INTRODUCTION
Problem statement
The banking crisis is a persistent issue in modern economies, extensively studied from various perspectives Research has evolved through three main trends: the initial qualitative descriptions by Friedman and Schwartz (1963) regarding historical U.S crises, followed by econometric analyses utilizing panel data based on sufficient banking crisis observations, and culminating in studies post-2007 global financial turmoil These research trends have identified crucial macroeconomic and banking indicators, such as reserves, current accounts, and real exchange rates (Kaminsky et al., 1998) While logistic regression has been instrumental in predicting crisis signals and timing, its reliance on quantitative models has introduced noise that may undermine its effectiveness, prompting further investigations into new methodologies and critical variables.
As suggested, there have been many criteria to help researchers with banking crisis identification Amongst, money market pressure index from the work of Hagen and
Ho (2007) builds upon the work of Eichengreen, Rose, and Wyplosz (1995, 1996a, 1996b) regarding currency crises, offering a useful framework for understanding and collecting data on banking crisis symptoms This index identifies periods when banking systems face liquidity issues by analyzing the simultaneous occurrence of high demand for central bank reserves and fluctuations in short-term real interest rates Ultimately, the index serves as a criterion to determine the presence of a crisis within the analyzed scope.
Banks relevant data, to some extent, seems to be difficult to obtain precisely due to h
This research utilizes macroeconomic indicators to assess banks' health, as highlighted by Quagliariello (2008), who noted that analyzing these variables aids banking supervisors Key macroeconomic and financial variables, including inflation, monetary base growth, depreciation, real interest rates, private credit growth relative to GDP, deposit growth over GDP, and M2 growth over reserves, are examined in line with recommendations from previous studies (Hagen and Ho, 2007) Additionally, institutional signals (Kaufmann et al., 2008) are incorporated to predict vulnerability and crisis occurrence, enhancing the model's effectiveness as suggested by Kaminsky et al (1998) Building on Breuer et al (2006), this research evaluates the influence of six updated world governance indicators (Kaufmann, 2013)—voice and accountability, government effectiveness, political stability, rule of law, regulatory quality, and control of corruption—on the banking system's crisis dynamics Finally, the inclusion of a 12-month lagged term of banking crises in the regression model (Falcetti and Tudela, 2006) provides a significant assessment of these relationships.
Most research on banking crises focuses on identifying the causes of their occurrence, often overlooking why some countries experience stability during certain periods While understanding and predicting crises is crucial, future studies should also explore the significance of non-crisis situations within the banking system, as these scenarios play a vital role that is frequently undervalued (Kauko, 2014).
Despite numerous studies on banking crises, there is a noticeable lack of research that simultaneously examines the health of government, macroeconomic factors, and financial conditions within a single model This gap highlights the need for a comprehensive approach to understanding the interconnectedness of these elements in the context of banking stability.
This thesis aims to utilize a combination of the MMP index approach and updated data from the IMF-IFS for the years 2001-2010 to analyze the overall phenomenon of banking crises It focuses on the impacts of the macroeconomic environment, financial conditions, and institutional indicators This approach is expected to yield valuable insights for banking crisis analysis and enhance awareness among authorities regarding banking sector management and broader economic implications.
Research objective
This thesis aims to provide updated research on benign periods of banking systems by analyzing banking crises It focuses on identifying macroeconomic, financial, and institutional factors that help explain the occurrence of banking system crises.
Research question
Which are the macroeconomic, financial and institutional indicators that provide awareness for the crisis time of banking system?
Structure of the thesis
After the finish of Chapter 1 about thesis introduction, the rest of this thesis will be categorized as following chapters:
Chapter 2 introduces banking crisis definition, relevant literature reviews of trends of banking crisis researches, money market pressure index which will be applied for banking crisis dependent variable identification
Chapter 3 states the methodology, model choice and specification and data scope used This chapter also gives readers clear arguments on explanatory variables used, suggested statistical diagnostics of significance of model and variables Simultaneously, data scope and sources together with model conceptual framework and analytical framework are also declared
Chapter 4 interprets results and findings of thesis regression model h
Chapter 5 concludes with policy recommendation, thesis limitation and further research suggestion h
LITERATURE REVIEW
Defining banking crisis
According to the IMF's definition from 1998, a banking crisis occurs when bank runs and widespread failures force banks to halt the convertibility of their liabilities, leading to significant government intervention in the banking system.
In another work of Demirgtic-Kunt and Detragiache (1998), the concept of banking crisis was defined as event method whose conditions are that one or the entire following phenomenon holds:
1) The existence of at least 10% of the ratio of non-performing assets over total assets in the banking system
2) Cost of the rescue packages reached at least 2% of GDP
3) Extensive nationalization of banks due to banking sector problems
4) Governmental regulation of deposit guarantee, large-scale bank runs, long holidays of banks, deposit freeze
The definition of a banking crisis has notable limitations Primarily, the costs associated with government rescue packages remain ambiguous until a crisis unfolds, resulting in delayed identification of the issue Additionally, measures such as extended bank holidays and the nationalization of banks tend to occur only after the economy has already suffered significant damage.
The extent of government intervention in banking crises is critical, as authorities may act early or late, leading to uncertainty in identifying accurate intervention dates (Caprio and Klingebiel, 1996a) Furthermore, the event method used to classify crises relies on significant market events, which can introduce bias due to selective event identification This bias limits the effectiveness of determining the underlying causes of banking crises and undermines their analytical value.
With the attempt to contribute an alternative identification for banking crisis, the money market pressure index (MMP) was built up in the work of Hagen and Ho
In their 2007 analysis inspired by Eichengreen's 1995 work on currency crises, Hagen and Ho define a banking crisis as a period marked by excessive liquidity demand in the money market This concept arises from the traditional view that short-term interest rates negatively influence the banking sector's demand for central bank reserves They propose that a banking crisis is characterized by a significant increase in the aggregate demand for these reserves, which can be examined through three key reasons.
- Banks confront with increasing non-performing loans and/or significant decline in bank loans quality leading to illiquidity, hence, a rising in demand of reserves to retain liquidity
- When sudden withdrawals occur, there will be a pressure for banks to deal with interbank market and central bank to be refinanced
- Government bonds and other more guaranteed assets are favored by financial institutions rather than lending to those in troubled leading to “a drying up of inter-bank lending”
With the attempt to react to the increasing demand for reserves, central bank, who is the last lender, will enact two basic policies on either bank reserves targeting or h
Short-term interest rate targeting can lead to two scenarios: an increase in short-term interest rates or an injection of reserves into the banking system through open market operations (OMO) or discount window lending The presence of sharply rising short-term interest rates or a significant amount of central bank reserves indicates that the money market is under considerable pressure Consequently, the money market pressure index can effectively highlight vulnerabilities within the banking sector.
The index is determined by the weighted average of fluctuations in the ratio of reserves to bank deposits, along with variations in the short-term real interest rate, where the weights correspond to the sample standard deviations of these two components (Hagen and Ho, 2007).
The reserves to bank deposits ratio, denoted as "where," increases during periods of high tension in the money market, particularly when the central bank injects reserves into the banking system or when depositors withdraw funds The variable "r" represents the short-term real interest rate, while the terms and their respective standard deviations reflect the variability of these two components.
The judgment for banking crisis (BC) will be shown below:
This chapter concludes the discussion on banking crises by outlining three key research trends that offer a comprehensive overview of existing empirical studies Additionally, it includes a thorough review of the methodologies and findings from the research conducted by Hagen.
Ho (2007) as a conjunction for the Chapter 3.
Trends of banking crises researchtogether with crises mechanism
The history of banking system fragility highlights significant crises, beginning with the qualitative analysis by Friedman and Schwartz in 1963, which detailed the US banking crisis This analysis paved the way for further research, culminating in the establishment of the first comprehensive banking crisis database by Caprio and Klingebiel in 1996, which cataloged various banking crises and their impacts.
Banking crises, as highlighted in the seminal works of Demirguc-Kunt and Detragiache (1998) and Kaminsky and Reinhart (1999), typically arise from similar underlying factors, including economic health, government stability, and the inherent fragility of banking systems, often exacerbated by external contagion effects Despite these commonalities, each crisis period necessitates distinct assessment methods tailored to the available data, analytical techniques, and statistical software support This article aims to provide a comprehensive overview of banking crisis research, detailing existing trends and methodologies, while also discussing Kauko's (2014) categorized suggestions on these approaches.
Description of specific historic events is mainstream of the first trend of banking crisis analyses The below words introduce some authors of this trend
Friedman and Schwartz (1963) in the work of “Monetary history of the United
Between 1867 and 1960, observations indicated that an increase in short-term interest rates coincided with a decline in the deposit-to-currency ratio, leading to bank runs According to Waldo (1985), banks often resorted to prematurely selling long-term securities to ensure withdrawal capabilities, resulting in higher yields on short-term assets This tradeoff between maintaining withdrawal readiness and liquidating securities before maturity forced some banks to default on deposits, prompting depositors to convert their funds into cash as a self-protective measure against potential bank runs The banking crisis of October 1930 exemplified this issue, as bank failures triggered a public rush to cash out deposits, ultimately leading to a widespread collapse of the US banking system by December 1930.
