Department of Econometrics
Faculty of Econometrics and Operational Research University of Amsterdam
Master’s thesis
The predictability of economic crises in emerging market
economies: A panel binary fixed-effect logit model.
R.W. Sabelis
Date : 25-04-2014 Studentnr. : 0335223 Supervisor : H. van Ophem 2nd Supervisor: K. van Garderen Field : Econometrics
Table of content
1. Introduction 3
2. Economic crisis in EME during the 1990s: Common fundamentals 5 2.1 The Asian currency and banking crisis 5
2.1.1 The banking sector: financial liberalisation
and the prudential framework 5 2.1.2 The banking sector: foreign liabilities and
domestic credit 7
2.1.3 The external sector 12
2.2 The Mexican crisis 14
2.2.1 The banking sector: financial liberalisation
and the prudential framework 14 2.2.2 The banking sector: foreign liabilities and
domestic credit 14
2.2.3 The external sector 17
2.3 The real sector 19
2.4 Common fundamentals 20
2.5 Other research on leading indicators 21
3. Data 22
3.1 Sample 22
3.2 Crisis identification 22
3.3 The banking sector 24
3.4 The external sector 25
3.5 The real sector 26
4. The predictability of the eruption of a crisis 27
4.1 Panel binary logit model 27
4.2 Individual-specific effects 28
4.3 Assumptions 29
4.4 Results 30
Appendix 6: Estimation results using expected values
Appendix 7: In-sample predicted probabilities using expected values Appendix 8: Estimation results using different types of crisis
Table 1: Binary crisis index based on occurrence banking crisis Table 2: Binary crisis index based on occurrence currency crisis Appendix 9: In-sample predicted probabilities
Chart 1: Banking crisis Chart 2: Currency crisis
1. Introduction
Pervasive currency turmoil, particularly in Latin America in the late 1970s and early 1980s, gave impetus to a flourishing literature on the causes of economic crises. As stressed in this literature, especially in the seminal paper of Paul Krugman (1979), a crisis is caused by excessively expansionary fiscal and monetary policies, which resulted in a persistent loss of foreign-exchange reserves that ultimately forced the authorities to abandon the exchange rate parity. With calmer currency markets in the mid- and late 1980s, interest in this literature languished. The decade of the 1990s was marked by an unusual number of financial and economic crises, such as the attack on the European Monetary System in 1992-1993, the Mexican peso crisis in 1994-1995, the Asian crisis in 1997, the Russian default in 1998 and the Argentinian crisis in the beginning of this millennium. This rekindled interest in the causes of economic crises. Yet, the focus of this literature has shifted. The new theoretical literature stresses that banking sector problems give rise to economic crisis.
The focus of this thesis is on the crises that erupted in emerging market economies (EME) during the 1990s. The Mexican and Asian crises are extensively reviewed and compared in order to find common factors that could be seen as leading indicators signalling economic crisis in emerging market economies. For this objective especially the papers of Corsetti, Pesenti and Roubini (2001) and Sachs, Tornell and Velasco (1996) are used frequently. For comparative purposes each crisis review is divided into three parts. Each part corresponds to the developments in a particular sector, namely the banking, external and real sector, in the years preceding the onset of the crisis. It is expected that a number of macroeconomic, external, financial and/or institutional variables will be found that exhibit the same behaviour prior and during the crises under investigation.
Subsequently, a panel binary logit model is used to assess whether these common variables have a significant impact on the likelihood of a crisis and to what extent the probability of a crisis can be predicted by these variables. The dependent
coefficients and it is assessed whether the explanatory variables (i.e. the common fundamentals) have the expected sign and a significant impact on the likelihood of a crisis. A general-to-specific methodology is applied in order to get the best specification for the probability that a crisis occurs. Moreover, a Hausman test will be performed in order to investigate whether fixed effects are present. In search of the best specification this paper will also distinguish between a binary crisis index that is based the occurrence of a banking and/or a currency crisis, just a banking crisis and just a currency crisis. Secondly, this model in combination with the estimated regression coefficients and the sample data are used to estimate the in-sample probability that a crisis occurs at a particular time in a particular country. On the basis of these estimated probabilities it can be assessed to what extent the model could have predicted the occurrence of a crisis in a particular country at a particular period in time. The predictive power of the model will not only be assessed for emerging market economies but also for developing and advanced economies.
2. Economic crises in EME during the 1990s: Common fundamentals
Both East-Asia and Mexico experienced a simultaneous banking and currency crisis during the 1990s. This section reviews the developments in the banking, external and real sector in the years preceding the onset of the crisis for each crisis individually. The primary objective of these reviews is to find fundamentals that displayed a common path in run-up to each crisis, which might be an indication that these factors could signal an economic crisis in emerging market economies.
