Leading indicators of financial crisis Which variables are most suitable?

Leading indicators of financial crisis

Which variables are most suitable?

by

Wouter Engelsman S1324748

University of Groningen Faculty of Economics and Business Msc International Business and Management

And

Leading indicators of financial crisis

Which variables are most suitable?

Abstract:

This paper examines which indicators, mentioned in literature about early warnings systems, are more likely involved in predicting a financial crisis than other indicators. In order to make this distinction data from 10 countries, which all experienced a financial crisis between 1994-2002 were used. This paper finds that domestic credit/GDP, equity prices, GDP growth and current account/ GDP are the best performing indicators.

Table of contents

1. Introduction p. 3

2. Literature Review p. 6

Financial crisis p.7

- First generation currency crisis model p.9

- Second generation currency crisis model p.10

- Financial Liberalization p.11 Leading indicators p.13 - Current account p.15 - Capital account p.17 - Financial sector p.19 - Real sector p.21 - Global economy p.22

Identification financial crisis p. 24

3. Methodology p. 26

Time periods p. 27

Crisis dates p. 29

Country coverage p. 31

Data collection indicators p. 33

1. Introduction

Detragiache (2000) both advance the logit approach, but base their warnings on different indicators for upcoming financial crises.

The main research question for my paper is therefore:

Which variables of the domestic and external sector can best be used in predicting a financial crisis?

In order to answer this question I will conduct an analysis of variance to test if certain variables, obtained from literature about EWSs, show significant difference between a normal period and a period prior to a financial crisis. A normal period in this paper refers to a period in a country when it is not faced with the dealings of a financial crisis. If there is indeed a significant difference between these periods, consistent across several historical crises, future variations could be harbingers for financial crises. The variables from both the domestic and external sector refer to the variables obtained from literature concerning EWSs. The structure of the paper is as follows. The next chapter deals with the definition of a financial crisis and then especially a currency crisis, which is essential to identify periods prior to a crisis. This is followed by several financial crisis theories, which will provide more insight into why certain variables deviate from normal behaviour prior to a crisis. The next step is to identify the variables which are expected to show signs of deviation prior to a crisis. I will restrict myself to the use of variables used in literature about EWS. Certain literature has shown that their variables contain predictive abilities under certain conditions and it can be expected that the best performing variables are among those indicators. The third chapter will discuss the methodological issues of the research; difference between the two previously mentioned periods, identification of the crisis dates, selection of the sample countries and finally an overview of the methods used to collect the indicators. The focus in chapter four is on the analysis of the research which is followed by a discussion of the results in chapter 5. Chapter 5 discusses the results of the research integrated within the theoretical background. The last chapter gives a report of the conclusions and limitations of the research.

2. Literature Review

This section of the paper deals with the literature used in papers concerning EWS. Central to this is understanding why the movement of a given variable will increase the probability of a financial crisis. The structure in order to answer this question is as follows. First the term financial crisis is defined and subsequently theories about the coming about of a financial crisis are discussed. These theories form an essential part for the explanation why the value of indicators can change prior to a financial crisis. This aspect is covered in the second section of this chapter called ‘the leading indicators’ and explains the inclusion of 40 indicators contained within eight articles used for this paper. The last part of this section will provide an overview of the methods used in order to identify the starting date of a financial crisis.

Financial crisis

focuses mainly on currency crises. A currency crisis occurs when participants in an exchange rate market come to perceive that an attempt to maintain a pegged exchange rate is about to fail, causing speculation against the peg that hastens the failure and forces a devaluation. The value of the currency changes quickly, undermining its ability to serve as a medium of exchange or to store value. In an article by Kaminsky and Reinhart (1999) the link between banking and currency crises is analyzed. They conclude that banking and currency crises are closely linked with banking crises, in general, preceding the currency collapse. That one type of crisis occurs before the other does not necessarily imply causality. Aziz, Caramazza and Salgado (2000) state, in contrast to the findings of Kaminsky and Reinhart (1999), that when the financial sector is poorly supervised and regulated, banking system fragilities may be fully revealed only after a run on a currency had undermined confidence. Although the main focus of this paper is on currency crisis there will be a strong focus on other financial indicators, ensuring enough measurement of the stability of the financial system. A further explanation about the identification of a currency crisis is provided in a later segment. The next section deals with the currency crisis theories.

Currency Crisis Theories

models. In these models the interaction between expectations and actual outcomes are central, in which market expectations directly influence macroeconomic policy decisions (Pesenti & Tille, 2000). The third theory refers to the liberalization of the financial system as a cause for a financial crisis.

First Generation Model

regime. This line of thought formed the development for a newer generation of currency crisis models, called the second generation model.

Second Generation Model

consequences for other policy targets, as illustrated in the above example with public debt, resulting in an increased probability of a speculative attack. Another action taken by the government which can enhance the probability of a currency crisis is the liberalization of the financial system.

Financial Liberalization

What is financial liberalization? Financial liberalization is the deregulation of interest rates, privatization of the banking sector and increase in the international capital movements ensuring stability in the financial system (Arican, 2007). The effects of financial liberalization are heatedly debated. According to Aizenman (2007) there is solid evidence that dismantling of capital controls in emerging countries can lead to the greater probability of a financial crisis. Aziz et al. (2000) agree and state that hastily or badly sequenced financial reforms that open up the economy to an increased volume of intermediation without adequate regulations have been an important factor in causing extreme currency instability. On the other hand, if these crises force the country to deal with its structural deficiencies, in the long run the financial liberalization can induce higher growth rates (Rancière, Tornell, & Westermann, 2005). In the short run the expansion of the financial system in emerging markets can be seen as a precursor for a financial crisis however, ultimately the restrictions on capital mobility are bound to bring sub optimal results (Aizenman 2007). But how does the dismantling of capital controls result into a currency crisis?

reserves is, as explained in the previous section, that it engenders expectations of devaluation which are self fulfilling.