9 period from March to June 1931, the second wave of crisis occurred more severe because the banking system had been unhealthy during the former crisis
Herrala (2011) contributes a description on Finnish crisis within the scope of 1865
In 1998, a study analyzing the profitability of banks in Finland revealed that the events leading to the banking crisis align with previous international research The study defines a banking crisis as the occurrence of negative profitability within the banking sector and seeks to identify critical characteristics and crisis cycles that may adversely affect financial stability By utilizing available statistical data, the research compares findings from international banking crises and examines indicators that precede a banking crisis, focusing on periods with healthy financial conditions Key factors influencing bank profitability, including total assets, real GDP growth, investment levels, inflation, export volume changes, stock money, exchange rates, interest rates, and the ratio of bank deposits to loans, are also analyzed.
Gorton (1988) provided econometric evidence on the determinants of banking panics in the U.S National Banking Era (1863-1914) by analyzing banking panics and depositor behavior The study highlighted the role of moral hazard and agency issues, emphasizing that changes in depositor risk perceptions could trigger banking panics Key indicators considered included the deposits ratio and liabilities.
Such econometric based researches made a link between the first trend and the second trend which will be introduced below h
Recent econometric research on banking crises has utilized panel data to analyze macroeconomic and financial factors, relying on a wealth of observational data However, much of this analysis predominantly focuses on developed countries, often overlooking emerging economies Moreover, the studies typically examine banking crises in a binary manner, categorizing situations strictly as either crisis or non-crisis, which reflects a dichotomous approach highlighted in various academic discussions.
The phenomenon of banking crises has garnered significant attention, particularly through the pioneering work of Caprio and Klingebiel (1996a, 1996b), who established the first comprehensive banking crisis database, detailing crisis dates, countries, and economic variables alongside policy observations Their research primarily focused on bank insolvency, utilizing data from 69 countries between 1970 and 1996, including factors such as GDP, inflation, and trade balances In-depth expert interviews were conducted to document crisis episodes However, Caprio and Klingebiel emphasized the need for additional bank performance and development indicators to enhance the accuracy of predicting crises for individual banks and the broader financial system Furthermore, they suggested that political economy research on bank insolvencies could serve as a valuable resource for governments.
The twin crisis refers to the simultaneous occurrence of a currency crisis and a banking crisis, analyzed through an econometric lens This concept utilizes a signal-to-noise approach to assess crisis situations, determining whether alarm signals are triggered based on threshold values of specific indicators.
This study utilizes a “by-indicator basis” approach, as proposed by Kaminsky and Reinhart (1999), to analyze banking and twin crises across 20 countries from 1970 to mid-1995 By employing 16 indicators from the financial, external, real, and fiscal sectors, the research aims to minimize the signal-to-noise ratio, although it acknowledges potential drawbacks related to incorrect signaling The selected countries include both industrial and developing nations, such as Denmark, Finland, and Argentina, allowing for an examination of 76 currency crises and 26 banking crises, as documented by Caprio and Klingebiel (1996) Additionally, out-of-sample testing was conducted on the twin crises in Asia during 1997, providing valuable insights for authorities regarding early warning signals.
Dermirguc-Kunt and Detragiache (1998) conducted a comprehensive study across developed and developing countries from 1980 to 1994 using a multivariate logistic model to identify key factors leading to systemic banking crises Their findings revealed that such crises were more likely to occur in weak macroeconomic conditions characterized by high inflation and low growth Additionally, elevated real interest rates and vulnerable balance of payments significantly contributed to banking sector issues The research also highlighted that institutional factors, including the presence of deposit insurance and inadequate law enforcement, posed risks to banking systems Furthermore, the study underscored the importance of low GDP growth, which could jeopardize the banking sector's stability While banks, as financial intermediaries, are inherently inclined to take risks, they may overlook domestic credit risks linked to economic fluctuations.
The banking sector's activities in developed countries have both benefited and pressured developing nations to enhance their regulatory frameworks, aiming to mitigate the fragility induced by cross-border banking volatility There is ongoing debate regarding the impact of financial liberalization on banking stability, with studies indicating weak evidence of increased banking crises during periods of controlled real interest rates However, limitations in the estimation models and the interplay between macroeconomic, institutional, and financial factors necessitate further research Future studies should focus on structural banking indicators, including bank capitalization, market concentration, competition, interbank and bond market liquidity, ownership structures, and the effectiveness of regulatory supervision.
Broad new and old samples of banking crisis over different countries have been combined in some researches However, once again, these analyses on focused on developed countries
Bordo and Meissner (2012) conducted an analysis of 14 advanced countries from 1880 to 2008 to explore the relationship between credit booms, inequality, housing policy, and banking crises Their research indicates that credit booms, while not fully understood, likely play a significant role in contributing to banking instability Utilizing a logit model with and without fixed country effects, the study found a positive correlation between credit booms and banking crises Notably, although the one-year lag term of credit growth showed a low probability of triggering a banking crisis, the findings revealed a significant positive relationship indicating that banking crises tend to follow an increase in real credit growth, particularly when considering lagged terms.
Research indicates that factors like rising real income and declining interest rates are crucial for understanding credit booms However, the existing data reveals limited evidence that increasing income inequality and housing redistributive policies significantly impact the likelihood of financial crises.
Schularick and Taylor (2012) explored the significant impact of previously overlooked credit expansion on the global economy, emphasizing that the stable relationship between money and credit established post-Great Depression and World War II continues to influence modern crises They noted that factors such as the rise of fiat money and the banks' role as lenders of last resort have contributed to this credit expansion Over time, structural changes in the financial system have elevated the importance of credit in the macro-economy, making it increasingly vital Despite debates regarding the constructive role of credit in monetary policy, historical lessons suggest that credit accumulation poses risks that have been largely ignored by researchers and policymakers The potential for credit booms to exacerbate future financial crises raises concerns, although some argue that credit expansion can drive real economic growth Ultimately, while the predictive power of credit booms remains contentious, the historical context of credit expansion and its link to financial instability warrants further investigation into its role in the macro-economy.
Jorda et al (2011) analyzed the financial fragility in the relationship with external imbalanced situations of the economy such as deteriorations of the current account, h
This study analyzes the growth of loans, interest rate volatility, inflation, and GDP growth over 140 years across 14 developed countries using a logistic country fixed effect model It combines descriptive statistics with a logit model to explore financial fragility The findings reveal that loan growth significantly accelerates crises on both national and global levels, while deteriorating current accounts contribute to crisis escalation A suppressed natural interest rate signals impending crises, particularly during four major global events: 1890, 1907, 1930-1931, and 2007-2008 Additionally, real interest rates and inflation exhibit similar predictive trends The research underscores the need for policymakers to focus on the buildup phase of crises by monitoring external macroeconomic imbalances, highlighting the significant interplay between credit growth and current accounts as predictors of financial instability.
Money Market Pressure (MMP) Index (Hagen and Ho, 2007)
The MMP index is utilized to identify banking crises by applying two key criteria: first, the index must exceed the 98.5 percentile of its sample distribution for each country; second, the index's growth rate must be at least 5% According to Hagen and Ho (2007), the first criterion ensures that only significant episodes are classified as crises, while the second criterion accounts for countries without crises during the sample period Empirical evidence suggests that relaxing the first criterion may result in an overestimation of crises, whereas tightening it could lead to the omission of genuine crises.
The analysis reveals that altering the percentile values has diminished the explanatory power of the regression model, particularly at the 95th percentile, while the 99.5th percentile remains largely unaffected The stricter conditions of the second criterion have resulted in the omission of some genuine crisis episodes It is important to note that the definition of a crisis is country-specific; thus, applying a uniform definition across all countries by pooling data may lead to overlooking true crises, especially in nations with lower fluctuations in the MMP index Ultimately, the use of percentiles is favored over multiple standard deviations for the first criterion due to the non-normal distribution characteristics of the MMP index.
The definition of banking crises presents several drawbacks: firstly, while banking crises are often seen as asset-driven, this thesis does not address the increased demand for reserves resulting from deteriorating bank assets Secondly, the MMP index may not be applicable in countries where interest rates are controlled by central banks; however, it remains advantageous as it operates independently of interest rate flexibility when central bank policies are market-based Lastly, the MMP index effectively indicates the onset of a crisis but fails to determine its conclusion, as identifying when a banking crisis ends is a complex issue in empirical literature (Kaminsky and Reinhart, 2000), which falls outside the scope of this thesis.
The analysis covered a sample of 47 countries from 1980 to 2001, selected based on data availability from the IMF Notably, Argentina and Brazil were excluded from the study due to their unusually high inflation and interest rates.
