Determinants of currency crises in emerging economies in 1996 2005, an early warning system approach

VIETNAM- NETHERLANDS
PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

Determinants of currency crises in
emerging economies in 1996-2005: An
Early Warning System approach
A thesis submitted in partial fulfillment of the requirements
for the degree of
Master of Arts in Development Economics

By
.

'

Trtrang Hong Tuan

• BQ GIAO DVC FJAO TAO
TRUONG fJ~f HQC KINH TE
TP.HCM

THliVIEN
Thesis supervisor:
Dr. Vii Thanh Ttr Anh

•

Ho Chi Minh city, October 2009

r- ~1

•


Certification

I declare that the thesis hereby submitted for the Master degree at the Vietnam-Netherlands
Programme for M.A in Development Economics is my own work and has not been previously
submitted by me at another university for any degree. I cede copyright of the thesis in favor of the
Vietnam- Netherlands project for M.A programme in Development Economics.

Ho Chi Minh City, October 2009

Truong H6ng TuAn.

Page 2 of 52


Abstract
Theories of currency crisis consisted of 4 generations of models suggest that economic and
institutional variables can be employed in early warning system models to predict currency
crisis for the purpose of prevention policy. This study incorporates 5 variables from Berg and
Pattillo (1999b) model (Real exchange rate overvaluation, Foreign reserves loss, Export
growth, Current account deficit, Short-term external debt/Foreign reserves) and additional
Domestic credit growth, 6 institutional variables adopted from Worldwide Governance
Indicators (Kaumann et al., 2008) (Voice and Accountability, Political Stability and Absence of
Violence, Government Effectiveness, Regulatory Quality, Rule of Law, Control of Corruption)
into a simple logit model with dataset of 15 emerging market economies in the period 1996:012005:09. The new finding is the high statistical significance of the variable 'Voice and
Accountability' (represents freedom of speech, free media and ability to participate in selecting
government of a country citizen) on reducing probability of currency crisis. 'Regulatory
Quality' (measures government ability to implement efficient policies promoting private sector

development) also shows its statistical significance at a lower level in the model. This study
also reconfirms other studies that Domestic credit growth and Current account deficit precede
currency crisis.

Page 3 of 52


TABLE OF CONTENTS

Certification
Abstract
Table of Contents
Chapter 1 Introduction
Chapter 2

Theories of curren

Chapter 3 Typical early warning systems and empirical currency crisis models A brief review of
Chapter 4 Methodological issues - Empirical framework
Chapter 5 Empirical results
Chapter 6 Policy implications and conclusion
Notes
References
Appendix 1 Specification of 05 empirical currency crisis models
Appendix 2 Logit regression results by Eviews (Probability of currency crisis)
Appendix 3 Robustness test of the result by running logit regression on regions

Page 4 of 52

Chapter 1
Introduction
1.1 Statement of the Problem

On the way to development, currency crises are very costly for emerging economies. Currency
crises can lead to banking crises, loss of GDP, high unemployment rate and loss of
development momentum. In the Asian crisis in 1997-98, Thailand lost 10.5% of GDP,
Indonesia 13.1% & Malaysia 7.4%.
In May 2008, Morgan Stanley issued a report on Vietnam named "Beyond the tipping point",
comparing Vietnam then with Thailand in 1997 and warning a 38% -55% depreciation of VND
against USD in the next 12 months.
In June 2008, State bank of Vietnam widened trading band for foreign exchange (USD) from
1% to 2%. In November 2008, it raised the band to 3% and in March 2009 to 5%. Domestic
credit growth rate is 50% in 2007, 34% in 2008 and estimated 30% in 2009 by Economist
Intelligence Unit.
After the booming in stock and real estate market in 2007 with capital inflow mainly for
portfolio investment, a crash of more than 70% in stock market broke out in 2008 against its
peak in October 2007. Capital inflow and export shrink, larger current account deficit (13.6%
in 2008) is putting pressure on the peg regime of VND to USD. It seems that the scenario of
Asian crisis repeats in Vietnam.
Unlike Asian crisis countries of crony capitalism, Vietnam's relationship-based system has
even weaker institutions. So do institutions play any role in setting stage for a currency crisis?
Now is October 2009. Fortunately, the Morgan Stanley's forecast failed, but whether the
Vietnam currency crisis is coming soon? Thus, currency crisis is a burning issue in Vietnam
for the time being.
1.2 Objective of the Research

Page 5 of 52

Vietnam is an emerging market economy. Understanding what caused currency crises in other
emerging market economies in recent years would be a good reference for further
comprehensive researches on what Vietnam should do to prevent its own currency crisis.
1.3 Research Questions

This study aims to answer the following questions:
1. What are the key determinants of currency crises in emerging economies in the period
1996-2005 in the light of early warning system approach? (especially, current account
deficit, domestic credit growth and institutional factors)
2. What are the policy implications to prevent a currency crisis?
1.4 Research Methodology