Leading Indicators

Table 1: Overview of the variables used in this paper

closely linked to each other which make the extensive description of each indicator laborious. Appendix A gives an additional, more extensive overview of all the indicators and the method how they are gathered. The data for the indicators come from one data base viz. the International Monetary Fund, which is further explained in the methodology section.

Current account

The current account refers to a country's transactions arising from flow of goods, services, and investment income to and from other countries. Deterioration in the current account can trigger a currency crisis, because deficits need to be paid either with net capital inflows or with running down international reserves, which according to the financial theories can lead to a currency crisis.

The current account is also a standalone variable which can be used as an indicator for a financial crisis. This variable is one of the classic indicators of structural imbalances leading to a financial crisis according to the first generation model. In the current account goods, services, incomes and current transfers are recorded. A current account deficit implies net borrowing from the rest of the world. This can put a country in a vulnerable position. For example, a deterioration of the terms of trade of a country can considerably reduce its ability to repay its debt (Pesenti & Tille, 2000). Foreign investors might then decide not to extend further lending and call in their loans. The government is forced to help out the private institutions if the foreign creditors stop providing credit. Economic agents observing the weaknesses of the private sector can see that the government will be forced to adopt an expansionary fiscal stance in the future to finance the costs of bailout intervention (Pesenti & Tille, 2000). This action will increase the inconsistency between excessively expansionary fiscal policies and the proposition of a fixed exchange rate regime, which according to first generation models enlarges the possibility of a financial crisis. The worsening of the current account leads thus to a higher possibility of a financial crisis. To make sure the variable is measurable across countries, the current account is divided by GDP. The current account is denoted in dollars, so transformation using the exchange rate is necessary to divide it by GDP (see Appendix A).

by a deterioration of the export performance (Aziz, et al., 2000). An increase in the level of imports will have an adverse affect on the current account, making the economy less stable, and more receptive to a financial crisis. The data for import differs between import under Free On Board (F.O.B.) and import under Cost, Insurance and Freight (C.I.F.), which is dependent on the data availability in the IMF database. The C.I.F. import value is higher than the F.O.B. import value because under C.I.F. rules the country is obliged to pay the costs and freight to bring the goods to the port of destination, instead of the port’s loading cost only. The usage of either F.O.B. or C.I.F. is consistent for the sample in a country and therefore does not affect the outcome.

Another related variable is the terms of trade. Terms of trade is the relative price, on world markets, of a country's exports compared to its imports, which is calculated using unit value of exports divided by the unit value of imports (see Appendix A). Any increase in the terms of trade should strengthen a country’s balance of payments position and hence lower the probability of a crisis (Lestano, et al., 2003). This means that fewer exports have to be given up in exchange of a given volume of imports and vice versa. A big fall in the terms of trade signifies a reduction in real living standards since imports of goods and services have become relatively more expensive. As explained above any deterioration in the terms of trade may precede a financial crisis.

Capital account

Capital account is a country's international transactions arising from changes in holdings of real and financial capital assets. It includes foreign direct investment, plus changes in private and official holdings of stocks, bonds, loans, bank accounts, and currencies (Deardoff, 2006).

The value of M2 relative to international reserves captures to what extent the liabilities of the banking system are backed by foreign reserves. In the event of a currency crisis, individuals may rush to convert their domestic currency deposits into foreign currency, so that this ratio captures the ability of the central bank to meet their demands (Lestano et al., 2003). Abiad (2003) not only looks upon the level of M2 relative to the international reserves but captures the adequacy of the banks reserves also by looking at the growth rates. As explained earlier, expansionary monetary policy or sharp decline in reserves are associated with the onset of a crisis (Edison, 2000). The indicator short term debt/ international reserves resembles this connection. Other articles show the worsening of the level of foreign reserves in relation with imports (Frankel & Rose, 1996; Kumar, Moorthy, Perraudin, 2003).

The following variable measures the international level of debt (Frankel & Rose, 1996). The international level of debt equals the sum of debt securities liabilities and other investment liabilities measured in U.S. dollars for every country. This variable increases the changes of a financial crisis according to the second generation crisis models in the following manner. The response from a government towards market expectations of currency depreciation is usually to raise domestic interest rates. This will attract more investment, increasing the demand for the currency. The downside is that highly indebted governments with mostly short-term or floating-rate nominal debts will find their fiscal burden increased (Obstfeld, 1996). Increase in debt thus increases the changes of a financial crisis.

and prospects, which would indicate a flourishing economy, and lower the possibility of a crisis (Kamin, Schindler & Samuel, 2001). The last variable is government budget divided by gross national product (GNP) which Frankel & Rose (1996) use as a crude measure for fiscal policy. A relatively high budget deficit could easily tilt the market sentiment against the domestic currency (Brüggemann & Linne, 2002).

Financial sector

This sector entails the leading indicators related to the financial sector, especially focused on the fragility of the banking sector. As mentioned in the section about financial crisis theories, financial liberalization reforms in a country are often accompanied with problems evocating a financial crisis. These changes in financial intermediation, ‘financial liberalization’, are often measured by changes in M1 or M2 levels (Aziz et al., 2000). The first four variables refer to these changes (see table 1). It is widely accepted that narrow money (M1) and broad money (M2) are the most conventional measures of liquidity (Feridun, 2006). According to the first generation model a money supply that exceeds the demand for domestic currency will influence market agents to increase the demand for foreign currency in a response to the excess liquidity on the domestic money market. This can fuel speculative attacks on the currency thus leading to a currency crisis. Lestano et al., 2003).

confidence in banks by looking at the ratio between banks deposits and M2. The last variable is a rough measurement of capital adequacy; the ratio of banks reserves to bank assets. Bank assets consist out of claims on the central government, claims on state and local governments, claims on official entities and claims on the private sector (see Appendix A). This indicator shows in which matter a bank’s capital can cope with potential losses. A lower value would indicate less protection for depositors, thereby decreasing confidence in the financial system.

