Developing an early warning system to predict currency crises in emerging markets

In the general, these studies used the macroeconomic and financial indicators to predict the currency crises such as foreign reserves, export and import, real interest rate, real exchang

HO CHI MINH CITY VIETNAM

THE HAGUE THE NETHERLANDS

VIETNAM – THE NETHERLANDS PROGRAMME FOR M.A IN

DEVELOPMENT ECONOMICS

DEVELOPING AN EARLY WARNING SYSTEM

TO PREDICT CURRENCY CRISES

IN EMERGING MARKETS

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

By HOANG THUY HONG NHUNG

Academic Supervisor Assoc Prof NGUYEN VAN NGAI

I would also like to thank Dr Truong Dang Thuy for his professional advices, and

Mr Truong Hong Tuan and Mr Luong Duy Quang, former students, for their valuable comments

My gratefulness is also extended to all of my lecturers and staff of the Netherlands Program, particularly, Assoc Prof Nguyen Trong Hoai and Dr Pham Khanh Nam for their assistance during the first days when I started this program

Vietnam-I wish to thank my family for their encouragement and support during my study as well Without them, I would not have a chance to finish the thesis

Finally, I would like to thank all my friends and other people who have had any help and support for my thesis but are not above-mentioned

ABSTRACT

This thesis develops a new early warning system (EWS) model to predict the currency crises in emerging markets by using the logit regression According to the results, the macroeconomic variables and the institution variables are valuable indicators which play important roles in EWS model for predicting the currency crises It shows that the real exchange rate, export growth, import growth, current account surplus/GDP, short-term debt/reserves have correct sign and are statistically significant at 5% level It also shows that the law and order, external conflict have correct sign and are statistically significant at 1% In addition, this thesis also applies credit-scoring method to get the optimal cut-off threshold in order to have a more accurate probability of predicting currency crises Since then, the policy-makers can consider taking the effective pre-emptive actions to prevent the currency crises occurring in the future

Key words: currency crisis, early warning system, emerging market, logit model

TABLE OF CONTENTS

CHAPTER 1: INTRODUCTION 1

1.1 Problem statement 1

1.2 Research objectives 4

1.3 Research questions 4

1.4 The scope of the thesis 4

1.5 The structure of the thesis 5

CHAPTER 2: LITERATURE REVIEWS 6

2.1 Definition of currency crisis 6

2.2 Theoretical literatures of currency crises 7

2.2.1 First generation models of currency crises 7

2.2.2 Second generation currency crisis theoretical model 9

2.2.3 Third generation currency crisis theoretical model 10

2.2.4 “Fourth generation” currency crisis theoretical model 12

2.3 Empirical studies of currency crises 14

2.3.1 Indicators of currency crisis 14

2.3.2 Existing methods approach in EWS model of currency crisis 16

2.3.3 Summary of recent empirical findings 19

2.4 Conceptual framework 26

CHAPTER 3: RESEARCH METHODOLOGY AND DATA 28

3.1 The EWS model specification 28

3.1.1 Dating the currency crisis and define the dependent variable 28

3.1.2 Explanation variables choice and hypothesis testing 29

3.1.3 Methodology research 36

3.2 How to choose the optimal cut-off threshold 39

3.3 Data collection 42

3.4 Estimation strategy and statistical tests of the model 43

CHAPTER 4: RESEARCH RESULTS 45

4.1 The descriptive statistic of the sample 45

4.2 Empirical results 50

4.2.1 Effected by macroeconomics factors 51

4.2.2 Effected by institution factors 53

4.3 Choosing the optimal cut-off threshold 55

4.4 Predicting the currency crisis 58

4.4.1 Asian Crisis 1997-1998 59

4.4.2 Turkey crisis in 1994 and 2001 59

4.5 Robustness test 62

4.5.1 Out-of-sample test of Latin America case 62

4.5.2 Choosing optimal cut-off threshold of EWS model in Latin American 64

4.6 Compare results with other empirical studies 65

CHAPTER 5: CONCLUSIONS AND RECOMMENDATIONS 68

5.1 Main findings 68

5.2 Policy implications 69

5.3 Research limitations 71

5.4 Suggestions for future researches 72

REFERENCES 73

APPENDIX A: LITERATURE WORKSHEET AND DATA SOURCES 77

APPENDIX B: RESULTS OF CHOOSING CUT-OFF THRESHOLDS AND PREDICTING VALUE OF EWS MODEL IN ASIA 90

APPENDIX C: RESULTS OF ROBUSTNESS TEST 93

APPENDIX D: DISCRIPTIVE STATISTIC 96

APPENDIX E: COMPARISON OF TWO MODELS: MACROECONOMIC VARIABLES ONLY AND INCLUDING INSTITUTIONS VARIABLES 108

LIST OF FIGURES

Figure 2.1: The flowchart of developing an EWS model to predict currency crises 19

Figure 2.2: Conceptual framework 27

Figure 3.1: Logit and probit cumulative distributions 37

Figure 3.2: The optimal cut-off identification 42

Figure 3.3: Research processing 44

Figure 4.1: Optimal cut-off threshold of 12-months EWS model in Asian countries 57

Figure B.1: The fitted and predicted value of EWS model in Asian countries 90

Figure C.1: Optimal cut-off threshold of 12-months EWS model in Latin America 93

Figure C.2: The fitted and predicted value of EWS model in Latin America countries 94

Figure D.1: Reserve loss 96

Figure D.2: Export growth 96

Figure D.3: Import growth 98

Figure D.4: Short-term debt/Reserves 99

Figure D.5: GDP growth 100

Figure D.6: Current account/GDP 101

Figure D.7: Real exchange rate growth 102

Figure D.8: Government stability 103

Figure D.9: Corruption 104

Figure D.10: Law and order 105

Figure D.11: External conflict 106

Figure D.12: Internal conflict 107

LIST OF TABLES

Table 3.1: Summary expected sign of explanation variables 36

Table 4.1: The summary of sample used in the regressions 47

Table 4.2: The multicollinearity between independent variables 48

Table 4.3: The correlation between independent variables 49

Table 4.4: The empirical results of logit regression of 12-month EWS model 51

Table 4.5: Specification error test 51

Table 4.6: Probability of predictability of 12-months EWS model (cut-off =13.27%) 57

