Determinants of banking crisis in developing countries

In the wake of these recent crises, issue of what determinants affecting to occurrence of banking crisis has been a hot topic for economists in both developed and developing countries ar

VIETNAM- NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

DETERMINANTS OF BANKING CRISIS

IN DEVELOPING COUNTRIES

A thesis submitted in partial fulfillment of the requirements for the degree of Master of Arts in Development Economics

By LuO'ng Duy Quang

I am grateful to my all teachers and staffs of the Vietnam-Netherlands Program, particularly, Assoc Prof Nguyen Trong Hoai for his assistance during the first days I study in this program

Many thanks are respectfully sent to my manager, Mr Nguyen Tu Han, and

my colleagues for their encouragement and support during my writing thesis duration

Finally, I am indebted to my family, especially my parent and others who give me great encouragement and support for my study

TABLE OF CONTENTS CHAPTER 1: INTRODUCTION

1.1 Problem statement: 5

1.2 Research objective: 6

1.3 Research question~ 6

1.4 Research hypothesis: 6

1.5 Structure of the thesis: 6

CHAPTER II: LITERATURE REVIEW 2.1 Key Definition: 8

2.2 Banking crisis theory: 9

2.3 Empirical Studies; 14

2.4 Chapter remarks: 17

CHAPTER III: MODEL SPECIFICATION AND DATA 3.1 Model choice: l8 3.2 Theoretical model and model specification: 20

3.2.1 Growth~ 21

3.2.2 Real short-term interest rate: 21

3.2.3 Exchange rate~ 22

3.2.4 Inflation: 23

3.2.5 Terms ofTrade; .23

3.2.6 Speculative attack: 22

3.2.7 Deposit Insurance: 22

3.2.8 Liberalization~ 22

3.3 Estimation strategy and statistical tests of the model: 24

3.4 Data sources: 26

3.5 Data filler process; 27

CHAPTER IV: REGRESSION RESULTS AND TESTING HYPOTHESIS 4.1 Accessing the best unbiased model: 28

4.2 Significance of explanatory variables: 33

4.3 Minimization of banking crisis frequency: 36 CHAPTER V: WELL-KNOWN CRISES AND PREDICTION POWER OF THE MODEL

5.1 Thai Banking Crisis~···39 5.2 Uruguay's crisis: Victim of contagious effect 40 5.3 Implementing the best unbiased model to Thailand and Uruguay cases: failure to explain the crisis originating in balance of payment crisis 43 CHAPTER VI: CONCLUSION AND POLICY RECOMMENDATION

LIST OF TABLES Table 1: Regression results ofbanking crisis determinants: the panel eliminating

observations following end year of crisis 31 Table 2: Regression results of banking crisis determinants: the panel eliminating

observations following first year of crisis 33 Table 3a: Average marginal effects of determinants case for those countries without deposit insurance system -37 Table 3b: Average marginal effects of determinants case for those countries with deposit insurance system 37 Table 4: Banking indicators before the crisis erupt (December 31,2001 } 41

LIST OF FIGURE Figure 1: Breakdown structure of liability side of Uruguay banking system before crisis erupt 41 Figure 2: Peso/US$ exchange rate and evolution of US$ reserves -43 Figure 3: Predicted probability of banking crisis in Uruguay and Thailand with

specification 5a 44

CHAPTER I: INTRODUCTION 1.1 Problem statement:

The banking crises that erupted in US in 2007 are the latest in a series of such episodes that have been experienced by economies in various regions of the world in recent years In the 1990s, banking crises have occurred in Europe (the 1992-93 crises in the European Monetary System's exchange rate mechanism), Latin America (the middle of 1990s), as well as in East Asia (the 1997-98 crises in Indonesia, Korea, Malaysia, the Philippines, and Thailand) These crises have been costly in varying degrees both in lost output and in the fiscal expense to rescue fmancial sectors Through fmancial system, their significant spillovers spreads internationally and in a number of situations international financial assistance have required to mitigate their severity, costs and spillovers to other countries

In the wake of these recent crises, issue of what determinants affecting to occurrence of banking crisis has been a hot topic for economists in both developed and developing countries around the world Accessing this question is quite significant because it does not merely help authorities have confident base for their policies making process but it also is necessary to build up early warning system so that the crisis can be prevented beforehand However, understanding determinants

of banking crisis is not an easy task Financial innovations and the increased integration of global financial markets are driving forces that make us harder to deal with this issue Such factors push financial system to evolve rapidly and generate new risks for financial system One critical thing is that development of fmancial market seems so complicated and exceeds our knowledge and prediction What we can do as crisis comes is waiting for crisis wave and witnessing its effects on our life Thus, widely spreading influence of financial crisis, especially in banking sector, and its possible consequences obviously show important role of researches about determinants of banking crisis Even though this issue is not new, it always deserves to pursue

This paper, thus, will focus on examining theoretical paths that lead to occurrence of banking crisis From that place, early warning system is developed to deal with crisis However, because of data limitation, the scope of study narrows in developing countries during period 1974-2002

1.2 Research objectives:

Some objectives of the thesis are to identify:

(i) Determinants of banking crisis

(ii) Threshold values of determinants which minimize banking crisis

occurrence probability

(iii) Suggest policies recommendation to prevent banking crisis

1.3 Research question:

(i) What are determinants of banking crisis?