Real Sector

rates in the event of a speculative attack if there is high unemployment, since this will only exacerbate the unemployment problem.

The last variable in the real sector is the index of equity prices. A feature often preceding a financial crisis is the bursting of asset price ‘bubbles’. This was recently the case with the ‘dotcom bubble’ in which the value of new internet based companies collapsed within a few months of joining the stock markets. A devalue weakens the balance sheets of companies, making them unable to repay their loans and suffer loss or bankruptcy. This weakens the economy and may facilitate the upcoming of a financial crisis.

Global economy

The global economy refers to indicators measured outside the examined country. They are indicators with a more global nature and are similar for all the countries used in this research. One such indicator is the GDP growth of G7 countries. If these countries experience a period of economical growth, this will result in higher foreign import growth, hence strengthening exports and thus reducing the probability of a crisis (Kamin et al., 2001). Vice versa, one could argue that when these countries experience a decline in GDP, this will result in a decline of export to these countries, consequently leading to a worsening of the current account. Although the effects of these indicators are indirectly visible on domestic variables, they are mentioned separately in literature (Edison 2000; Kamin et al., 2001).

associated with regressions, and thus viable for the prediction of a financial crisis. A distortion in the price level of world oil prices is thus the last variable.

Identification financial crisis

A critical element in the development of an EWS scenario is to identify periods of financial crisis. A great diversity of methods for identifying the crises appears in the literature and follows that different periods can be identified as being a crises situation (De Vicente, Alvarez, Perez, & Caso, 2008). An apparently small change in the identification of a financial crisis could be of significant impact in the amount of correctly indentified financial crises. A common element in the literature on predicting financial crisis is the pressure on the exchange rate.

Early work is based on identifying significant changes in constructed speculative pressure indexes (SPI). A SPI tries to detect episodes of strong turbulences in the currency market in which high devaluations take place. Hence, they are calculated as the weighted sum of the exchange rate, interest rates and international reserves variations (volatilities of the three components being equalled) (Perez, 2005). Crisis is declared when the correspondent SPI is above an arbitrary threshold. One of the earliest indexes is the one by Eichengreen, Rose, & Wyplosz, (1995), which is based upon the weighted average of exchange rates, interest rates and variations in international reserves. A problem with this approach is the lack of data in respect of market interest rates for a large number of emergent countries. Many authors disregard this component of the index; Kaminsky et al. (1998), Berg & Pattillo, (1999) and Edison (2000).

the weighted average of two-month changes in real exchange rate and reserves changes more than 1,75 (country specific) standard deviations above the mean. As can be imagined, a small variation can have considerable consequence for the identification of the crisis.

3. Methodology

The ANOVA will be used to test if the level of a variable will significantly differ between a normal period and a period preceding a financial crisis. The definition normal will be further specified in the next chapter. The expectations are, in line with the currency crisis theories, that the level of a variable would differ between the two mentioned periods. For example, according to the first generation crisis model it is expected that the level of foreign reserves is lower prior to a financial crisis than in normal times. If the test confirms that there is indeed significant difference this would indicate that the variable shows different behaviour preceding a financial crisis. A distinction between the variables can be made with this method; dividing them between variables which show significantly different behaviour prior to a crisis and variables which show no significant signs prior to a crisis. The level of significant difference will be in line with other literature and be set on 1 % (Berg and Pattillo, 1999).

The remainder of this chapter is organized as follows. The first step is to identify the difference between the two periods, followed by the identification of the crisis dates and the selection of the countries. Next I will discuss the data collection and data transformation, which is accompanied with an overview of the descriptive statistics of the data set. The last section will elaborate on the tests.

Time Periods

policy maker when using an EWS is to extract as many correct signals about an upcoming crises and identify at the earliest stage possible. The problem arising from this preference is that the longer the period of investigation prior to a crisis, the fewer missed crisis, but the more false alarms, and vice versa. This is referred to as type 1 errors, smaller periods and more missed crises, and type 2 errors, longer periods and more false alarms. This paper will rely on both an 18 and 24 month period prior to a crisis. There is a preference for periods longer than 12 month, because type 2 errors may cause less severe problems than type 1 errors. Argumentation in favour of this preference is twofold. The first argument, adopted from the work of Fuertes & Kalotychou (2007), is based upon costs. The cost of missed investment opportunities or those of adopting pre-emptive policies following false warning are often less severe than the losses accompanied by a missed crisis. The second argument relates to the warning ability of false alarms. For instance, when a variable crosses a certain threshold, for example 2 times its standard deviation, this would indicate that there is an increased risk of a financial crisis. When this warning is giving there must be a financial crisis within a certain threshold, for instance 24 months, otherwise the warning is called a false alarm. But what if this warning simply reflects that although there were severe economic weaknesses, suitable policy actions were taken and a crisis was avoided. See the following simplified example to illustrate this point. Imagine that the level of the variable M2 is significantly higher than its normal position. Authorities, warned by the signal, take conservative measures regarding their excess liquidity, which result in fewer speculative attacks and as a result there is no financial crisis. With the length of the period prior to a financial crisis identified the next step is to focus on the tranquil period.

considered normal if it is two years away from a financial crisis in both directions. The length of this period is of less concern amongst authors than the period prior to a financial crisis, but also differs greatly. Some base there research on 1 year (Zhang, 2001) and others on 20 years (Frankel & Rose, 1996). To make the test as reliable as possible I will rely on different timeframes for the normal period. Obviously, an indicator which shows significant behaviour on three different timeframes is more reliable than an indicator which does not. Given the basis of my classification for a normal period it is not feasible to look at a stretch of 20 years, because this restricts me to a limited number of countries. Therefore I will use 36, 60 and 84 months as periods for the normal period. These periods will give enough measurements to conduct a reliable test. The next step is to determine the beginning of a financial crisis.