Table 4.7: EWS model performance with different cut-off point 58

Table 4.8: Robustness test of Asian countries in 1994, 1997, 2001, 2007 60

Table 4.9: Performance of EWS model in Asian countries when cut-off = 13.27% 62

Table 4.10: The results of EWS model in Latin American countries 63

Table 4.11: Results of explanation variables compare with other empirical studies 67

Table A.1: The summary references of explanatory variables of the model 77

Table A.2: Summary data, sources and period time of explanation variables 79

Table A.3: The literature worksheets of empirical studies 80

Table B.1: Identify optimal cut-off in Asian countries by Credit-scoring approach 90

Table C.1: Probability of predictability of 12-months EWS model (cut-off =12.02%) 93

Table C.2: Performance of EWS model in Latin American countries, cut-off = 12.02% 93

Table E.1: Comparing 12-month EWS predicting of 02 models in Asia: 1992 - 2011 109

Table E.2: Nested model test 110

Table E.3: Specification test of macroeconomic model 110

Table E.4: Specification test of full model 110

CHAPTER 1: INTRODUCTION

1.1 Problem statement

There were a lot of financial crises which occurred during the 1990s: the crises of European in 1992-1993, Mexico in 1994-1995, the crises of Asia in 1997-1998, Brazil in 1999, Turkey in 2001, Argentina in 2002 and the economic crises over the world in 2008-2009 These financial crises have strong influences on economy, politics and society They caused the economic uncertainty which suffered from high inflation, slow growth, high unemployment and poverty It made the GDP growth rate is negative, the abrupt changes in nominal exchange rate with over 50% devaluation In Argentina, it lost 20% of GDP growth and the real wages decrease match with it percentage The policy-makers were all under the pressure of implementing new policies in order to recover the affected economy Moreover, the cost of crises was very high, which led to an increase in the number of empirical studies with the aim of constructing the monitoring tools to predict the crisis occurrence These studies were often called early warning system (EWS)

There are three common types of financial crises: currency crisis, banking crisis and debt crisis However, The EWS model in this thesis only focuses on the currency crises like most of EWS models in previous empirical studies

EWS models for currency crises were first built by Krugman (1979) and enhanced

by Flood and Garber (1984) They proved that reserve loss is an important indicator

to predict crises Obstfeld (1994, 1996) has proposed a different model for predicting currency crises He stated that the currency crises occurred due to the expectation of speculators However, the model failed to take time matter into account therefore it could not predict the time when crises occurred After Asian crisis in 1997, it has created the foundation to develop a new model for currency crises Kaminsky and Reinhart (1999) built the models of the EWS for twin crises that combine banking crises and currency crises They also stated that, banking

crisis often occurred prior to currency crisis, when the currency crisis occurred, this deepened the banking crisis; as the result the economy is in twin crises In the general, these studies used the macroeconomic and financial indicators to predict the currency crises such as foreign reserves, export and import, real interest rate, real exchange rate, M2/reserves, M2 multiplier, current account deficit (or surplus)

to GDP ratio, short-term debt/reserve (Kaminsky et al.,1998, Frankel and Rose,

1996, Berg and Pattilo, 1999) In the recent years, some economists concerned about institutional factors such as bureaucratic quality, government stability, government effectiveness, voice and accountability, rules of law, democracy, election, control of corruption and so on (Block, 2003, Shimpalee and Breuer, 2006, Leblang and Satyanath, 2008) that were used to predict the probability of imminent crises

Besides selecting the potential indicators, several methods have been suggested The most popular and suitable one is logit models that were applied by Frankel and Rose (1996), Berg and Pattillo (1999) And the second is the signal approaches that were proposed by Kaminsky et al (1998) and applied by Edison (2003), Bruggemann and Linne (2000), Subbaraman, Jones and Shiraishi (2003)) Some alternative approaches are cross-country regression models which proposed by Sachs et al (1996), Ordinary least square (OLS) method such as Tornell (1999), Brussiere and Mulder (1999), Markov-switching method applied by Martinez-Peria (1999), Abiad (2003), and Artificial Neural networks (ANN) method applied by Nag and Mitra (1999)

Nevertheless, most of the EWS models only focus on identifying the indicators, which are statistically and economically significant, that should be included in the models to predict the currency crises, the problem raised is that the ability to predict

of those EWS models were unexamined In order to solve the problem, the optimal cut-off threshold is chosen to evaluate the EWS model and minimizing the crisis risk If the chosen cut-off point is low, it will give more signals of crises, therefore

more crises will be detected, however, resulting in false alarms (having the signals but no crises happen – type 2 error) increase Conversely, if the chosen cut-off point

is high, the fewer correct crises will be detected; thus, the missing signals (the crises occur but no preceding alarm – type 1 error) will decrease Kaminsky et al (1998) developed the method named the noise-signal-ratio (NSR) to choose the optimal threshold that minimized the ratio of false signals to good signals Berg and Pattilo (1999) used the Quadratic probability score (QPS) and the Log probability score (LPS) to prove that their EWS models have better forecasting ability than the Kaminsky et al (1998) Bussiere and Fratcher (2002) based on the Damirguc-Kunt and Detragiache (1999) idea to build the loss function for policy-maker to predict the currency crises They said that the choice of optimal cut-off thresholds and the predictive periods were based on the risk-aversion degree In the recent years, Candelon et al (2012) who were the first ones to summarize many methods to choose the absolute optimal cut-off points that highly contributes to evaluate the EWS forecast performance They concluded that Credit Scoring approach and Accuracy measure are better than given cut-off point method or the noise-to-signal ratio of Kaminsky et al (1998)

Following the trend of this development and enhancement of EWS models, this thesis will use seven macroeconomic variables that suggested by many previous studies such as Kaminsky et al (1998), Berg and Pattillo (1999), Bussiere and Fratzscher (2002) combine with five institutional indicators used by Shimpalee and Breuer (2006); simultaneously, use the logit approach to develop EWS models in terms of predicting the probability of currency crisis occurrence in emerging markets In order to evaluate the predictability of EWS models, this thesis will apply the Credit-scoring approach according to Candelon et al (2012) This is also one of earlier paper studies applied Credit-scoring approach to evaluate the EWS models performance to predict the currency crises

To reach the research objectives, the following questions should be answered

- What are crucial indicators that will be used in the EWS models to forecast the probability of currency crisis occurrence in emerging markets?

