 -0.58 -0.34 2.88 0.55 0.07
p-value* 0.71 0.43 0.73 0.00 1.00 1.00 0.00 0.90 1.00
(2) Exchange rate (period avg.)
Break point 1986.5 1965.25 1986.5 1971.25 1968.25 1989.25 1971.25 1969.25 1990.75
supF 11.08 11.57 9.94 43.57 0.77 3.49 43.59 4.84 5.62
p-value 0.00 0.00 0.00 0.00 0.47 0.03 0.00 0.01 0.00
p-value* 0.11 0.09 0.18 0.00 1.00 1.00 0.00 0.77 0.66
expF 2.99 4.01 3.10 17.81 0.11 0.72 17.80 0.70 1.23
p-value* 0.16 0.05 0.14 0.00 1.00 1.00 0.00 1.00 0.72
avgF 1.32 1.86 1.61 2.76 -1.54 0.13 2.70 -0.02 0.53
p-value* 0.42 0.22 0.39 0.00 1.00 1.00 0.00 1.00 1.00
Notes:
1/ Volatility1 is calculated assuming parameters do not vary with time.
2/Volatility2 is calculated assuming parameters may vary with time.
3/ P-value* is calculated using Andrews methodology
86
87
Table 6 (continued).
Structural Break Test: AR1 Coefficient Structural Break Test: Volatility1 Structural Break Test: Volatility2
1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2
fixed ER Floated ER fixed ER Floated ER fixed ER Floated ER
(3) Exports
Break point 1986.5 1967.75 1986.5 1965.5 1965.5 1982.25 1964.5 1969.25 1989
supF 6.84 47.79 7.04 92.31 6.26 3.31 94.89 14.54 3.16
p-value 0.00 0.00 0.00 0.00 0.00 0.04 0.00 0.00 0.05
p-value* 0.50 0.00 0.47 0.00 0.57 1.00 0.00 0.02 1.00
expF 1.37 20.18 1.31 42.29 1.93 0.69 43.78 3.93 0.62
p-value* 0.66 0.00 0.68 0.00 0.42 1.00 0.00 0.06 1.00
avgF 0.76 2.41 0.75 3.34 1.11 0.10 3.37 1.23 0.01
p-value* 0.90 0.00 1.00 0.00 0.73 1.00 0.00 0.23 1.00
(4) Imports
Break point 1986.5 1965.75 1986.5 1969.25 1962 1977.5 1969.25 1969.25 1977.5
supF 12.99 7.54 12.87 88.29 6.70 5.07 90.63 5.60 5.98
p-value 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.01 0.00
p-value* 0.05 0.40 0.05 0.00 0.51 0.74 0.00 0.67 0.61
expF 2.71 2.30 3.11 40.75 1.25 0.65 41.68 0.83 0.85
p-value* 0.20 0.30 0.14 0.00 0.71 1.00 0.00 0.89 0.88
avgF 1.09 1.36 1.40 3.56 0.55 -0.16 3.59 0.13 0.06
p-value* 0.50 0.62 0.39 0.00 1.00 1.00 0.00 1.00 1.00
Notes:
1/Volatility1 is calculated assuming parameters do not vary with time.
2/Volatility2 is calculated assuming parameters may vary with time.
3/ P-value* is calculated using Andrews methodology
87
88
Table 6 (continued).
Structural Break Test: AR1 Coefficient Structural Break Test: Volatility1 Structural Break Test: Volatility2
1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2
fixed ER Floated ER fixed ER Floated ER fixed ER Floated ER
(5) Volume of Exports
Break point 1988.25 1967.25 1988.25 1968 1968 1978.75 1968 1968.75 1978.75
supF 10.30 12.01 11.26 130.24 66.05 13.54 136.53 31.10 14.03
p-value 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
p-value* 0.15 0.08 0.10 0.00 0.00 0.04 0.00 0.00 0.03
expF 2.64 4.35 3.40 60.74 29.59 4.11 63.64 12.14 4.29
p-value* 0.22 0.04 0.10 0.00 0.00 0.05 0.00 0.00 0.04
avgF 1.06 1.92 1.41 3.94 3.08 1.68 3.91 2.49 1.70
p-value* 0.52 0.17 0.33 0.00 0.00 0.20 0.00 0.00 0.18
(6) Volume of Imports
Break point 1985.5 1969.25 1985.5 1970.75 1964.25 1994 1970.75 1962 1990.5
supF 5.66 26.71 5.11 76.07 7.59 18.09 77.56 11.18 14.92
p-value 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00
p-value* 0.66 0.00 0.74 0.00 0.39 0.00 0.00 0.11 0.02
expF 1.47 10.01 1.07 33.96 1.53 6.77 35.37 3.04 6.01
p-value* 0.61 0.00 0.78 0.00 0.58 0.00 0.00 0.15 0.00
avgF 0.90 2.15 0.49 3.64 0.83 2.45 3.63 1.03 2.36
p-value* 0.86 0.00 1.00 0.00 0.85 0.03 0.00 0.41 0.05
Notes:
1/Volatility1 is calculated assuming parameters do not vary with time.
2/Volatility2 is calculated assuming parameters may vary with time.
3/ P-value* is calculated using Andrews methodology
88
89
Table 6. (continued).