2.1 The Asian banking and currency crisis
In 1997 the Asian crisis erupted. Several countries in East-Asia experienced one of the most severe financial and currency crisis after the Great Depression. At some point foreign investors lost confidence in the Thai economy and pulled their capital out of the country, which putted large pressure on the Thai bath. Thailand was forced to let its national currency float vis-à-vis the US dollar on 2 July 1997, which eventually led to a large devaluation of the Thai national currency. These pressures soon spread over to Malaysia, South Korea, the Philippines and Indonesia, which subsequently also experienced a devaluation of their national currencies. Hong Kong, Singapore and Taiwan were hit by strong speculative pressures as well, but these countries were able to successfully avoid severe currency crises. The crisis caught everybody by surprise. The affected countries seemed to have done everything right, avoiding the mistakes made by other countries hit by a crisis in the previous years, both in Latin America and Europe. In particular, the macroeconomics appeared sound, the countries had balanced budgets, monetary policy was not expansionary and inflation was low. However, as will be described below, this seemingly good performance was based on large short-term capital inflows from abroad, which were invested long term and excessively risky. This made the Asian crisis countries very vulnerable to external shocks and capital reversals.
markets. The financial system is critical to the health of the economy, because it performs the essential function of channelling funds from savings to individuals or firms that have productive investment opportunities (Mishkin, 2000). It is widely believed that financial deregulation improves both the level of savings and the efficiency of the banking system in allocating resources, which should lead to higher economic growth. Liberalisation of the domestic financial system implies, among others, the removal of (extraordinary) high reserve ratios imposed by governments on banks. In the past banks were required to keep a large amount of funds in deposits at the central bank, which, in turn, were used to finance the large fiscal deficits. This type of financing of fiscal deficits acted as a tax on the banks’ reserve holdings at the central bank. In order for banks to make an even modest profit they had to recoup the tax from their depositors and borrowers by widening the interest rate spread. Eventually, this resulted in a suboptimal level of savings. Additionally, the amount and quality of investment decreased as only borrowers with high risk-return projects were willing to pay the high interest rates on loans, that is, the exposure of the banking system to the adverse selection problem was larger compared to the current situation in which the Asian crisis countries had deregulated their financial markets (Hallwood and MacDonald, 2000). That is, the financial deregulation will reduce the interest rate spread as a result of lower reserve requirements on banks which will induce investors with ordinary risk-return project to apply for a loan as these projects become profitable. So the probability that a bank selects investors that are most likely to default on it will be reduces by the deregulation of financial markets.
However, despite the reduced exposure to adverse selection, it is found that financial liberalisation often precedes banking crisis, especially when the financial sector lacks a satisfactory prudential framework. In the 1980s and early 1990s the Asian crisis countries had improved banking regulation and supervision. Indonesia had put in place an advanced version of the U.S.-inspired capital, asset, management, equity, and liquidity, (CAMEL) rating system. Commercial banks in Korea were required to reach a minimum of 8 percent capital adequacy ratio at the end of 1995. Thailand created the Financial Institutions Development Fund, a separate legal entity within the Bank of Thailand with a mandate to restructure, develop and provide financial support (liquidity and solvency) to financial institutions. However, despite these reforms shortcomings in the regulatory and supervisory framework remained. These shortcomings included weak prudential rules, or the enforcement of those rules,
on loan concentration, which led banks to build up excessive exposure to particular sectors such as the property market. Another significant problem was the lack of strict loan classification criteria and weak rules on provisioning (Balino et al., 1999).
Additionally, the existence of implicit government guarantees in the crisis countries weakened market discipline in the absence of an adequate prudential framework. This is a moral hazard problem, which results from asymmetric information, a situation in which one party to a financial contract has less (accurate) information than the other party. It occurs after the transaction takes place, because the lender is subjected to the hazard that the borrower has incentives to engage in activities that make it less likely that the loan will be paid back. This is a crucial impediment to the efficient functioning of the financial system, because the conflict of interest between the borrower and the lender stemming from moral hazard implies that many lenders will decide that they would rather not make loans, so that lending and investment will be at suboptimal levels. Additionally, uncertainty about the health of the banking system could result in runs on banks, and the failure of one bank can hasten the failure of others. Both the unwillingness of lenders to make loans and the reluctance of economic agents to deposit their funds could be prevented by the imposition of a government safety net. However, such government guarantees will weaken market discipline on banks, which, in turn, induces banks to take on greater risk than they otherwise would. Because depositors know that they will not suffer losses when a bank fails they will not withdraw deposits when they suspect that a bank is taking on too much risk. The incentive of banks to engage in excessively risky activities is even greater through the expectation of bailout interventions in case bank investment and lending practices fail. To limit the moral hazard that the government safety net creates, an adequate prudential framework, in which the government establishes regulations to reduce risk taking and monitors banks to see that they are complying with these regulations, is needed to foster market discipline and to ensure the safety and soundness of the financial system (Mishkin, 2000).
reserves) was as much as 2.1, 1.7, and 1.5 in Korea, Indonesia and Thailand respectively.
Short-term external debt (June 1997) (relative to foreign exchange reserves)
Liberalisation of the capital account, in combination with the deregulation of the financial system discussed above, allowed banks much greater latitude to borrow from abroad (Corsetti, Pesenti and Roubini, 1999). In addition, the exchange rates of the Asian crisis countries were mostly pegged to the U.S. dollar. The key reason for many Asian countries to pursue a policy of an effective and credible peg against the dollar was to lower the cost of borrowing by reducing the currency risk premium charged by international investors. In effect, the pegged exchange rates were seen as government guarantees of exchange value (IMF, 1998).
Apart from the enlarged ease with which the Asian crisis countries could borrow from abroad and the greater willingness of the Asian crisis countries to borrow from abroad, foreign countries were also willing to lend funds to the Asian crisis countries for a number of reasons. The perception that short-term interbank cross-border liabilities were effectively guaranteed by either the government or the IMF made international banks more willing to lend large amounts of funds to the crisis countries (Corsetti, Pesenti and Roubini, 1999). Moreover, foreign capital flowed into the Asian crisis countries as a result of the search for higher yields by
international investors at a time when investment opportunities appeared less profitable in Europe and Japan owing to sluggish economic growth and low interest rates (IMF, 1998).