- Which is the optimal cut-off threshold of the EWS models using to predict the currency crises in emerging markets?

1.4 The scope of the thesis

This thesis uses monthly data of 5 emerging markets in Asian area during the period 1992M1 – 2011M3, it included Indonesia, Malaysia, Philippines, Thailand and Turkey It focuses on the emerging markets because of two reasons Firstly, several paper studies on emerging markets have been done after the wave of financial crises

in the 1990s and succeeded in determining the indicators that lead to the currency crises as well as constructing to the EWS models for preceding currency crisis occurrence Secondly, those countries are emerging markets, so it is reasonable to assume that they have the same characteristics The dataset started in January 1992

because this thesis would include all the crises from 1990s, the dataset ended in March 2011 because of the limitation of monthly economic data from IFS CD-ROM (2011)

1.5 The structure of the thesis

The following sections of this thesis are organized as follow Chapter 2 discusses the literature review of the currency crises and the EWS models Chapter 3 presents the data and methodology researches Chapter 4 describes the results obtained from logit regression model, and credit-scoring approach, the robustness test and comparison results with previous studies And, Chapter 5 is the conclusion and recommendations

CHAPTER 2: LITERATURE REVIEWS

This chapter presents three main parts First of all, it introduces the definition of currency crisis Secondly, it presents the “four” theoretical generations which are the theoretical literature of currency crisis Thirdly, it summarizes existing empirical researches, it includes (i) how to define currency crises, (ii) providing some existing EWS models of currency crises, (iii) the indicators that were used in previous studies, (iv) the findings of those researches Finally, it presents the conceptual framework

2.1 Definition of currency crisis

To build the EWS models, it is essential to identify the definition of a currency crisis In general, a currency crisis is defined as a situation in which speculative attacks on the currency that leads to exchange rate depreciation or forces the government to defend the currency depreciation by increasing the real interest rate

or selling the foreign reserves

However, the currency crises need to be measured and transformed into the value Eichengreen, Rose, and Wyplosz (1995) made an important landmark of building a method to measure the pressure of the currency and to date currency crises They enhanced the monetary model of Girton and Roper (1977) to construct the index of exchange market pressure (EMP) that included the weight of nominal exchange rate, foreign exchange and interest rate The exchange rate is under pressure when this index exceeds the certain threshold

However, the method of Eichengreen et al (1995) did not satisfy other researchers; therefore, alternative methods were developed The Frankel and Rose (1996) dropped the foreign reserves and interest rate from EMP index to define the currency crash index They defined the currency crises occurring when the depreciation of currency was greater than 25% and the rate of depreciation

increased at least 10% However, Berg and Pattillo (1999) argued that this definition could lead to false positives because some countries were regularly experiencing a fluctuation larger than 25% in exchange rate which were not be problems when taking the 25% in depreciation

Kaminsky et al (1998, 1999) method was closed to the methods of Eichengreen et

al (1995) They, however, dropped the interest rate because they argued that in some countries of their samples, the interest rate would be controlled by the central banks Moreover, while Eichengreen et al (1995, 1996) define the threshold when EMP exceeds 1.5 standard deviations of mean, Kaminsky et al (1998, 1999) used 3 standard deviations

Bussiere and Fratzscher (2002) method was similar with Eichengreen et al (1995), except they used “real” variables They take “the real exchange rate and the interest rate are intended to account for differences in inflation rate across countries and over time” (Bussiere and Fratzscher, 2002, p.9)

This thesis, thus, will follow the newest method which is based on Bussiere and Fratzscher (2002) to date the monthly of currency crisis occurrence

2.2 Theoretical literatures of currency crises

Based on histological evidence of many currency crises occurred in the past, the theoretical literatures of the currency crises are summarized in four generation models as below:

2.2.1 First generation models of currency crises

The first-generation model was built by Krugman (1979) and enhanced by Flood (1984) after the crises occurred during periods prior 1990s Krugman stated that the government could fixed the foreign exchange rate by several ways “it can use open-market operations, intervention in the forward exchange market, and direct

operations in foreign assets to defend exchange parity” (Krugman, 1979, p.311) Moreover, his model assumes that, there are two kinds of asset available: domestic and foreign currency, then the only way to keep the exchange rate fixed is to sell out the foreign reserve Therefore, the government has to trade-off between two cases: when they want to prevent the domestic currency depreciation, they have to sell out the foreign reserves until it is exhausted; and when they want to prevent the domestic currency appreciation, the government has to print money to increase money supply, resulting in high inflation occurs Then, when the economy is in high inflation, it causes the composition of speculators’ asset portfolio change, the proportion of foreign reserves will be increased and proportion of domestic reserves will be decreased In order to keep the pegged exchange rate, government has to sell out it reserves on the way finance budget deficit With such expectation, the speculators sell domestic currency quickly and reserve run out faster When the reserves exhausted push government is not able to defend fixed exchange rate anymore When the government is no longer preventing the fixed foreign exchange rate, the currency crises take place

The first-generation models stated that the problems of balance of payment are the main reasons of financial crises; it explained well the crises of Latin America countries in 1980s that have the fundamental problems are macroeconomic such as fiscal deficit, monetary excesses, and high inflation Therefore, variables that are usually used to precede the crisis in this period are international reserve loss, money supply, international interest rate, budget deficit growth, current account deficit growth, domestic credit growth and exchange rate growth

On the other hand, there were many crises occur in European countries in

1992-1993 that could not be explained by the first-generation model Although the macroeconomics fundamental were believed to be applied in this area, crisis still occurred The currency crisis models were needed to be enhanced and developed to

adapt with the current economic condition Therefore, the second-generation models were built up

2.2.2 Second generation currency crisis theoretical model

The second generation models were built after crises in European countries in

1992-1993 According to Obstfeld (1994, 1996), the crises still occurred while European countries have such the healthy macroeconomics fundamental; thus, it could not be explained by Krugman’s model Therefore, there were some other factors such as the effect of high interest rate or unemployment growth that made the government considered to response to the crises during 1992-1993 Then, once one doubt that the government was willing to maintain the foreign exchange rate by borrowing international reserves and used other policy options to deal with the crises and the question was raised: “Why does government decide to abandon a pegged exchange rate?” He stated that there is the trade-off among variety of government’s policies