Structural Break Test: AR1 Coefficient Structural Break Test: Volatility1 Structural Break Test: Volatility2
1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2
fixed ER Floated ER fixed ER Floated ER fixed ER Floated ER
(7) Unit Value of Exports
Break point 1986.5 1968.25 1986.5 1967.75 1967 1978 1967 1968 1978
supF 9.09 15.51 9.85 68.86 1.91 3.19 75.65 2.41 4.36
p-value 0.00 0.00 0.00 0.00 0.16 0.04 0.00 0.10 0.01
p-value* 0.23 0.02 0.18 0.00 1.00 1.00 0.00 1.00 0.83
expF 2.50 5.12 2.94 30.49 0.27 0.37 33.94 0.42 0.43
p-value* 0.25 0.02 0.16 0.00 1.00 1.00 0.00 1.00 1.00
avgF 1.38 1.15 1.58 3.33 -0.72 -0.48 3.29 -0.26 -0.47
p-value* 0.56 0.10 0.44 0.00 1.00 1.00 0.00 1.00 1.00
(8) Unit Value of Imports
Break point 1986.5 1959.25 1986.5 1974 1959.5 1991 1971.5 1959.25 1991.25
supF 11.68 15.07 11.36 86.83 11.07 17.05 79.78 21.41 22.47
p-value 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
p-value* 0.09 0.02 0.10 0.00 0.11 0.00 0.00 0.00 0.00
expF 2.64 4.68 3.14 38.75 3.08 6.17 36.93 7.23 8.77
p-value* 0.22 0.03 0.14 0.00 0.14 0.00 0.00 0.00 0.00
avgF 1.09 1.96 1.60 3.55 1.40 1.97 3.54 1.79 2.37
p-value* 0.52 0.14 0.38 0.00 0.40 0.05 0.00 0.02 0.00
Notes:
1/Volatility1 is calculated assuming parameters do not vary with time.
2/Volatility2 is calculated assuming parameters may vary with time.
3/ P-value* is calculated using Andrews methodology.
89
90
Table 7. Structural Break Tests for Non-NER-related Variables
Structural Break Test: AR1 Coefficient Structural Break Test: Volatility1 Structural Break Test: Volatility2
1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2
fixed ER Floated ER fixed ER Floated ER fixed ER Floated ER
(9)Industrial Production
Break point 1975 1962.75 1990.75 1966 1964.75 1991.5 1965.75 1962 1991.5
supF 7.72 3.50 4.23 90.64 12.78 21.33 93.45 12.06 22.16
p-value 0.00 0.04 0.02 0.00 0.00 0.00 0.00 0.00 0.00
p-value* 0.37 1.00 0.85 0.00 0.05 0.00 0.00 0.07 0.00
expF 1.41 0.71 0.94 41.69 4.67 7.99 43.28 4.19 8.30
p-value* 0.64 1.00 0.84 0.00 0.03 0.00 0.00 0.05 0.00
avgF 0.86 0.19 0.45 3.79 2.04 2.36 3.84 1.93 2.40
p-value* 0.88 1.00 1.00 0.00 0.14 0.01 0.00 0.19 0.01
(10) Wages
Break point 1974.5 1961 1982.75 1972 1968.75 1976.5 1964 1959.75 1997.5
supF 37.54 82.40 26.53 25.11 6.51 2.00 76.15 47.47 6.22
p-value 0.00 0.00 0.00 0.00 0.00 0.14 0.00 0.00 0.00
p-value* 0.00 0.00 0.00 0.00 0.54 1.00 0.00 0.00 0.58
expF 14.25 37.60 11.56 10.05 1.36 0.38 34.10 21.31 1.13
p-value* 0.00 0.00 0.00 0.00 0.66 1.00 0.00 0.00 0.76
avgF 2.94 3.26 2.98 2.14 0.64 -0.39 3.03 2.92 0.07
p-value* 0.00 0.00 0.00 0.00 0.90 1.00 0.00 0.00 1.00
Notes:
1/Volatility1 is calculated assuming parameters do not vary with time.
2/Volatility2 is calculated assuming parameters may vary with time.
3/ P-value* is calculated using Andrews methodology
90
91
Table 7 (continued).
Structural Break Test: AR1 Coefficient Structural Break Test: Volatility1 Structural Break Test: Volatility2
1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2
fixed ER Floated ER fixed ER Floated ER fixed ER Floated ER
(11) Employment
Break point 1991.5 1962.25 1992.25 1964 1961 1983.75 1988.75 1959.25 1988.25
supF 330.90 3.73 41.87 4.15 12.24 5.39 16.26 12.55 18.74
p-value 0.00 0.03 0.00 0.02 0.00 0.01 0.00 0.00 0.00
p-value* 0.00 1.00 0.00 0.86 0.07 0.70 0.01 0.06 0.00
expF 160.61 0.85 18.16 0.49 4.14 1.58 5.64 4.20 7.01
p-value* 0.00 0.88 0.00 1.00 0.05 0.56 0.01 0.05 0.00
avgF 4.23 0.42 2.81 -0.25 1.63 0.94 1.88 1.74 1.81
p-value* 0.00 1.00 0.00 1.00 0.20 0.83 0.07 0.19 0.02
(12)GDP
Break point 1992 1959.25 1992 1965 1965.5 1986.75 1965.25 1959.25 1986.75
supF 16.90 13.63 15.67 41.36 9.13 2.08 44.38 21.51 2.92
p-value 0.00 0.00 0.00 0.00 0.00 0.13 0.00 0.00 0.06
p-value* 0.01 0.04 0.02 0.00 0.23 1.00 0.00 0.00 1.00
expF 5.55 4.48 5.43 17.32 3.04 0.32 18.66 7.46 0.49
p-value* 0.01 0.03 0.02 0.00 0.15 1.00 0.00 0.00 1.00
avgF 1.51 1.90 1.65 2.19 1.59 -0.64 2.30 2.10 -0.22
p-value* 0.07 0.16 0.08 0.00 0.41 1.00 0.00 0.02 1.00
Notes:
1/Volatility1 is calculated assuming parameters do not vary with time.
