creased. It could be explained by the fact that the weakened banks are forced to hold high capital in relation to assets.
In contrast, the longer-term leading indicators of future failure would be more useful as they can be used by regulators or central bank in order to develop action for prevention policy. For this reason, data have been examined from the pre-recession period, before banks actually weakened. The most statistically significant variable from the earlier period regression is loan growth. In this estimation period, high credit growth is positively correlated with failure and it is proved to be a significant precursor of the banking crisis. The rapid expansion of bank loans, at the peak of the previous boom, reflects loan quality problems in the next recession period because of a poor selection of creditworthy customers.
The predictive performance of indicators based on the regression with short time horizon has been encouragingly accurate in identifying subsequent failures. Appropriate variables have been selected in order to minimize the likelihood of multicollinearity and therefore, more accurate results have been exacted. However, the predictive ability of the model deteriorates when a longer-term period is explored.
The estimation of a multivariate logit econometric model, based on a large sample of developed and developing countries for 1980-94, indicates that banking crises tend to occur when the growth rate of GDP is low and inflation is high. Also, high interest rates and balance of payments deficits create deterioration in the bank condition. High exposure to foreign exchange risk tends to increase the likelihood of a banking crisis. Similarly, the bank profitability is negatively affected by an unexpected depreciation of the domestic currency, particularly when banks borrow in foreign currency and lend in domestic currency. The ratio of M2 money supply to the central bank's holdings of foreign-exchange reserves is introduced as an explanatory variable because vulnerability in the banking sector appears to be associated with sudden capital outflows. Moreover, the presence of deposit insurance scheme creates incentives for excessive risk-taking and therefore more fragile banking sector. Other factors significantly associated with increased vulnerability in the banking sector are low liquidity, a high share of credit to the private sector and past credit growth. Finally, countries with weak institutions are more likely to experience crises.
Secondly, the signaling approach is another major predictive method that has been adopted in the literature. Kaminsky (1998) and Borio and Lowe (2002) are examples of studies based on the signaling approach. Signal extraction model compares the behavior of single variables in periods of tranquility with periods of banking crisis. An indicator provides a signal of future crisis when it exceeds a predetermined threshold value which distinguishes between normal and abnormal behaviour for each variable. Nevertheless, there are two types of signalling errors, a type I error represents failure to identify a crisis and error type II represents a false prediction of failure.
The performance of signaling approach depends on the ability to accurately call crises and non-crises episodes. In common with the results from logit regression, the signalling approach reports that high real interest rates, rapid domestic credit growth, low output growth and falls in the terms of trade divided by real exchange rate are associated with banking sector problems. The most suitable approach for a global early warning model is the logit model whereas for a country specifi