(ii) How do determinants affect on probability of banking crisis occurrence? (iii) How is marginal effect of each variable on probability of banking crisis evaluated?

1.4 Research hypothesis:

exchange rate would have positive effects on probability of banking crisis

(ii) Margial effect of each variable on probability of banking crisis is helpful

to reduce risk of crisis

1.5 Structure of the thesis:

The thesis includes 6 chapters:

Chapter 1: The introduction of the thesis

Chapter 2: Literature review

Fundamental components of banking cns1s such as key defmition, conceptual framework, theories and empirical studies will be discussed in this chapter Firstly, defmition of banking crisis is clarified Then, theoretical part will review the relationship between dependent variable {banking crisis) and explanatory

variables (fmancial variables, institutional variables, macro-economic variables) Empirical work is the last one mentioned in this chapter

Chapter 3: Model specification and data

The top priority of this chapter is to develop necessary steps to obtain the best unbiased model for data analysis steps later on To fulfill this purpose, advantages and disadvantages of potential models is initially analyzed Next, strategy to obtain model specification from originally conceptual one will be presented Other important issues relating to data such as data filler process, data sources are also mentioned in this chapter

Chapter 4: Regression result and testing hypothesis

In chapter 4, second objective of this paper will be fulfilled The chapter begins with analytic steps to access the best unbiased model Then, hypotheses relating to sources of banking crisis are tested to provide basis of policy recommendation The last part of the chapter will investigate marginal effects of the most significant variables on probability of banking crisis From that place, policymaker could proceed to maximize effectiveness of macro-policies on frequency of the crises

Chapter 5: Well-known crises and prediction accuracy of model

and Uruguay) in Latin America and Asia will be reviewed For convenient purpose, events are arranged in chronology with two obvious parts: prior to the crisis and afterwards the crisis Then, prediction power of model will be discussed after analyzing results and information collected from prediction model and descriptive evidences of crisis

Chapter 6: Conclusion and policy recommendation

CHAPTER 2: LITERATURE REVIEW

This chapter initially focuses on clarifying the defmition of banking crisis Then, various theories of banking crisis are introduced in order to get hypotheses on how determinants affect on probability of banking crisis occurrence Finally, the chapter closes with empirical studies of banking crisis that help the readers understand how other researchers have approached their research questions, what dataset they have collected as well as which models and statistical methods they have used

2.1 Key Definition:

Banking crisis: Effects of banking crisis is always huge and costly to resolve Despite economies may experience different kinds of crisis, one thing ruled out is that if the collective effects of financial collapse is large enough, the government is forced to intervene Therefore, Ergungor and Thomson (2005), as cited by Caprio and Klingebiel (1996b ), suggest that when central bankers think that a particular shock to the financial system could develop into systemic problem, and the monetary authorities begin to respond, banking system is considered as crisis In other words, banking crisis can be defmed in terms of behaviors of central banks Kaminsky and Reinhart ( 1996) share this perspective in his study by clarifying two policies of central bank in crisis period Under this view, banking crisis links closely with two types of events (1) bank runs that lead to the closure, merging, or takeover

by the public sector of one or more financial institutions (as in Venezuela in 1993); and (2) if there are no runs, the closure, merging, takeover, or large-scale government assistance of an important financial institution (or group of institutions) that marks the start of a string of similar outcomes for other financial institutions

As discussed by Kaminsky and Reinhart (1996), this approach is not without drawbacks It could date the crises too late, because the financial problems usually begin well before a bank is finally closed or merged; it could also date the crises too early because the worst of crisis may come later Moreover, data of banking crisis in

•

terms of their approach is available at limited level Kaminsky and Reinhart (1996) just list system banking crisis of more than 20 countries This makes it harder for researchers to expand the scope of study

Given data limitation, this paper, thus, follows definition of banking crisis developed by Caprio and Klingebiel (1996b) Accordance with his perspectives, banking crisis is the case in which the net worth of the banking system has been almost or entirely eliminated This creates something contradiction as banking problem in many nations is still to appear when a banking system has positive net worth However, Caprio and Klingebiel (1996b) indicate that the problem would be much easier if defmition of banking crisis above linked with insolvency of banking system More obviously, that is, bad loans are strong enough to "blow out" system's capital Based on data of banking sector in developing countries during 1980s, Caprio and Klingebiel (1996b) suggest that nonperforming loans must account for

at least 5 percent of total loans so that loan loss would be sufficient to wipe out banking system Under this definition, Caprio and Klingebiel (1996b) track more than 80 systemic banking crises around the world This dataset is updated through various studies including Caprio and Klingebiel (1996c) and Honohan et al (2005)