2/Volatility2 is calculated assuming parameters may vary with time.
3/ P-value* is calculated using Andrews methodology
91
92
Table 7 (continued).
Structural Break Test: AR1 Coefficient Structural Break Test: Volatility1 Structural Break Test: Volatility2
1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2
fixed ER Floated ER fixed ER Floated ER fixed ER Floated ER
(13) Household Consumption Expenditure
Break point 1973 1963 1997 1964 1967 1977.75 1964 1959.25 1977.75
supF 5.19 18.33 5.17 65.73 6.49 8.75 69.62 8.50 13.37
p-value 0.01 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00
p-value* 0.73 0.00 0.73 0.00 0.54 0.26 0.00 0.28 0.04
expF 1.19 6.84 1.35 28.80 2.00 1.68 30.30 2.68 3.56
p-value* 0.73 0.00 0.66 0.00 0.39 0.52 0.00 0.21 0.09
avgF 0.77 2.29 0.91 3.20 1.18 0.79 3.38 1.46 1.38
p-value* 1.00 0.03 1.00 0.00 0.71 0.80 0.00 0.51 0.30
(14) Government Spending Expenditure
Break point 1965.5 1959.75 1980 1973.5 1966.25 1981 1964 1959.25 1980.75
supF 24.96 18.29 7.28 13.22 8.66 4.26 40.84 13.80 5.81
p-value 0.00 0.00 0.00 0.00 0.00 0.02 0.00 0.00 0.00
p-value* 0.00 0.00 0.43 0.05 0.27 0.84 0.00 0.04 0.64
expF 9.83 6.56 1.32 4.34 2.92 1.02 17.35 4.28 1.33
p-value* 0.00 0.00 0.68 0.04 0.17 0.81 0.00 0.04 0.67
avgF 2.10 2.10 0.16 1.37 1.60 0.45 2.50 1.85 0.61
p-value* 0.00 0.04 1.00 0.17 0.44 1.00 0.00 0.18 1.00
Notes:
1/Volatility1 is calculated assuming parameters do not vary with time.
2/Volatility2 is calculated assuming parameters may vary with time.
3/ P-value* is calculated using Andrews methodology
92
93
Table 7 (continued).
Structural Break Test: AR1 Coefficient Structural Break Test: Volatility1 Structural Break Test: Volatility2
1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2
fixed ER Floated ER fixed ER Floated ER fixed ER Floated ER
(15) Fixed Capital Formation
Break point 1991 1959.25 1991 1966 1964.5 1987.25 1966 1959.5 1987.25
supF 16.87 9.54 13.06 59.71 9.20 6.75 63.05 26.30 6.15
p-value 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
p-value* 0.01 0.20 0.05 0.00 0.23 0.51 0.00 0.00 0.59
expF 5.27 2.06 4.04 25.87 3.48 1.19 27.68 10.18 1.06
p-value* 0.02 0.38 0.05 0.00 0.09 0.73 0.00 0.00 0.79
avgF 1.60 0.96 1.39 2.69 1.86 0.37 2.67 2.43 0.28
p-value* 0.09 0.69 0.22 0.00 0.31 1.00 0.00 0.00 1.00
(16) Consumption of Fixed Capital
Break point 1964.5 1960.5 1976.75 1971.75 1964 1981.75 1964 1960 1981.75
supF 41.57 36.75 7.28 71.86 9.71 36.22 105.53 53.96 47.22
p-value 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
p-value* 0.00 0.00 0.43 0.00 0.19 0.00 0.00 0.00 0.00
expF 16.64 15.47 1.94 32.65 2.34 15.83 48.44 23.51 20.61
p-value* 0.00 0.00 0.42 0.00 0.29 0.00 0.00 0.00 0.00
avgF 1.51 2.96 1.05 4.00 1.08 3.21 4.19 2.88 3.43
p-value* 0.00 0.00 0.73 0.00 0.61 0.00 0.00 0.00 0.00
Notes:
1/Volatility1 is calculated assuming parameters do not vary with time.
2/Volatility2 is calculated assuming parameters may vary with time.
3/ P-value* is calculated using Andrews methodology
93
94
Table 7 (continued).