In order to defend the currency devaluation, the policy-makers have to consider between the cost and benefits, and when the costs exceed the benefits they are willing to decide to abandon a pegged exchange rate target Consequently, his theory suggests that government policies are affected by the market’s expectations and the expectations of the market are affected by the government’s policies Causality of both ways leads to the existence of multiple equilibriums

The expectation of investors and speculators on whether the pegged foreign exchange rate should be kept or not will have effect on government decisions Because when the people expected the domestic currency depreciates, the expectations linking together reflect the pessimistic of the investors and the public about the policy of government, the herd behavior of investors, they could convert the domestic currency into the foreign currency Consequently, their actions would cause the depreciation of domestic currency Due to multiple equilibriums, second generation models explained crises as self-fulfilling outcomes

The second-generation models can explain well the European countries crises in 1992-1993 These crises did not come from weak macroeconomic fundamentals but from self-fulfilling expectation of speculators and the existence multiple equilibriums

These models also suggest that any factor that is likely to influence the government’s decision whether to maintain or abandon the peg exchange rate might contain information on the likelihood of a crisis occurring For the purpose of econometric model, those factors have been constrained to the following variables: level of unemployment, inflation, the amount and composition of debt and financial sector stability

However, the second-generation models could not detect the time of crisis occurring Moreover, the reality crises that happened in 1997-1998 led to develop the new generation crisis model

2.2.3 Third generation currency crisis theoretical model

The third generation models were built after the financial crises of Asian countries

in 1997-1998 According to Krugman (1999), the crises occurred because of moral hazard and asset bubble He stated that, Asian crises came from the moral hazard in the financial system: the financial institutions who received the implicit guarantee invested on the risky lending; moreover, they prefer to invest into highly risky projects which are expected to generate higher returns even if with a very low probability The consequence of moral hazard and the risky investment is raising asset prices Moral hazard encourages financial institutions to increase investment projects even low expected return If the investment focuses on the supply of assets fixed (such as real estate that most finance companies investing in Thailand), the asset price will rise An expectation- circle arises People invest in real estate because real estate prices expected to rise When the amount of the investment increases the demand, the prices are going up as expected The bubble asset was

created The financial institutions lend investors to invest in real estate are also willing to continue lending as seen “too safe”

Bubble price grows excessively until it booms when the investors realized that the real situation or the real price was not reality, it was different from the expectation The reverse action occurred They investors sell their assets because they expect the price to decrease The supply will then increase rapidly and become greater than the demand, this excess supply will cause the price to fall; and as price falls the banks realized the value of collateral for loans decrease and no longer lend for new loan, also try to recover of old loans

However, Radelet and Sachs (1998) said that the Asian crises occurred because investors lost their confidence; they did not believe the foreign reserves were enough for repaying the short-term debt A bank is in normal trading and suddenly depositors continuously make a substantial withdrawal of money that pushes the bank into a difficult situation At the beginning state, asset values remain greater than the value of liability; it means the bank is still stable However, when large amount of money is withdrawn simultaneously, there will be insufficient cash flow

to cover the increasing demand Thus, banks are no longer leaving the solvency (or loss of liquidity) The bank claimed bank loans, refused to rollover and stopped to new lending The broker sold their stocks to exchange the foreign currency and transferred money to foreign countries From their perspective, the lack of resolution mechanism of enterprise debt, bank debt forth both good and not good enterprises have falling into trouble and depression the financial crises

According to Kaminsky and Reinhart (1999), the models of this period are the early warning system for “twin crises” that combine banking crises and currency crises They also stated that, bad loan from domestic debt, and short-term debts from foreign bank are the causes of the banking crises and it often occurs to precede the currency crises; when the currency crises occurring, this deepened the banking crises; the result caused the economy being twin crises

Different from prior two generation models, the third generation models showed that before the crises, Asian countries which have macroeconomic background are relatively good as high GDP growth, low unemployment, inflation and government spending be controlled, the budget deficit is low, the flow in of foreign investment

is large and political and economic are stable However, behind the bright picture are the problems of the banking system such as bad debt from borrowers in the country with short-term foreign currency loans The moral hazard relative with strong political relation between government and finance institution, bank, corporation (it called crony capitalism) when suggesting the implicit guarantees from government And, the bubbles of asset and stock prices were soaring and plugging preceding the Asian crises

Third generation models suggest some variables such as: real interest rate, lending

or deposit rate growth, domestic credit growth, M2/reserves, bank deposit, bank cash/ bank asset, non-performing loan

2.2.4 “Fourth generation” currency crisis theoretical model

Beside three generations models mentioned above, there seems to have additional approaches to investigate the causes of the currency crises Although, there were no specific events that might attribute to this generation of currency crises, the occurrences of currency crises following the 1997-98 Asian Financial Crisis such as Russia (1998), Turkey (2000-2001), and Argentina (2001-2002) have rose up the interested to find all possible causalities and linkages, which different from factors already known in three theoretical models of currency crises, could lead to a currency crises Uncertainly, it may be called “fourth generation theoretical model”

In these models, weak institutions worsen problems associated with economic growth and contribute to causing currency crises

According to Li and Inclan (2001), institutions affected currency crises in two ways First, institutions tend to have an impact and relative with the health of the national

economies Then, the institutions can cause the bad economy and contribute to currency crisis occurrence or institutions can create the good economic fundamental and remove some reasons to cause the currency crisis occurrence Second, the institutions are the informative The institutions sign to market agents about the future economic fundamentals; thereby, it can shape market expectations Then, when institutions correlate with bad economies cause the instability market expectation, increase the market uncertainty; cause more capital outflow and increase the likelihood of currency crises On the contrary, when institutions correlate with good economies cause the stability market expectation, decrease the market uncertainty, cause the capital outflow less and reduce the likelihood of currency crises

Acemoglu et al (2002) stated that countries have poor macroeconomic policies also have weak institutions, weak investor’s property rights, widespread corruption, and

a high degree of political instability They implied that the poor macroeconomic policies are the symptom of institutions problems rather than the instable economic;

it stated “weak institutions cause volatility through a number of microeconomic, as well as macroeconomic channels”