2.2 Banking crisis theory:

A number of studies have been developed around the world to provide insight view as well as explain logic behind crisis trouble in banking sector1 One interesting thing is that there is no hope to find full theory in any study The theory

is supplemented across study to study For that reason, in order to catch whole picture of banking crisis, it is quite useful to examine various studies This paper, thus, introduces banking crisis theory mainly referring to works of Ergungor and Thomson (2005) and Demirguc-Kunt and Detragrache (1998a) because they are the most updated and easily understandable version At the beginning, the theory starts

1

Eichengreen and Arteta (2000) list more than ten studies about systemic banking crisis, See Eichengreen

at classic view, next Demirguc-Kunt and Detragrache (1998a), finally Ergungor and Thomson (2005)

In the classic view, systemic banking crisis is recognized as a result of series

of macro-economy instability events Under this view, insolvency problems at one bank probably cause runs by depositors on other banks in the system If there is no protection or guarantee by monetary authorities, the fear of bank insolvency would spread through the banking system, and hence, liquidity pressure would lead a systemic collapse The occurrence of bank run needs to satisfy three conditions First of all, asymmetric information must exist so that depositors can not distinguish healthy banks Consequently, a run on weaker banks may transmit itself to healthy ones Secondly, sequential withdrawals must be paid at par until the bank is closed The final condition is that lack of precautionary plans for providing liquidity to the bank that faces the runs Considered from that viewpoint, systemic banking crisis is depositors' rational respond against information shock Ergungor and Thomson (2005) argue that the classic view fairly explains well the logic inside banking crises, but this theory is less useful to explain source of the banking crises of the last

20 years

In their research about determinants of banking crisis, Demirguc-Kunt and Detragrache (1998a) contribute to banking crisis theory by focusing on role of bank system in economy as well as problems on its balance sheet The theory shows that banks are financial intermediaries whose responsibility is to provide liquidity for real economy The bank serves its intermediary role by mobilizing short-term deposits from savers and lending that money, usually in long-term loans, to borrowers Therefore, as value of bank's asset does not match value of liabilities side, the bank gets into trouble A decline in value of asset probably comes from deterioration of borrowers' balance sheet This leads to borrower being unable or unwilling to pay back money (credit risk) The credit risk can be mitigated in many ways including diversification loan portfolio by making loan toward various risk sectors, or requiring loan collateral Actually, diversification loan portfolio is unable

to eliminate entirely credit risk for the bank that focus on a particular industry or operate in narrow market Also, loan collateral is not efficient to deal with default risk because its value is costly to monitor and usually subject to fluctuations Thus, credit risk always exist despite effort of bank's authorities This theory predicts that any adverse shock that influences to balance sheet performance of borrowers has positive correlation with banking crisis The shocks linking with banking crisis episodes could be cyclical output downturn, terms of trade deterioration, devaluation of asset price (Gorton, 1998; Caprio and Klingebiel, 1996c; Lindgren, Garcia, and Saal, 1996; Kaminsky and Reinhart, 1996)

Another determinant leading to deterioration of bank balance sheet mentioned in this theory is that rate of return on asset side mismatch cost for mobilizing capital on liability side Bank balance sheets should deteriorate because banks typically collect short-term deposits to fund long-term projects When interest rates in contracts are fixed, an increase in interest rates will quickly be reflected in the cost of (short-run) deposits, whereas it will take more time to pass on the increase in the cost of funds to borrowers Moreover, negative effect of high short-term interest rate probably passes to borrowers as the banks require higher lending rate Borrowers' balance sheet should be deteriorated, and hence, non-performing loans increase A surge in short-term interest rate may come from various ways such as tightened monetary policies, or financial liberalization (Gabis 1993), or fight against speculative attack (Kaminsky and Reinhart, 1996)

Demirguc-Kunt and Detragrache (1998a) also consider foreign deposit risk is likely a determinant that makes bank's balance sheet get worse As banks borrow foreign currency for domestic loans, a sharp devaluation of exchange rate threatens bank profitability In this case, banks eliminate exchange rate risk by following dominantly domestic deposits However, there is no hope that the risk disappears completely because foreign exchange rate risk is transferred to borrowers Unpredicted depreciation of domestic currency can deteriorate borrowers' profit, and thus, affect banks' balance sheet though an increase on bad loans

Like the classic view mentioned above, Demirguc-Kunt and Detragrache (1998a) agree that a trigger of banking crisis may come from a shock of information As cited by Diamond and Dybvig (1983), self-fulfilling may cause bank run if depositors realize the deterioration in quality of banks 'asset Run on one bank should not lead to bank crisis, but in condition, depositors can not discriminate healthy bank with strong balance sheet state Bank run turns into bank panic Demirguc-Kunt and Detragrache (1998a), therefore, argue that deposit insurance declines the risk of bank panic However, its correlation with banking crisis is still under controversy, as Kane (1989) has shown, the presence of deposit insurance usually come with moral hazard phenomenon that create strong incentive for the banks taking on more risk In the countries following financial liberalization and deposit insurance policies, occurrence of baking crisis's probability tend to be double because liberalization creates good conditions for taking on more risk

