This study measures the severity of a banking crisis by using its duration and the cost. Using this new methodology, we find that the factors associated with a severe banking crisis are not quite the same as those associated with a simple banking crisis. An ordered logit model and a large panel data set were used for this study. One of our major findings is that there exists a four-year time lag between an economic boom, or financial system liberalization, and the occurrence of a severe banking crisis in a country. This indicates that banking problems start much earlier than the time when they are revealed as banking crises. This study also finds that the lower the remains of a past banking crisis, the higher the probability of a severe banking crisis. It could be due to less-attentiveness of banking sector policy-makers with elapsed time. A high rate of inflation, existence of an explicit deposit insurance scheme, and a weak institutional environment are found to be common factors positively associated with both simple and severe banking crisis.
At first, this paper proposes a new methodology to measure the severity of banking crisis, and then finds the factors associated with a severe banking crisis. Most of the previous studies1 defined a banking crisis incidence depending on meeting one or more of the criteria such as a threshold level of non-performing loans, a threshold amount of assets of insolvent banks, a threshold level of cost of rescue operation from a crisis, or other kinds of banking sector problems. Those studies used a dummy variable (0, 1) to define the incidence of a banking crisis, and then found the factors associated with the crisis. However, all banking crises are not alike. They are different in terms of severity. Hence, unlike previous studies, we define a banking crisis with its severity level by taking an arithmetic average of two factors―crisis-duration and cost-ranking as determined by the percentage of GDP loss due to the banking crisis (see
A number of studies have examined the factors associated with a banking crisis. Demirguc-Kunt and Detra- giache [
To our knowledge, as of today no study has defined the severity of a banking crisis by taking both the cost and the duration of the crisis into account while examining the factors associated with a banking crisis; or has shed light on the lag time between a banking problem and the revelation of a banking crisis. Hence, we felt the need for this study.
In order to determine the factors associated with a severe banking crisis, we examine various macroeconomic, financial, and institutional factors. Section 2 provides a brief theoretical underpinning for using a set of explanatory variables/factors in our model. These explanatory variables are GDP growth rate, terms of trade, depreciation rate, rate of inflation, ratio of M2 to foreign exchange reserves, growth rate of real domestic credit in the private sector, using of an explicit DIS, per capita GDP, and the remains of the past crisis. Formulation of these explanatory factors and their data sources are reported in
We have used a panel data set for a large number of countries from around the world for the period of 1980- 2003 (we could not consider the recent banking crises experienced by the US and the European banking sectors for lack of necessary information to determine the severity of the banking crises). Unlike previous studies that used bi-variatelogit or probit model, we use an ordered logit model and find that the probability of a severe banking crisis is positively associated with an economic boom and a liberalized financial system when both variables are lagged by 4 years. It implies that banking problems start much earlier than the time when they are revealed. A possible explanation of that could be the fear of negative economic repercussions from early revelation. We also find that the longer the time-lag after the last crisis, the lower the remains of the crisis, and the higher the probability of a severe banking crisis. The factors such as the rate of inflation, the existence of an explicit DIS, and the country’s weak institutional environment that other studies found to be associated with banking crisis, are also found to be positively associated with the severity of a banking crisis.
The remainder of this paper is structured as follows. The next section explains the data and the variables with a brief theoretical underpinning. Section three presents the estimation results and their robustness. Section four concludes.
We use Laeven and Valencia [
The variable “Severe banking crisis” takes the value 0, if a country didn’t experience any banking crisis in a year; or else it takes the severity value as calculated. In order to avoid the simultaneity bias, we dropped the data for the duration of a crisis.
The contractionary phase of the business cycle can cause a banking sector problem [
Many countries use liberalized financial system to stimulate domestic investment and economic growth by moving away from financial repression that reduces private credit [
The institutional environment in a country is an important factor for a banking crisis. The institutional environment indicates the country’s capability to take necessary measures to prevent or contain a banking crisis [
An explicit DIS seems to have two opposing effects on a banking crisis. According to Diamond and Dybvig [
Kindleberger [
The high rate of inflation in a country can cause its financial market and economic condition to be fragile [
The real interest rate is another influencing factor of financial market, and a contributing factor to banking crisis [
The large volume of currency in circulation relative to the foreign exchange reserve is also a strong indicator of the vulnerability of a country’s financial market [
As our dependent variable “Severe banking crisis” takes multiple values because of different severity levels of the crises, we use an ordered logit model for data analysis. Following is a brief description of our fitted model. See Green [
In the ordered logit model, the dependent variable
In a population, the continuous latent variable
where
We use the estimated twelve cut-off values to calculate the probabilities of 13 values of
The ordered logit model estimates the probability that the latent variable
When interpreting the regression results of an ordered logit model, it is important to remember that an estimated coefficient does not indicate the direct increase in the probability of a crisis given a one-unit increase in the corresponding explanatory variable. Instead, the coefficient reflects the effect of a change in the explanatory variable on the probability function as mentioned above. However, the sign of the coefficient does indicate the direction of the change.