The study utilized 21 indicators across 697-726 observations, identifying 34-38 crisis episodes, which corresponds to approximately 5% of the sample population experiencing crises Explanatory indicators were selected from existing literature and available data, employing a conditional fixed effects model to analyze banking crises as indicated by the MMP index Key regressors included macroeconomic variables such as real GDP growth rate, exchange rate valuation, real interest rates, inflation rates, and fiscal deficits, alongside financial variables like private credit to GDP and stock market changes Institutional factors considered were GDP per capita, explicit deposit insurance, and financial liberalization The findings reveal that a slowdown in real GDP, lower real interest rates, extreme inflation, significant fiscal deficits, and overvalued exchange rates often precede banking crises, while the impact of monetary base growth on the likelihood of such crises was deemed negligible.
Chapter summary
After all related literatures are introduced this section gives brief summary in Table 2.1 according to the order of structure written above
The banking crisis, as defined in this thesis, is characterized by fluctuations in the MMP index, which highlights the vulnerabilities within the banking sector This index is calculated by assessing the weighted average of changes in the ratio of reserves to bank deposits, alongside variations in short-term real interest rates (Hagen and Ho).
Figure 2.1 illustrates the mechanisms that contribute to banking crises, highlighting the roles of macroeconomic conditions, financial vulnerabilities, and institutional inefficiencies A banking system operating within a weak macroeconomic environment, coupled with fragile financial health among banks and ineffective institutions, faces a heightened risk of experiencing a banking crisis.
Table 2.1 Summary of literature reviewed
Author(s) Key indicators Methodology Data sample Findings
Bank run, short-term interest rate, bank deposit
The stability of banks is often assessed by their ability to withdraw funds, which may involve selling long-term securities prematurely This can lead to an increase in the yields of short-term assets, potentially causing some banks to default on deposits As a result, depositors may feel compelled to convert their deposits into cash to safeguard themselves against the risks associated with a bank run.
- The later wave of banking crisis occurred more severe because the banking system had been unhealthy during the former crisis
The analysis of macroeconomic indicators, including real GDP growth, export volume changes, investment levels, inflation rates, money supply, exchange rates, interest rates, and variations in total assets and deposits relative to loans, utilized a descriptive statistics approach to identify the characteristics of banking crises.
- Banks’ profitability over total assets has been taken into account to analyze for the deterioration of banks’ financial condition which may lead to banking crisis cycles Gorton
Banking panics and the depositors’ behaviors
The research highlights that banking panics may stem from shifts in depositors' risk perceptions, with key indicators including the deposits ratio and liabilities.
1996b) insolvency of banks, GDP, inflation, monetary growth, fiscal balances, trade balances, real deposit rate, financial deepening, real credit/GDP
69 countries over the period of late 1970-1996
- Been considered to be the first banking crisis database with crisis dates, countries and some economic explanatory variables together with observations on policy measures
- Weak macroeconomic background and health of banks tend to lead to banking crisis
16 indicators from financial sector, external sector, real sector and fiscal sector signal-to- noise approach
20 countries for the period 1970-mid-
- This study gives the opportunity to study
76 currency crises and 26 banking crises following the database in the work of Caprio and Klingebiel (1996) Out of sample testing was examined with the twin crises in Asia of 1997
- Volatilities of such indicators from financial aspect, external and real sector seem to contribute to the probability of banking crisis occurrence
Dermirguc- Growth rate of multivariate a large - The research found that weak h
(1998) real GDP, change in term of trade, depreciation, real interest rate, inflation,
GDP, M2 over reserves, private credit over GDP, bank cash over assets, growth rate of domestic credit, dummy for deposit insurance, GDP per capita, law and order
(ICRG) logistic model sample of 45-65 countries in IFS database from both developed and developing countries over their scope from
1980 to 1994 macroeconomic background of the economy such as low growth rate of GDP tends to trigger banking crises due to the risk taking nature of banks
Banking crises can arise from maturity transformation issues within the banking sector, particularly during periods of high nominal interest rate fluctuations and elevated inflation While inflation control measures, such as restrictive monetary policies, aim to stabilize the banking sector, they can inadvertently increase the risk of banking crises through the channel of high real interest rates Therefore, weak banking systems must exercise caution when implementing inflation control and monetary policies.
- Moreover, institutional decision of deposit insurance scheme seems to increase the likelihood of banking crisis Bordo and
(2012) credit booms, inequality and housing policy to banking crises logit model with and without countries fixed effect
14 advanced countries over the scope of 1880-2008
- Banking system instability, which may lead to banking crisis, was evidently found to have a strong positive relationship with lagged term of two to five years of credit booms
Inequality and housing policy are critical economic factors examined for their influence on the likelihood of financial crises However, analysis of the current dataset revealed no evidence supporting a correlation between these factors and the occurrence of such crises.
Credit expansion plays a crucial role in driving real economic growth, yet it also poses significant risks for future financial crises These risks often stem from operational failures and regulatory shortcomings within the financial system, highlighting the need for careful management of credit booms to ensure stability.
- The role of such credit booms from the perspective of macroeconomics should be further studied as there have been the historic lessons of credit expansion and financial fragility
(2011) the current account, growth of loans, volatility of interest rate, inflation, growth of GDP logistic country fixed effect model, descriptive statistics
The growth of loans significantly contributes to the acceleration of banking crises, impacting both national and global economies Understanding the mechanism of macroeconomic indicators is crucial in analyzing how these factors interplay to trigger financial instability.
The deterioration of current accounts has clearly contributed to the escalation of crises, affecting both global markets and individual nations Additionally, the natural interest rate being significantly suppressed signals a phase leading up to these crises.
Real interest rate and inflation also gave similar predicting signal to this trend Kauko
(2012) deficit current account to banking vulnerabilities with main concentration in banking sector, deterioration of credit quality,
Non- performing loans no banking crisis probability calculation model, OLS regression cross-section of 34 advanced countries according to the classification of IMF 2009
The outcome showed that credit growth in the combination with current account deficit act as an important predictor for financial crises
GDP ratio, an interaction term between international reserves-to-
GDP ratio and a recipient of a swap line dummy variable, factors of external exposure, institutions, financial development, banking sector health and competition ordinary least square regression
107 countries together with their financial crises episodes from 2008-
The spillover effect of this crisis on other countries reveals significant dynamics and relevant factors Research indicates that these two types of stresses exhibit common characteristics, highlighting the interconnectedness of global challenges.
Countries exhibiting higher levels of de facto openness experienced more significant economic shocks, while those with competitive banking systems demonstrated reduced vulnerability, particularly when their banks were well-capitalized and effectively supervised Additionally, nations with substantial international reserves faced increased external stress, whereas commodity-exporting countries encountered less internal stress.
Berkmen et al (2012) wide range of variables of
Trade linkages, financial linkages, vulnerabilities/ financial structures, policy framework descriptive statistic evidence and country cross- sectional regression
43 countries from Consensus Forecast, 141 countries from WEO database over the year of
The contagion effect of financial crises in advanced countries significantly impacts economies worldwide While the intensity of these crises varies across nations, the macroeconomic environment and institutional policies are crucial in determining each country's financial system vulnerabilities during such shocks.
(2011) financial crisis in the linkage with labor market cross-country model
56 countries over the scope of 2007-2009 first quarter
Controlling for variables like trade and capital market integration, financial development, monetary and fiscal policy, institutional differences, and population growth, our findings indicate that reduced hiring costs significantly lessen output loss, particularly in high-income countries Conversely, in low-income countries, lower dismissal costs are associated with prolonged crisis durations.
7 factors of cross-country crisis severity, i.e the GDP growth over times, growth of consumption,
Eight key indicators commonly utilized by researchers include exchange rate regime, current account balance, growth of trading partners, credit market regulation, short-term external debts, fluctuations in house prices, growth of bank credit, and levels of international reserves.
METHODOLOGY, MODEL SPECIFICATION AND DATA
Model selection
The judgment outcome of banking crisis is in binary format, i.e it takes the value of
In analyzing banking crises, empirical studies typically utilize a binary classification system, assigning "1" for a crisis and "0" for no crisis Among the methodologies reviewed, the logit regression model is preferred due to its binary nature and ease of interpretation, as highlighted in studies by Dermirguc-Kunt and Detragiache (1998), Hagen and Ho (2007), Jorda et al (2011), and Bordo and Meissner (2012) This thesis will adopt the logit regression technique, following the trend in existing literature The mathematical principles underlying logit regression are detailed in the following sections.