As a result of this mutual willingness to borrow/lend funds, under-capitalized banks borrowed heavily from abroad at very short maturities and then, in turn, lent these borrowed funds on to domestic firms (Corsetti, Pesenti and Roubini, 1999). This rapid growth of bank lending to the private sector can be observed from the table below. According to the table financial claims on the private sector stood well above 100 percent of GDP in Korea and Thailand.
Financial claims on private sector (in percent of GDP)
Thus, the large capital inflow during the 1990s helped fuel a credit expansion in the Asian crisis countries, which might explain the remarkable macroeconomic performance in this period. When the economic growth is indeed largely explained by the large capital inflows during that time, the fact that the macroeconomics in the Asian crisis countries appeared sound might be misleading, because a sudden reversal of the capital inflows could bring these countries into serious trouble as will be explained below.
The borrowings from abroad were at short maturity, while the lending to domestic firms by domestic banks, which were largely financed by the foreign liabilities, was often at long maturity. Therefore the capital inflows carried significant maturity mismatches with them. In normal times this may not cause concern, as short-term foreign liabilities are easily rolled over. However, these maturity mismatches
Many loans made by banks were also excessively risky due to the absence of an adequate prudential framework, as described above. The lending and investment boom was primarily concentrated in the real estate sector (Corsetti, Presenti and Roubini, 1999). The growing dependence of the banking sector on property lending can be observed for Thailand from the figure below. The percentage share of property loans increased from 15 percent to about 24 percent while manufacturing loans steadily declined from 22 percent to 15 percent during 1988-1996. So there was a marked shift in the composition of Thai finance companies’ loan portfolio. In Thailand the growth of lending by finance companies to the property sector averaged 41 percent per annum compared with total lending growth of 33 percent per annum during 1990-1995. In Indonesia loans to the real estate sector grew at an annual rate of 37 percent during 1992-1995, compared with 22 percent for total bank credit (Miller and Luangaram, 1998).
Loans to real estate and manufacturing sector by Thai finance companies (share of total outstanding loans)
This resulted in inflated asset prices. In fact, the Asian economies were in a self-reinforcing cycle. The asset price inflation that resulted from the credit expansion encouraged even more lending to the real estate sector, which further inflated collateral values until an asset price bubble was created (Balino et al., 1999).
Under a fixed exchange rate regime and capital mobility the Asian authorities had lost monetary autonomy due to the incompatible trinity. In other words, the prolonged maintenance of the pegged exchange rates as well as the openness of the Asian crisis countries to international capital flows complicated the monetary policy response of the Asian authorities to the asset price bubble (Corsetti, Presenti and Roubini, 1999). Ideally, the Asian authorities wanted to bring inflation down by raising the interest rate. However, this higher yield would have attracted international investors. Under capital account openness this would have led to huge capital inflows, which would result in an upward pressure on the nominal exchange rate. In order to defend the exchange rate peg the Asian authorities would have to intervene in the foreign exchange market by supplying its national currency in exchange for the foreign currencies. This would have increased the domestic money supply, or equivalently reduced the interest rate, which would results in the resurgence of inflationary pressures. This inability to simultaneously target the exchange rate, allow full capital mobility and run an independent monetary policy increased the vulnerability of the financial system to a sudden reversal of capital inflows. When capital starts flowing out of a country the authorities have to intervene in the foreign exchange market in order to defend the exchange rate peg, which will reduce the supply of money, or equivalently raise the interest rate. With respect to the real estate sector, for example, this higher interest rate will gradually result in a reduction in the demand for housing, which will eventually lead to a fall in house prices. In turn, these lower house prices reduce the collateral of homeowners which deteriorates both the amount and terms of credit available to households, that is households face tighter credit constraints. Thus, the sudden reversal of capital inflows, which will reduce the amount of funds available for lending, and the downturn in house prices might lead to a credit crunch. Additionally, among other factors as increasing unemployment, the fall in house prices might force more and more households to default on their mortgages, which will deteriorate the balance sheet of financial institutions. That is, the sharp drop in the value of real estate left the banking sector in the crisis countries
table below for Indonesia and Thailand. As can be seen from this table the share price of property companies in Thailand fell significantly by about 68 percent during the four quarters before the abandonment of the Thai baht on 2 July and by the year property companies in Thailand were worth only 10 percent of their value 24 month before. Although, the property share index in Indonesia showed an upward trend from September 1996 until June 1997, it sharply declined by 58 percent during the second half of the year (Miller and Luangaram, 1998).
Share value of property companies in stock markets
2.1.3 The external sector
In the above the effect of capital account liberalisation on the foreign liabilities was already described in detail. Reference was also made to maturity mismatches and the importance of foreign exchange reserves as a device to defend the exchange rate peg. Referring back to the figure on short-term external debt on page 8, during most of the 1990s the growth of the short-term foreign-currency debt outpaced the growth in foreign exchange reserves, which increased the vulnerability of the financial system to capital flow reversals. When external creditors become unwilling to roll-over existing short-term loans and foreign-exchange reserves are insufficient to cover short-term liabilities, foreign investors will lose confidence in the credibility of the fixed exchange rate regime. A currency devaluation will raise the value of external debt, so when investors lose confidence in the credibility of the fixed exchange rate regime they might fear that the Asian crisis countries can not repay their debt anymore. Investors might realize that, if other investors stop lending to the Asian crisis countries and were unwilling to roll-over existing short-term loans, the government
would be unable to repay its debts as they fall due. Therefore, each individual investor can do no better than to withdraw his funds when other investors start to withdraw their funds. This self-fulfilling panic will put pressure on the currency and eventually, when foreign-exchange reserves are depleted, could result in large devaluations.