The estimation results are reported in
Regression-1 | Regression-2 | ||||
---|---|---|---|---|---|
Dep. var. = Severe banking crisis | Coefficient | Std Error | Dep. var. = Severe banking crisis | Coefficient | Std Error |
GDP growtht-4 | 0.0276* | 0.0150 | GDP growth | −0.1487*** | 0.0429 |
Inflation | 0.0017** | 0.0008 | Inflation | 0.0009 | 0.0008 |
Terms of tradet-4 | 1.0140** | 0.4780 | Terms-of-tradet-4 | 0.9800*** | 0.3791 |
Real interest | −0.0001 | 0.0004 | Real interest | 0.0002 | 0.0003 |
M2/reserves | 0.0004 | 0.0008 | M2/reserves | 0.0013 | 0.0012 |
Credit growtht-4 | 0.0152** | 0.0070 | Credit growtht-4 | 0.0148*** | 0.0061 |
Per capita GDP | −0.0895** | 0.0381 | Per-capita-GDP | −0.1003*** | 0.0385 |
Remains of the past crises | −5.7562* | 3.0821 | Remains of the past crises | −6.2215** | 3.1266 |
Existence of DIS | 0.8685*** | 0.35 | Existence of DIS | 0.9667*** | 0.3647 |
Obs | 1408 | Obs | 1406 | ||
Log likelihood | −308.7 | Log likelihood | −299 | ||
Wald chi2 | 69.08 | Wald chi2 | 82.44 | ||
Prob > chi2 | 0 | Prob > chi2 | 0 | ||
Pseudo R2 | 0.0511 | Pseudo R2 | 0.0806 |
***, ** or * indicates that the coefficient is significant at 1%, 5% or 10% level respectively.
We find that the correlation coefficient between contemporaneous GDP growth rate and the severity of a banking crisis is negative (significant at 1% level). It supports the theoretical view [
As theory suggests [
We also find that the more the time elapses after a banking crisis, the lower the remains of the crisis, and the higher the probability of a severe banking crisis. A plausible explanation of this is that as time elapses after a banking crisis, policy-makers become more and more passive. That eventually leads to a severe banking crisis.
The rate of inflation is another factor that is positively associated with the severity of a banking crisis (significant at 5% level). With a high rate of inflation, the lenders incur financial losses. As a result, many lenders/banks recall their loans or increase interest rates. These actions cause many firms to fail and the number of non-performing loans to increase, which contributes to a banking crisis [
Khan, et al. [
We also find that the existence of a DIS increases the probability of a severe banking crisis (significant at 1% level). This finding is consistent with the findings of several other studies about simple banking crisis that used bi-variatelogit/probit model [
To check for the robustness of our results, we need to test for simultaneity bias or reverse causality problem for the contemporaneous variables (i.e. the variables without any lag or lead). We focus on the variables that are statistically significant in Regression-1 of
Replacing per capita GDP with the level of corruption as an indicator for institutional environment in Regression-1 can solve one possible simultaneity bias problem. For the level of corruption in a country doesn’t change frequently. It is very unlikely that the level of corruption would change immediately after a banking crisis. Using corruption3 as an independent variable, we obtain the same relationship as before between the institutional environment and the probability of a severe banking crisis in a country (results can be obtained from the authors).
The adoption of DIS and a banking crisis may happen simultaneously, which creates causality problem. One can identify the causality run by using a two-stage estimation method. In the first stage, estimate the predicted value of the adoption of a DIS based on a set of explanatory variables, and call it DIS-predict. In the second stage, replace the DIS-dummy with the variable DIS-predict to estimate the probability of a severe banking crisis. See Khan and Dewan [
To check for the causality run between inflation and banking crisis, we use a two-stage estimation method again. Stage 1: Using the OLS method, we estimate the predicted value of inflation based on a set of independent variables, where at least one variable is theoretically related to inflation but not to severe banking crisis. That variable is “Excess population growth rate,” which is the difference between the population growth rate and the GDP growth rate.