Based on theories, empirical models of currency crises and available variables/data, regression
with logit model is carried out to recognize the determinants of currency crisis. The statistic
software Eview 4.1 is employed in this study.
1.5 Scope of the Research

Based on availability of data of institutions and review of recent history of currency crises in
emerging market economies, 15 economies are selected and the period of study is limited in
1996-2005. 15 economies are: Argentina, Brazil, Colombia, Czech, Ecuador, India, Indonesia,
Korea, Malaysia, Philippines, Russia, Slovakia, South Africa, Thailand and Turkey.
1.6 Organization of the Research

The first chapter of this study hereby presents introduction of the issue. The second chapter
will look through theories of currency crisis. The third introduces some typical early warning
systems (EWS) and empirical currency crisis models employed at IMF, Goldman Sachs and in
a few academic studies. The fourth mentions methodology for designing a EWS model
including variables of domestic credit growth and institutions. The fifth represents merits of
the focused variables in the model and the sixth delivers policy implications and concludes.

Page 6 of 52


Chapter 2
Theories of currency crises - A brief review of the literature
In the literature of financial crises, there are banking crises, (sovereign) debt crisis and
currency crises. Banking crises are recognized as the insolvency of the banking system that
occurs with high ratio of non-performing loan to assets. Demirguc-Kunt and Detragiache
(1997) considered one of event or combination of the following as banking crisis: (1)
nonperforming assets/total assets ratio in the banking system exceeds 10%; (2) the cost of
rescue at least 2% of GDP; (3) large scale nationalization of banks; and (4) extensive bank runs
or other emergency measures executed by the government. Debt crisis is defined as a national
government fails to meet a principal or interest payment on the due date (Reinhart and Rogoff,
2008). This study focuses on currency crises only.
This chapter consists of 2 parts. Part 1 presents the identification of a currency crisis, part 2
reviews 4 generations of currency crisis models.
What cause a currency crisis? It is an exciting question since the collapse to the Breton Wood
system. Especially, the heavy costs of currency crisis in Mexico, Asia, Russia, Argentina ...
provoke attention of several economists.
2.1 Identifying currency crises

An abrupt drop in a country currency value is regarded as currency crisis. It is called 'crisis'
because it bring about negative economic effects. They includes shrink in GDP, investment
and job loss, banking and business failures, inflation. Currency crisis can be brought about by
currency speculators or government action, or a mix of both.
The ideal way to define a currency crisis is as Bussiere and Fratzscher (2002) that incorporate
moves in exchange rate, interest rate and foreign reserves (initially set out by Girton and
Roper, 1977). They identified a crisis as the exchange market pressure (EMP) of a specific
country exceeds its mean by 2 standard deviations. EMP is constructed as a weighted average
of the change of the real effective exchange rate (RER), the change in the real interest rate (r)

and the change in foreign exchange reserves (res). Real values are considered to avoid
different inflation rate across countries. Interest rate is involved in case the central bank
defends the domestic currency by increasing its interest rate.
Page 7 of 52


The EMPi,t for defining a currency crisis for each country i and period t in formula is as
follows:

_
EM~.~

The weights

-

ffiRER, ffir

and

ffires

are computed as the inverse of the variance of each variable for

itself so as to give a larger weight to the variables with less volatility.
Due to lack of data, most of the cases currency crisis is defined with changes in exchange rate
and foreign reserves only like Kaminsky et al. ( 1998) (KLR) with a little bit difference in the
threshold of number of standard deviations. KLR model recognized a crisis as EMP goes
beyond its mean by 3 standard deviations.
In another simple way, Frankel and Rose (1996) defined a currency crisis as a depreciation of

the nominal exchange rate by at least 25% that also exceeds the previous year's depreciation
by at least 10%. Thus, they did not consider speculative attacks failed by government
intervention via selling foreign reserves as a currency crisis.

The theories of currency crisis have developed over the time. It seems that after a series of
currency crises occurred, a new-generation crisis model emerges. A brief review of currency
crisis literature can trace out 4 generations of currency crisis model.
2.2 Four generations of currency crisis models
The first generation crisis models
The first generation crisis model is firstly presented by Krugman (1979). In a small open
economy, government would defend the fixed exchange rate regime with limited foreign
reserves. Perfect foresight private investors hold asset portfolio in two kinds of assets:
domestic and foreign currency. In an attempt to maintain the fixed rate, government has to sell
out reserves until it is exhausted. Government finances its budget deficit by printing money
that increases money supply.
Page 8 of 52