Besides large short-term external debt, which outpaced the growth in foreign exchange reserves, the Asian crisis countries also had large current account deficits prior to the crisis. These deficits, or equivalently these higher levels of import than export, resulted in a net demand for foreign currencies, which represented an additional burden on the foreign exchange reserves. The large current account deficits in the Asian crisis countries were the result of a number of factors. Wide swings in the dollar/yen exchange rate caused large shifts in the international competitiveness of the Asian crisis countries. In particular, the appreciation of the U.S. dollar from mid-1995 against the yen and the associated losses of competitiveness in the countries with dollar-pegged currencies, contributed to their export slowdown in 1996-1997 and their wider external imbalances (IMF, 1998). Furthermore, the stagnation of the Japanese economy in the 1990s, which was the most important trading partner of the Asian crisis countries, and the increasing weight of China in total exports from the region also contributed to a worsening of trade balances in the Asian crisis countries between 1996 and 1997. Finally, prices of important export goods, such as semi-conductors and other manufactured goods, fell sharply in 1996. As a result several East Asian countries had experienced significant negative terms of trade shocks (Corsetti, Pesenti and Roubini, 1999). These large trade imbalances until 1996/1997 is shown for Thailand by the current account deficit in the figure below.
5000 10000 15000 20000 25000
Current Account Balance
2.2 The Mexican crisis
Mexico also experienced a currency and banking crisis in the 1990s. Similar to the Asian crisis countries the economic environment of Mexico seemed sound. Despite this seemingly good economic performance foreign investors fled the country and Mexico was exposed to a self-fulfilling panic in December 1994. As will be described below the causes and consequences of this panic are more or less the same as for the Asian crisis countries.
2.2.1 The banking sector: financial liberalisation and the prudential framework
By the end of 1994, Mexico was an economy where financial deepening had vastly enlarged the stock of liquid assets, which could be turned into dollars in short notice. The ratio of M2 to GDP had increased from 25 percent in 1989 to over 33 percent by the end of 1993. The Mexican financial sector underwent a substantial liberalisation implemented by the central bank, which instituted a zero legal reserve requirement for the banks (Sachs, Tornell and Velasco, 1996). Furthermore, bank deposits were implicitly or explicitly insured by the government. As explained above, such government guarantees will weaken market discipline on banks, because depositors know that they will not suffer losses when a bank fails and as a result will not withdraw deposits when they suspect that a bank is taking on too much risk. In turn, this lack of market discipline induces banks to take on greater risk than they otherwise would (Sachs, Tornell and Velasco, 1996).
2.2.2 The banking sector: foreign liabilities and domestic credit
As can be observed from the figure below foreign investment in the Mexican economy increased significantly prior to the crisis in 1994. A number of factors contributed to the resurgence of capital inflows into Mexico. Firstly, Mexican law was changed in 1990 to allow foreigners to hold government bonds and to buy (non-voting) shares in almost all sectors of the economy. These economic reforms led to the resumption of economic growth, which averaged 3,1 percent per year between 1989 and 1994, and brought inflation down to single-digit levels (Gil-Diaz, 1998).
Secondly, an important part of this phenomenon is explained by the fundamental economic and political reforms that had taken place, including the restructuring of external debts held by Mexico (Calvo, Leiderman and Reinhart, 1993). As its reforms advanced, Mexico began to attract more foreign capital, a significant part of which was short-term (Gil-Diaz, 1998). However, domestic reforms alone cannot explain the resurgence of capital flows into Mexico. Finally, the renewal of capital flows to Mexico could be due to external factors, such as falling interest rates and a continuing recession in the United States.
The massive capital inflows that took place prior to the crisis, which are shown by the upward trend in the capital account on page 18, were used to finance the increase in private consumption. As can be observed from the figure below total loans as a percentage of GDP increased from 24 percent to 38 percent between 1991-1994.
The banking system intermediated the capital flows of resources and converted them into real estate and consumption loans (Sachs, Tornell and Velasco, 1996). Mortgage loans, for example, rose at an annual rate of 47 percent, all in real terms (Gil-Diaz, 1998). This transformation of short-term capital inflows into peso loans and the concentration of these loans in the real estate sector were responsible in part for the fragility of the banking system (Sachs, Tornell and Velasco, 1996). The real estate loans resulted in inflated asset prices, which encouraged even more lending to the real estate sector, which further inflated collateral values until an asset price bubble was created. Thus, as concluded by Lindgren et al. (1996), financial liberalization and other economic reforms can result in over-indebtness, rapid growth of bank credit and asset price bubbles (Gil-Diaz, 1998). As explained for the Asian crisis, the monetary policy response of the Mexican authorities to the asset price bubble was complicated due to the incompatible trinity. As was also explained before, the inability to simultaneously target the exchange rate, allow full capital mobility and run an independent monetary policy increased the vulnerability of the financial system to a sudden reversal of capital outflows, because this would result in a credit crunch and a downturn in asset prices. The collapse of the national stock market and the sharp drop in the value of real estate left the Mexican banking sector with a huge amount of non-performing loans as can be seen from the figure below which shows a sudden peak in 1995-1996.