Crisis dates

The starting point of the identified periods is dependent on the starting point of the financial crisis. But when does a financial crisis exactly start?

As mentioned in the literature review, there is a broad spectrum of crisis identification used throughout literature. The most common method for the identification of a financial crisis is that authors construct their own measure of speculative pressure indexes (SPI) to identify periods of greater than normal pressure (currency crisis). Kaminsky et al. (1998) based their index on weighted averages of exchange rates and reserves, which is generally the same index used by Eichengreen et al. (1995) except that the latter includes interest rates changes. Recent studies questioned the inclusion of interest rate changes when determining a crisis date, due to the lack of market-determined data (Abiad, 2003). Another method of crisis identification considers the sudden and unusual changes in exchange rate (Frankel & Rose (1996).

confidence that any particular dating system has correctly identified the crises in a sample. It would be even more difficult to create my own measurement of exchange rate pressure given that results are not guaranteed.

Country coverage

The countries included in this research must comply with several qualifications. These qualifications are; (1) crisis dates can be obtained from previously mentioned literature, (2) be an upcoming market, (3) they should have had a financial crisis in the last 15 years (4) availability of data (5) the countries should be comprised from a selection from around the world in different time zones. The first requirement relates to the availability of currency crisis dates from the literature of either Edison (2000) or Kaminsky (2006). Edison (2000) gives an overview of 28 industrial and developing countries and Kaminsky (20006) of 20 countries. Many countries are mentioned both in the research by Edison (2000) and by Kaminsky (2006). The identification of a crisis in the wrong month ruins the reliability of the tests and their aggregated results. For instance, when a crisis is identified three months prior to the occurrence of the actual crisis, these last three months are not taken in the measurement, which could be crucial in determining if a variable significantly differs in value prior to a crisis or not. The second requirement is in line with other research which shows that financial crises in developed economies tend to be of a fundamentally different nature than those in upcoming markets (Bussière & Fratzscher, 2006). One of the reasons mentioned is that industrial countries generally have a more developed financial infrastructure than the emerging markets. This would imply that emerging markets are more receptive to financial liberalization, see chapter 2, and consequently to a financial crisis. This reduces the amount of countries from 28 to 21 and 20 to 15.

markets. Reason was the investors need for portfolio diversification, which fuelled the exposure of these institutional investors in emerging countries.

Qualification four is pretty straightforward and is further explained in the data collection section. This reduces the amount of countries respectively to 11 and 9. The last criterion is included to ensure that the results are not based upon the characteristics from one crisis or one region. For instance, if the selected countries all come from Asia certain characteristics of the Asian market are included in the outcome of the tests. To ensure that this does not occur the sample countries consist of a variety of countries from around the world. In total, including the countries merged from both the Edison (2000) and Kaminsky (2006) selection, there are ten countries which complied with all the criteria; three South American countries, five Asian countries, one African country and Turkey. Table 2 provides an overview. This is an average which is comparable with other authors. Kaminsky et al. (1998), Berg & Pattillo, (1999) and Kumar et al. (2003) use 20 countries and whilst there are other examples of authors focusing on a specific region and using a smaller amount of countries; Lestano et al. (2003) and Abiad (2003) for instance both focus on only 6 countries.

Data collection indicators

The methods by which Kaminsky et al. (1998) and Frankel & Rose (1996) identify the starting data of a currency crisis are both based upon monthly data as are the leading indicators. This limits the availability of data, because some indicators are only available on a quarterly or annually basis. In order to get around this problem, I have used linear interpolation to transform certain data. Linear interpolation is a method of constructing new data points within the range of a discrete set of known data points. There are different methods of interpolation which vary from easy to use but with a higher error margin, to more difficult in handling but presenting a lower error margin. The linear interpolation used in this paper belongs to the former category. With linear interpolation the error is proportional to the square of the distance between the data points (Blu, Thévenaz & Unser, 2004). To reduce the error estimate as much as possible the linear interpolation is only used to interpolate quarterly data into monthly data. Furthermore, interpolation from annual into monthly data would affect the test in a disproportional way. The test is conducted on several time periods, the smallest being a period of 60 month; a change in one year (12 months) has too much consequences.

Although in general authors collect their data from various sources, this paper uses data from the International Monetary Fund, specific to the International Financial Statistics section. The reason is threefold. First the data should be free to access. The paper by Frankel & Rose (1996) uses a World Bank’s World Data CD-rom which is only available in return of a usage fee. Secondly, most data sources do not give long enough time series. For example, Edison (2000) uses data from the Bank for International Settlement, but they only provide data dating back towards the year 1990. The last argument relates to the previous section in which the usage of annual data, quarterly data and monthly data is analysed. Most sources provide annual data, which is not useful for this research.

respect to its level a year earlier. This ensures that the indicators are comparable across countries. Furthermore, this method makes the indicators stationary and takes into account the change over time. For example, the level of GDP is expected to increase with the development of the economy. For example, the ANOVA test for South Korea will compare the level of the tranquil period (5 years), November 1990- October 1995, with that of the period prior to the crisis (2 years), November 1995-November 1997. The trend increase of GDP every year would cause a considerable difference in value within a time span of 7 years. The proposed method would look upon the increase of the variable in relation to the value a year earlier. To apply this method an additional year is necessary to collect the percent change in the first year of the normal period. This extension does not violate the conditions which are stated to identify a normal period for all 10 countries. Certain variables are free from this exclusion, such as indicators associated with the interest rate and indicators in the form of fraction. These indicators are also variable over time, but they do not continuously grow in accordance with the size of the economy. For example, in contrast with the level of international reserves which is expected to increase with the development of the economy, the real interest rate can stay stationary over time. The level form of these indicators is maintained. The transformation of these indicators into 12 month percentage change will show false values.