First, those may take a look at the logit model in its mathematic equation below:
Let = , the formula above can be rewrote as:
It is easy to recognize that while takes the values from -∞ to +∞, will receive values ranging between 0 to 1 h
Supposed that is the probability of banking crisis occurrence, then the state of non-crisis will be described by the term (1- ) = 1- (2) Take (1) divide by (2):
= (3) After all, take natural log for both sides of equation (3) the logit model will be obtained as:
= ln ( )= = (4) Where denotes the coefficients of each explanatory variables separately
Noted that = ln ( ) Therefore, taking the antilog of the estimated logit, we get
The odds ratio can be understood by taking the antilog of the j-th slope coefficient in a regression model By subtracting 1 from this value and multiplying the outcome by 100, you can determine the percentage change in the odds associated with a one-unit increase in the j-th regressor.
The formula (4) under the conditions that time and entities are considered together can be rewritten as follow:
Gujarati (2004) highlighted that estimation (5) is influenced by assumptions regarding the intercept, slopes of coefficients, and the error term There are five key assumptions: (1) the intercept and slopes remain constant across time and entities (countries), with the error term accounting for differences; (2) the slopes are constant while the intercept varies among countries; (3) both slopes and intercepts vary over countries and time; (4) both intercept and slopes vary across countries; and (5) both intercept and slopes change over time and countries These assumptions reflect increasing complexity and align more closely with real-world scenarios For this analysis, only the first two assumptions will be examined further.
The pooled regression model, often referred to as assumption (1), offers a straightforward regression approach but comes with significant limitations, potentially distorting the overall relationship between the dependent and independent variables.
As common sense, people are more interested in the specific nature of each country This turns to the second case described by the estimation rewritten from (5) as below:
The fixed effect model (FEM) allows for varying intercepts across countries while maintaining time invariance, highlighting unique characteristics of individual nations This adaptability makes FEM a more suitable choice than pooled regression for the analysis presented in the thesis.
In addition, a formal statistical test, i.e the restricted F-test, is suggested for the choice between restricted pooled model and FEM
The Fixed Effects Model (FEM) is straightforward to apply; however, it can become complex with numerous cross-sectional units due to the limitations on degrees of freedom (Gujarati, 2004) This raises the question of exploring alternative approaches that address the unknown information of the error term, rather than solely focusing on the intercept in FEM As a result, the Random Effects Model (REM) was developed to provide a more comprehensive analysis.
The Random Effects Model (REM) posits that while individual countries share a common mean intercept, variations arise from a random error term with a zero mean and a specific variance This approach emphasizes the importance of accounting for unobserved heterogeneity among countries in statistical analyses.
By substituting (7) into estimation (6), the REM estimation obtained below:
Hausman (1978) introduced a formal test to determine the appropriate model between Fixed Effects Model (FEM) and Random Effects Model (REM) Given that FEM consistently provides unbiased estimations, it is often preferred in analytical contexts.
Model specification
The selection of variables in regression models, as highlighted by Hagen and Ho (2007) and Gujarati (2003), should integrate both theoretical frameworks and empirical research However, the potential for specification errors remains due to factors such as the exclusion of relevant variables or the inclusion of irrelevant ones, as well as the possibility of an inaccurate functional form This thesis asserts that a logistic regression model is suitable for analyzing banking crises To address the risk of specification errors, it aims to incorporate a comprehensive set of relevant variables, drawing on previous research and available data The primary goal is to identify determinants that significantly contribute to the likelihood of banking crises, utilizing the logit model as outlined in the works of Dermirguc-Kunt and Detragiache (1998), Hagen and Ho (2007), Jorda et al (2011), and Bordo and Meissner (2012).
A comprehensive framework for predicting banking crises should encompass all relevant variables Drawing on banking crisis theory and the availability of data, the proposed model is outlined as follows.
The regression model includes a constant item and an error term, with coefficients representing each independent variable in the logit model It assesses the banking crisis occurring 12 months prior in country i, alongside inflation and the growth rate of the monetary base to GDP at time t Additionally, it examines the depreciation of the domestic currency, short-term real interest rates projected 36 months ahead, and the growth rates of domestic credit and deposits to GDP over the past 12 and 6 months, respectively The model also analyzes the growth rate of M2 relative to reserves, as well as governance indicators such as voice and accountability, political stability, government effectiveness, regulatory quality, rule of law, and control of corruption.
This thesis classifies relevant variables into three groups: macroeconomic indicators, financial indicators, and banking crisis history Key macroeconomic factors include inflation, monetary base growth, currency depreciation, and previous banking crises, as weak macroeconomic conditions may heighten the risk of banking crises (Berkmen et al., 2012) Financial indicators focus on metrics such as M2 growth relative to reserves, short-term real interest rates, and credit growth.
GDP and the growth of bank deposits relative to GDP are key indicators of financial system health Additionally, institutional indicators, such as Voice and Accountability, Political Stability and Absence of Violence, Government Effectiveness, Regulatory Quality, Rule of Law, and Control of Corruption, provide insights into governmental health The subsequent sections will discuss the impact of each explanatory variable and their anticipated effects.
Inflation serves as a key indicator of government mismanagement in the economy, often coinciding with high interest rates, currency deterioration, and weakened bank balance sheets While inflation's adverse effects are crucial for understanding banking crises, empirical studies indicate that its relationship with such crises is weaker in developed countries compared to emerging markets Despite this, inflation remains a prominent factor for analysis due to its comprehensive reflection of the macroeconomic environment, warranting its inclusion in this thesis, even if its correlation with banking crises remains uncertain.
Past banking crisis (lagged 12 months)
Identifying precise periods of banking crises is challenging due to delayed recognition and asymmetric information regarding recovery signals Analyzing past banking crises can aid in predicting current crises, as suggested by Falcetti and Tudela (2006), who incorporated similar lagged terms in their model for banking crisis determinants This approach supports the hypothesis that historical data can inform present-day banking stability.
34 last period of banking crisis may lead to the crisis in present; hence, the expected sign of this relationship is positive
The monetary base is considered to reflect the monetary expansion (Hagen and Ho,
The growth of the monetary base is a crucial indicator for evaluating the health of the banking system, as it serves as a key source of liquidity A robust monetary base is likely to mitigate the risk of financial crises, suggesting an expected negative correlation with potential vulnerabilities in the banking sector.
Banks often rely on overseas borrowing to finance domestic lending, which can jeopardize their profitability if the domestic currency depreciates (Demirgỹỗ-Kunt and Detragiache, 1998) In response, many countries have implemented regulations to restrict foreign borrowing by banks; however, these regulations are frequently circumvented Some banks even issue domestic loans in foreign currencies, transferring the foreign exchange risk to borrowers Consequently, any depreciation of the exchange rate can lead to increased bad loans, ultimately harming bank profits Historical narratives support these findings (Demirgỹỗ-Kunt and Detragiache).
Research from 1998 indicates that foreign currency loans contributed to banking system issues in several countries, including Chile in 1981, Mexico in 1995, the Nordic countries in the early 1990s, and Turkey in 1994 Consequently, a positive correlation is anticipated between this indicator and the occurrence of banking crises.
The monetary aggregate serves as an indicator of central bank reserves, with a deficiency potentially triggering financial crises Jorda et al (2011) highlighted that crises were often preceded by a four-year peak in the money-to-nominal GDP ratio Conversely, Drehmann et al (2011) presented conflicting findings, suggesting that this ratio is not a reliable predictor of crises This divergence in empirical evidence underscores the complexity of predicting financial instability.
35 results, the thesis, once again, let the data speak by leaving the impact expected direction ambiguous
Real short-term interest rate
High interest rate often results in insolvencies for the borrowers; hence leading to banks may suffer non-performing loans Demirgỹỗ-Kunt and Detragiache (1998,
Research by Jorda et al (2000, 2002) indicates that banking crises are often preceded by periods of high real interest rates However, there is ongoing debate regarding the significance of high versus low real interest rates in influencing the likelihood of banking crises.
A study conducted in 2011 indicated that the explanatory power of certain factors was minimal, suggesting that their influence only became evident over time in individual countries Additionally, low real interest rates facilitate credit expansions, which can increase banks' exposure to crisis risks, as noted by Bordo and Meissner in 2012 Similarly, Hagen and Ho (2007) concluded that high real interest rates negatively affect banking stability Consequently, the relationship between these variables remains ambiguous within the thesis.
Growth of credit to GDP
Rapid credit growth can signal potential risks within the banking system, as highlighted in various studies on banking and financial crises This phenomenon directly impacts the performance and profitability of banks Historically, significant increases in credit levels have been observed preceding banking crises, suggesting a positive correlation between credit growth and the likelihood of such crises occurring.
Growth of deposit to GDP
Deposit of all kinds may reflect the belief of the public in health of banking sector,
A higher ratio of deposits to assets contributes to a safer banking system, as noted by Kauko (2014) This viewpoint is supported by researchers analyzing both the first and second trends in banking crises, particularly in the absence of any implicit or explicit deposit guarantee regulations Consequently, the growth of this ratio relative to GDP will be incorporated into the thesis model, where a negative relationship is anticipated.