A sudden withdrawal of foreign deposits after a long period inflow 1s recognized as source of bank sector instability Demirguc-Kunt and Detragrache (1998a), as cited by Khamis (1996), capital inflows increase is due to the combination effect of two driving forces, i.e liberalization and high domestic rate

As foreign investors lost confidence or international interest rate increase, foreign depositors withdraw their deposit As result of this, domestic banks become illiquid

In countries where domestic financial investors try to flee national currency

as expectation of currency devaluation (speculative attack) that will take place soon,

of course, contributes to risk of crisis This idea is firstly mentioned in Krugman's study about model of balance of payment crisis Under his view, to avoid capital losses financicd investors tend to use national currency to purchase exchange rate According to Sachs et al (1996), the problem for the banks in that their deposits will decline dramatically due to massive withdrawal Consequently, bank run will occur

as the bank system can not meet such sudden demand of liquidity

(2005) show that periods of expansionary monetary and fiscal policy and typically

include some form of financial liberalization are main driving forces of banking crisis Internationally widening financial system, of course, is a highly risky action for banking system Giannetti (2006) discuss that after financial liberalization, corporations, which were previously dependent on bank loans, have with greater access to financial markets by using corporate bonds and commercial paper On the supply side, uninformed international investors rationally provide large amounts of funds at low cost and cool down demand of capital Under higher competition, the deposit market becomes smaller, and hence, banks have to pay more to attract depositors from their competitors Consequently, revenue decreases, cost rises, poor institutions will run at loss and they could face the prospect of closure by banking regulators

Another critical factor relating to occurrence of banking crisis is that a long period of expansionary monetary policy with low interest rate Excessive monetary growth usually comes with an increase in value of assets such as real estate, shocks, and consumer loans Banks respond rationally to these changes by increasing their market share to these markets because increasing price of assets raise a good signal that return of investment on the market is rising while risk is falling This tendency continues until one believes that the asset prices will continue to grow However, even when bankers realize that the trend is unsustainable, they may continue to lend with the expectation that they can extricate themselves from these loans and investments before the market collapse (overconfidence bias) In a study whose name is "Bubbles in Real Estate Markets", Herring and· Wachter (2002) provide another incentive to loan to risky investment as asset market booming, that is, large economic shocks occur so infrequently that banker often underestimate shock probabilities Despite overconfidence bias or underestimation of unpredicted shock plays a more crucial role in systemic banking crises One lesson can be learned is that the end result is the same In the absence of a shock, the lending continues, coupled with increasing asset prices and a booming economy As discussed by Ergungor and Thomson (2005), highly inflationary economy that comes from a long

period of rising asset prices and booming credit create strong incentive for the governments to intervene In this situation, higher interest rate and restriction on loan policies are introduced to cool down the economy As a consequence, economic growth slows, depressing asset prices and lowering borrowers' ability to pay Eventually, declining margins and increasing loan defaults erode banks' capital

A strong system of bank supervision Is acknowledged as a critical determinant that mitigates risk as well as damage of banking crisis Ergungor and Thomson (2005) indicate that bank supervisor's task is to supervise and maintain healthy banking system, but this purpose is not always fulfilled In some situations, budgetary and staffing constraints are main barriers limiting their control ability Another frequently mentioned reason influencing quality control of banking supervisor is political pressure, especially under booming period of economy For instance, as fmancial liberalization come with promoting economic growth The bank supervisor is not quite easy to close insolvent bank and regulate the bank pursuing highly risky projects because these projects are usually profitable Such reasons make bank supervision system get worse and trigger eruption of banking crisis In their work about theory of banking crisis, Demirguc-Kunt and Detragrache (1998a) also discuss that in countries with fmancial banking sector, weak law, and low quality of supervision system, banking crisis may occur because of "looting behavior" Bank manager is able to invest to risky project, but they can use the money for personal purposes as well

2.3 Empirical Studies:

In one research about models of economic and financial crisis, Roberto et al (2005) point out that one of approaches for prediction of banking crisis is logic or probit model with value of warning indicators The model requires a crisis dummy variable that acts as dependent variables and constructs explanatory variables that are fmancial variables, external sector and fiscal variables According to Roberto et

al (2005), this method is relatively popular than the others and has advantage of estimating statistical effect of various explanatory variables Moreover, it also allows estimation of probability ofbanking crisis occurrence in future

Demirguc-Kunt and Detragrache (1998a) had applied this approach for their study about determinants of banking crisis The multivariate logit approach allows the authors to link vector of n explanatory variables to probability of non-occurrence or occurrence of banking crisis The dummy banking crisis variable take value of 1 (crisis occurs) or 0 (no crisis) The parameters of logic function are obtained by maximum likelihood estimation Thus, coefficients obtained by this method do not have constantly marginal effect on dependent variables