The final subject concerning the transformation from data in indicators is related to the real exchange rate. As was briefly discussed in the literature section, the overvaluation of the real exchange rate is calculated by looking at deviations from a trend. The trend is based upon the average of the real exchange rate in different periods. These periods differ, because they are calculated using different criteria. For example, when using 60 months for a normal period and 24 months for the period prior to a crisis, the real exchange rate trend will be based upon the average of 84 months. See appendix A for the precise calculation for the real exchange rate.

Table 3: Descriptive statistics of all the countries

The next striking feature is the level of variance. The level of variance is a measure of statistical dispersion among involved values. The higher the level of variance, the more the separates values differ among each other. Two indicators have a distinct higher value: FDI and the trade balance. Both indicators are stated in their real value instead of the percentage change, which makes the countries less comparable among each other. The values of these indicators are much more tied to the countries characteristics than other indicators. Domestic credit/GDP is also stated in its real value instead of percentage change. The level of equity price is very volatile and large differences among monthly fixtures are common, let alone the difference among periods of 12 months. This explains the outlier for the maximum statistic 20 times the standard deviation. Other outliers, for instance outliers in the M1 balance, are more incidental and do not affect the variance level in a disturbing matter.

ANOVA

4. Analysis

For the indicators there are 3 different periods for the normal period, 2 separate periods for the period prior to a crisis and in 6 of the 10 countries there were differences between the classification of the starting date of the crisis between the SPI method by Kaminsky et al. and the Frankel & Rose method. Table 4 gives an overview in which percentage of the cases an indicator has a significant different value prior to a crisis compared with a normal period of 84 months. This distinction is based upon an ANOVA analysis, which is explained in the previous chapter. Differences generated by these variances will be explained in the subsequent section. The result are combined in table 6, where after the discussion of the outcome commences.

The four different columns in table 4 show two differences: one based upon the difference in crisis dating method and the other based upon the difference in period prior to a financial crisis. KLR 24 months in table 4 means that the SPI method by Kaminsky et al. (1998) for the crisis identification is used, the period prior to the crisis is based upon 24 months and the normal period is 84 months. The percentages in the table are constructed using appendix B. Appendix B gives a complete overview of the separate significant indicators of every county. For example, appendix B table 2 ,FR 24 months, shows that 2 of the 10 countries have a significant different value prior to the crisis based upon the indicator export. This is also visible in table 4, the value for the indicator export when based upon FR 24 months is 20%. Not every indicators is based upon 10 countries, due to missing data, which explains the variations among indicators. Table 5 is in the same way constructed as table 4, the only difference is the amount of months for the normal period, which is 60 or 36 months.

The differences in table 4 & 5 are based upon the three distinctions mentioned earlier; difference in crisis date identification, difference in the normal period and difference in the period prior to a crisis. Although the difference between the starting date of a financial crisis can be as large as 3 months according to the two methods of crisis date identification (see Table 2, Malaysia), this does not generate a lot of different results. Usually there is no difference at all between the KLR 24 months and FR 24 month’s columns or between the KLR 18 months and the FR 18 month’s columns. If and when there is a difference it does not exceeds 16,67 %, which represents in most cases a difference of one observation only. The two other distinctions have a far greater impact on the results.

According to Fuertes & Kalotychou (2007) the amount of indicators signalling the upcoming event of financial crises in EWS models increases when the examined period prior to a crisis becomes longer. The results from the tables support this proposition. The crude indicator ‘total’ shows that 5 out of 6 times the amount of significant indicators is higher for a 24 month period prior to a crisis than a 18 month period. However, the results are ambiguous. The indicators ‘real interest rate’ and ‘growth of world oil price’ both show a distinct difference between 60 months KLR 2 and KLR 1,5 (table 5). This suggests that the value for ‘real interest rate’ and ‘growth of world oil price’ both steeply ascend prior to a financial crisis. When the difference is compared between ‘growth of world oil price’ and ‘real interest rate’ between 36 months KLR 2 and KLR 1,5 the data suggest other results (table 5)

the changes of higher values rises. Despite a small number of clear differences, in general the samples are comparable. No cases were found which having a high percentage on the basis of 84 months and non or very low values for the other periods, or vice versa.

From table 4 and 5 it can be concluded that alterations of the conditions can change the outcome to more desirable results. For instance, the increase of world oil prices is significantly higher when the investigated period prior to a crisis is reduced from 24 to 18 month. The same indicator does not give sign of any significant result for 84 month normal period and gives moderate significant results for the 36 month period. However, in the last mentioned period the outcome between 24 month and 18 month is reverse to the 60 month period. The explanation for these diverse outcomes lies in the length of the chosen periods and the capriciousness of the chosen data. This example is not illustrative for the entire sample, but it does reflect the proposition that different conditions generate different outcomes.

5. Discussion results

Table 6 provides an overview of the results and shows which percentage of the time the indicators have significant different values prior to a crisis compared with normal periods in accordance with the currency crisis theories from chapter 2. The top 7 indicators from table 6 have significant different behaviour in more than halve the cases. The results demonstrate that typically prior to a currency crisis an economy will show the following characteristics: a significantly expansionary monetary policy resulting in higher amounts of domestic credit, sharply declining equity prices, decreasing growth figures and current account, a worsening of the trade balance and of the capital adequacy of national banks and final an increase in debt.