The model derives that as long as there is budget deficit and inflation, investors change the
composition of their asset portfolio by increasing the proportion of foreign exchange, reducing
the proportion of domestic currency. Government has to run down its reserves to retain the
fixed rate on the way to finance its budget deficit. Exhaustion of reserves pushes government
to abandon the pegging. Such expectation makes investors advance the date of their
speculative attacks, the leading speculators sell domestic money even earlier and so on,
reserves run out faster and currency crisis breaks out.
The model seems to have highly simplified assumptions on two asset portfolio holding, the
tools government uses to intervene in foreign exchange market, only selling reserves, perfect
foresight speculators, the time of crisis is unclearly identified.
Flood and Garber (1984) refined Krugman (1979) model by developing it into 2 models. The

first one is a perfect-foresight, continuous-time model that relaxes two-asset portfolio to 4asset portfolio (included domestic and foreign bond) and adding domestic credit into the
model. They found out the exact timing of the peg collapse, and timing has a positive
relationship with the size of reserves and a negative one with the domestic credit growth rate.
The first model also shows that currency crisis can emerge under arbitrary speculative
behavior of investors but they assumed that this effect is zero for simplifying analysis. The
second model is a discrete time, stochastic one (relax assumption of perfect foresight) that
incorporate uncertainty to study the forward exchange rate of a peg regime as a response to
reality that forward rate may exceed the fixed rate for long period of time. Using the concept
the shadow exchange rate, the rate that would prevail after the speculative attack, the second
model yields an endogenous probability distribution over the crisis time and produces a
forward exchange rate that is greater than the fixed rate (capture the real world).
The policy conflict in above-mentioned models is the financing fiscal deficit and retaining
fixed exchange rate regime. Dooley (1997) proposed an insurance model, in which the policy
conflict is the desire of a credit-constrained government to hold reserve assets as a form of
self-insurance against shocks to national consumption and the government's desire to insure
financial liabilities of residents. The first objective is pursued by accumulation of foreign
reserves while the second objective depletes it. Once the domestic yield is greater relative to
international returns, it generates a private gross capital inflow. Capital inflow increases to an
extent that there is not enough foreign reserves to insure deposits, extra deposits have risk
exposure. This gap will ignite a speculative attack to minimize loss that runs up to crisis.

Page 9 of 52


Dooley model offers a capital inflow/currency crisis follow view, that not budget deficit,
money supply or higher international interest rate is blamed for currency crisis.
The first generation models explained well the crisis of Latin America countries in the 1980s
that have macroeconomic fundamental problems such as fiscal deficit, hyperinflation, foreign
loan, current account deficit, capital flight.
The first generation models suggest that macroeconomic fundamentals are the causes of

currency crises. Variables used in early warning systems could be budget deficit, money
supply, domestic credit, current account deficit, international interest rates, capital
inflow/outflow (capital control).
In the early 1990s, there were several currency crises, such as the European Monetary System

crisis of 1992-93, that could not be explained by the first generation models. Europe countries
at that time had sound macroeconomic fundamentals, but currency crises still occurred. The
currency crisis model then evolved to confront the new reality. The second generation came
out.
The second generation crisis models

The second generation crisis models, initially developed by Obstfeld (1994, 1996): selffulfilling and contagious crises.
The first generation models constrained government's tools in intervening foreign exchange
market to selling limited reserves only. The second· generation models relax this point, let
foreign reserves can be freely borrowed in the world capital market, subject only to the
government's intertemporal budget constraint.
In a setting of purposeful reaction by the government, self-fulfilling crisis is taken into

account. Speculative anticipations depend on government responses, which subject to how
price changes, that in turn are driven by expectations. This dynamic circle suggests a potential
for crises that would not have occurred, but that do because the market players expect them to.
They are self-fulfilling crises.
Obstfeld (1994) raised a question: Why does the government like to abandon the fixed rate?
He described 2 models in the paper to answer. It is because government debt denominated in
domestic currency and unemployment problem. In one model, he mentioned the role of
Page 10 of 52


nominal interest rate. Devaluation expectations feed into high nominal interest rate that pushes
government further to give up the peg would have been viable under reverse private

expectations. Maturity structure of the government's domestic obligations and the currency
composition of the overall public debt would decide the effect of interest rate in devaluation.
Large portion of debt burden denominated in domestic currency, higher nominal interest rate
will lead to higher devaluation to lessen the government debt burden. Perfect foresighted
speculators would try to get out of the domestic currency ahead of that devaluation. In another
model, expectations feed into wages and competitiveness, raising unemployment. The country
suffers from unemployment/competitiveness problem due to demand shock and/or pre-set
nominal wage, would like to abandon the fixed rate. Negative expectation of government's
willingness to tolerate unemployment can trigger a devaluation that would not have occurred
under opposite. expectations. In contrast to the first generation models, the loss of reserves is
not the factor triggering currency crisis.
The models propose that a speculative attack can occur even with the absence of poor
macroeconomic conditions in the pre-crisis time. Currency crisis can result from self-fulfilling
attack in which speculative anticipations and herding behavior play a role.
Once unemployment rate is on upward trend and government policy is to maintain a fixed
exchange rate. Speculators can perceive that high political cost of the maintaining of the fixed
exchange rate facing the future rising unemployment. They recognize the devaluation is likely.
Even they don't know when, they start selling domestic currency now. Herding behavior sets in
and full-scale speculative attack breaks out even before unemployment becomes a problem.