Demirguc-Kunt and Detragrache (1998a) based on theory of banking crisis

to determine factors affecting on risk of crisis including macro-economic variables (growth of real GDP, change of terms of trade, rate change of exchange rate, real interest rate, rate change of GDP deflator, ratio of government budget surplus to GDP), fmancial variables (ratio of M2 to foreign exchange reserve, ratio of domestic credit to private sector to GDP, ratio of bank liquid reserves to bank assets, growth of real domestic credit), and institutional variables (deposit insurance scheme, real GDP per capita, quality of law enforcement) Then, macro data of developing and developed countries over period of 1980-1994 are collected for their model (data source mainly come from International Financial Statistic database) The study begins with list of IMF's members Some countries are eliminated regularly due to lack of data, data outliers, central planed economy, or economy in transition After the data filler process, the sample closes at 45 countries in speculation regression The main result of regression is that low growth rate, high inflation and weak macroeconomic promise a high risk of crisis High real interest rate relates banking problem too Explicit insurance and weak law enforcement also link closely with crisis risk

The feedback effects of dependent variable discussed by Demirguc-Kunt and Detragrache (1998a) is also worthy for reference Demirguc-Kunt and Detragrache

(1998a) argue that the crisis could cause various chain effects on behavior of explanatory variable that may confuse the model An illustration of this is the credit

to GDP ratio This ratio tends to decline after onset of banking crisis, and a decrease

of credit pulls down other variables such as GDP Another example for this problem

is real interest rate As theory noted earlier, banking crisis may come from low real interest rate that is probably a result of long time loose monetary policies or banking sector rescue operations Demirguc-Kunt and Detragrache (1998a), thus, suggest that the regression should be excluded all observation following crisis period An alternatives approach is that set of regression involves all observation with end date

of the crisis The shortcoming of this approach is the end date of crisis is not easy to

be affirmed because the end date may be determined too late or soon Kunt and Detragrache (1998a) believe that the probability that crisis occur in one country had experienced problem is higher than those that have not ever experienced For this reason, different estimators are added to the model such as quantity of past crisis, duration of last crisis

Demirguc-Another technique of predicting banking crisis mentioned by Roberto et al (2005) is signaling approach The approach requires to build a threshold value for each explanatory variables A concern for banking crisis will rise if indicator variables exceed critical thresholds and vice versa Ideally, the threshold value should be chosen so that indicator variables exceed its threshold ahead of crisis One of difficult problems of this method is how to build threshold value as benchmark for early warning system Kaminsky and Reinhart (1996) access this matter by minimizing signal-to-noise ratio The calculation of this ratio is obtained

by using expression as follow

[signal - to - noise _ ratio ] =

T2: variable issues signal but crisis does not occur

This ratio is defmed as the ratio of the false rate of crisis predictions relative

to the successful alarm rate Thus, the smaller the ratio become, the more effective the variable predicts crisis This logic leads to next step as Kaminsky and Reinhart (1996) replace various thresholds to find out minimum value of the ratio Roberto et

al (2005) argue that this method is very easy to access source of crisis However, beside advantage side, it ignores strong correlation among variables and provides no framework for statistical testing or calculation of crisis probabilities in the future, and is still open to misclassification errors that can bias the conclusions of the analysis

2.4 Chapter summary:

In short, possible determinants of banking crisis could be classified into 3 separate large groups including macro-group (deterioration of borrower's balance

(speculative attack, financial liberalization), and lastly the institutional group (quality of banking system supervisor, deposit insurance) Though various fmancial channels, such determinants contribute to risk of crisis in many ways and sometime

we see the crossroads in that two or three determinants show simultaneously their effects on one trigger-crisis variable (i.e real interest rate) This makes the problem become puzzled and limits considerably our ability to approach separate effect of each determinants Thus, further discussion about the methods that address power of each determinant on banking crisis should be focused in next chapter Besides, reviewing empirical studies also provides at least two solutions (logit model and signal-to-noise ratio) to approach objectives of this paper One important thing is that each method has its own advantages and disadvantages Therefore, model choice should pe also carefully considered in next chapter

UEff -ISS

a._ , i nne t.a lw

CHAPTER 3: MODEL SPECIFICATION AND DATA

In chapter 2, hypothetical relationship between banking crisis and various macroeconomic variables was discussed in banking crisis theory Besides, this chapter also reviewed previous empirical works about banking crisis around the world

Based on background mentioned in chapter 2, model specification and data chapter is going to suggest suitable model, data collection, statistical method and other related sections to deal with research questions Outline of this chapter, thus, will be arranged into 3 parts involving model choice, model specification, and data source

3.1 Model choice:

Banking crisis theory of previous chapter has fulfilled a very fundamental problem of this study, that is, it clarifies likelihood of various macro variables and risk of crisis This is particularly significant because it provides framework of applying statistical methods for prediction of crisis occurrence