Leblang and Satyanath (2006) developed the framework model to link the institution, speculation’s expectation and crises; in addition, the empirical proved the better of their approach in terms of predicting This study stated that, the divided government and the government turnover increase the uncertain of speculator’s belief and raise the likelihood of currency crisis It showed that, the focal point for the speculators’ predication of economic variables is the forecast of government However, speculators expectation will rely on their own predict if the government announcement is not credible Therefore, the new government needs time to develop their accuracy forecast During this time, the speculators deprived the common of focal point It implied that, following the government turnover is the

period of uncertain of speculator’s belief and increase the probability of currency crisis even the economic fundamentals is unchanged

The fourth generation crisis models created more choice of variables in early warning system for currency crises Some variables suggested for these models are: Prudential supervision, accounting and disclosures requirements, legal and judicial systems, bureaucratic quality, government stability, absence of corruption, law and order, absence of external conflict, election, absence of internal conflict, exchange rate, capital control, absence of ethnic tensions, central bank independence, deposit insurance, financial liberalization and legal origin

2.3 Empirical studies of currency crises

An EWS model included the specific definition of currency crisis and given the structure to predict the likelihood of currency crisis occurrence According to Glick and Hutchison (2011), the contents of EWS model to predict the currency crises typically require three parts as follows: (i) a method to define or date of the currency crises as discussed in section 2.1, (ii) a set of explanation indicators, and (iii) the statistical methods This section will present some potential important indicators that common used in the EWS models to predict the currency crises; then, this section will present some methods that usually used in previous researches, thereafter this thesis will summarize the findings of some empirical studies

2.3.1 Indicators of currency crisis

In fact, in order to EWS models operate effectively, the selections of indicators using in the model are very crucial because they contribute to the accuracy prediction of currency crisis occurrence

The theoretical literature review and the previous empirical studies have identified a large of potential variables regarding to currency crisis occurrence Some

representative studies such as Eichengreen et al (1996), Frankel and Rose (1996), Kaminsky et al (1998), Kaminsky and Reinhart (1999), Berg and Patillo (1999), Peltonen (2006), Shimpalee and Breuer (2006), Leblang and Satyanath (2008) They identified many variables of macroeconomic fundamentals such as the growth

of real exchange rate, the growth of broad money, domestic credit growth, current account surplus (or deficit)/ GDP ratio, reserve loss, export growth, import growth, Short-term debt/reserve and institution factors such as government stability, control

of corruption, law and order, external conflict, internal conflict, voice and accountability, regulation quality

There are many variables that may possible to enter the predicting model for currency crises These indicators are classified into categories as below:

 Capital account: M2/foreign reserves, foreign reserves growth, gross external

debt/ export and short-term debt/foreign reserves

 Current account: the growth of real exchange rate, export growth, terms of

trade, import growth, and current account/GDP,

 Domestic and public real sector: public debt/ GDP, change in stock price, GDP

growth, an index of equity prices

 Financial sector: the growth of M1 and M2, M2 multiplier, domestic

credit/GDP, domestic real interest rate

 Institution factors: openness, exchange control, changes of government, degree

of political instability, legal or illegal executive transfer, and lack of control corruption, external or internal conflict, voice and accountability, regulation quality

 Global economy: US interest rate, growth of world oil prices

2.3.2 Existing methods approach in EWS model of currency crisis

During 1990s, there were many researchers such as Kaminsky et al (1998) Kaminsly and Reinhart (1999), Berg and Pattillo (1999), Frankel and Rose (1996), Edison (2000), Bussiere and Fratzscher (2002) who pursued development of models that have statistically and economically significant of predicting currency crisis occurrence, known as “EWS model” Two methods are common approached in EWS model There are the Signal approach (Kaminsky et al., 1998, 1999, Edison, 2003) and the Logit/Probit model (Frankel and Rose, 1996, Eichengreen et al.,

1995, Berg and Pattillo, 1999)

Crisis occurs Non-crisis occurs

Good signal

B False alarm

Missing signal

D Bad signal

In which,

- A: good signal - a number of months the indicators issue the signal, the crisis

occurs in following

- B: false alarm - a number of month the indicators the signal but the crisis does

not occur (Type 2 error) or called it “noise”

- C: missing alarm - a number of month the indicators were not issued a signal

precede the crisis imminent (Type 1 error)

- D: bad signal - a number of month the indicator “refrain” from the crisis occur,

it did not issue the signal and the crisis also does not occur

She stated that, the predictors can observes only in A and D cell It means, it will issue the signal and the crisis will occur in following (within 24 months), thus A>0, C=0 It will not issue the signal and the crisis will not occur in following (within 24 months), thus D>0, B=0

They defined “optimal” threshold at minimize the noise-to-signal ratio (NSR), that

is the false signals to good signals ratio and calculate as equation below:

NSR = [B/(B+D)] / [(A/(A+C)]

Moreover, Kaminsky et al (1998) suggested another way to interpret the signal result of indicators by comparing the conditional probability of crisis [A/(A+B)] with unconditional probability of crisis [(A+C/(A+B+C+D)] of the indicators from above matrix In order to the indicators have the good information; the conditional probability has to higher than unconditional probability In addition, in the purpose early warning currency crisis in 24 months, their research also ranked indicators according to their ability predict a crisis at the first signal

Zhang (2001) said that this method is very useful It is easy to applied and also easy

to interpret the problem regarding to the abnormal exhibit of each indicators Although it is the most basic application for EWS model to predict the currency crisis, it has some shortcoming Glick and Hutchison (2011) stated that due to

evaluating each indicator separately, it cannot examine the relative of indicators that contribute to likelihood of currency crises Second, this method did not consider the correlation between indicators Besides that, Abiad (2003) also stated that the signaling approach did not measure the marginal effect of indicator on the probability of currency crises and the multicollinearity among indicators Because it also not to exam the statistical testing of its own as well as comparing with other methods

2.3.3.2 Logit/ probit approach

The logit/ probit approach solved some disadvantages of signaling approach It estimate the relationship between the dependent is the binary (e.g, equal 1 if a currency crisis occur and equal 0 if a currency crisis does not occur) and explanation variables It takes into account the correlation among the explanation variables, it measures the marginal effect of each variable on probability of crisis, and it can also test the statistic significant of individual variables And, it also gives the probability of future crises (Glick and Hutchison, 2011)