Eichengreen et al. (1996) show that contagion takes its effect once the devaluation of a
country's currency may reduce its trading partners' competitiveness enough to make their
currencies subject to devaluation too. They pose hypothesis that there are 2 channels for
contagion taking effect. Trade linkage between 2 countries can motivate one country to
devalue its currency to increase its international competitiveness one the other country
devalued previously. Based on the current and prospective international competitiveness of the
countries concerned, speculators ignite attacks. The similar macroeconomic conditions across
countries also are foundations for advancing speculative attacks. Using a panel of quarterly
data for 20 industrial countries for the period 1959-1993 to test for contagious currency crises,
they find evidence of contagion. Contagion appears to spread to countries which have close

international trade linkages rather than to countries in similar macroeconomic conditions.

Page 11 of52


Calvo (1995) suggest that the basic cause of a currency crisis may be investors' behavior. Risk
averse investors invest in several countries. Financial diversification and lender's information
have an interesting relationship. Investors with highly diversified portfolio have lower
incentives to learn about individual countries than investors with few diversification
opportunities. Diversification encourages ignorance and, in that context, rumors could result in
massive capital flows from a country. So investment into or away from a country is highly
sensitive to news in a world of highly diversified investors. Diversification magnifies herding
behavior by making investors more sensitive to rumors. If the composition of portfolio is
mainly short-term capital, the capital flight may be quick, causing an abrupt crisis.
If self-fulfilling crises are a real possibility, what stimulates them? The answer is that anything
could in principle be the driver. They would be expected unemployment, public debt,
international competitiveness in trade, political factors ... This rationale gives a spacy room for
empirical early warning system models in predicting crises.
Second-generation models can explain the European Exchange Rate Mechanism (ERM) crisis
in the 1990s. These crises didn't emerge by the poor macroeconomic fundamentals but by
inconsistent policy and political events. In Europe, the crises occurred in Britain, Italy, France
because of the inconsistent policy with their committed peg to the German mark. Retaining
peg to German mark, they have to maintain high interest rate on local currency and face slow
growth, gloomy export and increasing unemployment. Speculators such as Soros recognized
the tradeoff government confronts and expected the government under political pressure will
give up the peg. Their speculative attacks succeeded.
In fact, behind the scene of claimed sound macroeconomic fundamentals, potential economic
instability laying aside in Europe incites speculative attacks. Macroeconomic fundamentals
actually still play their role to a certain extent in the second generation model of crisis.
In 1997-1998, new reality of crisis emerged again that requires currency crisis model to

continue to evolve to deal with new generation of crisis.
The third generation crisis models

The Asian crisis of 1997-98 led to the third generation models: twin crisis, a mix of banking
and currency crisis. In fact, Velasco ( 1987) proposed models of interaction between banking
problems and currency crisis. Credit boom, bad debt, government spending on bail out
Page 12 of 52


(budget shrinks) and stop of capital inflow all lead the way to abandonment of fixed exchange
rate regime. Velasco (1987) represented experiences in South America with these features.
Their models received little attention until the Asian crisis in 1997 broke out.
What Velasco did is to extend Krugman model to a situation where foreign assets pay interest,
including the presence of banks to find out the dynamics of banking crisis and currency crisis.
There is an asymmetry in the liquidity and riskiness of bank assets and liabilities. A bank
usually guarantees the nominal value of the deposits it accepts while allocating the money to
investments with a variable return. Normally, bank deposits are highly liquid, while bank
investments are low liquid and long-term. It takes time and cost to liquidate bank investments.
There is an assumption that all bank deposits are implicitly or explicitly guaranteed by the
government. Domestic and foreign depositors/lenders believe in the government umbrella, still
put money into the banking system in spite of banking operation loss. Banks play the Ponzi
game until the problem become serious. Government steps in to help, depletes its budget and
foreign reserves. Currency crisis of Krugman type eventually arrived here.

Using monthly data of 20 countries for the period 1970-mid1995, Kaminsky & Reinhart
(1999) also found that problems in the banking sector typically emerge before a currency
crisis. The currency crisis then worsens the banking crisis, making a vicious spiral. Financial
liberalization often precedes banking crises. The progression of these events suggests that
crises occur as the economy starts to enter a recession, following a boom in economic activity
that is fueled by credit, capital inflows, associated with an overvalued currency.