2.2.3 The external sector
Firstly it should be noted that the bulk of the external borrowing in the years preceding the crisis was done by the private sector. The central bank sterilized the capital inflows by issuing short-term peso-denominated debt, which was called Cetes. These gross liabilities issued to sterilize the monetary effects of the capital inflow were matched until late 1993 by accumulated foreign exchange reserves (Sachs, Tornell and Velasco, 1996).
Unexpected shocks and an inadequate response to those shocks made Mexico financially vulnerable. In the aftermath of the assassination of the presidential candidate Luis Donaldo Colosio in March 1994, net capital inflows ceased for much of the remainder of the year. As a result, the exchange rate experienced a nominal depreciation of around 10 percent, reaching the edge of the band (i.e. the range in which the nominal exchange rate of Mexico was allowed to fluctuate by the monetary authorities) Mexico had long been operating, and interest rates increased by around 7 percentage points. Despite the increase in the interest rates international investors did not restrain from fleeing the country, because they probably recognized the weak economic situation in Mexico, such as the creation of an asset price bubble and the low level of foreign exchange reserves w.r.t. short-term foreign liabilities, and therefore might have lost confidence in this country. The capital outflows resulted in a fall in foreign exchange reserves in order to keep the exchange rate within the band. Because of the perceived fragility of the banking system the authorities tried to prevent further interest rate increases through an expansion of domestic credit to the private sector and to the government and by converting the short-term peso-denominated government liabilities (Cetes) falling due into dollar-peso-denominated bonds (Tesobonos). The Mexican private sector used the pesos generated by the credit expansion to fund the current account deficit, which was caused by the nominal depreciation of the exchange rate and is shown in the figure below.
Since the exchange rate was pegged against the dollar and the current account deficit could not be funded by capital inflows anymore, because international investors were fleeing the country, the central bank sold the Mexican private sector the dollars needed to cover their excess of imports over exports. Foreign exchange reserves thereby fell even further in the amount of the current account deficit. Thus, the fall in foreign exchange reserves was caused by a combination of the reduction of capital inflows and the monetary policy response, which was designed to limit further increases in domestic interest rates that were considered dangerous given the vulnerable situation of the banking system. This fall in the foreign exchange reserves can be observed from the figure below, which shows a sharp decline from February 1994 onwards. As a result of these policy responses the authorities ended up illiquid, that is, foreign-exchange reserves were insufficient to cover short-term dollar-denominated liabilities (Tesobonos), which increased the financial vulnerability of Mexico (Sachs, Tornell and Velasco, 1996).
Although the Mexican authorities could still have borrowed, albeit at high cost, they chose to devalue the nominal exchange rate in November and early December 1994, because the authorities perceived the high cost of defending the exchange rate peg in terms of high interest rates and domestic recession. The devaluation contributed to a shift in expectations. Because the devaluation raised the value of the short-term dollar-denominated debt, investors feared that Mexico would default on its debt. These adverse expectations about Mexico’s financial conditions generated a self-fulfilling panic in the market for government securities. Investors realized that, if other investors stopped lending to the Mexican government and were unwilling to roll-over existing short-term loans, the government would be unable to repay its debts, particularly the dollar-denominated Tesobonos, as they fell due. Therefore, each individual investor could do no better than to withdraw his funds when other investors started to withdraw their funds. This exposed Mexico to a self-fulfilling panic, which resulted in a currency crisis at the end of 1994.
2.3 The real sector
For the Mexican case as well as the Asian case a growth in money supply and domestic credit was observed to precede both crisis. Firstly consider the growth in money supply. This growth would, according to economic theory, lead to larger demand by consumers for goods and services which raises the price level. A higher price level will cause the real exchange rate to depreciate which will result in a widening of the current account imbalance. Higher imports than exports mean excess demand for foreign currency, which must be supplied with the foreign exchange reserves of the central bank. This additional burden on the foreign exchange reserves increases the vulnerability of a country to capital outflows and to a crisis.
Secondly, consider the increase in domestic credit. This stimulates investments according to economic theory. The increased level of consumption as well as the increased level of investment raise the level of GDP, which increases the
appropriate explanatory variable. Therefore, this causal effect will not be considered anymore in the remainder of this paper.
2.4 Common fundamentals
Besides the growth in GDP and the increased inflation rate found in the previous section, the reviews of the Asian and Mexican crisis suggest a number of fundamentals that could be seen as a signal of an unfolding crisis. Firstly, deregulation of financial markets, which implies, among others, the removal of (extraordinary) high reserve ratios imposed by governments on banks, encourages excessive risk taking when the financial sector lacks a satisfactory prudential framework. In both crises excessive risk taking took the form of large exposures to particular sectors such as the property market. In the Asian crisis countries this excessive exposure was the result of weak prudential rules, and the enforcement of those rules, on loan concentration. Additionally, the existence of implicit government guarantees in Mexico as well as in the Asian crisis countries led to weakened market discipline, which, in turn, resulted in excessive risk taking (moral hazard problem).
Secondly, the liberalisation of the capital account, among other factors such as slow growth and low interest rates in the US, Europe and Japan, led to an increase in short-term foreign liabilities. These capital inflows fuelled a credit expansion. The lending to domestic firms was often at long maturity and therefore the capital flows carried significant maturity mismatches with them. Moreover, as already mentioned, the lending and investment boom was primarily concentrated in the real estate sector, which created an asset price bubble. An asset price bubble could lead to serious financial problems, i.e. an increase in non-performing loans, when the capital inflows reverse, especially in countries that lost their monetary autonomy as a result of having a fixed exchange rate regime and capital mobility (incompatible trinity).