Structural Break Test: AR1 Coefficient Structural Break Test: Volatility1 Structural Break Test: Volatility2
1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2 1957-2002 pre1971:2 post1971:2
fixed ER Floated ER fixed ER Floated ER fixed ER Floated ER
(17) Change in Inventories
Break point 1973.25 1959.25 1997.75 1974.75 1965.5 1976.25 1974.75 1966 1976.25
supF 10.34 16.62 11.62 9.03 4.35 8.22 9.31 4.99 12.50
p-value 0.00 0.00 0.00 0.00 0.02 0.00 0.00 0.01 0.00
p-value* 0.15 0.01 0.09 0.24 0.83 0.31 0.22 0.75 0.06
expF 3.16 5.64 3.12 2.20 1.16 1.73 2.41 1.31 3.03
p-value* 0.13 0.01 0.14 0.33 0.75 0.50 0.27 0.68 0.15
avgF 1.68 1.75 1.53 0.77 0.68 0.47 0.98 0.83 0.88
p-value* 0.38 0.07 0.39 0.65 1.00 0.79 0.59 1.00 0.41
(18) Net Primary Income from Abroad
Break point 1967.5 1968.75 1981 1971.75 1959.75 1987.25 1967 1959.5 1987.25
supF 10.23 4.46 4.00 70.05 13.40 33.15 89.51 17.95 33.64
p-value 0.00 0.02 0.02 0.00 0.00 0.00 0.00 0.00 0.00
p-value* 0.16 0.82 0.87 0.00 0.04 0.00 0.00 0.00 0.00
expF 2.85 0.86 0.90 32.14 3.88 13.87 41.65 6.16 14.47
p-value* 0.18 0.87 0.86 0.00 0.06 0.00 0.00 0.00 0.00
avgF 1.44 0.42 0.45 3.86 1.33 2.87 4.03 1.56 2.92
p-value* 0.46 1.00 1.00 0.00 0.24 0.00 0.00 0.05 0.00
Notes:
1/Volatility1 is calculated assuming parameters do not vary with time.
2/Volatility2 is calculated assuming parameters may vary with time.
3/ P-value* is calculated using Andrews methodology
94
95
3.4.1 NER-related Variables: RER, Exports and Imports
As seen in Table 6, the conditional mean of RER appears to be time-
invariant, as no structural break can be found when tested under each regime or
under all sample period. All three statistics suggest that RER displays a volatility
break when tested under all sample period. However, when RER is tested
separately for each regime, no volatility break is detected regardless of the
restriction on the conditional mean, indicating that RER has been quite stable in
both regimes. This behavior suggests an existence of a one-time break at some
point during the sample period, and since the supF-statistic suggests that break date
is in the second quarter of 1971, it is possible that the one-time break is
synchronous with the time of exchange rate regime switch. This finding holds
regardless of the restriction on the conditional mean.
Similar to RER, the conditional mean of exports also appears to be time-
invariant as no structural break can be found when the whole sample period is
considered. However, the conditional mean of exports is found to be unstable
when testing is done only within the period of fixed regime. Under the time-
invariant conditional mean assumption, exports volatility shows a pattern similar to
the RER volatility. More specifically, an evidence of a volatility break is found
when tested across regimes but none can be found when each regime is considered
separately.
The same is true for the time series of imports, for which no evidence of
structural break in the conditional mean is found. Similar to the time series of
96
exports, evidence of volatility break can be found when all periods are considered
but not when each separate regime is considered separately. However, unlike RER,
the suggested date of volatility break of both exports and imports are in the late
1960s, which are earlier than the time of exchange rate regime change in early
1970s.
3.4.2 NER-related Variables: Exports Volume & Exports Unit Value
The conditional mean of the exports volume series appears to be time-
invariant, because a structural break can be found neither under each separate
regime nor under all sample period. The opposite can be said for volatility of the
exports volume. A volatility break in exports volume is detected both when tested
in each separate regime or across both regimes.
The conditional mean of the unit value of exports also appears to be time-
invariant, as no structural break can be found when tested under all sample period.
When the test is done separately for each regime, two out of three statistics suggest
that the conditional mean of the unit value of exports is unstable under the fixed
exchange rate period, while all three statistics agree that it is stable during the
floating rate period. Volatility of the unit value of exports exhibits similar patterns
to the exports volatility as well as the RER volatility. More specifically, volatility
break is found when tested across regimes but none is found when tested in each
regime separately. This finding supports a theory of a single break point rather
than volatility instability in general. Similar to export and imports, the suggested
97
date of volatility break of both exports volumes and unit value are also in the late
1960s, which are earlier than the time of exchange rate regime change in 1971.
3.4.3 NER-related Variables: Imports Volume & Imports Unit Value
The conditional mean of the imports volume series appears to be time-
invariant, because no structural break can be found when the test is conducted
across both regimes. When the test is done separate for each regime, its conditional
mean appears to be unstable in the fixed period but stable in the floating period.
The opposite is true for volatility stability. Volatility of imports volume is unstable
when all periods are considered. However, it seems stable in the fixed period but is
unstable in the floating period. This is always true whether or not the restriction is
imposed on the conditional mean.
The results of structural break tests for unit value of imports are similar to
those of imports volume discussed above. There is no sign of instability in its
conditional mean when all periods are considered. However, two out of three
statistics indicate that its conditional mean is unstable in the fixed period but stable
in the floating period. Again, the opposite is true for its volatility. Instability in its
volatility exists when all periods are considered. Based on separate structural break
tests for each regime, the volatility is found to be stable during the fixed period but
unstable during the floating period. This finding holds regardless of the restriction
on conditional mean. The supremum statistics suggest the period of volatility break
98
of imports volume to be in the third quarter of 1970 and of import unit value to be
in the first quarter of 1974.