This approach requires dependent variable is crisis dummy variables (equal 1 or 0) that can be identified as above discussed in section 2.3.1 In order to anticipate a crisis at the first time of crisis, the definition of crisis is transformed into an exclusion window or the predicting horizon time (e.g., 1 month, 3 months, 12 months, 18 months or 24 months) It also requires the set of explanation variables that possible to be early warning indicators suggesting by previous literatures and describing in section above

Then, the probability of a crisis occur is calculated by the equation following

( ) ( )

And the probability of crisis not occur is (1- P) The outcome of model is the

probability of a crisis occurs in given value in the next k months (predicting time)

The logit/ probit approach were applied by many studies such as Eichengreen et al (1995), Frankel and Rose (1996) and Berg and Pattillo (1999)

Figure 2.1: The flowchart of developing an EWS model to predict currency crises

2.3.3 Summary of recent empirical findings

There are many empirical studies that research on the currency crises This thesis just summarizes the studies that focus on the EWS model to predict the currency crises and the explanation variables that they suggest as follows:

Eichengreen, Rose and Wyplosz (1995) identified crisis by the EMP index that calculated by the weight of nominal exchange rate, foreign reserves and interest rate and exceeded the mean and one and a half standard deviation of it mean Even though using the same a definition of crises, they also have many differences monthly of crises by changing the number of standard deviation from one and a haft

to two or three standard deviations They used quarterly data of 20 industrial

countries from 1959 to 1993 and probit method to estimate the contagious of currency crises and they found that they are significant

Frankel and Rose (1996) used annual data of 105 developing countries from 1971

-1992 to identify the currency crisis crash They defined crises when the depreciation

of exchange rate change larger than 25%, and at least 10% change compare with previous depreciation They used sixteen variables that were classified into four categories: foreign variables, macroeconomic variables, external variables and composition of capital inflow variables Thay applied probit approach to estimate the EWS models with 3 years exclusive window They found that the crises occurred due to low FDI inflows, foreign reserves, high domestic credit growth, interest rates and the real exchange rate overvaluation

Kaminsky et al (1998) suggested the EWS model specifically (called KLR model) This paper used 15 monthly indicators of 20 countries during 1970 - 1995 to issue signal of crisis when they cross the certain threshold They applied the signal approach to find the probability of the crises within next 24 months was estimated

by signal of indicators at that moment They asserted that this model successfully applied to predict the crises in 1997

The IMF World Economic Outlook (May, 1998) used monthly data of 53 countries

of industrial and developing countries spanning from 1975 – 1997 It used probit approach to estimate the probability of currency crises in the next 18 months against with 12 macro variables It found that M2/reserves rises, real domestic credit growth rises, world interest rate, real exchange rate rises, inflation, reserve loss, stock price, term of trade declines, private credit growth increase somewhat are significant They also found that Thailand and Malaysia are vulnerable, Korea and Indonesia somewhat vulnerable, but Philippine is not

Esquivel and Larrain (1998) used annually data of 30 emerging markets and industrial countries from 1975 – 1996 periods They applied panel probit with random effect approach to determine the currency crises They used 3 groups of

variables: stock variables, flow variables and contemporaneous variables in their model They found that macroeconomic variables are the important components in order to determine of crises This also supposed the rule that predicting the currency crises (Pi,t) when the probability of currency crises exceeds the certain cut-off threshold (P*) It means, when the Pi,t > P*, is the time of crises, and otherwise is the time of tranquil Then they found the cut-off threshold value is P = 0.3 where the total of crises predict correctly (both in crisis and tranquil time) is maximum

Berg and Pattillo (1999a, b, c) are in a list of studies developing the existing early warning model for using at the IMF Berg and Pattillo (1999a) re-estimate the KLR model to predict the Asian crises They used 15 indicators in KLR model and add two more variables are M2/reserve (level) and current account/ GDP After eliminating some variables with incorrect sign and insignificant at 10% level, there exists 5 significant variables: real exchange rate, export loss, reserve loss, M2/reserves (level) and current account deficit/GDP (named BP model) They also compare the effect of two crises prediction models (signal approach based on Kaminsky et al (1998) and probit approach) to employ the Asian crises in 1997 and they found the explanation variables power in probit model

Berg and Pattillo (1999b) try to enhance the BP model (Berg and Pattillo, 1999a) by adding one more variable, which is a short term external debt/reserves based on Redelet et al, 1998a They found that the model was highly significant However, M2/reserves became incorrect sign and insignificant, they eliminate it from the model Therefore, the BP model (1999b) includes 5 variables as following: real exchange rate, export loss, reserve loss, M2/reserves, current account/GDP and short term debt/reserves

Berg and Pattillo (1999c) evaluated three models for forecasting the probability of currency crises occur that were suggested before 1997 Three models were suggested by Frankel and Rose (1996), Sachs, Tornell and Velasco (1996) and Kaminsky et al (1998) They try to answer two questions: (i) whether the IMF can

use these models in late 1996, (ii) have these models been predicted the Asian crises well They realized that Frankel and Rose (1996) and Sachs, Tornell and Velasco (1996) fail to use to predict, while Kaminsky et al model can forecast but still have some information not reliable

Glick and Moreno (1999) define a crisis as change of exchange rate percentage over mean plus two standard deviations and the inflation was not larger than 150% in the last 6 months They used the multivariate probit approach to estimate the likelihood

of currency crisis in Asia and Latin America separately They used the monthly data from 1972:01 to 1997:10 of two group variables: Money and credit variables and the competitive and trade variables They found that, the foreign reserves, real domestic credit are significant to predict the crises in Latin America while the real exchange rate is more significant in Asia

Kamin Schindler and Samuel (2001) used the exchange rate pressure to define the currency crisis; exchange rate pressure construct by weight average of two-month changes of real exchange rate and foreign reserves; and the crises occur when this index exceed 1.75 standard deviation above the mean They used the annual data of

13 variables of three groups: domestic variables, external balance and external shock variables in 26 emerging market from 1981 to 1999 and pooled probit approach to predict the currency crisis With one year lagged of explanation variables, they used the values of variables in this year to predict the crises in the next year They found that the external balance and external shock variables have more effective than the domestic variables in terms of predicting the currency crises within one year