Compared to first and second generation models, Asian crisis appears to be differently.
Macroeconomic fundamentals are sound with· high GDP growth rates, low unemployment
rate, low inflation, low budget deficits, manageable current account deficits, strong capital
inflows and prevailed political stability. However, inside the bright picture, there are problems
in the banking sector - bad loans from domestic borrowers and unhedged, short-term
borrowing from foreign banks. The bad loan accumulation is the consequence of over lending.
Krugman (1998) described a moral hazard/asset bubble view. There was a boom-bust cycle in
the asset markets preceding the Asian currency crisis. Prices of stock and land were soaring
and plunging before the crisis. Besides, moral hazard involved as strong political relations
between government and finance companies/banks/corporations (crony capitalism) suggested
implicit guarantees from government for lenders and depositors. The lenders become much
less prudent in lending to inefficient investment projects because of expectation on
government bailout.
Page 13 of52


Third generation model implies an additional set of variables related to financial liberalization
and banking problem for early warning system. Name a few as: Real interest rate, lendingdeposit rate ratio, M2/reserves, bank deposit, non-performing loan.
Recently, we faced the re-emergence of institutional economics (economics Nobel prize 2009
is an award for institutional economics) that pays more attention to the foundations for market
functioning that shed new light on currency crises.
The fourth generation crisis models
Behind the scene of phenomena like government budget and current account deficits,
hyperinflation, self-fulfilling attacks, herding behavior, excessive lending, deeper questions
should be raised. What institutional factors set conditions for these pictures? In these models,
institutional factors are incorporated and emphasized with a wide range: law framework,
property right, enforcement of contract, financial regulations, bureaucratic quality, government
stability, democracy and corruption ... Breuer (2004) coined these models as the fourth
generation.
This institutional focus has foundations on the theory of relationship between institutions and

economic growth/crisis presented by Acemoglu et al. (2002), Rodrik et al. (2002), Fukuyama
(1995), Sen (1999), Johnson et al. (2000) and asymmetric information problem in financial
market by and Mishkin (1996, 2001) as typical ones among several other authors.
A defmition of institutions is helpful here. What are institutions? By words ofDouglass North
(1993) in his Nobel prize lecture:
"Institutions are the humanly devised constraints that structure human interaction.
They are made up of formal constraints (rules, laws, constitutions), informal
constraints (norms of behavior, conventions, and self imposed codes of conduct), and
their enforcement characteristics. Together they define the incentive structure of
societies and specifically economies."
Incentives and expectations play a vital role in the financial world that is infamous for
uncertainty. Institutions form the foundations and facilitate a well-functioning currency and
banking system. That's reason why the fourth generation models of currency crisis incorporate
institutional variables.
Page 14 of 52


Fukuyama (1995) proposed the role of trust in society that lowers administration cost,
increases institutional reliability and promotes large and efficient organizations. Low trust
society usually has corruption and trade with influences and tends to maintain small and
inefficient organizations. Trust should promote economic development. It is vital for financial
market whether investors have trust or loss of confidence in them.
Sen (1999) presented concisely some of his best-known work on famines. They are usually
occurred by a lack of purchasing power or entitlements, not by food shortage. He claimed that
large-scale famines never happened in a democracy. They can only happen in authoritarian
regimes lacking openness of information and transparency. His analysis inspired similar
approach to the Asian crisis in 1997.
Johnson et al. (2000) proved that the effectiveness of protection for minority shareholders
explain the extent of declines in exchange rate depreciation and stock market better than
standard macroeconomic measures do. They modelized the conflict of interest between

insiders (managers) and outsiders (equity owners). Weaker corporate governance rules and a
weaker legal system reduce the cost of stealing (expropriation) of managers. The manager
compares the marginal cost and marginal benefit of stealing for their reaction. For a given rate
of return, if the manager invests less their own money, they have more incentives to steal.
However, if the manager has more shares in the firm, an increase in the return on investment
persuades himto invest more into the project and, therefore, to steal less. On the other hand, if
the manager owns more of the firm, but the return on investment reduces, then he steals more.
The stealing of manager shrinks value of firm. Less value of the firm, less confidence of
foreign and domestic investors stays in the stock market. Loss of confidence in the stock
market triggers capital outflow or stop of capital inflow. The capital outflow affects foreign
exchange market, ignites currency crisis. In order to study 27 emerging markets in the end of
1996 to January 1999, Johnson et al. employed data measure institutional variable such as:
shareholder protection, credit right, accounting standards, enforceability of contracts Gudicial
efficiency, corruption, rule of law, corporate governance); economic variable like: fiscal and
monetary policy, current account and reserves. Running linear regression of these institutional
and economic variables on exchange rate and then stock price, they found that the regression
results support their theory. Lower quality of shareholder protection and enforceability of
contract have high statistical significance and influence in explaining the depreciation of
domestic currency and decrease in stock price in emerging markets.