Review of empirical works has opened 2 alternatives to reach the second objective of this study, identifying threshold value of each variable which minimizing banking crisis occurrence The first method is signal-noise ratio As noted earlier, minimization of signal-noise ratio by replacement of different threshold allows Kaminsky and Reinhart (1996) to limit frequency of crisis Even though this approach does not fulfill completely second objective (marginal effect

of each variable on probability of crisis are not issued), its result is relatively acceptable (banking crisis occurrence minimization) Perhaps the best approach is multivariate logit model This model provides background for estimation of marginal effect of each variable on crisis occurrence so that probability of crisis can

be reduced Moreover, logit model receives a great support from econometric software such as Sata, Eviews, ect Therefore, data process is always shortened in few minute even dataset contains thousands of observation For reasons discussed above, multivariate logic model is employed to access research questions

In accordance with Gujarati (2003), logit model is based on logistic distribution function The model can be formulated mathematically as follow:

+e

We rewrite formulation above as

oftranquility state (1-Pi) is

Lis called the logit, and hence the name logic model for equation (3.2)

One thing should be remembered is that an increase of one unit parameters does not reflect change of banking crisis probability Equation (3.2) shows that the coefficients indicate the effect of one-unit change of explanatory variables on log of odds ratio Therefore, the marginal effects of each variable changes over time

depending on start level of the crisis and their coefficients Besides, the form of cumulative logic function (S-shape) reveals an interesting thing, that is, a change of explanatory variables has different effect on probability of crisis occurrence depend upon initial probability State differently, if initial probability is low or high, marginal effect of independent variable on probability is not considerably, but the same marginal effect may have larger change on the probability if the countries are

on middle of range

3.2 Theoretical model and model specification:

In his book about econometric, Gujarati (2003) states that construction of any regression model always depends on a combination of economic theory, basic human behavior, previous research, and past experience However, specification error may occur because of two following reasons: (I) missing relevant variable or including irrelevant variable or (II) incorrect function form To access type II of specification bias, the author assumes that logistic function is the correctly specified model for probability function of banking crisis For type I, the author will apply general-to-simple modeling strategy to get the model that can provide qualitative and confident results The key point of this strategy is that the first model should include all possible regressors Then, the best model will be obtained after insignificance variable elimination process According to Gujarati (2003), general-to-simple modeling strategy leads to presence of irrelevant variable in the model However, this problem does not raise concern because the model still gives unbiased and consistent estimates of coefficients in the true model Furthermore, error variance is correctly estimated and hypothesis-testing methods are still valid

A disadvantage of this strategy is that addition of unnecessary variables will cause less precise in estimators In contrast, if the model excludes a relevant estimator, parameters become biased, the error variance is not correctly estimated, and hypothesis-testing methods turn to be invalid Under this view, general-to-simple modeling strategy seems to be the best choice Thus, general framework for

predicting the occurrence of banking crisis would compnse the all possible variables that we have Based on banking crisis theory and data availability, the model is suggested as follows:

4 = L{_!LJ = ~ + j3prowth+ A Rea/interest+ AEx+ ~Inflation+-f3sTOT + fi,M2/Reserve

1-P;

+ jJ.,Deposit+ f3sCreditgrowh+U;(3.3)

Where Growth is vector of growth factors, Real interest is vector of interest rate factor, Ex is vector of exchange rate factors, Inflation is vector of inflation factors, TOT is vector of terms of trade factors, M2/Reserve is vector of speculative attack factors Deposit is vector of deposit insurance factors Credit growth is vector

of liberalization factors, and ui is vector of summarizing unobservable factors that contribute to risk of crisis

3.2.1 Growth:

The first variable mentioned in my model is rate of growth of real GDP Theory of banking crisis indicates that negative shocks to whole economy affects fmancial situation ofbank's borrowers Therefore, this will increase non-performing loan and facilitate banking crisis occur In many situations, diversifying loans and investments to various sectors is wise response to reduce risk, but this method also takes no effect if whole economy is under distress To capture macroeconomic shock, this study use rate of growth of real GDP

3.2.2 Real short-term interest rate:

The second variable in my model is real short -term interest rate As discussed earlier, a sudden increase in real short-term interest rate will deteriorate balance sheet of both borrowers and lenders Therefore, it affects positively to risk

of crisis According to theory, a surge in short-term interest rate informs that the bank system is under liquidity pressure which is probably consequences of speculative attack or long-term expansionary monetary policies or fmancial liberalization In other words, using this variable to predict banking crisis is very

helpful because it captures many economic determinants that cause banking crisis

It is strongly believe that this ratio will get highly significant

3.2.3 Exchange rate:

Exchange rate only increases risk of crisis in condition liability side of banks

devaluation of domestic currency probably doubles their debt obligation Consequently, fmancial sectors may be breakdown To test this hypothesis, this paper uses the rate of devaluation of exchange rate

3.2.4 Inflation:

Another indicator introduced in my model to predict probability of crisis occurrence is inflation Demirguc-Kunt and Detragrache (1998a) use this ratio as a proxy for government mismanagement that affects economy though various ways But perhaps, the most convincible reason is inflation is highly associated with nominal interest rate that causes deterioration on balance sheet of banks as well as the firms Accordingly, it contributes risk of crisis