Estimation strategies and relevant model diagnostics
3.3.1 Calculation of MMP for banking crisis assessment
The MMP index is calculated using the equation introduced in Chapter 2, where 'y' represents the reserves to bank deposits ratio, 'r' indicates the short-term real interest rate, and their respective terms and standard deviations are denoted accordingly.
This thesis utilizes monthly data from the IMF International Financial Statistics CD-ROM, updated in July 2011, covering the period from 2001 to 2010 Total deposits are derived from the sum of demand deposits, time and savings deposits, and foreign liabilities of deposit money banks Following the methodology of Hagen and Ho (2007), borrowed reserves—defined as loans from monetary authorities to financial institutions—are used as the reserves aggregate instead of total reserves The nominal interest rates are represented by money market rates, with alternatives such as treasury bill rates, government bond yields, deposit rates, lending rates, and discount rates applied when money market rates are unavailable The inflation rate is calculated using the consumer price index to determine the real interest rate.
- Obtain the BC time based on the 2 criteria for MMP index shows in equation
(2) stated in Chapter 2 as below:
The percentile command in Excel calculates the 98.5 percentile of the MMP distribution once all index values are obtained from the previous step This metric represents the growth rate of the MMP index between period t and the preceding period.
3.3.2 Model estimation steps and diagnostics
- Estimate the logit model using all possible explanatory variables from macroeconomic, financial and institutional aspects
- Putting lagged terms of indicators into the estimation basing on relevant empirical studies to obtain better regression model with larger value of Likelihood Ratio between models
The validity of logistic regression analysis relies on several key assumptions, and any violation of these assumptions can lead to biased coefficient estimates and inflated standard errors, ultimately resulting in invalid statistical conclusions Therefore, it is crucial to conduct diagnostic assessments of the model before drawing any statistical inferences The essential assumptions for logistic regression include: (1) the dependent variable must represent true conditional probabilities of the independent variables; (2) there should be no omission of important variables; (3) irrelevant variables must be excluded; (4) independent variables should be measured without error; (5) observations must be independent; and (6) independent variables must not exhibit multicollinearity.
- Test for model specification error
In building a logistic regression model, it is assumed that the logit of the outcome variable is a linear combination of the independent variables This involves two key aspects: first, the logit function is deemed the appropriate link function for the outcome variable; second, it is assumed that all relevant variables are included in the model without any extraneous variables, ensuring that the logit function accurately reflects a linear combination of the predictors.
Using the logit function as the link function may not be appropriate if the relationship between the outcome variable and independent variables is not linear, leading to potential specification errors in the model While misspecification of the link function can be an issue, it is generally less critical than the choice of alternative link functions like probit, which relies on the normal distribution In practice, the primary focus should be on ensuring that the model includes all relevant predictors and that their linear combination adequately represents the data.
The Stata command linktest is utilized to identify specification errors following the logit or logistic command The premise of linktest is that a properly specified model should not yield any additional statistically significant predictors except by chance After running the regression, linktest employs the linear predicted value (_hat) and its square (_hatsq) to reconstruct the model The variable _hat is expected to be a significant predictor, indicating the model is well-specified, while _hatsq should lack predictive power unless the model is misspecified A significant _hatsq suggests potential issues such as omitted relevant variables or incorrect link function specification, indicating that the model may require further refinement.
Multicollinearity arises when two or more independent variables in a regression model are closely related through a linear combination of other variables Its impact can vary, with perfect multicollinearity leading to the inability to derive unique estimates for regression coefficients In such cases, software like Stata automatically removes the variable that is a perfect linear combination of others, ensuring that only independent variables that are not exact linear combinations remain in the model for accurate regression analysis.
When using Stata for model analysis, it's crucial to understand that the variable excluded from the model may not necessarily be the "correct" one to omit Instead, a solid theoretical foundation should guide the decision on which variable to leave out, ensuring the model's integrity and relevance.
Moderate multicollinearity is common in regression analysis, as any correlation among independent variables indicates its presence Severe multicollinearity can inflate standard errors for coefficients, leading to unreliable logistic regression estimates The command "collin" is utilized to detect multicollinearity, providing insights into the strength of interrelationships among variables Two key measures are tolerance, which indicates the extent of multicollinearity a regression can tolerate, and the variance inflation factor (VIF), which assesses the inflation of standard errors due to multicollinearity Tolerance is calculated as 1 minus the R² from regressing other variables on a specific variable, while VIF is the reciprocal of tolerance When variables are orthogonal, both tolerance and VIF equal 1; however, if a variable is closely related to others, tolerance approaches 0, resulting in significantly increased variance inflation.
Data scope and sources
Data availability plays a crucial role in selecting the countries included in the sample, which encompasses 18 nations, primarily from ASEAN, as well as China, Japan, Korea, and several Latin American countries This selection is based on the fact that these countries have faced significant banking crises during their economic development Additionally, many of them are emerging economies with relatively immature banking systems, making them more susceptible to banking crises.
This thesis utilizes data from the International Financial Statistics (IFS) by the IMF, the World Bank (WB), and the World Governance Indicators (WGI) to calculate both dependent and independent variables The IFS dataset provides monthly data on inflation, monetary base growth, depreciation, and real interest rates, while the WB database offers annual data for indicators such as M2 growth relative to reserves, credit growth compared to GDP, bank deposit growth over GDP, and various institutional variables from the WGI.
The data analyzed spans from January 2001 to December 2010, reflecting the availability of relevant information during this period This analysis utilizes a combination of cross-country and time series data, resulting in a comprehensive panel dataset Detailed sources for the data can be found in Tables 3.1 and 3.2 below.
Table 3.1 Data for MMP index calculation
Demand deposits IFS line 24 monthly Time and saving deposits IFS line 25 monthly
Foreign liabilities of deposit money of banks
Total reserves Borrowed reserves used instead IFS line 26G monthly
Nominal interest rate Money market rates IFS line 60B monthly
Inflation rate Obtained from consumer price index IFS line 64 monthly
In countries where money market rates are unavailable, alternative financial metrics will be utilized in the following order: Treasury bill rates, government bond yields, deposit rates, lending rates, and discount rates.
Table 3.2 Data and sources of explanatory variables
Variable name Definition Expected sign Sources
Inflation (%) Inflation rates Ambiguous IFS line 64 monthly
MB_gr Growth of monetary base Negative IFS line 14 monthly
Past BC BC in last period Positive 12-month lagged term of
Depre (%) Depreciation: changes of nominal exchange rates Positive IFS line RF monthly
M2RES_GR Growth of M2 over reserves Ambiguous World Bank
Nominal interest rates are from IFS line 60b; Inflation rates are from IFS line 64 monthly
CrdGDP_GR Growth of private credit to
DEPOGDP_GR Growth of deposit to GDP Negative World Bank
VA Voice and Accountability Negative WGI Annually
PS Political Stability Negative WGI Annually
GE Government Effectiveness Negative WGI Annually
RQ Regulation Quality Negative WGI Annually
RL Rules of Law Negative WGI Annually
CC Control of Corruption Negative WGI Annually h
Conceptual framework
Growth of banking deposits over GDP (-)
Banking Crisis (Money market pressure index approach)
Hagen and Ho, 2007 Facetti and Tudela, 2006 h
Research Process
Thesis content: determinants of banking crisis
Introduction of three main trends of banking crisis analyses over time
In-depth introduction of Money Market Pressure Index (Hagen and Ho,
Choice of explanatory variables from macroeconomic, financial and institutional aspects
Conditional country fixed effect using
Apply available explanatory indicators from macroeconomic, financial and institutional aspects and arguments on effects of those variables on banking crisis
Model specification test, goodness of fit test, multicollinearit y test Methodology applied in the thesis
Thesis results, findings, policy recommendation, limitations and future research suggestions
Relevant data from IFS – IMF and WB for countries in Asia and Latin America over the period of 2001-2010 h
RESUTLS AND FINDINGS
Descriptive statistics of explanatory indicators
To provide a comprehensive understanding of the regression analysis, it is essential to first introduce the statistical characteristics of the dataset Table 4.2 presents the mean, standard deviation, minimum, and maximum values for each indicator, organized into three distinct categories: macroeconomic, financial, and institutional groups.
The analysis reveals significant fluctuations in macroeconomic and financial indicators, with M2RES_GR exhibiting a minimum value of -67.88314 and a maximum of 216.7543 Similarly, R_i, CrdGDP_GR, and DEPOGDP_GR demonstrate considerable variability, as indicated by their high spreads between minimum and maximum values, alongside their mean and standard deviation These substantial gaps suggest potential vulnerabilities within the economy from both macroeconomic and financial perspectives, highlighting the impact of a weak economic backdrop.
49 increase banking crisis probability; there is likelihood that banking crisis could occur according to big gaps in fluctuation magnitudes of relevant indicators
Institutional indicators, ranging from -2.5 to +2.5, show significant fluctuations, suggesting potential issues within government health A weakened government, as indicated by these metrics, could jeopardize the stability of the banking system, potentially leading to a crisis.