3.4.4 Non-NER related Variables
Based on the structural break tests, the conditional mean and volatility of
most non-NER related variables appear to be regime-dependent. Industrial
production and consumption expenditure are the only two variables with time-
invariant conditional mean, because the test across regimes finds no structural
break in their conditional means. Change in inventories is the only non-NER
related variable with time-invariant volatility since no structural break can be found
either considering the whole sample period or in each separate regime,
A structural break is absent in the conditional mean industrial production
whether or not the sample period is divided based on the exchange rate regime. A
volatility break or volatility instability across the two regimes is present in
industrial production. However, since volatility breaks are also detected when the
tests are conducted separately for the fixed and the floating exchange rate regimes,
the hypothesis of instability in industrial production volatility is more possible than
a single volatility break.
The conditional mean of wages appears to be time-varying, as structural
breaks are detected in both sub-periods. Under the assumption of time-varying
conditional mean, the volatility of wages appears to be unstable under the fixed
regime but stable under the floating regime. A structural break is present in the
99
conditional mean of employment. In particular, it is stable during the fixed regime
but unstable during the floating regime. Volatility of employment displays
structural break patterns that are similar to its conditional mean. GDP volatility
and fixed capital volatility is unstable during the fixed regime but stable during the
floating regime.
3.5 Conclusions
It is noteworthy that all test statistics detect volatility break, or volatility
instability, in every variable considered in our study, except changes in inventories,
when the test encompassed all sample periods. When tested separately under each
regime, all NER-related variables appear to have stable volatility under each regime
with the exception of the volume of exports during the periods of fixed regime and
the volume of exports and imports during the periods of floating regime. Since
volatility of most NER-related variables exhibit instability across regimes but
appears to be stable when considered in each regime separately, this lends some
support to the hypothesis that volatility of NER-related variables experienced a
one-time break rather than instability during the periods under our study. NER-
related variables that fall into this pattern are real exchange rates, exports, imports
and unit value of exports. Volume of exports and net income from abroad are the
only NER-related variables that appear to have unstable volatility throughout the
sample periods as its volatility tested positive for instability under both regimes.
100
The same pattern that suggest a one-time break in volatility can be found
only in one non-NER related variable, namely the household consumption
expenditure. Government consumption expenditure may fall into this category if
we consider only Andrew’s average statistics, AvgF . However, the other two test
statistics suggest that the volatility of government consumption expenditure is
unstable under the fixed regime as well. The volatility of industrial production,
employment, and consumption of fixed capital, appear to be unstable throughout
sample periods because it tested positive for instability under both regimes.
Change in inventories is the only variable that appears to have no break in volatility
when considered either under all periods or under each regime.
We also found that supremum statistics (supF) and exponential statistics
(expF) show similar test results which are quite different from the average statistics
(avgF), which tend to be more accepting to the null hypothesis of no structural
break. Our test results only suggest that there is an evidence of one-time break in
volatility of some Japanese variables considered during the period under our study.
However, we cannot determine, with the results alone, that these breaks were
caused by the exchange rate regime change since there are many other factors that
might cause these volatility breaks during the periods under our study such as
major incidents like the oil crisis in 1973. Therefore, by testing all these variables
with the same test statistics, a pattern of one-time break in volatility emerges that
appears to be more prevalent in the NER-related variables than in the non-NER
related variables.
101
CHAPTER 4
MAIN CONCLUSION
We have now come to a conclusion on the impact on exchange rate regime
changes in Asia with examination employed through econometric theory and
modeling analysis. After having utilized Thailand and Japan as the studied
countries that both went through involuntary exchange rate regime changes, we are
able to make some conclusion analysis.
With the study of corporate debt restructuring in Thailand that occurred
mainly following the abrupt governmental decision in July 1997 to change the
nation’s exchange rate regime from a fixed or pegged rate to a floating rate regime,
we have been able draw the conclusion that there is no strong evidence that a
favorable or non-favorable outcome came about as a result of the corporate debt
restructuring process. This conclusion was made based on extensive econometric
modeling and analysis as detailed in Chapter 2 using both the matching estimator
and endogenous switching regression model. With the exception of the ROA
measure, all other financial measures failed to effectively illustrate any evidence.
Financial measures resulted in either conflicting, inconsistent or ambiguous results.
As for Japan, discussed in Chapter 3, we have been able to conclude,
through usage of the structural break test model, that after the floating of the Yen in
the early 1970s, there was a strong pattern of one-time break in volatility for most
variables that are related to the nominal exchange rates, such as real exchange rates,
102
exports, imports and unit value of exports. The same pattern can be found only in
volatility of some of the variables that are not related to the NER, namely the
household consumption expenditure, while volatility of most non-NER related
variables appears to be instable throughout the periods of our investigation. The
Japanese government’s decision to change exchange rate regime from a fixed to a
floating regime occurred in early 1973 and although past literature indicated that
there was no effect on real variables (with the exception of the actual exchange
rates), our study has shown evidence that might suggest otherwise in terms of
volatility stability of some other real variables. It is noted that our results alone
cannot determine whether these possible one-time breaks were caused by the
exchange rate regime change since there are many other factors that might cause
these volatility breaks during the periods under our study.
103
BIBLIOGRAPHY
Abadie, A. (2005). Semiparametric difference-in-differences estimators. Review
of Economic Studies 72, 1-19.