3.2.5 Terms of Trade:

As mention earlier, a suddenly shock of terms of trade against bank's customers also raises a concern for banking crisis When terms of trade turns dramatically, bank's customer find it harder to service exiting loans This index is particularly highly correlated with financial crisis in developing countries in which the banks just focus on few industrial sectors Caprio and Klingebiel (1996c) point that 75 percent of the countries with banking crises in their database, the terms of trade fell by more than 10 percent in the years preceding the crisis

3.2.6 Speculative attack:

Sachs et al ( 1996) suggest that perfect foresight of exchange rate devaluation probably trigger a run on banking system From this perspective, low foreign reserve creates high incentive for domestic currency holders to convert their

domestic currency into foreign currency The banking system will be in danger if this tendency leads depositors to withdraw their money out of the system As a consequence, the banks turn to be insolvency This logic leads Demirguc-Kunt and Detragrache (1998a) to choose ratio M2 (currency plus saving deposit in commercial bank) to foreign reserves to evaluate effects of such speculative attack

on banking system Calvo (1996) states that this ratio is a good determinant for predicting balance payment crisis

The next choice is deposit msurance Demirguc-Kunt and Detragrache (1998a) point out that deposit insurance may exist in 2 types: explicit or implicit Deposit insurance is called explicit if the bank purchase full or partial insurance for its depositors or the government commits to guarantee benefit of depositors despite how worse bank's balance sheet is performing In the case, the depositors believe that the government will correct bank problems as banking crisis appears, that is, implicit one The author defmes appearance of deposit insurance as explicit one This variable takes value of 1 if a country establishes explicit system and 0 for otherwise As pointed above, the relationship between deposit insurance and banking crisis is ambiguous It is convinced that the presence of deposit insurance reduces dramatically fear of bank panic However, effect of deposit insurance would

be negative if moral hazard phenomenon exists

3.2.8 Liberalization:

The theory of banking crisis shows that after financial liberalization deposit market become comparative strictly as uninformed international investors willingly provide large amounts of funds at low cost This turns banks to increase cost as well

as revenue to keep its consumers As profit reduce, the bank with poor balance sheet can not pay their contract on time, investors stop holding bank debt and banks default Allegret et al (2003) identify 3 channels though which fmancial liberalization could influence on banking stability including fmancial system

opemng, interest rate deregulation, and bank loan deregulation This means that financial liberalization can be evaluate though effects of these channel on financial markets Galbis (1993) found that real interest rate may be good proxy for fmancial liberalization because interest rate deregulation usually leads to sharp rise in real interest rate However, as mentioned in the theory section, real interest rate could be result of various determinants such as speculative attack (Kaminsky and Reinhart, 1996), or restrictive monetary policies (Ergungor and Thomson, 2005) Therefore, real interest rate may not evaluate well fmancialliberalization as such phenomenon take place To capture fmancialliberalization process, this paper evaluates effect of the two remaining variables by using related credit variables Pill and Pradhan (1995) suggest ratio of domestic credit to GDP Demirguc-Kuntand Detragrache (1998a) interchange domestic credit to GDP and domestic credit growth Because of data limitation, this study uses domestic credit growth as an explanatory variable for fmancialliberalization in the model

3.3 Estimation strategy and statistical tests of the model:

best way to get best unbiased model More specific, this strategy could be classified

in 4 steps

- Use economic theory, previous research, and experience to specify a general model that includes all possible relevant regressors

- Estimate the model

- Leave any of parameters are statistically insignificant out model This step should begin with least significant variable and estimate again The effect of eliminating insignificant variable probably improves significance level of remaining ones Therefore, variables are omitted one-by-one

-Use Chi-square statistic to test the final logic model

The final model after testing with Chi-square statistic test will be used to clarify hypotheses as well research objectives mentioned in previous sections As stated by Demirguc-Kunt and Detragrache (1998a), onset of banking crisis might

generate "feedback" effects of banking crisis itself on remaining explanatory variables This must puzzle our model when we would like to observe how independent each variable contributes to risk of crisis To deal with problem, Demirguc;-Kunt and Detragrache (1998a) leave out all observations following banking crisis One disadvantage of this approach is that the panel gets very small because many observations are eliminated Another method to access feedback effects is that the full duration of banking crisis must be identified, and then, our regression model includes all observations that following end date of crisis The unnecessary drawback is end date of banking crisis is still under controversy Therefore, it is not quite easy to identify end date For these reasons, this study will run 2 regressions with 2 different approaches on banking crisis The fmal conclusion will be provided depends on the comparison between 2 regression results

G~jarati (2003) suggests three criteria to access quality of logic model: statistic, MacFadden R square, and likelihood ratio (LR) statistic Because logic model is e:stimated by maximum likelihood method, instead of using to t statistic to measure significance of a parameter, this study introduces Z-statistic One of popular measure of goodness of fit is R square However, it does not take effect on binary regressand models Eviews provides an alternative, the MacFadden R square Like R square, this test also ranges from 0 to 1 and it is calculated as