Thirdly, the reversal of capital inflows could be even more harming when the short-term foreign-currency debt outpaced the growth in foreign exchange reserves, because confidence in the credibility of the fixed exchange rate regime will be lost. As a result investors are unwilling to roll-over the short-term foreign currency denominated debt and they will pull their capital out of the country. Eventually this will lead to large devaluations when foreign-exchange reserves are depleted, which raises the value of foreign debt and makes the government unable to repay its debt.
In the case of Mexico it was indeed found that the fall in foreign exchange reserves was caused by the reversal of capital inflows, but also by the monetary policy response in the form of a domestic credit expansion which was designated to limit further increases in domestic interest rates that were considered dangerous given the vulnerable situation of the banking system. The Mexican private sector used the pesos generated by the credit expansion to fund the current account deficit, which was caused by the nominal depreciation of the exchange rate. Since the exchange rate was pegged against the dollar and the current account deficit could not be funded by capital inflows anymore, because international investors were fleeing the country, the central bank sold the Mexican private sector the dollars needed to cover their excess of imports over exports. Foreign exchange reserves thereby fell even further in the amount of the current account deficit. Therefore, a large current account deficit could be seen as signalling the eruption of a crisis. For the Asian case, current account imbalances, due to a deterioration in international competitiveness, were also found to preceed the eruption of the crisis.
2.5 Other research on leading indicators
The leading indicators of a crisis in EME found in this paper by comparing the Asian crisis and the Mexican crisis are largely supported by other research conducted in this field. Calvo, Izquierdo and Talvi (2003) illustrate that crises are preceded by a sudden stop or reversal of capital inflows, draining a country’s foreign currency reserves. Kaminsky and Reinhart (1999) argue that financial liberalisation helps to predict the occurrence of banking crisis. They stress the importance of proper sequencing and management of financial liberalisation in their paper. With respect to the financial sector, Ottens, Lambregts and Poelhekke (2005), IMF (2004), Borio and Lowe (2002) and Eichengreen and Arteta (2000) show that rapid lending growth is an important leading indicator of banking sector problems. In the context of institutional factors, Demirguc-Kunt and Detragaiche (1998) find a significant positive relationship
3. Data
In the previous section a number of factors were found in the banking, the external and the real sector that behaved similar in the run-up to the Mexican and the Asian crisis and, therefore, could be seen as leading indicators signalling economic crisis in emerging market economies. In this section the data is briefly described. This description primarily focuses on the period that will be covered, the countries that are included, the ways that a crisis can be identified, and the way the leading indicators are measured. Reference will also be made to the expected effect of these leading indicators on the probability of a crisis.
3.1 Sample
The sample that is used in the research presented in this paper includes 113 countries (developing, emerging and developed) and covers the period 1988-2012. That is, for each country data is collected on the dependent and the explanatory variables for each year in the period 1988-2008. So this paper uses panel data, which is defined as repeated observations on the same cross section (a country in this case) observed for several time periods (years in this case). A major advantage of panel data is increased precision in estimation. This is the result of an increase in the number of observations owing to pooling several time periods of data for each country.
3.2 Crisis identification
The literature on economic crisis defines and identifies crisis in many different ways. Most of these definitions have several shortcomings. The events method of the IMF, for example, relies to heavily on the observation of government intervention. The cost of a bailout is available only after a crisis and with a time lag, so it tends to identify a crisis too late. Moreover, there are few objective standards for deciding whether a given policy intervention is “large” (Hagen and Ho, 2007).
The problem of crisis identification is best illustrated by comparing the crises identified in different studies. Hagen and Ho (2007) show that there are large differences in the timing of crisis between different studies. Different studies sometimes identify the onset of the same crisis with a difference of more than two
years. This poses obvious problems for empirical research into the determinants of crises (Hagen and Ho, 2007).
In view of these problems, the basis for crisis identification in this thesis is the paper written by Reinhart and Rogoff (2009), most notably because of their extensive dataset which even makes the distinction between currency crisis, banking crisis and other types of crisis. A currency crisis is defined as a period in which an annual depreciation versus the relevant anchor currency of 15 percent or more took place. A banking crisis is marked by two events: (1) bank runs that lead to the closure, merging, or takeover by the public sector of one or more institutions; and (2) closure, merging, takeover, or large-scale government assistance of an important financial institution (or group of institutions), if there were no bank runs, that marks the start of a string of similar outcomes for other financial institutions.
Of course this approach to identifying the beginning of a banking crisis is not without drawbacks. It could date a crisis too late, because the financial problems usually begin well before a bank is finally closed or merged. In order to verify the years in which Reinhart and Rogoff identify a currency and/or banking crisis four additional papers are used. Besides verification it should be noted that these papers are also used to extract information on crisis periods in countries that were not included in Reinhart and Rogoff in order to extend the sample.