Stage 2: Using the ordered logit estimation method, we estimate our benchmark regression equation (Regression-1) by replacing inflation rates with the predicted values of inflation (Inflation-predict) in the equation.
The estimation result shows that the coefficient of Inflation-predict has the expected sign, and it is a highly significant factor for a severe banking crisis. It proves that a high inflation causes a severe banking crisis, not vice-versa (results can be obtained from the authors).
Instead of considering all banking crises alike, in this study, we differentiate among the crises based on their severities. The severities are measured by the costs and the durations of the crises. Using an ordered logit model, we identify the factors that are associated with a severe banking crisis.
The most significant finding of this study is that a severe banking crisis is associated with an economic boom and financial system liberalization when both variables are lagged by 4 years. It implies that the process of a banking crisis starts much earlier than when it is revealed [
This study suggests that the policy-makers ought to be careful during an economic boom, especially when the country has a liberalized financial system. Sustainable economic growth with a secured and non-vulnerable financial system is certainly better than a rapid but unsustainable economic boom with a shaky liberalized financial system. Many firms take advantage of liberalized financial channels and invest money in risky projects. As a result, productions increase to have an economic boom, but the economic boom does not last long as the productions from the projects are not sustainable and the projects eventually fail. In most cases, projects struggle for a long time before they fail. The financial institutions that lent money to the struggling projects start to feel the bites during that period. Once the projects fail, the financial market’s problems get worse. The policy-makers’ timely and prudent measures about a liberalized financial system are essential to avoid imminent problems in the financial sector.
Another important finding of this study is that the more the time elapses after a banking crisis, the less the remains of the crisis, and the higher the probability of the recurrence of a severe crisis. We argue that as the past crisis gets older, the policy-makers are less vigilant about the stability of the financial market, which causes another banking crisis to start. We suggest that once a country experiences a banking crisis, efforts for a stable banking sector should be continued even after the crisis is resolved. Prudential financial market regulation, timely supervision with necessary stringent actions against problematic financial institutions should be inevitable part of the financial market management policy for all the time.
A few other findings of this study are not different from those in other studies. For instance, we also find that a high rate of inflation, a weak institutional environment, and the existence of a DIS are associated with the severity of a banking crisis. Therefore, we recommend price stability, transparent institutional environment, and a prudentially regularized and supervised DIS for a stable banking sector.
One limitation of this study is that we could not include the most recent banking crises experienced by the US and the European banking sectors due to the lack of complete data on crisis-cost and crisis-duration. We will extend this study when that data becomes available.
Country | Crisis began | Crisis-duration (in yrs.) | Crisis-cost | Cost-ranking | Severity of crisis |
---|---|---|---|---|---|
Argentina | 1995 | 1 | 0 | 0 | 0.5 |
Argentina | 1980 | 3 | 58 | 3 | 3 |
Argentina | 2001 | 4 | 71 | 4 | 4 |
Bolivia | 1994 | 1 | 0 | 0 | 0.5 |
Bolivia | 1986 | 1 | 49 | 3 | 2 |