Page 15 of 52


Acemoglu et al. (2002) observed that countries pursuing poor macroeconomic policies also
have weak institutions, including political institutions that do not constrain politicians, weak
property rights for investors, widespread corruption, and a high degree of political instability.
They proposed that macroeconomic policies are more likely to be symptoms of underlying
institutional problems rather than the main causes of economic volatility, weak institutions can
cause volatility through a number of macroeconomic and microeconomic channels.
Mishkin ( 1996, 200 1) approached financial crises in the light of asymmetric information with

the moral hazard problem. Emerging market economies such as Mexico, Ecuador, East Asian
crisis countries and Russian are well-known for weak financial regulations and supervision.
When financial liberalization facilitate international capital inflow and opportunities to take
more risky lending, these weak regulatory/supervisory system could not limit the moral hazard
problem created by the government safety net. Once government has signals offering bailouts
to protect banks & corporations, excessive risk taking is one result, increasing the probability
of financial crisis embracing banking crisis and currency crisis. Mishkin also proposed 12
areas of policy to prevent financial crisis: 1. Prudential supervision, 2. Accounting and
disclosures requirements, 3. Legal and judicial systems, 4. Market based disciplines, 5. Entry
of foreign banks, 6. Capital control, 7. Reduction of control of state-owned financial
institutions, 8. Restrictions on foreign-dominated debts, 9. Elimination of too-big-too-fail in
the corporate sectors, 10. Sequencing of the financial liberalization, 11. Monetary policy and
price stability, 12. Exchange rate regimes and foreign exchange reserves. From this list, we can
see some institutional factors that cause financial crisis.
Rajan, R. G. and Zingales, L. (1998) looked through the Asian crisis, pointed out that
relationship-based systems (that are inefficient and corrupt) tend to attract short-term external
capital inflows that make them excessively prone to shocks. A relationship-based system
distorts the price system and the signals it provides. As a result, it can misallocate capital.
Majority of external capital inflow is typically from foreign investors that have little
contractual rights or power in a relationship system. They understand the potential for
misallocation and keep their claims short term for easy withdrawing.
Due to the definition of institutions, they may include factors of law system, legal practices,
financial regulations, politics, customs, culture. Thus the fourth generation crisis models spare
large room for the choice of variables usable in early warning system for currency crises.

The US crisis 2008: a new crisis generation or an old story?

Page 16 of 52

The existing crisis in the US is in fact a banking crisis caused by excessive lending on subprime mortgages (Ellis, 2008). The sub-prime mortgages are cut into pieces in securitization
process and the risk become opaque, then are sold out, spread in the US and over the world.
Like the scenario is described in the third crisis generation (twin crisis), banking crisis come
first, government implements giant bailouts of banks/corporations and then currency crisis
follows. There are holes in financial regulations in the US financial system that promotes such
excessive risk lending (no job-no problem housing loans). Thus, the fourth generation factors
can also be used to explain the US crisis. Although FED's newly-issued treasury bonds (for
bailouts) are supported by huge international reserves from China, the US dollar is being
depreciated against other main currencies (EUR, GBP, AUD). So we can consider the US crisis
is kind of third and fourth generation combination.
Theories on the causes of currency crises set foundation for several early warning system and
empirical currency crisis models coming out. The four generations of currency crisis models
suggest what variables should be employed in empirical early warning system and currency
crisis models. While the first and the third generation models point out the specific economic
variable such as budget deficit, current account deficit, money supply, capital flow, bad debts
as indicators of currency crisis, the second and the fourth generation give more room for
choices ofvariables. It is because expectations can be based on a wide range of indicators and
institutions involve a lot of factors related to laws, politics, culture.

Page 17 of 52


Chapter 3
Typical early warning systems and empirical currency crisis
models - A brief review of the literature
The Mexican and Asian crises took the international community somewhat by surprise and
since then raised the attention on methods that could forecast crises of highly vulnerable
countries in the timely manner.
Several early warning systems (EWS) empirical currency crisis models have been. designed
and some of them are being used by IMF and private financial institutions like Goldman Sachs

for policy and speculative purposes. The Table 1 hereunder summarizes the main features of
typical EWSs, including EWSs used in IMF like Kaminsky et al. ( 1998), Berg and Pattillo
(1999b), in Goldman Sach like GS-WATCH (Ades et al., 1998), and EWSs presented in other
academic studies like Peltonen (2006), Shimpalee and Breuer (2006), Leblang and Satyanath
(2008).
3.1 Signal approach

Kaminsky et al. ( 1998) (KLR) proposed observation of 15 indicators that give warning
signals prior to a crisis. KLR's crisis definition is previously mentioned in chapter 2. of this
study (Theories of currency crisis). KLR selected 15 indicators based on prior theories and on
the monthly data available. 15 variables are listed below:
1. Real exchange rate deviation
2. Banking crisis
3. Export
4. Stock prices
5. Growth rate of M2/ international reserves
6. Output
7. 'Excess' real M1 balances (Residual from regression of real M1 on real GDP, inflation,
and a deterministic trend)
8. International reserves
9. M2 multiplier
10. Domestic credit/GDP
11. Real interest rate