In order to verify the years in which a country was hit by a currency crisis according to Reinhart and Rogoff, the paper of Glick and Hutchison (2005) is used. In this paper a currency crises is identified when “ large” changes are observed in an index of currency pressure, defined as a weighted average of monthly real exchange rate changes and monthly (percent) reserve losses. The other three papers are used to verify the years in which a country was hit by a banking crisis according to Reinhart and Rogoff. The paper of Laeven and Valencia (2012) lists banking crisis episodes for a number of countries based on a large number of criteria ranging from bank runs to significant interventions in the form of gross bank restructuring costs of at lest 3 percent of GDP. The IMF paper of Demirguc-Kunt and Detragiache (1998), which
The reasoning behind this indicator starts from the conventional assumption that the banking sector’s aggregate demand for central bank reserves depends negatively on the short-term interest rate, the immediate opportunity cost of holding reserves. The main conjecture is that a crisis is characterised by a sharp increase in the banking sector’s aggregate demand for central bank reserves. This may be due to a sharp decline in the quality of bank loans or an increase in non-performing loans, causing a loss of liquidity in the banking sector. Another reason for the increased demand for central bank reserves by the banking sector may be a sudden withdrawal of deposits by the non-bank public, forcing banks to turn to the inter-bank market and the central bank to refinance themselves. The central bank, as a monopolistic supplier of bank reserves, can react to this increase in the demand for reserves in two basic ways. If bank reserves are the operating target, the total supply of bank reserves is kept constant and the short-term interest rate will rise. If, instead, the central bank targets the short-term interest rate, it must inject additional reserves into the banking system through open market operations. In this case the central bank buys government bonds held by the banking sector in exchange for reserves. Thus, a banking crisis is characterized by a sharp increase in the short-term interest rate, a large reduction in the volume of central bank reserves, or a combination of both, indicating a high degree of tension in the money market (Hagen and Ho, 2007).
3.3 The banking sector
Banking sector weaknesses can be explained by institutional factors. Banks have obviously more scope to indulge in excessive risk behaviour in case of weak supervision and regulations. The security of property rights could potentially be used in trying to capture the strength of a country’s prudential framework. As explained in the previous section, deposit insurance may enhance financial stability by lessening the risk of bank runs, but it may also increase the likelihood of financial problems by weakening market discipline and encouraging excessive risk taking. However, because of a lack of data it is impossible to evaluate the effect of institutional factors on the probability that a crisis occurs.
Excessive risk taking took the form of large growth in credit, which was primarily concentrated in the property sector and eventually led to an asset price bubble. The financial indicator “Domestic credit provided by the banking sector (% of
GDP)” from the World Bank is used to measure the provision of credit in a country over the period 1988-2012. This definition includes all credit from monetary authorities, deposit money banks and other banking institutions (such as saving and mortgage loan institutions) to various sectors except the central government. Unfortunately, data on asset prices for most countries included in the sample was not available for the period 1988-2012. But, assuming that a credit expansion was primarily concentrated in the property sector (which was shown to be the case in the Asian crisis countries by the figure on page 10), an excessively high level of credit provisioning makes the economy more vulnerable to the eruption of a crisis, so the expected sign of domestic credit is positive.
3.4 The external sector
These credit expansions were fuelled by large capital inflows, which, in turn, were the result of low interest rates and slow growth in the rest of the world and of capital account liberalisations. Short-term foreign liabilities are extracted from the Oxford Economics database, which contains short-term outstanding debt measured in current US dollars. It is expected that a large increase in short-term foreign liabilities make a country more vulnerable to a crisis, so the expected sign of short-term capital inflows is positive.
The monetary authorities were obliged to buy the inflow of foreign capital in order to defend the exchange rate peg, which raised the money supply. This monetary expansion is measured by the financial indicator “Money and quasi money growth (annual %)” from the World Bank database. Money and quasi money comprise the sum of currency outside banks, demand deposits other than those of the central government, and the time, savings and foreign currency deposits of resident sectors other than the central government. This definition is frequently called M2. It is expected that a large increase in the money supply raise the probability of a crisis, so the expected sign is positive.
loans. Unfortunately, data on the level of non-performing loans is not available for most countries in the period 1988-2012. Secondly, it will lead to large devaluations when foreign exchange reserves are depleted, which raises the value of foreign debt and makes banks and the government unable to repay its debt. Data on foreign exchange reserves are obtained from the Oxford Economics database. A declining level of foreign exchange reserves will increase the probability that the reserves will be depleted in the event of a capital outflow and therefore will increase the probability of a crisis. So it is expected that the sign of foreign exchange reserves is negative.
For the Mexican case it was found in the previous section that the depletion of foreign exchange reserves was also the result of a monetary policy response in the form of an expansion of domestic credit, which was used by the private sector to fund a current account deficit. Since the exchange rate was pegged against the dollar and the current account deficit could not be funded by capital inflows anymore, because international investors were fleeing the country, the central bank sold the Mexican private sector the dollars needed to cover their excess of imports over exports. Foreign exchange reserves thereby fell even further in the amount of the current account deficit. For the Asian case, current account imbalances, due to a deterioration in international competitiveness, were also found to preceed the eruption of the crisis. Therefore, a large current account deficit is expected to raise the probability that a country will experience a crisis. Data on the current account balance are extracted from the Oxford database and are measured in US dollars.
3.5 The real sector
In both Mexico and the Asian crisis countries higher inflation rates, which according to economic theory were the result of a growth in money supply, were found to signal an upcoming crisis. A higher price level will cause the real exchange rate to depreciate which will result in a widening of the current account imbalance. This excess of imports over exports leads to an excess demand for foreign cuurency, which must be supplied with the foreign exchange reserves of the central bank. This lowers the level of foreign exchange reserves, which increases the vulnerability of a country to a crisis. Thus, a higher inflation rate is expected to increase the probability that a crisis occurs. The inflation rates for each country in the sample in the period 1988-2012 are obtained from the World Bank database.