Brazil | 1990 | 5 | 62 | 4 | 4.5 |
Cameroon | 1987 | 5 | 106 | 6 | 5.5 |
Chile | 1981 | 5 | 9 | 1 | 3 |
Colombia | 1982 | 1 | 47 | 3 | 2 |
Colombia | 1998 | 3 | 43 | 3 | 3 |
Finland | 1991 | 5 | 70 | 4 | 4.5 |
India | 1993 | 1 | 0 | 0 | 0.5 |
Indonesia | 1997 | 5 | 69 | 4 | 4.5 |
Jamaica | 1994 | 3 | 38 | 2 | 2.5 |
Japan | 1991 | 5 | 45 | 3 | 4 |
Jordan | 1989 | 3 | 106 | 6 | 4.5 |
Kenya | 1985 | 1 | 24 | 2 | 1.5 |
Kenya | 1992 | 3 | 50 | 3 | 3 |
Korea | 1997 | 2 | 58 | 3 | 2.5 |
Kuwait | 1980 | 4 | 143 | 8 | 6 |
Malaysia | 1997 | 3 | 31 | 2 | 2.5 |
Mexico | 1994 | 3 | 14 | 1 | 2 |
Mexico | 1981 | 5 | 27 | 2 | 3.5 |
Morocco | 1980 | 5 | 22 | 2 | 3.5 |
Nepal | 1988 | 1 | 0 | 0 | 0.5 |
Nicaragua | 2000 | 2 | 0 | 0 | 1 |
Nigeria | 1990 | 5 | 0 | 0 | 2.5 |
Norway | 1987 | 3 | 5 | 1 | 2 |
Panama | 1988 | 2 | 85 | 5 | 3.5 |
Paraguay | 1995 | 1 | 15 | 1 | 1 |
Peru | 1983 | 1 | 55 | 3 | 2 |
Philippines | 1997 | 5 | 0 | 0 | 2.5 |
Philippines | 1981 | 4 | 92 | 5 | 4.5 |
Senegal | 1988 | 4 | 6 | 1 | 2.5 |
Sierra Leone | 1990 | 5 | 34 | 2 | 3.5 |
Sri Lanka | 1989 | 3 | 20 | 1 | 2 |
Swaziland | 1995 | 5 | 46 | 3 | 4 |
Sweden | 1991 | 5 | 33 | 2 | 3.5 |
Thailand | 1983 | 1 | 25 | 2 | 1.5 |
Thailand | 1997 | 4 | 109 | 6 | 5 |
Togo | 1993 | 2 | 39 | 2 | 2 |
Tunisia | 1991 | 1 | 1 | 1 | 1 |
Turkey | 2000 | 2 | 37 | 2 | 2 |
United States | 1984 | 1 | 0 | 0 | 0.5 |
Uruguay | 2002 | 4 | 27 | 2 | 3 |
Uruguay | 1981 | 5 | 38 | 2 | 3.5 |
Venezuela, Rep. Bol. | 1994 | 5 | 1 | 1 | 3 |
Yemen, Republic of | 1996 | 1 | 16 | 1 | 1 |
Zimbabwe | 1995 | 5 | 10 | 1 | 3 |
Severity of crisis is the average of crisis-duration and cost-ranking. Crisis-cost is measured in terms of the percentage of GDP lost during the crisis. For cost-ranking, we used 1 for 1% - 20% GDP loss, 2 for 21% - 40% GDP loss, and so on.
Variable | Formulation | Source |
---|---|---|
GDP growth | IFS | |
Terms of trade | IFS | |
Inflation | IFS | |
Real interest | IFS | |
M2/Reserves | IFS | |
Credit growth | IFS | |
GDP per capita | Ratio of GDP (in US Dollar) to total population | IFS |
Existence of DIS | 1 if explicitly formed DIS exists in a country, 0 otherwise | Demirguc-Kunt et al. [ |
Remains of the past crisis | 0 if there wasn’t any past crisis; otherwise, the inverse of the period (in years) between two crises. Note that the past crises considered are the crises that took place in the 1980s and the 1990s. | Caprio & Klingebiel [ |
IFS = International Financial Statistics.
List of the countries used for Regression-1 in
Argentina, Armenia, Australia, Austria, Bahrain, Bangladesh, Barbados, Belgium, Belize, Benin, Bhutan, Bolivia, Botswana, Brazil, Burkina Faso, Burundi, Cambodia, Cameroon, Canada, Cape Verde, Chile, Colombia, Croatia, Cyprus, Côte d’Ivoire, Denmark, Dominican Republic, Egypt, El Salvador, Estonia, Fiji, Finland, France, Germany, Greece, Guatemala, Haiti, Honduras, Hungary, Iceland, India, Indonesia, Ireland, Israel, Italy, Jamaica, Japan, Jordan, Kazakhstan, Kenya, Korea, Kuwait, Kyrgyz Republic, Lesotho, Lithuania, Luxembourg, Madagascar, Malawi, Malaysia, Malta, Mauritius, Mexico, Mongolia, Morocco, Myanmar, Namibia, Nepal, Netherlands, New Zealand, Nicaragua, Nigeria, Norway, Oman, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Poland, Portugal, Rwanda, Saudi Arabia, Senegal, Seychelles, Sierra Leone, Singapore, Slovenia, South Africa, Spain, Sri Lanka, St. Lucia, Swaziland, Sweden, Switzerland, Thailand, Togo, Trinidad and Tobago, Tunisia, Turkey, United Kingdom, United States, Uruguay, Venezuela, Yemen, and Zimbabwe.
NB: Countries are dropped from the regression if they are centrally planned or socialist states4, subservient states5, states affected by civil war6, or if the data are missing on the variables that we use. Furthermore, parts of the study period for some countries are dropped because of their very transitional state of nature7.