**5. Conclusions**

Currency and banking crises have centuries of history, yet it is still difficult to mitigate future crises. The origins of currency and banking crises can be divided into three groups, which are rational actions, panic and contagion effect.

The first-generation model of currency crises started in the early 1980s with the speculative attack model which suggests the rational expectations of investors as the source of currency crises. In this model, investors are assumed to doubt the government's ability to manage a fixed exchange when there is a current account deficit. When the foreign reserve is drying in order to keep the exchange fixed, investors will attack the currency, leading to the breakdown of the fixed exchange rate regime.

When the speculative attack model could not explain the de facto breakdown of the European exchange rate mechanism which led to currency crises in Europe in 1992–1993, the second-generation model of currency crises which focuses on the selffulfilling model emerged. In this model, the herd behaviour of investors may cause panic and lead to the withdrawal of assets. As a result, the exchange rate tends to depreciate and translates into a crisis. Many aspects of herd behaviour such as coordinated action of investors, sequential observation of other investors' actions and information cascade are investigated. However, herd behaviour is not the whole story since investors are unlikely to ignore their new information where potential capital gain does not depend on other investors' actions.

On the other hand, early generations of banking crises focused on random withdrawal. This model shows that depositors can do a bank run due to random events such as sunspots or economic projections due to the lack of information held by the depositors.

Unclear triggers of the bank run in the random withdrawal model encourage the emergence of the information-based model. This model shows that the bank run is a logical consequence of a rational change of risk in bank portfolios.

However, in the aftermath of the Asian financial crises of 1997–1998, new models emerge that claim currency and banking crises can still occur in the absence of panic and a well-perform economy due to contagion effects.

According to the model, failure in one bank can spread to the whole banking system through money market currency and banking crises still can occur in the absence of panic and a low-risk environment due to a contagion effect.

The systemic risk claims that interbank lending increases the systemic risk for banks. Therefore, failure in one bank can lead to the failure of many other banks. Furthermore, due to the interconnectedness of financial markets, the twin crisis model shows that a currency crisis can easily translate into a banking crisis or vice versa.

In terms of the identification of crises, early studies of currency crises generally use the depreciation of the exchange rate at a certain level as a basis for determining the crisis. However, this may be biased when the central bank intervenes so that the exchange rate does not depreciate despite considerable pressure on the currency. For that reason, most recent studies use Exchange Market Pressure Index as the basis for the determination of currency crises. This model illustrates that the pressure on the exchange rate is not only reflected in the depreciation but also in the amount of central bank intervention through the spot market (and sometimes through interest rate).

On the other hand, the definition of a banking crisis is more complicated than a currency crisis. There are various methodologies to define a banking crisis. These methodologies consider various factors such as bank performance, government bailout, widespread bank failures, extensive bank runs and professional analysis to

specify bank crises. To address the complexity of the event approaches, the money market pressure index was developed to help determine the banking crisis.

In terms of methodology, while the multivariate logit model is arguably the most popular methodology for analysing currency and banking crises, it fails to provide a good forecast. On the other hand, while the signalling method is considered the most successful method to forecast financial crises, it is difficult to interpret the result as it is highly variable. The most recent study employs innovative techniques such as Markov switching models, artificial neural networks and genetic algorithms, and binary recursive trees. However, while they are much more complicated, the projection powers are still not significantly improve.

Furthermore, the empirical studies suggest that the currency and banking crises are typically preceded by a real appreciation, which is often represented as financial sector indicators (e.g. M2 multiplier, domestic credit/GDP, real interest rate) and a lending boom, which is often represented as external sector indicators (e.g. export, term of trade, real exchange rate, import, international reserve).