Page 18 of 52


12. Terms of trade
13. [Real interest differential
14. Imports

15. Bank deposits
16. Lending rate/deposit rate]
An indicator gives out a signal whenever it moves beyond a specific threshold. Thresholds are
defmed in relation to percentiles of the distribution· of observations of the indicator. An optimal
threshold for a given predictor, such as export loss, might be 70, for example, that is uniform across
countries, the corresponding country-specific thresholds would likely differ (for example 20% in a
country, 30% in another). The performance of each indicator is considered in the following matrix:

Table 3.1: Matrix of measuring performance of indicators

Signal was issued
No signal was issued

In this matrix, A represents the number of months in which the indicator issued a good signal, B is
the number of months in which the indicator issued a bad signal or
'noise,' C is the number of months in which the indicator failed to issue a signal which would have
been a good signal, and D is the number of months in which the indicator did not issue a signal that
would· have been a bad signal. The optimal percentile threshold is the one that minimizes the bad
signal to good signal ratio [(B/(B+D)]/[A/(A+C)]. Four indicators (Real interest differential,
Imports, Bank deposits, Lending rate/deposit rate) that produce excessive noise are eliminated from
the KLR model.

As described by Berg and Pattillo (1999a), Kaminsky

1

later developed a single composite

indicator of crisis that is computed as weighted sum of the indicators, where each indicator is
weighted by the inverse of its bad signal/good signal ratio.

Then a probability of crisis for each value of the composite index is calculated by the number of
months having a given value of the index is followed by a crisis within 24 months.

Page 19 of 52


Because of the shortcoming of signal approach in testing statistical significance of each
indicator, most of the EWSs are probit/logit based models.
3.2 Probit/Logit approach
Berg and Pattillo (1999a) modified and carried out out-of-sample test for the KLR model to
predict Asian crisis 1997. Berg and Pattillo added 2 additional variables (Current account
deficit/GDP ratio, level ofM2/Reserves) also replaced 5 developed countries in the sample by
8 emerging market countries. Then they transformed KLR's signal approach to probit based
model with 05 variables only (Reserves growth rate, Export growth rate, Real exchange rate
deviation, Current account deficit/GDP and M2/Reserves). Probit and modified KLR
performed well in predicting Asian crisis 1997. However, probit model with 5 variables can
test the statistical significant of individual indicators with ease while the KLR model (with 17
variables) can not do it. (Table 3.2)
Berge and Pattillo (1999b) (BP model) presented a simple probit model over a panel of
developing countries through 1995 to predict Asian crisis 1997. They found that 5 variables
(Reserves growth rate, Export growth rate, Real exchange rate deviation, Current account
deficit/GDP & Short-term debt/Reserves)-measured in percentile- performed well in
forecasting the crisis 1997. They regarded the first 4 variables as first generation ones and
short-term external debt/reserves as second generation one. Short-term external debt/Reserves
ratio is used to measure vulnerability to panic, suggested by Radelet and Sachs (1998). Once
foreign lenders become convinced that other lenders would not roll over their loan, there are
not enough reserves to cover the mature loans. Panics tum into self-fulfilling. (Table 3.2)
Goldman Sach's GS-WATCH model, designed by Ades et al. (1998), is the logit model
employing 09 variables (Table 3.2) included 01 political risk indicator. Like BP model, their

definition of crisis involves reserves loss and nominal exchange rate move, but they have
specific thresholds for specific countries by using Self Exciting Threshold Autoregression
technique (also used to identifY recession in business cycle literature). Different from BP
model, GS-WATCH incorporates stock prices, real interest rate in G7 economies, contagion,
credit (to private sector) growth and political risk - a fourth generation variable. For their
purpose of private institutional investor, the predicting horizon is 3 months only. Their insample test proved their model worked well in predicting currency crisis in developing
countries. (Table 3.2)

Page 20 of 52


Peltonen (2006) model identified a currency crisis similarly to BP model but he used the
threshold of 2 standard deviations beyond the mean of the specific country's crisis index. His
choice of variables falls in a range of 3 generations, from budget deficit, contagion to, stock
price, real interest rate (Table 3.2). He also used several dummy variables for the world's
regions and for different exchange rate regimes. After testing in-sample and out-of-sample,
Peltonen also claimed that his model did a good predicting work.
3.3 Incorporating institutional variables
Shimpalee and Breuer (2006) extended EWS by adding institutional variables to previous
economic EWSs in the literature. They selected 3 models of Eichengreen et al. (1995),
Kaminsky and Reinhart (1999), Frankel and Rose (1996) and incorporated 13 institutional
variables into these 3 models. Most oftheinstitutional variables are measured by International
Country Risk Guide. The additional variables are briefly described as follows:
1. Bureaucratic quality: measures the strength and quality of civil service and bureaucrats
and their ability to manage political problems without interruption of services.
2. Government stability: captures the government ability to continue its announced
programs and to stay in office.
3. Absence of corruption: higher index means less corruption.
4. Law and order (quality): measures the strength and impartiality of the legal system
and popular observance of the law.