Burkart and Coudert (2002) used quarterly data of 15 emerging market from 1980 –

1998 They used 34 indicators and dropped until six variables: reserve/M2 ratio, reserve/debt ratio, inflation, short-term debt/total debt, real overvaluation and regional contagion They used discriminant model and predict correctly four crises out of total five crises

Kumar, Moorthy, and Perraudin (2003) define currency crises when the depreciation over the 5%, 10% or 15% cut-off level They used the monthly data spanning from 1985 to 1998 of 32 emerging market economics and the logit approach They found export growth, real GDP, reserves losses, inflation variables are significant in their model; also the contagion, fiscal balance, and commodity prices has explanation power

Edison (2003) used 15 variables of KLR model plus seven indicators included US interest rate, oil price, G7 output, US output, M2/reserves (level), short-term debt/reserves (percentage change), and short-term debt/reserves (level) They also add eight countries in original 20 countries in KLR model in their data spanning from January 1970 to December 1999 They applied the signal approach and found that six variables (reserves, exports, real exchange rate, M2 multiplier, M2/reserves, and imports) are useful for forecasting

Peltonen (2006) define a currency crisis as weight average of exchange rate and foreign reserve exceed the two standard deviation plus mean They construct the new model to predict the currency crisis; it is artificial neural networks (ANN) They applied this model for 24 emerging markets from 1980 to 2001 with monthly data of fourteen variables and found that contagion effect have largest marginal effect on probability of currency crises He also stated that the current account/GDP rises, budget deficits/GDP rises, the real GDP growth rate decline are the good indicators to predict the currency crises

While most empirical studies in the past used macroeconomic fundamental to predict the currency crisis, just recent researches focus away from macroeconomic factors to focus on the institutions factors

Shimpalee and Breuer (2006) research is the rare research studies about the deeper institution factors with currency crisis They define the currency crises based on Eichengreen, Rose and Wyplosz (1995), Frankel and Rose (1996) and Kaminsky,

Lizonro and Reinhart (1998) and used 13 institution variables to estimate the effect

of institution factor on crises They also add control variables by eight macroeconomic variables to control a currency crisis affected by only institution variables They found that instable government, weak law and order, more corruption, and fixed exchange rate regime increase the onset of currency crises while the modest of capital control and bank independence reduce the currency

Leblang and Satyanath (2008) based on three models: Frankel and Rose (1996) and Kaminsky, Lizonro and Reinhart (1998), Bussiere and Fratzscher (2002) to define the crisis They also used variables suggested by three models above and add some institutions into three models to exam the effect of institution factors on currency crisis occurrence The institutions variables included: government turnover, unified government, democracy, number of checks and balance and political polarization They found that the political variables play a good role, specially, government turnover and divided government

Tuan, T.H (2009) based on Peltonen (2006) to define the currency crises He added domestic credit growth indicators into BP model (Berg and Pattillo, 1999b) to create the economic factors groups In order to exam the effect of institutions factors, he used six variables that suggested by Kaufman (2008) This paper applied the logit regression with monthly data of 15 emerging markets in period 1996-2005 and found that the current account/GDP, reserves loss, export loss, domestic credit growth, voice and accountability and regulatory stability have significant impact on the currency crises occurring in the next 18 months This study proved the effect of institution factors on the currency crises occurring and reaffirmed the contribution

of economic indicators to predict the currency crises However, due to data’s limitation of institutions variables that available from 1996, this paper used the short period time (1996-2005) that did not cover all the crises occurring during 1990s such as the crises of Mexico, Turkey in 1994, even the shorten period before the Asian crises in 1997-1998 with the predicted time is 18 months Moreover, the

author just focused on determining which the key indicators effect on the currency crisis occurrences; thus, the predictive ability of this model was not considered in this paper Because the outcome of logit regression is the probability of crisis occurrence, the author did not mention which probability needed to be concerned, for instance, when the probabilities of crisis occurrence exceed 20%, 30% or even 70%, 80%, it issued signals of crises occurring in the next 18 months

Most of empirical studies above focused on determining what indicators cause the currency crises The recent studies focused more on how to evaluate the EWS models they compare with different methods to choose the optimal cut-off threshold

in order to improve the predictive ability of EWS model (Candelon et al., 2012) or they compare with different methodology such as Comelli, F (2014)

Candelon et al (2012) suggested a curial tool to evaluate EWS models This tool is free methods that can use the logit/probit approach or Markov switching approach and it can be applied for different EWS model to obtain the forecast abilities Moreover, it can be used for any types of financial crises such as: banking crises, currency crises, debt crises and so on In addition, it proposed the different standard

to evaluate the predictive ability of EWS model through credit-scoring approach and accuracy measures to get the optimal cut-off threshold and suggested different tests such as “Quadratic Probability Score (QPS)” and “The area under the ROC Curve (AUC)” Furthermore, this EWS model can apply for both in sample and out-of-sample Using the data of 12 emerging markets in South-Asian and Latin American during period 1980-2010, they found that the cut-off point is important element for the decision-makers to know the country is being crises or calm period

at the given time And, when adding the cut-off points in criteria and tests, it refines the predictability better

2.4 Conceptual framework

Upon above theoretical and empirical literature review and the available of data, this thesis will use 12 potential early warning indicators and divided the explanation variables in two groups: macroeconomic and institution factors In which,

 Macroeconomic factor : reserve loss, export growth, import growth, real

exchange rate growth, current account surplus/GDP, short-term debt/reserve and the GDP growth

 Institution factor: government stability, corruption, law and order, external

conflict and internal conflict

This thesis is going to test relationship between currency crises occurring in the next 12 months and the explanation variables in emerging markets as the following:

Figure 2.2: Conceptual framework

CHAPTER 3: RESEARCH METHODOLOGY AND DATA

This chapter included four sections: First section presented how to construct the EWS model with three parts: dating the currency crises, description the set of explanation variables and present the methodology research The second will present the method to choose the optimal cut-off threshold The third section will describe data collection and the last section will discuss the estimation strategy which will employ to answer the research questions

3.1 The EWS model specification

As discussed in section 2.3, in order to construct the EWS model, it consists of three parts: dating the currency crisis, selecting the set of indicators, and choosing the statistical method