Abadie, A., Drukker, D., Herr, J.L., & Imbens, G.W. (2001). Implementing
matching estimators for average treatment effects in Stata. The Stata
Journal 1, 1-18.
Amemiya, T. (1985). Advance Theory of Econometrics. Cambridge, MA:
Harvard University Press.
Andrews, D. (1993). Test for parameter instability and structural change with
unknown change point. Econometrica 61, 821-856.
Andrews, D. & Ploberger, W. (1994). Optimal tests when a nuisance parameter is
present only under the alternative. Econometrica 62, 1383-1414.
Basu, S. & Taylor, A.M. (1999). Business cycles in international historical
perspective. Journal of Economic Perspectives 13(2), 45-68.
Baxter, M. & Stockman, A.C. (1989). Business cycles and exchange rate regime:
Some international evidence. Journal of Monetary Economics 23, 377-400.
Bélanger, D., Gutiérrez, S., Racette, D., & Raynauld, J. (1992). The impact of
exchange rate variability on trade flows: Further results on sectoral U.S.
imports from Canada. North American Journal of Economics and Finance
3, 888-92.
Bubula, A. & Otker-Robe, I. (2002). The Evolution of Exchange Rate Regimes
since 1990: Evidence from de Facto Policies. Working Paper No. 02/155,
International Monetary Fund.
Claessens, Klingebiel, & Laeven. (2003). Financial restructuring in banking and
corporate sector crises: What policies to pursue. In Michael P. Dooley and
Jeffrey Frankel (Eds.), Managing Currency Crisis in Emerging Markets.
Chicago, IL: University of Chicago Press.
Dehejia, R.H. & Wahba, S. (1999). Causal effects in non-experimental studies:
Reevaluation of the evaluation of training programs. Journal of the
American Statistical Association 94, 1053-1062.
104
Dehejia, R.H. & Wahba, S. (2002). Propensity score matching methods for non-
experimental causal studies. Review of Economics and Statistics 84(1),
151-161.
Eichengreen, B. (1994). International Monetary Arrangements for the 21
st
Century. Washington, DC: Brooking Institution.
Fisher, S. (2001). Exchange rate regimes: Is the bipolar view correct? Journal of
Economic Perspectives 50(2), 3-24.
Flood, R.P. & Rose, A.K. (1995). Fixing exchange rates: A virtual quest for
fundamentals. Journal of Monetary Economics 36(1), 3-37.
Ghosh, A.R., Gulde, A., & Wolf, H.C. (2002). Exchange Rate Regimes: Choice &
Consequences. Cambridge, MA: The MIT Press.
Ghosh, A.R., Gulde, A.M, Ostry, J.D. & Wolf, H.C. (1997). Does the Nominal
Exchange Rate Regime Matter? NBER Working Papers 5874, National
Bureau of Economic Research, Inc.
Goldfeld, S.M. & Quandt, R.E. (1973). The estimation structural shifts by
switching regression. Annals of Economics and Social Measurement 2,
475-485.
Heckman, J.J. (1979). Sample selection bias as a specification error.
Econometrica 47(1), 153-161.
Heckman, J.J., Ichimura, H., Smith, J.A. & Todd, P. (1998). Characterizing
selection bias using experimental data. Econometrica 66(5), 1017-1098.
Heckman, J.J., Ichimura, H. & Todd, P. (1997). Matching as an econometric
evaluation estimator: Evidence from evaluating a job training program.
Review of Economic Studies 64(4), 605-654.
Heckman, J.J., Ichimura, H. & Todd, P. (1998). Matching as an econometric
evaluation estimator. Review of Economic Studies 65(2), 261-294.
Imbens, G.W. (2005). Nonparametric estimation of average treatment effects
under exogeneity: A review. Review of Economics and Statistics 86(1), 4-
29.
Klein, R. & Spady, R. (1993). An efficient semiparametric estimator for binary
response models. Econometrica 61(2), 387-342.
105
Levy-Yeyati, E. & Sturzenegger, F. (2002). Classifying Exchange Rate Regimes:
Deeds versus Words. Mimeo., Universidad Torcuato Di Tella. Also
available via the Internet at: www.utdt.edu/~fsturzen
McConnell, M.M. & Pérez-Quirós, G. (2000). Output fluctuations in the United
States: What has changed since the early 1980s? American Economic
Review 90, 1464-1476.
Maddala, G.S. (1983). Limited Dependent and Qualitative Variables in
Econometric Cambridge, UK: Cambridge University Press.
Maddala, G.S. & Nelson, F. (1975). Switching regression models with exogenous
and endogenous switching. Proceedings of the American Statistical
Association (Business and Economic Section), 423-426.
Mussa, M. (1986). Nominal exchange rate regimes and the behavior of real
exchange rates. Carnegie-Rochester Conference Series on Public Policy
25, 117-213.
Newey, W.K. (1999). Two Step Series Estimation of Sample Selection Models.
Working Paper, MIT
Obstfeld, M. & Rogoff, K. (2000). New directions for stochastic economy models.
Journal of International Economics, 117-154.
Rosenbaum, P.R. & Rubin, D.B. (1983). The central role of the propensity score
in observational studies for causal effects. Biometrika 70, 41-55.
Rosenbaum, P.R. & Rubin, D.B. (1984). Reducing bias in observational studies
using sub-classification on the propensity score. Journal of the American
Statistical Association 79, 516-524.