How do these results relate to the theories discussed in chapter 2? The introduction mentions the diversity among models predicting financial crises (EWS). The top indicators from table 7 are mentioned in other researches as best performing indicators, but never together in one research, which further emphasizes the diversity among EWSs. Of course the indicators are gathered from papers about EWS, but this does not necessarily mean they are among the best performing.

The second best indicator is a sharp decline in equity prices. Falling equity prices can be the messenger that heralds lower economic activity. Lower economic activity weakens the debt-servicing capacity of domestic borrowers and contributes to increasing credit risk which results in fewer investments in the economy. This is especially dangerous after a long rise of asset prices. Rising asset prices may encourage banks and financial markets to expand the availability of credit, and enable firms and households to increase their purchases of capital goods (Brealey & Vila, 1998). Yet, when asset prices fall substantially, those additions to capital may seem in retrospect unwarranted. Personal consumption and corporate investments may be reduced and the repayments of the loans that supported the earlier capital acquisitions are under pressure. Due to their overstretched balance sheets financial institutions cannot lend more money, nor are they willing to lend to other banks who are also greatly overstrained and hence not able to repay loans. The fall in equity prices is an indicator which in most cases can be linked with an upcoming financial crisis, which is supported by Edison (2000), who states to look upon the equity price with more scrutiny.

In accordance with Obstfeld’s (1986) second generation currency crisis model low growth rates and a high debt usually force the government to abandon a fixed exchange rate. The two indicators both have a high score, which could indicate that countries are more receptive to financial crises when the growth rates decline and foreign debt increases. Interesting in that matter is the semi significant increase in world interest rates. One would think the combination of high indebtedness and an increase in world interest rate would be particularly devastating for economies, but this case is not wholeheartedly supported by the result. (See appendix B)

thrive on investments from abroad, but it also makes them more dependable. A cessation of foreign capital, ‘sudden stop’, can trigger a currency crisis (chapter 2). Table 6 shows that not every form of foreign investment can be associated with the occurrence of a crisis because in most cases there was no significant decrease in foreign direct investment prior to a currency crisis. A possible reason maybe that FDI is directly tied to real investment in plant, equipment and infrastructure, which does not add to the destabilizing effect of foreign capital. The last indicator, which value is significant lower in more than halve the cases, is the amount of bank reserves relative to total bank assets. This lower value would indicate less protection for depositors, thereby decreasing confidence in the financial system.

One of the striking results of table 6 is the absence of the real exchange rate exchange as one of the top indicators. A possible explanation is the method in which this paper calculates the overvaluation of the exchange rate. The overvaluation is based upon the deviation from the average value for the real exchange rate for the entire sample. The entire sample from this paper is smaller than of other papers (Kaminsky et al., 1998), which could have influenced the outcome.

The indicators showing the most significant difference prior to a currency crisis and thus most likely to be involved in the prediction of currency crises, cannot be classified in one category. They come from variables originating from the current account, capital account, financial and real sector. Interesting to see is that the indicators which perform badly are usually mentioned in one paper. The indicators based upon growth rates mentioned in the paper by Abiad (2003) seem to yield no fruitful results nor does the indicator inflation mentioned by Aziz et al. (2000). These indicators can work in the specific setting accompanied by the model from Abiad (2003) or Aziz et al. (2000), but are otherwise not very likely to generate results.

financial crises, 5 are linked to the Asian financial crisis: Indonesia, Malaysia, Philippines, South Korea and Thailand. To check the sensitivity of this specific crisis, these countries are excluded from the test. The sensitivity for the choice of countries is tested by looking at the effect of the exclusion of one country. In this case the exclusion of Argentina. The last test is an out of sample test. An out of sample test measures if the results shown in table 6 are not only an outcome of the chosen conditions (crisis choice, country choice), but also if they are achieved under other conditions. To resemble other conditions I have looked if the same leading indicators would show significant difference prior to a crisis in Hungary and Romania in October 2008. October 2008 is the starting date of the global financial crisis and in this month both Hungary and Romania’s currency devaluated by more than 15 %. Furthermore, both Hungary and Romania are geographically located in another part of the world compared to the countries in the sample. The outcomes of these 3 additional tests are almost fully comparable with the outcome shown in table 6. The top 4 from table 6 are also in the top 5 of the other tests, which can be checked in Appendix C.

6. Conclusion

This paper attempts to find out which variables of the domestic and external sector are best to be used in predicting a financial crisis. The research is based upon 37 variables, previously mentioned in eight papers about EWS. EWS are models with the aim of anticipating if and when individual countries may be affected by a financial crisis. The 37 variables are divided in different periods to asses if their value prior to a financial crisis deviates from normal. If this is the case it is likely that deviations in the future can also be harbingers for an upcoming crisis. The difference between periods is tested by an analysis of variance. To ensure the results are not specific to one region, the tests are conducted on indicators from 10 upcoming economies around the world. The top performing indicators are; domestic credit/ GDP, the index of equity prices, GDP growth and current account / GDP. Other indicators which also performed well are trade balance, debt / GNP and bank reserves/ total banks assets.

This does not mean other variables cannot be used in order to predict a financial crisis. It means the behaviour of these indicators prior to a crisis is in most cases significantly different compared with normal periods. Close monitoring of these indicators would give a better indication and more reliable results. In short, these indicators are more suitable. These findings are also supported by earlier literature, which use different methods than this paper. The EWS used by Aziz et al. (2000) find that the indicator ‘domestic credit/ GDP’ is one of the best indicators for an upcoming crisis, whilst Kumar et al. (2003) sees a weakening of the real activity as a more precise indicator. Notable is that many of the existing literature finds a link between the overvaluation of the exchange rate, but this is not supported by the findings of this paper.

the outcome from table 4 & 5. Conditions can be manipulated in such distinct matter that they will generate desirable outcomes. In order to keep the conditions as reliable as possible I used different methods to identify the crisis dates, different sample sizes for the two periods and conducted the tests in countries scattered around the world. Despite these attempts it is conceivable to think the outcome would be different when one of the conditions changes. One of the conditions which should be scrutinized is the way a normal period is defined. Using this classification a country can cope with political turmoil, natural disasters or even economical instability, but if these events do not lead to a financial crisis they are not taken into consideration.