L(O)

all parameters are equal to 0 and L(/3*) is the value of log likelihood function has been maximized Likelihood ratio statistic is the next test to test significance for whole model Like F-test in linear regression, LR is used to test null hypothesis that all slopes are equally zero The last test mentioned in this part is Chi-square statistic test This test helps us to make sure whether k explanatory variables should be dropped from the logit model Chi-square statistic test is formulated as

xi= -2ln(LR I LuR) = -2ln(lnLR -lnLuR) LuR is the maximum of likelihood function

when maximized with respect to all the parameters and LR is the maximum when

is greater than the critical

not simultaneously equal zero

3.4 Data sources:

The most updated database of Caprio-Klingebiel (1996b) about systemic banking crisis presented by Honohan et al (2005i is the main source to obtain data for banking crisis variable This paper presents a new database on the timing of

systemic banking crises for the period 1974-2002, with specific information on policy respond of over 40 banking crisis around the world

Demirgii~-Kunt et al (2006b), and International association of deposit insurance (2008) are mainly data source for deposit insurance dummy variable

Demirgii~-Kunt et al (2006b) provide very specific on deposit insurance in each country including the year of insurance system is introduced and names of the governing institutions, along with sources for our data on deposit insurance coverage and other design features Moreover, available data in their study has been complemented by contacting individual deposit insurers or related agencies (such as

with the list of more than 180 countries following explicit and implicit guarantee Data for this study continue to be updated by survey of International association of deposit insurance (IADI) 2008 IADI conduct a questionare of more than 160 questions covering 14 areas of deposit insurance such as background information,

infrastructure, etc Based on partial and completed reports of IADI 's survey, 11 countries are added to explicit insurance group involving Hong Kong, Jordan, Malaysia, Morocco, Nicaragua, Paraguay, Russia, Slovenia, Uruguay, VietNam,

2

The electronic version of the banking crisis database is available at

http://wwwl.worldbank.org/finance/html/database sfd.html or http://www.luclaeven.com/Data.htm

and Isle of Man Eventually, dataset of explicit deposit insurance finally close with data of 180 countries

Data from International Financial Statistic (IFS) is mainly used for other variables in this study IFS has been published monthly since January 1948 In

1961, the monthly was supplemented by a yearbook, and in 1991 and 2000,

country tables for most Fund members, as well as for Aruba, the euro area, and the Netherlands Antilles The country tables normally include data on a country's exchange rates, fund position, international liquidity, money and banking accounts, interest rates, prices, production, international transactions, government accounts, national accounts, and population Selected series, including data on international reserves, and international trade, are drawn from the country tables and published in area and world tables as well The full sample details and guide for international

http://www esds.ac uk/international/support/user guides/imf/ifs.asp

In this study, the author has drawn out from IFS published in Apirl 2009 including data of exchange rate, inflation, real interest rate, growth rate, private credit over GDP, M2 over GDP Information on the focused variables is represented

in appendix 1

3.5 Data filler process:

included in the sample Data filler process begins with list of 180 countries collected for deposit insurance variable Firstly, 75 countiies are excluded because

of lack of macroeconomic or fmancial data Next, 22 countries belongs to OECD are left out because this study just focus on developing countries Accordingly, 3 countries (Argentina, Brazil and Bolivia) are eliminated because 2 variables (inflation and real interest rate) are outliers After elimination process, sample for our specification model is limited at 80 developing countries The list of countries and data filler results is represented in Appendix 2

CHAPTER 4 REGRESSION RESULT AND TESTING HYPOTHESIS

4.1 Accessing the best unbiased model:

When constructing any regression model, one of the critical purposes is to find out the best unbiased model for final conclusions The path to final model needs a harmonization of economic theory and statistical test In other words, the expected model must satisfY not only theoretical base ( p 's get expected sign), but also necessary conditions of statistic test Hence, before estimation steps it would be useful to review potential sign of coefficients

According to banking crisis theory, growth rate of economy is quite important because a negatively macroeconomic shocks to whole economies in which the bank system is not internationally diversified usually leads to a surge of non performing loans, and hence, the crisis occurs Put differently, it is expected

expansionary monetary policies and financial liberalization potentially result in race

of interest in short run among banks to face with liquidity pressure Thus, short-term interest rate gets high expectation of explaining bank distress, this means f3 2 relate positively to dependent variable As international capital market lost confidence to the local market, there would be high probability that a sudden number of depositors run out bank system and contribute to financial instability In this case,

deregulation of financial sector, which increase competition in the banking system Under competitive pressure, the poor banks can not fmish their obligation on

liberalization increases bank fragility As a result, f3 4 should be positive In theory section, inflation is introduced not only as variable that captures government's macroeconomic management ability, but it also is a part of nominal interest rate Therefore, its coefficient ( f3 5) shares the same trend with ·real interest Rate change

of exchange rate and terms of trade are, in turn, determinants that promise high