Roy, A.D. (1951). Some thought on the distribution of earnings. Oxford
Economic Paper 3, 135-146.
Sopraseuth, T. (1999). Exchange Rate Regimes and Business Cycle Properties:
Some Stylized Facts. Papers 1999.51, Paris I - Economie Mathematique et
Applications.
White, H. (1980). A heteroskedasticity-consistent covariance matrix and a direct
test for heteroskedasticity. Econometrica 50, 1-26.
Wiwattanakantang, Y. & Polsiri, P. (2004). Business Groups in Thailand: Before
and After the East Asian Financial Crisis. Working Paper, Hitotsubashi
University.
106
A P P E N D I C E S
107
APPENDIX A
TYPES OF THAI INDUSTRY AND THE NUMBER OF THAI FIRMS ENTERING
DEBT RESTRUCTURING PROGRAM IN EACH YEAR IN EACH INDUSTRY
Number of firms entering debt restructuring program
Industry No. of firms 1996 1997 1998 1999 2000 2001 2002 2003 2004 TOTAL
1 Agribusiness 21 1 1 2
2 Building and Furnishing Materials 21 2 1 2 1 6
3 Chemical 12 2 2
4 Commerce 10 1 1
5 Communication 12 1 1 2 1 5
6 Electrical Products and Computer 8 2 1 1 1 5
7 Electronic Components 7 -
8 Energy 5 1 1
9 Food 20 1 1 2
10 Health 12 2 1 3
11 Hotels and Travel Services 12 1 1
12 Packaging 13 1 1
13 Printing 8 -
14 Property Development 29 1 2 3 2 8
15 Textile 16 1 1 2 4
16 Transportation 6 1 1
17 Vehicles and Parts 7 1 1
TOTAL 219 1 - 2 8 11 11 5 2 3 43
107
108
APPENDIX B
DERIVATION OF ERROR TERMS IN PERFORMANCE EQUATION AS A
FUNCTION OF ERROR TERMS IN SELECTION EQUATION UNDER
THE JOINT NORMAL DISTRIBUTION ASSUMPTION
The last term in equation (2.11), [ ] γ ε
i i i
Z u E − > |
1
, denotes the unknown
selection factor for the treated. To obtain this term when ) , (
1 i i
u ε are correlated,
Heckman (1979) suggests that if ) , (
1 i i
u ε are joint normally distributed and
homoskedasticity with
⎥
⎦
⎤
⎢
⎣
⎡
⎥
⎦
⎤
⎢
⎣
⎡
⎥
⎦
⎤
⎢
⎣
⎡
⎥
⎦
⎤
⎢
⎣
⎡
11 1
1
1
1
,
0
0
~
σ σ
σ ε
ε
ε
N
u
i
i
Given a joint normal distribution of a pair ( )
2 1
, z z
⎥
⎦
⎤
⎢
⎣
⎡
⎥
⎦
⎤
⎢
⎣
⎡
⎥
⎦
⎤
⎢
⎣
⎡
⎥
⎦
⎤
⎢
⎣
⎡
22 21
12 11
2
1
2
1
, ~
σ σ
σ σ
μ
μ
N
z
z
The conditional normal distribution of
1 2
| z z is
() [ ]
12
1
11 21 22 1 1
1
11 21 2 1 2
, ~ | σ σ σ σ μ σ σ μ
− −
− − + z N z z
This implies ( ) ξ μ σ σ μ + − + =
−
1 1
1
11 21 2 2
z z where [ ]
12
1
11 21 22
, 0 ~ σ σ σ σ ξ
−
− N is
independent of
1
z .
In our problem, ε =
1
z ,
i
u z
1 2
= , 0
2 1
= = μ μ and 1
11
= σ , therefore, we
have
i i i
u
1 1 1
ξ ε σ
ε
+ = as the error term equation where [ ]
2
1 11
, 0 ~
ε
σ σ ξ − N
109
APPENDIX C
DERIVATION OF SELECTION CORRECTION TERMS UNDER THE
JOINT NORMAL DISTRIBUTION ASSUMPTION
Suppose z has a standard normal distribution, then the left-truncated or
below-truncated moments of z are
[]
()
() [] c
c
c z z E
Φ −
= >
1
|
φ
and []
( )
() c
c
c z z E
Φ
= − >
φ
|
while the right-truncated or above-truncated moments of z are
[]
()
() c
c
c z z E
Φ
−
= ≤
φ
| and []
( )
() c
c
c z z E
Φ −
−
= − ≤
1
|
φ
where () ⋅ φ and () ⋅ Φ are ‘standard normal’ probability density function and
cumulative ‘standard normal’ distribution function respectively. The details are as
follows. Given a variable with standard normal distribution, [] 1 , 0 ~ N z , with
density ()
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛ −
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
2
exp
2
1
2
z
z
π
φ . Since [ ] ( ) c c z Φ − = > 1 Pr , the conditional
density of c z z > |is
()
() c
z
Φ − 1
φ
. We then have the conditional expectation as follows;
[] c z z E > |
( )
()
∫
∞
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
Φ −
=
c
dz
c
z
z
1
φ
() c
dz
z
z
c
Φ −
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛ −
⎟
⎠
⎞
⎜
⎝
⎛
=
∫
∞
1
2
exp
2
1
2
π
110
() c
dz
z
z
c
Φ −
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛ −
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
−
∂
∂
=
∫
∞
1
2
exp
2
1
2
π
() c
z
c
Φ −
⎥
⎦
⎤
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛ −
⎟
⎠
⎞
⎜
⎝
⎛
−
=
∞
1
2
exp
2
1
2
π
( )
() c
c
Φ −
=
1
φ
.