Another limitation is the dependence on one data source only, IMF. The data for certain indicators could not be found and some of the data has been interpolated from quarterly to monthly data, which both can affect the outcome of the tests. This transformation from quarterly data into monthly data was necessary to ensure enough data points, but this method leads to an estimation which is by definition less accurate than the actual data would lead to.

References

Abiad, A. 2003. Early-warning systems: A survey and a regime-switching approach,

IMF working paper, 1-60

Arias, G., (2003) Currency crises: what we know and what we still need to know

C.E.F.I. Working Paper, 1-73

Arican, E. 2007. Relation between financial liberalization and foreign currency crises in Turkey: an applications in terms of foreign currency crises. Journal of American

Academy of Business, 7: 236-246

Aizenman, J. 2007. Financial Crisis, Princeton University Press, 1-17

Aziz, J., Caramazza, F., & Salgado, R. 2000. Currency crisis: In search of common Elements. IMF working paper, 1-55

Berg, A., Borensztein, E., & Pattillo, C. 2005. Assessing early warning systems: how have they worked in practice?. IMF Staff Papers, 52(3): 462-502

Berg, A., & Pattillo, C. 1999. Predicting currency crises: The indicators approach and an alternative. Journal of International Money and Finance, 18(4): 561-586. Berger, W., & Wagner, H. 2005 International monetary fund interdependent

expectations and the spread of currency crises. IMF Staff Papers, 52(1): 41-54 Bird, G., & Rajan, R. S. 2001. Banks, financial liberalization and financial crises in

emerging markets. World economy, 24(7): 889-911

Brealey, R., & Vila, R. 1998. Equity prices and financial stability. Financial Stability

Brüggemann, A., & Linne, T. 2002. Are the Central and Eastern European transition countries still vulnerable to a financial crisis? Results from the signals approach.

Discussion Papers, 5: 1-24

Blu, T., Thévenaz, P., & Unser, M. 2004. Linear interpolation revitalized. IEEE

Transactions On Image Processing, 13(5): 710-718

Bussière, M., & Fratzscher, M. 2006. Towards a new early warning system of financial crises. Journal of International Money and Finance, Vol. 25: 953-973

Bussière, M., & Fratzscher, M. 2007. Low probability, high impact: policy making and extreme events. Journal of Policy Modelling, 30: 111–121

Deardoff, A. V. 2006. Terms of trade: Glossary of international economics, World Scientific Publishing Company.

Dermirguc-Kunt, A. & Detragiache, E. 2000. Monitoring banking sector fragility: a multivariate logit approach. World Bank Economic Review, 14(2): 287–307. De Vicente, S., Alvarez, P., Perez, J., & Caso, C. 2008. Does currency crisis

identification matter?. Applied Financial Economics, 18: 387–395

Edison, H. 2000. Do indicators of financial crises work? An evaluation of an early warning system. International Finance Discussion Papers, 1-74

Eichengreen, B., Rose, A. K., & Wyplosz, C. 1995. Exchange market mayhem: The antecedent and aftermath of speculative attacks. Economic Policy, 10(2): 251-312 Frankel, J. A., & Rose, A. K., 1996. Currency crashes in emerging markets: An empirical

Fuertes, A-M, & Kalotychou, E. 2007. Optimal design of early warning systems for sovereign debt crises. International Journal of Forecasting, 23: 85–100 Grenville, S. 1998. Capital flows and crises. Asian-pacific economic literature, 1-15 Huisman, M. 2009. Experimental Designs and Analysis of Variance. Harlow: Pearson

Education Unlimited

Jeanne, O. 2000. Currency crises: A perspective on recent theoretical developments.

Special papers in International Economics, 1-51

Jongwanich, J. 2008. Real exchange rate overvaluation and currency crisis: Evidence from Thailand. Applied Economics, 40: 373–382

Kamin, S. B., Schindler, J. W., & Samuel, S. L. 2001. The contribution of domestic and external sector factors to emerging market devaluations crises: An early warning systems approach. International Finance Discussions Papers, 1-56

Kaminsky, G. L. 2006. Currency crises: Are they all the same?. Journal of

International Money and Finance, 25(3): 503-527

Kaminsky, G., Lizondo, S., & Reinhart, C. 1998. Leading indicators of currency crises.

IMF Staff Papers, 45(1): 1-48.

Kaminsky, G., & Reinhart, M. 1999. The twin crises: the causes of banking and balance-of-payments problems. American Economic Review, 89(3): 473-500 Krugman, P. 1979. A model of balance-of-payments crises, Journal of Money, Credit,

and Banking, 11: 311-325

Krznar, I. 2004. Currency crisis: theory and practice with application to Croatia.

Kumar, M., Moorthy, U., & Perraudin, W. 2003. Predicting emerging market currency crashes. Journal of Empirical Finance, 10: 427– 454

Feridun, M. 2006. Impact of liquidity on speculative pressure in the exchange market.

Department of Economics Discussion Paper Series, 24: 1-12

Lestano, Jacobs, J., & Kuper, G. H. 2003. Indicators of financial crises do work! An early warning system for six Asian countries. International Finance, 1-39

Minsky, H. P. 1964. Longer waves in financial relations: Financial factors in the more severe depressions. American Economic Review, 54(3): 324-336

Minsky, H. P. 1995. Longer waves in financial relations: financial factors in the more severe depressions II. Journal of Economic issues, 29(1): 83-96

Obstfeld, M. 1986. Rational and self-fulfilling balance of payments crises.