Table 4.2 Summary statistics of variables used in the regression
Variable Observations Mean Std Dev Min Max
This section explores the potential correlations between the dependent variable, banking crisis (BC_MMP), and various independent variables presented in Tables 4.3a and 4.3b The correlation analysis reveals that past banking crises (BC_MMP(-12)), growth of the monetary base (MB_gr), depreciation (Depre), real interest rates (R_i), and the growth of credit relative to GDP are positively associated with the occurrence of banking crises Additionally, inflation (INF) also shows a significant relationship with the dependent variable.
A 50% growth in deposits relative to GDP is associated with a negative relationship between six institutional variables—VA, PS, GE, RQ, RL, and CC—and banking crises The analysis reveals a high correlation among these variables, particularly between RQ, RL, CC, and GE, indicating that the interplay of these institutional factors plays a significant role in mitigating banking crises.
The regression analysis reached a high correlation value of 0.9, indicating significant multicollinearity issues among the variables Consequently, not all highly correlated variables can be effectively included in the regression model In the following section of this chapter, specific remedies to address this multicollinearity problem will be discussed.
Table 4.3a The correlation on the sample observations
BC_MMP (-12) MB_gr Depre R_i
Table 4.3b The correlation on the sample observations
M2RES_GR VA PS GE RQ RL CC
Statistical tests for model
In analyzing banking crises, a "simple" logit regression model is employed, incorporating all relevant explanatory variables to assess their impact on the occurrence of such crises This model focuses on the relationship between regressors and the regressand, omitting lagged terms and fixed or random effects that are typically beneficial for panel data analysis The primary aim of this approach is to conduct a model specification diagnostic using the linktest command, rather than to delve deeply into the effects of the variables Consequently, initial statistical indices such as the likelihood ratio, probability, Chi-square, and R-squared may be set aside, despite their statistical significance.
The findings in Table 4.4 indicate that the logistic regression model is a suitable choice, supported by the inclusion of relevant indicators This is evidenced by the statistically insignificant item _hatsq (p-value of 0.276) and the significant item _hat at the 5% level.
Upon completion of this test, the issue of specification error can be effectively addressed, both in terms of statistical significance and through the robust application of foundational economic theories.
The goodness of fit is a crucial criterion for evaluating the effectiveness of a model, as demonstrated in Table 4.5 In Stata, the lfit command is used to assess this fit, with a higher Prob> Chi-squared value indicating better performance The results indicate a high goodness of fit index of 0.7524.
Multicollinearity is a significant concern in panel data analysis, as it can lead to biased regression results and diminish the reliability of explanatory indicators Utilizing the 'collin' command in Stata can effectively assist in addressing this issue during testing.
In Chapter 3, we explored the methodology for identifying multicollinearity using the Variance Inflation Factor (VIF) and tolerance indices A common guideline indicates that an indicator may be problematic if its VIF exceeds 10 or its tolerance falls below 0.1 The testing results presented in Table 4.6 reveal that three variables—GE (VIF 34), RL (VIF 13), and CC (VIF 68)—exhibit signs of multicollinearity To address these issues, several strategies can be employed, such as removing redundant variables or utilizing interaction terms among significant collinear variables The subsequent results in Tables 4.7, 4.8, and 4.9 demonstrate the VIF and tolerance values of the models after implementing these remedies.
The mean Variance Inflation Factors (VIFs) from the tests suggest that an optimal combination of variables for regression should have a minimum VIF value Consequently, the removal of GE, RL, and CC, as illustrated in Table 4.8, is justified, with the minimum VIF recorded at 1.44.
The regression results are presented for both the full model and the restricted model, which excludes variables exhibiting high collinearity Tables 4.10 and 4.11 illustrate the comparative outcomes between the full model regression and the restricted model following the removal of highly correlated variables.
The analysis reveals that by removing highly correlated variables from the full model, the thesis achieves an improved model characterized by more accurate expected signs of coefficients and enhanced statistical significance.
Following a comprehensive discussion of model diagnostics, the key findings are consolidated in the tables below, specifically Tables 4.12, 4.13, 4.14, and 4.15 in the Appendix, which facilitate a comparison among the logit, fixed effect, and random effect models.
Coefficients interpretation
Based on the results from normal logit regression, fixed effect model, and random effect model, all explanatory variables maintain consistent signs, with minimal fluctuations in coefficient values across models As discussed earlier in this chapter, the fixed effect model will be utilized for its ability to provide unbiased estimation results The fixed effect model is presented as follows:
Following the application of all necessary diagnostics and remedies, along with a robust estimation model, the reliability of the coefficients can be affirmed Subsequently, the indicators will be categorized into their respective groups and analyzed in relation to previous studies The discussion will commence with macroeconomic indicators, transition to financial counterparts, and conclude with an examination of institutional factors.
As having been classified, indicators of this group include inflation, past 12-month banking crisis, growth of monetary base and depreciation
Inflation plays a significant role in influencing the probability of a banking crisis, exhibiting a negative impact with a high significance level of 1% This finding contrasts with the conclusions of Demirguc-Kunt and Detragiache (1998), who argued that high inflation increases the likelihood of banking crises Typically, inflation peaks before or during a crisis, suggesting a complex relationship that warrants further investigation.
A study by 54 and Ho (2007) concludes that while high inflation rates do not significantly contribute to banking crises, their prolonged accumulation can signal potential crises The impact of inflation varies based on the country analyzed, with emerging and developing nations experiencing higher inflation during economic growth, which may reduce the likelihood of banking system crises This suggests that economic growth can lead to inflation due to resource scarcity in an inelastic supply environment Additionally, the analysis highlights the importance of reliable data sources and the methodology used, as well as the definition of crises (Davis and Karim, 2008) Furthermore, a 12-month lagged term of banking crises shows a strong correlation with current crises, aligning with findings from Falcetti and Tudela (2006) This prior crisis serves as a warning for banking authorities, as the regression results indicate a significant lagged term at a 0.001 level, suggesting that proactive measures can be effectively implemented to mitigate future crises.
The growth of the monetary base has reached a statistically significant level, but its sign is contrary to expectations Similarly, Hagen and Ho (2007) found that the sign of this variable fluctuates across various model specifications, leading them to conclude that its impact on predicting banking crisis probability is negligible Additionally, different datasets from varying groups of countries may yield divergent results, further complicating the analysis Despite this thesis indicating an incorrect sign for the growth of the monetary base, it highlights the complexities involved in assessing its predictive power.
The expected sign of 55 indicates a significance level accepted at 5%, highlighting the need for further observation and analysis This raises questions about the value of this indicator across different groups of countries and its potential positive or negative impact on crisis probability.
Davis and Karim (2008) highlight that depreciation plays a crucial role in predicting banking crises, particularly during periods of rapid credit growth and moderately high interest rates Their regression analysis indicates that depreciation, coupled with a significant increase in credit over the past 12 months and rising real interest rates over the previous 36 months, serves as a strong indicator of impending banking crises The p-values for depreciation, real interest rates, and credit growth to GDP are notably significant, at 0.009, 0.007, and 0.000, respectively, confirming the robustness of these findings at a 1% significance level.
This section presents the results and findings of explanatory variables from a financial perspective, highlighting the M2 to reserves ratio, real interest rates, and the growth rates of credit and deposits relative to GDP.
The M2 to reserves ratio, as highlighted by Falcetti and Tudela (2006), serves as a key indicator of the financial sector's liquidity capacity and reflects the vulnerability of financial institutions during sudden withdrawals or asset conversions by depositors A positive coefficient for this ratio suggests that a higher value correlates with an increased likelihood of banking crises, consistent with Falcetti and Tudela's findings However, this ratio failed to meet significant criteria, with a p-value of 95.6%, significantly exceeding the 10% significance threshold.
High real short term interest rate has found by Demirgtic-Kunt and Detragiache
(1998) to have contribution to the increasing likelihood of banking crisis Sharing h
Falcetti and Tudela (2006) identified a correlation between high real interest rates on bank deposits and an increased likelihood of banking crises This thesis highlights that a lagged real interest rate over three years is statistically significant at a 10% level (p-value = 0.011) These findings align with previous research by Demirgüç-Kunt and Detragiache (1998, 1999, 2000), Falcetti and Tudela (2006), and Kauko (2014), suggesting that elevated real interest rates can precede banking crises by up to four years Conversely, Hagen and Ho (2007) found that low short-term interest rates may raise the risk of banking crises within the same year, attributing this to contractionary monetary policies implemented to tackle inflation during such periods.