This implies []
( )
() c
c
c z z E
Φ
= − >
φ
| . Similarly, since [ ] () c c z Φ = ≤ Pr , the
conditional density of c z z ≤ |is
( )
() c
z
Φ
φ
, therefore
[] c z z E ≤ |
( )
()
∫
∞
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
Φ
=
c
dz
c
z
z
φ
() c
dz
z
z
c
Φ
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛ −
⎟
⎠
⎞
⎜
⎝
⎛
=
∫
∞
2
exp
2
1
2
π
() c
dz
z
z
c
Φ
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛ −
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
−
∂
∂
=
∫
∞
2
exp
2
1
2
π
() c
z
c
Φ
⎥
⎦
⎤
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛ −
⎟
⎠
⎞
⎜
⎝
⎛
−
=
∞
2
exp
2
1
2
π
( )
() c
c
Φ
−
=
φ
.
This implies []
( )
() c
c
c z z E
Φ −
−
= − ≤
1
|
φ
.
111
1960 1965 1970 1975 1980 1985 1990 1995 2000
50
100
150
200
250
300
350
400
CPI Indexed
Exchange rate(end of period)
1960 1965 1970 1975 1980 1985 1990 1995 2000
50
100
150
200
250
300
350
400
CPI Indexed
Exchange rate(period avg.)
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
100
200
300
400
500
600
CPI Indexed
Exports of Goods and Services
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
100
200
300
400
500
600
CPI Indexed
Imports of Goods and Services
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
20
40
60
80
100
120
140
Index
Volume of Exports
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
20
40
60
80
100
120
140
Index
Volume of Imports
1960 1965 1970 1975 1980 1985 1990 1995 2000
60
80
100
120
140
Index
Unit Value of Exports
1960 1965 1970 1975 1980 1985 1990 1995 2000
50
100
150
200
250
300
Index
Unit Value of Imports
Figure C1. Japanese Series of Interest before First Differencing Transformation
112
Figure C1 (continued).
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
20
40
60
80
100
120
Index
Industrial Production
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
20
40
60
80
100
120
Index
Wages
1960 1965 1970 1975 1980 1985 1990 1995 2000
40
50
60
70
80
90
100
110
Index
Manufacturing Employment
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
1000
2000
3000
4000
5000
6000
CPI Indexed
GDP
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
500
1000
1500
2000
2500
3000
CPI Indexed
HH Cons.Expend.
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
200
400
600
800
1000
CPI Indexed
Government Consumption Expend.
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
500
1000
1500
2000
CPI Indexed
Gross Fixed Capital Formation
1960 1965 1970 1975 1980 1985 1990 1995 2000
0
50
100
150
200
250
300
CPI Indexed
Consumption of Fixed Capital
113
Figure C1 (continued).
1960 1965 1970 1975 1980 1985 1990 1995 2000
-50
0
50
100
150
CPI Indexed
Changes in Inventories
1960 1965 1970 1975 1980 1985 1990 1995 2000
-20
0
20
40
60
80
100
CPI Indexed
Net Primary Income from Abroad
Abstract (if available)
Abstract
This dissertation is about the impact on exchange rate regime changes in Asia with examination employed through econometric theory and modeling analysis. The dissertation consists of two parts.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Essays on political economy of privatization
PDF
Mercantilism and marketization? Analysis of China's reserve accumulation and changes in exchange rate regime using the monetary policy trilemma
PDF
Two essays on major macroeconomic shocks in the Japanese economy: demographic shocks and financial shocks
PDF
The role of the timing of school changes and school quality in the impact of student mobility: evidence from Clark County, Nevada
PDF
Going offshore: studies of the maritime zone in East Asia from the eleventh to the fifteenth centuries
PDF
Three essays on the evaluation of long-term care insurance policies
PDF
Essays on the role of entry strategy and quality strategy in market and consumer response
PDF
Essays on the econometric analysis of cross-sectional dependence
PDF
Impacts of caregiving on wellbeing among older adults and their spousal caregivers in the United States
Asset Metadata
Creator
Poolvoralaks, Suriya
(author)
Core Title
Two essays on the impact of exchange rate regime changes in Asia: examples from Thailand and Japan
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Publication Date
02/15/2009
Defense Date
08/09/2006
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
corporate restructuring,endogenous switching,matching,OAI-PMH Harvest
Place Name
Japan
(countries),
Thailand
(countries)
Language
English
Advisor
Dekle, Robert (
committee chair
), Moon, Hyungsik Roger (
committee member
), Protopapadakis, Aris (
committee member
)
Creator Email
poolvora@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-m259
Unique identifier
UC1206924
Identifier
etd-Poolvoralaks-20070215 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-159137 (legacy record id),usctheses-m259 (legacy record id)
Legacy Identifier
etd-Poolvoralaks-20070215.pdf
Dmrecord
159137
Document Type
Dissertation
Rights
Poolvoralaks, Suriya
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Repository Name
Libraries, University of Southern California
Repository Location
Los Angeles, California
Repository Email
cisadmin@lib.usc.edu
Tags
corporate restructuring
endogenous switching
matching