American Economic Review, 76(3): 72-81

Obstfeld, M. 1996. Models of currency crises with self-fulfilling features. European

Economic Review, 40(3-5): 1037-1047

Ozkan, F. G., & Sutherland, A. 1998. A currency crisis model with an optimizing policymaker. Journal of International Economics, 44(2): 339-365

Perez, J. 2005. Empirical identification of currency crises: Differences and similarities between indicators. Applied Financial Economics Letters, 1: 41–46

Pius, 1999. Outsmarting another crisis: an early warning system for the Philippines.

Development Research News, 17(3): 1-3

Rancière, R., Tornell, A., & Westermann, F. 2008. systemic crises and growth

Quarterly Journal of Economics, 123(1): 359-406

Sbracia, M., & Zaghini, A. 2001. The role of the banking system in the international transmission of shocks. Temidi discussione, 1-40

Zhang, Z. 2001. Speculative attacks in the Asian crisis. IMF Working Paper, 1-21

Appendix A

Variables are retrieved from the International Monetary Funds data base expect for the LIBOR rate. Unless otherwise noted, a 12-month percent change measure is used for each variable.

Current account

Real exchange rate: The real exchange rate (RER) index is derived from a nominal exchange rate index (IFS-AE) , adjusted for relative consumer prices(measured in the same currency). Real exchange rate= AE*PF/P, where AE = nominal exchange rate (IFS –AE), P = domestic price (CPI), and PF = foreign price (US CPI). Deviation from trend is calculated using the following formula:[ RER/ (RER average entire sample period)]-1.

Current account/ GDP IFS line 78/ IFS line 99B. Both indicators are only available on a quarterly basis, linear interpolated to monthly data. Transformed to same currency with IFS line AE. (format value)

Trade balance Export/ Import (format value)

Export IFS line 70_D

Import IFS line 71_D

Terms of Trade Unit value of exports divided by the unit value of imports. Unit value of exports is IFS-74.D. Import unit value for country is IFS-75.D

Capital account

International reserves IFS line 1L.d

M2/ international reserves IFS lines 34 plus 35 converted into dollars (using IFS line AE) divided by IFS line IL.d.

Short term debt/ reserves Data unavailable

International reserves/ imports IFS line 1L.d / IFS line 71_D

Imports/ international reserves IFS line 71_D/ IFS line 1L.d

Total debt/ GNI: Total debt equals the sum of debt securities liabilities (IFS line 78bnd) and other investment liabilities (IFS line 78bid) measured in U.S. dollars. Gross National Income (IFS line 99a) is

transformed to $ using IFS line AE. Both indicators are linear interpolated to monthly data. (format value)

Foreign direct investment IFS line 78bed Linear interpolated to monthly data. (format value)

FDI/GDP IFS line 78bed / IFS line 99b. Both interpolated to

monthly data. (format value)

Government budget/GDP not available

Financial Account

M1 growth (money) IFS line 34

M2 growth (quasi money) IFS line 35

M1 and M2 growth IFS line 34 + IFS line 35

Change in M2-to-M1 ratio IFS line 35 divided by IFS line 34

Domestic credit IFS line 32

Domestic credit/ GDP IFS line 32 divided by interpolated IFS line 99b. (format value)

Excess real m1 balance Percentage difference between M1 (IFS-34)

deflated by CPI (IFS-64) and estimated demand for M1. Demand for real M1 is estimated as function of GDP (interpolated IFS line 99B), nominal interest rates (IFS-60L). (format value)

Real interest rate Deposit interest rate (IFS line 60l) deflated by consumer price inflation (IFS line 64). (format value)

Real interest differential Real interest differential is constructed as the difference between real rates for the domestic and foreign country, using real interest rates. The US interest rate is used as the foreign country interest rate. (format value)

Inflation Consumer price inflation (IFS line 64)

Bank deposits Demand deposit (IFS-24) plus time, savings and

foreign currency deposits (IFS-25) deflated by CPI (IFS-64)

Bank deposits/ M2 Bank deposits / M2

Bank deposits/ M2, growth rate Bank deposits / M2, (format value)

Lending rate/ deposit interest rate Lending interest rate (IFS-60P) divided by 6 month

time deposit rate (IFS-60L). (format value)

Bank reserves/ total bank assets Bank reserves (IFS line 20) divided by bank assets (IFS line 21 plus IFS lines 22a-f). (format value)

Real Sector

GDP growth IFS line 99b, interpolated to monthly data

Industry output IFS line 66, only for Indonesia IFS line 66a

Unemployment not available

Global Economy

U.S.A. interest rates IFS line 60C (format value)

LIBOR LIBOR is calculated by transforming historical

daily LIBOR rates from the BBA website in monthly rates (format value)

Foreign GDP growth The GDP (IFS line 99b) growth level is transformed for the G 7 countries in one single growth rate. First the data is for every country interpolated into monthly data, which is followed by growth rate calculation. The % growth is then transformed in the effect on the total amount of GDP growth for the G 7 denominated in $ by using (IFS line AE).

Appendix B

Table 1. Overview of indicators based upon the Kaminsky et al. (1998) crisis identification method and based upon a 24 month window prior to the financial crisis.

table 2. Overview of indicators based upon the Frankel & Rose (1996) crisis identification method and based upon a 24 month window prior to the financial crisis.

Table 3. Overview of indicators based upon the Kaminsky et al. (1998) crisis identification method and based upon a 18 month window prior to the financial crisis.

Table 4. Overview of indicators based upon the Frankel & Rose (1996) crisis identification method and

based upon a 18 month window prior to the financial crisis.

Appendix C

Tables are constructed in the same method as table 6 in the paper.