The regression analysis indicates that the 12-month lag of credit growth over GDP is significantly correlated with the likelihood of a banking crisis, with a coefficient of 0.037 and a p-value of 0.003, confirming its predictive power This finding aligns with previous research suggesting that lagged credit growth is a key indicator for forecasting banking crises (Dermirguc-Kunt and Detragiache, 2000; Jorda et al., 2011; Barrel et al., 2011; Schularick and Taylor, 2012; Kauko, 2014) Notably, the thesis supports the conclusions of Barrel et al (2011), while contrasting with Hagen and Ho (2007), who found no significant link between credit growth and banking crisis probability.
The growth of banking deposits relative to GDP is a crucial financial indicator, demonstrating a significant influence on the prevention of banking crises, particularly when analyzed with a 6-month lag This relationship is deemed reliable at a 5% significance level, aligning with findings from Falcetti and Tudela (2006), which suggest a correlation between deposit growth and reduced banking crisis probability.
57 increases under the circumstance of sharp decrease in growth of bank deposit over GDP Moreover, the result goes in line with the suggestion and result of Barrel et al,
(2010) which concludes that growth of bank deposit over GDP in lagged format is a good indicator for banking crisis
In summary, the analysis of macroeconomic and financial indicators indicates the overall health of the macro-financial environment beyond the banking system, aligning with established theories and empirical studies A fragile economic backdrop, coupled with unhealthy financial conditions, increases the vulnerability and instability of banking systems, potentially leading to a banking crisis.
In contrast to the previously discussed indicators, the three institutional variables derived from the Worldwide Governance Indicators (WGI) show that only two out of the three variables exhibit correct signs, although their significance levels remain insignificant, exceeding the 10% threshold.
The attempt to incorporate institutional variables into the model has not been successful However, it is highly advisable to include government-related indicators to assess their impact on the probability of banking crises Common sense suggests that government intervention can either help the banking system recover from potential crises or mitigate the effects of ongoing crises.
CONCLUSION, POLICY RECOMMENDATION AND LIMITATION
Conclusion
Since the early 1980s, the global banking system has faced numerous crises, notably the contagion effect triggered by the collapse of Lehman Brothers, which led to a worldwide financial crisis This has heightened the urgency to identify the factors contributing to banking crises, as the integration of economies has become more extensive and profound in both quality and quantity The banking crises of the 21st century appear to be increasingly serial in nature, necessitating a deeper understanding of their underlying causes.
Disasters in major countries can significantly impact others, leading to a substantial loss of output and a prolonged decline in global economic growth Numerous studies on banking crises have shown that both national and global crises often share common factors, which help explain the underlying logic of their occurrence.
The thesis effectively utilizes the MMP index from Hagen and Ho (2007) to analyze banking crises through macroeconomic and financial lenses However, incorporating WGI institutional indicators yields less favorable results than anticipated Given that government health significantly impacts the overall economy, the study suggests the need for a more suitable database, such as the ICRG data used by Breuer et al (2006) for currency crises This highlights the importance of integrating macroeconomic and financial factors with more relevant institutional indicators for a comprehensive analysis.
Policy recommendation
The author has found that results obtained from the regression model go in line with those from various prior studies that weak macroeconomic and financial h
59 background likely push banking system to the position of exposure to risks and vulnerabilities, which may trigger banking crises
Recent regression analysis suggests that a banking crisis occurring within the past 12 months may signal the potential for a current crisis to emerge or persist Consequently, it is crucial for authorities to closely monitor the situation and implement effective measures for either alleviating or eliminating the crisis By doing so, they can mitigate the detrimental effects on economic output and prevent the exacerbation of the banking crisis.
High real interest rates, credit growth over GDP, and deposit growth over GDP, along with their lagged terms and currency depreciation, are critical indicators for predicting banking crisis probabilities These indicators reflect the overall health of the banking system, necessitating that authorities monitor sudden fluctuations to implement timely interventions While their immediate effects may not be apparent, the cumulative impact over time can significantly increase the likelihood of a banking crisis Interestingly, high inflation during the period of analysis appears to correlate with positive economic growth in emerging markets, although this remains a topic of debate Thus, careful observation of inflation indicators is essential Additionally, inflation stability programs may induce volatility in real interest rates, affecting banks' profitability, which can heighten their vulnerability to crises Therefore, any inflation control measures must be designed with caution, balancing program effectiveness against the potential costs of a banking crisis (Dermirguc-Kunt and Detragiache, 1998).
Limitation of the research
Despite efforts to identify the determinants of banking crises, the findings remain imperfect This thesis emphasizes traditional analyses that focus on banking crises while neglecting non-banking crises, which are crucial for understanding why certain countries are more vulnerable to risks It argues for the establishment of criteria for non-crisis periods to enhance future studies A comprehensive view that combines factors leading to crises with those that help maintain stability is essential for effective banking system and economic management The research employs a banking crisis definition based on the money market pressure index and utilizes a fixed effect model regression, acknowledging potential data collection biases due to variable availability and access to updated data Additionally, some indicators may only show significance after delays, necessitating more data to avoid sample size reduction The thesis suggests that a trial-and-error approach for determining appropriate lag times is impractical given the number of variables, leading to reliance on previously suggested lags with minor subjective adjustments Unfortunately, institutional variables in the estimation model yield insignificant results, potentially due to an inappropriate dataset Therefore, further testing with more suitable databases, such as those used in recent empirical studies or the ICRG database, is recommended Ultimately, like many studies, this analysis does not provide evidence for the end of banking crises, focusing instead on their onset and related dates.
Aizenman, J., Pasricha, G.K., 2012 Dereminants of financial stress and recovery during the great recession.Int J Finance Econ 17, 347–372
Artha, I.K.D.S., de Haan, J., 2011 Labor market flexibility and the impact of the financial crisis.Kyklos 64, 213–230
Berkmen, S.P., Gelos, G., Rennhack, R., Walsh, J.P., 2012 The global financial crisis: explaining cross-country differences in the output impact J Int Money Finance 31, 42–59
Bordo, M D and Meissner, C.M (2012) Does inequality lead to a financial crisis?
Journal of International Money and Finance 31: 2147-2161
Breuer, J.B, Shimpalee, P.L 2006.Currency crises and institutions Journal of
Corsetti, G., Pesenti, P and Roubini, N (1999) Paper tigers?
A model of the Asiancrisis European Economic Review 43: 1211-1236 Chang, R., & Velasco, A (2000) Liquidity crises in emerging markets: theory and policy InNBER Macroeconomics Annual 1999, Volume 14 (pp 11-78)
Demirgtic-Kunt, A., &Detragiache, E (1998a) The determinants of banking crises in developing and developed countries IMF Staff Papers, 45, 81-109
Demirgtic-Kunt, A., &Detragiache, E.(1998b) Financial liberalisation and financial fragility.IMF Working Paper,No.WP/83/98
Eichengreen, B., & Rose, A K (1998) Staying afloat when the wind shifts:
External factors and emerging-market banking crises (No w6370).National
Eichengreen, B., Rose, A K., Wyplosz, C., Dumas, B., & Weber, A (1995)
Exchange market mayhem: the antecedents and aftermath of speculative attacks Economic policy, 249-312
Falcetti, E., &Tudela, M (2008) What do twins share? A joint probit estimation of banking and currency crises Economica, 75(298), 199-221 h
Frankel, J.A., Saravelos, G., 2010 Are Leading Indicators of Financial Crises
Useful for Assessing Country Vulnerability? Evidence from the 2008–2009 Global Crisis NBER Working Paper Series 16047
Glick, R., & Hutchison, M (2000) Banking and currency crises: how common are the twins? Financial crises in emerging markets.Cambridge University
González-Hermosillo, B (1996) Banking sector fragility and systemic sources of fragility, IMF Staff Papers, No.WP/96/12 (February)
In "Basic Econometrics" (4th edition, 2004), Gujarati emphasizes fundamental statistical methods essential for economic analysis Meanwhile, Kaminsky and Reinhart (1998) discuss the financial crises in Asia and Latin America, comparing historical contexts in their paper published in the American Economic Review Their insights highlight the recurring patterns of economic turmoil across different regions and times, underscoring the importance of understanding these dynamics for future economic stability.