3.1.1 Dating the currency crisis and define the dependent variable

As discussed above, there are many ways to define the currency crises It can observe the change of exchange rate only (Frankel and Rose, 1996), or the change

of both exchange rate and the international reserves (Kaminsky et al., 1998), or the change of nominal exchange rate, international reserves as well as the interest rate (Eichengreen et al., 1995, 1996, Bussiere and Fratzscher, 2002) or change of real exchange rate, international reserves as well as the real interest rate (Bussiere and Fratzscher (2002)

This thesis applied the Exchange Market Pressure (EMPi,t) index based on Bussiere and Fratzcher (2002) to identify the currency crisis of country i in period t The EMPi,t was calculated as weighted average of the change of three crucial factors: real effective exchange rate (RER), real interest rate (r) and foreign reserves (RES)

It presented in equation as below:

(

) ( ) (

)

According to Bussiere and Fratzcher (2002), a currency crisis (CC) was considered occurred at country i in period t if EMPi,t is greater than its average plus two standard deviations

{ ̅̅̅̅̅̅ ( )

After define the time when crisis occur, with the aim to predict the likelihood occurrence of currency crises or the currency crisis occurrence within the specific time horizon, this thesis transforms the contemporary currency crisis variables into

a forward dependent variable Yi,t, which is defined as

{

In other words, this model attempts to forecast whether a crisis will occur within the next 12 months It is the 12 month predicted because with the short-time forecast that could be treat policy-maker weaken and have no enough time to action when the economy comes to crisis closer However, if lengthen the forecast, when the policy-maker have pre-emptive actions, it could be lead to self-sufficient problem as the second-generate model That why this thesis used 12-month predicting that is the good trade-off between these two issues

3.1.2 Explanation variables choice and hypothesis testing

The choices of the explanatory variables which enter our EWS model are based on the previous literature on currency crises, the significant of indicators in those studies and the data availability Therefore, this thesis included 12 potential early warning indicators and divided the explanation variables in two sets: one set of macroeconomic factor and other is institution factor In which,

 Macroeconomic factor : includes 7 variables, there are: reserve loss, export

growth, import growth, real exchange rate growth, current account to GDP ratio, short-term debt to reserve ratio, the GDP growth

 Institution factor: includes 5 variables, there are: government stability,

corruption, law and order, external conflict and internal conflict

All variables were used based on many studies that presented in Table A.1 (Appendix), in which five variables of macroeconomic factors measurement based

on Berg and Pattllo (1999b) and two rest variables based on Kaminsky et al (1998) and all institutions variables measurement based on the Shimpalee and Breuer

(2006)

3.1.2.1 Reserve loss

The decrease of foreign reserves is reliable signal of devaluation of currency Because collapse of currency usually occur following the period of effort keeping the pegged exchange rate by declining the foreign reserve Thus, the large decreases

of foreign reserves cause the high probability of currency crisis occurrence Moreover, the total value of foreign reserve is also the factor that illustrates the difficulty of the economy Therefore, the loss of foreign reserves is the potential indicator leading to the currency crises Kaminsky et al (1998), Berg and Pattillo (1999), Glick and Moreno (1999), Bussiere and Fratzscher (2002), Edison (2003) and Tuan (2009) also found significant impact of this indicator on currency crises in their models The expected sign of this indicator is positive

 H1: Reserve loss has positive relationship with probability of predicting

currency crises

3.1.2.2 Export growth

Many studies such as Kaminsky et al (1998), Berg and Pattillo (1999), Glick and Moreno (1999), Kumar et al (2003), Edison (2003) and Tuan (2009) found that the

decrease of export growth may be caused by an overvaluation exchange rate Because, an overvaluation of exchange rate means the domestic currency can buy more foreign exchange, therefore the export goods is more expensive for foreigner and the demand of export good is decline It causes more unemployment Moreover, when the exchange rate is unstable, it leads to the speculative attack on currency Therefore, decrease export growth can be a leading indicator for currency crises occurring The expected sign is negative in this model

 H2: Export growth has negative relationship with probability of predicting

currency crises

3.1.2.3 Import growth

On contrary of export growth, the decrease of import growth may be caused by a devaluation exchange rate Because, exchange rate devaluated means the domestic currency can buy less foreign exchange, therefore the import goods is more expensive and the export goods is cheaper, then, demand of import good is decline and decrease the living standards Moreover, high import can cause the current account deficit that affecting on currency crisis occurrence (Kaminsky et al (1998); Berg and Pattillo (1999); Edison (2003)) Therefore, the expected sign is positive

 H3: Import growth has positive relationship with probability of predicting

currency crises

3.1.2.4 Real exchange rate growth

According to the Kaminsky et al (1998); Berg and Pattillo (1999); Glick and Moreno (1999), Kamin et al (2001); Bussiere and Fratzscher (2002), Edison (2003), with a pegged currency system, high inflation will increase real exchange rate growth which may put pressure on exports When the real exchange rate is growth or appreciation, it causes the currency devaluation and leads to high

probability of currency crises occurring The expected sign of coefficient in this model is positive

 H4: Real exchange rate growth has positive relationship with probability of

predicting currency crises

3.1.2.5 Current account surplus/GDP ratio

Current account surplus, it means that the export is higher than import that makes the capital inflow is higher than capital outflow Therefore, the ratio of current account surplus to GDP increase help decline the devaluation of currency and thus, decrease the likelihood of currency crises occurring This indicator was also used by Berg and Pattillo (1999), Kamin et al (2001), Bussiere Fratzscher (2002) and Tuan (2009) with the expected sign of this indicator is negative

 H5: Current account surplus/GDP has negative relationship with probability of

predicting currency crises.

3.1.2.6 Short-term debt/Reserve

This indicators suggest by Berg and Pattillo (1999b), they stated that, higher short term foreign debts/reserve give signals to outside lenders or investors to lend cautiously, thus, foreign currency supply reduces and cause the currency crises occurring And it also founded significant in model of Bussiere and Fratzscher (2002) and Leblang and Satyanath (2008) Accordingly, the expected sign is positive

 H6: Short-term debt/reserve has positive relationship with probability of

predicting currency crises.

3.1.2.7 The GDP growth