5. Absence of ethnic Tensions: rooted from racial, nationality, or language division and
gauge. The index measures how intolerant groups might be to compromise. Higher
number indicates lower ethnic tensions.
6. Absence of external conflict: ranges from trade restrictions and embargoes to
geopolitical disputes to incursions, insurgencies, and warfare.
7. Absence of internal conflict: reflects the extent of political violence.
8. Exchange rate regime: dummy variable. Code 1 with fixed exchange rate, and 0
otherwise.
9. Capital controls: controls on capital account transactions, or restrictions on capital
movements, especially inflows with a 1, and 0 otherwise.
10. Central bank independence: level of freedom from political pressure. It is measured on
a scale between 0 and 1 where 0 stands for the minimum level of independence and 1
for the maximum level.

Page 21 of 52


11. Deposit insurance: dummy, 1 with explicit deposit insurance protection systems, 0
otherwise.
12. Financial liberalization: real interest rates are used as a proxy for financial
liberalization because real interest rates are usually lower, or negative, in repressed
financial systems.
13. Legal origin: dummy. Civil law system are coded as 1 and common law countries as

0.
The rationale to include the institutional variables into the EWS model can be traced back to
theories of the four generation model of currency crisis.
The trial and error process, by dropping a few institutional variables in the 3 models,
suggested that, in most of the cases, government stability, absence of corruption, law and order
(quality), central bank independence play an important role in currency crises. (Table 3.2)

In the similar approach of Shimpalee and Breuer (2006), Leblang and Satyanath (2008)
employed 3 EWS models of Frankel and Rose (1996) (FR model), Kamin et al. (2001) (KSS
model), Bussiere and Fratzscher (2002) (BF model) and associated additional political
variables into them. Most of the political variables are measured by the World bank's Database
of Political Institutions (Beck et al., 2003), except for the Democracy indicator measured by
Adam Przeworski and his colleagues (Alvarez et al., 2000). The following are the extra
political variables: (Table 3.2)
1. Government turnover: extent of turnover of a government's key decision makers in
any one year.
2. Unified/Divided government: A divided government is defined as the party of the
chief executive does not control the legislative, then this dummy take the value of 1,
otherwise, 0.
3. Democracy: a dummy represents the absence/presence of a government actually
relinquishing office following an election.
4. Number of checks and balance: the number of actors whose permission is required to
change policy from the status quo.
5. Political polarization: captures the polarization level of zero as elections are not
competitive, or if the chief executive's party has an absolute majority in the
legislature.

Page 22 of 52


Leblang and Satyanath (2008) added 2 variables (Government turnover, Unified/Divided
government) into FR model, KSS and BF models first, then Democracy, Checks and balance,
Polarization.
For the FR model, Turnover and Unified government proved their correct signs and statistical
significance. A shift from unified to divided government or an increase in government turnover
does increase the probability of currency crisis.

In KSS and BF models, coefficients for unified government and government turnover are

correctly signed. Out of which, government turnover is statistically significant in KSS model
and vice versa in BF model. For the out-of-sample test, all 3 modified models claim its
improvement over 3 original models in predicting crises.
The FR model with additional Turnover and Unified government is considered as 'core FR
model'. Robustness tests are done on the core FR model by adding in turn each of 3 other
variables (Democracy, Checks and balance, Polarization) and Country and decade dummies.
Turnover and Unified government still proves correct sign and high statistical significance.
Democracy, Checks and balance and Polarization show low significance although Democracy
& Checks have correct sign.
(Refer to the Appendix 1 for specification of 05 currency crisis models of Eichengreen et al.
(1995), Frankel and Rose (1996), Kaminsky and Reinhart (1999), Kamin et al. (2001) and
Bussiere and Fratzscher (2002).)

Page 23 of 52


Table 3.2
Specification of typical Early Warning Systems and empirical currency crisis models

Crisis
definition

Predict time
horizon
Method

Dataset time
Countries

involved
Variables

Page 24 of 52


Table 1 (cont.)
Specification of typical Early Warning Systems and empirical currency crisis models
Crisis defmition

Predict time
horizon.
Method

Dataset time
Countries
involved
Variables

Dummy for de facto freely falling FX regime
Dummy for Latin America
Dummy for Europe
Dummy for Asia; Dummy for Africa

Page 25 of 52