Kaminsky, G., Lizondo, S., Reinhart, C.M., 1998.Leading indicators of currency crises.IMF Staff Pap 45, 1–48
Kaminsky, G L., & Reinhart, C M (1999) The twin crises: the causes of banking and balance-of-payments problems American economic review, 473-500 Kaufmann, D., Kraay, A., &Mastruzzi, M (2007).Governance Matters VII:
Aggregate and Individual Governance Indicators, 1996-2007 World Bank,
World Bank Institute, Global Programs Division, and Development Research Group, Macroeconomics and Growth Team
Klomp, J and de Haan, J (2009) Central bank independence and financial instability.Journal of Financial Stability 5: 321-338
Kokko, A (1999) The Asian crisis, many similarities with the Swedish crisis.EkonomiskDebatt 27: 81-92
Kauko, K (2012) External deficits and non-performing loans in the recent financialcrisis Economics Letters 115: 196-199
Kauko, K (2014) How to foresee banking crises? A survey of theempirical literature Economic Systems 38: 289–308 h
McKinnon, R I., & Pill, H (1996) Credible liberalizations and international capital flows: the “over-borrowing syndrome” Financial Deregulation and Integration in East Asia, NBER-EASE Volume 5 (pp 7-50) University of
McKinnon, R I., & Pill, H (1998) The overborrowing syndrome: are East Asian economies different? Managing Capital Flows and Exchange Rates:
Perspectives from the Pacific Basin, 322-55
Miller, V (1999).The timing and size of bank-financed speculative attacks.Journal of International Money and Finance, 18(3), 459-470
Mishkin, F S (1997) Understanding financial crises: a developing country perspective (No w5600) National Bureau of Economic Research
Noy, I (2004).Financial liberalization, prudential supervision and the onset of banking crises Emerging Markets Review 5: 341-359
Obstfeld, M (2012).Does the current account still matter? American Economic Review102: 1-23
Obstfeld, M (1995) The logic of currency crises (pp 62-90).Springer Berlin Heidelberg
Reinhart, C M., &Végh, C A (1995) Nominal interest rates, consumption booms, and lack of credibility: A quantitative examination Journal of Development
Rojas-Suárez, L., &Weisbrod, S R (1995).Financial fragilities in Latin America: the 1980s and 1990s (Vol 132) IMF
Rose, A.K and Spiegel, M.M (2011) Cross-country causes and consequences of thecrisis: an update European Economic Review 55: 309-324
Rose, A.K and Spiegel, M.M (2012) Cross-country causes and consequences of the2008 crisis: early warning Japan and the World Economy 24: 1-16 h
Rossi, M (1999) Financial Fragility and Economic Performance in Developing
Economies-Do Capital Controls Prudential Regulation and Supervision Matter?IMFWorking Paper, No WP/99/66 (May)
Velasco, A (1987) Financial crises and balance of payments crises: a simple model of the Southern Cone experience Journal of development Economics,27(1), 263-283
Von Hagen, J., & HO, T K (2007).Money market pressure and the determinants of banking crises Journal of Money, Credit and Banking, 39(5), 1037-1066 h
Table 4.1 Banking crisis dates retrieved from MMP index
Table 4.4Linktest for specification error of logit model logit BC_MMP INF l12 BC_MMP MB_grDepreR_iCrdGDP_GR DEPOGDP_GR M2RES_GR VA PS
Logistic regression Number of obs = 1943
Log likelihood = -411.64834 Pseudo R2 = 0.0389 - BC_MMP | Coef Std Err z P>|z| [95% Conf Interval] -+ - INF | -.3266457 1266713 -2.58 0.010 -.5749169 -.0783745 BC_MMP(12) | 1.237036 2891568 4.28 0.000 6702991 1.803773 MB_gr | 0254394 0167837 1.52 0.130 -.0074561 0583348 Depre| 0687579 0277797 2.48 0.013 0143107 123205 R_i| 0090064 0155471 0.58 0.562 -.0214654 0394781 CrdGDP_GR | 0226107 0113294 2.00 0.046 0004054 0448159 DEPOGDP_GR | -.0241729 0133022 -1.82 0.069 -.0502448 001899 M2RES_GR | -.0053557 0051703 -1.04 0.300 -.0154893 0047778
Logistic regression Number of obs = 1943
Log likelihood = -411.08473 Pseudo R2 = 0.0402 - BC_MMP | Coef Std Err z P>|z| [95% Conf Interval] -+ - _hat | 1.951366 891996 2.19 0.029 2030855 3.699646
Table 4.5 Goodness of fit test of model lfit, group(10) table
Logistic model for BC_MMP, GOODNESS-OF-FIT TEST
(Table collapsed on quantiles of estimated probabilities)
| Group | Prob | Obs_1 | Exp_1 | Obs_0 | Exp_0 | Total |
+ -+ number of observations = 1943 number of groups = 10
Tabel 4.6 Full model multicollinearity test result
Variable VIF SQRT VIF Tolerance R-Squared
Table 4.7 Dropping significantly high correlated variables GE, RL:
Variable VIF SQRT VIF Tolerance R-Squared
Table 4.8 Dropping high correlated variables GE, RL and CC
Variable VIF SQRT VIF Tolerance R-Squared
Table 4.9 Using interactive term of GE and RL
Variable VIF SQRT VIF Tolerance R-Squared
Table 4.10 Full model logit BC_MMP INF l12 BC_MMP MB_grDepreR_iCrdGDP_GR DEPOGDP_GR M2RES_GR VA PS
Logistic regression Number of obs = 1943
- BC_MMP | Coef Std Err z P>|z| [95% Conf Interval] -+ -
Table 4.11 Restricted model without GE, RL, CC
logit BC_MMP INF BC_MMP12 MB_grDepreR_iCrdGDP_GR DEPOGDP_GR M2RES_GR VA PS
Logistic regression Number of obs = 1943
- BC_MMP | Coef Std Err z P>|z| [95% Conf Interval] -+ -
Table 4.12 Fixed effect model with lags xtlogit BC_MMP INF l12.BC_MMP MB_grDepre l36.R_i l12.CrdGDP_GR l6.DEPOGDP_GR M2RES_GR VA PS RQ, fe
Conditional fixed-effects logistic regression Number of obs = 1511 Group variable: country1 Number of groups = 18 Obs per group: min = 83 avg = 83.9 max = 84
LR chi2(11) = 48.88 Log likelihood =-279.62085Prob> chi2 = 0.0000
- BC_MMP | Coef Std Err z P>|z| [95% Conf Interval] -+ -
INF | -.4326759 1718959 -2.52 0.012 -.7695856 -.0957662 BC_MMP12 | 1.144859 3351207 3.42 0.001 4880346 1.801683 MB_gr | 0448394 0189506 2.37 0.018 007697 0819818 Depre | 0983399 0376113 2.61 0.009 0246231 1720567 R_i(36) | 0279047 0110093 2.53 0.011 0063268 0494825 CrdGDP_G(12) | 0370456 012508 2.96 0.003 0125303 061561 DEPOGDP_GR(L6)| -.0354308 0153012 -2.32 0.021 -.0654207 -.005441
Table 4.13 Random effect model with lags xtlogit BC_MMP INF l12.BC_MMP MB_grDepre l36.R_i l12.CrdGDP_GR l6.DEPOGDP_GR M2RES_GR VA PS RQ, re
Random-effects logistic regression Number of obs = 1511 Group variable: country1 Number of groups = 18 Random effects u_i ~ Gaussian Obs per group: min = 83 avg = 83.9 max = 84
Wald chi2(11) = 48.62 Log likelihood =-313.92006Prob> chi2 = 0.0000
- BC_MMP | Coef Std Err z P>|z| [95% Conf Interval] -+ -
INF | -.413111 1467951 -2.81 0.005 -.7008241 -.125398 BC_MMP12| 1.159605 3339441 3.47 0.001 5050869 1.814124 MB_gr | 0441357 0189147 2.33 0.020 0070636 0812078 Depre | 0970619 0374037 2.59 0.009 0237521 1703717 R_i(36)| 0262877 0097476 2.70 0.007 0071827 0453927 CrdGDP_GR(12)| 0395713 0103906 3.81 0.000 0192061 0599365 DEPOGDP_GR(6)| -.0328716 014082 -2.33 0.020 -.0604719 -.0052713
RQ | -.0532946 2346824 -0.23 0.820 -.5132638 4066745 _cons | -3.156307 2033296 -15.52 0.000 -3.554826 -2.757788 -+ - /lnsig2u | -14.30817 43.4502 -99.46899 70.85265 -+ - sigma_u | 0007817 0169817 2.52e-22 2.43e+15 rho | 1.86e-07 8.07e-06 1.92e-44 1 - Likelihood-ratio test of rho=0: chibar2(01) = 0.00 Prob>= chibar2 = 1.000 h
Table 4.14 Simple logit model with lags logit BC_MMP INF l12.BC_MMP MB_grDepre l36.R_i l12.CrdGDP_GR l6.DEPOGDP_GR M2RES_GR VA PS RQ
Logistic regression Number of obs = 1511
Log likelihood = -313.92003 Pseudo R2 = 0.0722 - BC_MMP | Coef Std Err z P>|z| [95% Conf Interval] -+ -
INF | -.413118 1468117 -2.81 0.005 -.7008636 -.1253723 BC_MMP12 | 1.1596 333979 3.47 0.001 505013 1.814187 MB_gr | 0441349 0189171 2.33 0.020 007058 0812118 Depre | 0970666 0374072 2.59 0.009 02375 1703833 R_i(36)| 0262878 0097487 2.70 0.007 0071807 0453949 CrdGDP_GR(12)| 0395718 0103918 3.81 0.000 0192042 0599394 DEPOGDP_GR(6)| -.0328699 0140837 -2.33 0.020 -.0604735 -.0052662