4. The predictability of the eruption of a crisis
Now that it is clear how the indicators, that were found to signal the eruption of a crisis, are measured the model can be constructed. In order to accomplish the objective of predicting the eruption of a crisis in EME a panel binary logit model is used. This section describes the model. In this description reference will also be made to the possible presence of fixed effects in this model and how to proceed in order to get consistent results in this case. Subsequently, the model will be estimated and the results will be discussed.
4.1 Panel binary logit model
The crisis index yi,t for a particular country in a specific year is a binary outcome, because it takes one of two values
yi,t = 1 if a crisis occurs in country i at year t and
yi,t = 0 if no crisis occurs in country i at year t
The probability that the crisis index equals one is pi,t and the probability that the crisis index equals zero is (1- pi,t), so the crisis index yi,t is necessarily Bernoulli distributed if the sample (yi,t , xi,t) is assumed to be independent over i and t.
f(yi,t | xi,t , β , αi) = (pi,t)y (1- pi,t)1-y
For regression applications the probability pi,t will vary across individuals and time as a function of the regressors. A regression model is formed by parameterizing the probability pi,t to depend on a regressor vector xi,t and a Kx1 parameter vector β. The functional form that is used in this paper to model the probability pi,t is the logit
Combining the above and again assuming that the sample (yi,t , xi,t) is independent over i and t., the joint density for the ith observation yi = (yi,1,...,yi,T) is
f(yi I Xi , β , αi) = П f(yi,t I xi,t , β , αi) = П (pi,t)y (1- pi,t)1-y (1) = П [F(αi + xi,t’ β)]y [1- F(αi + xi,t’ β)]1-y
where the product is over t=1 until t=T and Xi is a KxT matrix
4.2 Individual-specific effects
The term αi in the formulas above is a time-invariant random variable that captures unobserved heterogeneity. The unobserved individual heterogeneity (country-specific effects) may be random or fixed. One variant treats αi as an unobserved random variable that is potentially correlated with the observed regressors xi,t.This variant is called the fixed effect (FE) model. In case of panel data and a non-linear model with fixed effects joint estimation of the fixed effects α1,….., αN and the other parameters β generally leads to inconsistent estimation of all parameters. Instead, Chamberlain (1980) proposed a method, which permits consistent estimation in the presence of fixed effects for the panel logit model. It eliminates the fixed effects and in this way permits consistent estimation of the other model parameters. This method eliminates αi by conditioning the joint density of yi given in (1) on the sufficient statistic Ʃt yi,t = c, in addition to the usual conditioning on regressors. It can be shown that the joint density of yi can then be expressed as
f(yi I Ʃt yi,t = c, Xi , β) = exp((Ʃt yi,t xi,t’) β) / ƩdB exp((Ʃt di,t xi,t’) β)’ (2)
where the set B = { di | Ʃt di,t = Ʃt yi,t = c } is defined to be the set of all possible sequences of 0s and 1s for which the sum of T binary outcomes Ʃt yi,t = c.
The alternative specification, known as the random effects (RE) model, assumes that the unobserved individual effects αi are random variables that are distributed independently of the regressors. Usually it is assumed that the individual effects are normally distributed with αi~N{0, Var(αi)]. Even if the unobserved
individual effects are uncorrelated with the regressors, it is not guaranteed that the estimates of the estimated parameters β are consistent because this paper uses a non-linear model. Therefore, the term αi should be eliminated in this case as well. This can be done by conditioning the joint density of yi given in (1) on the variance of αi. It can be shown that the the joint density of yi can then be expressed as
f(yi I Xi , β, Var(αi)) =
f(yi I Xi , β , αi) N(0, Var(αi)) d αi (3) where N(0, Var(αi)) represents the normal probability density function of αi. Eventually, model (2) and (3) can be used to estimate the regression coefficients by applying the maximum likelihood method.4.3 Assumptions
The assumption of independence of the sample (yi,t , xi,t) over time and countries is a very strong one in this economic context. Contagion effects, the phenomenon that a shock in a particular economy spreads over to other economies thorough some transmission channel due to interdependencies between the economies, is an example of dependence of economic crisis over countries. The collapse of the Thai baht spread over to the other East Asian crisis countries and triggered a crisis in most of these countries. Therefore the question arises whether it is possible to relax the assumptions of country- and time-independency.
In the linear case time-dependency and country-dependency of fixed effects can be incorporated into the model by introducing country dummies dj,it, which equals one if j=i and zero otherwise, and time dummies ds,it, which equals one if s=t and zero otherwise. So by permitting the intercept to vary across countries and time N country dummies and (T-1) time dummies are introduced. The Kx1 vector β is estimated by partitioned least-squares estimation and the resulting estimators can be used to obtain
paper does not introduce country- and time-dependent fixed effects in the non-linear model used in this paper.
4.4 Results
The main results of applying the maximum likelihood method to the panel binomial fixed effects logit model are presented in Appendix 1. Appendix 1 reports the results of 4 model specifications when the full sample of 113 countries is used. The quality of the specifications is assessed by the following 3 criteria: (i) model chi-squared, (ii) the sign and the significance of the coefficients, and (iii) accuracy of the in-sample predictions.
The chi-squared statistic is used to test the joint significance of all regressors. Appendix 1 shows that for all specifications the null hypothesis of all zero coefficients is rejected at the 1 percent level. However, based on the sign and significance of the coefficients only model 3 and model 4 are worth considering. All coefficients of model 4 are significant at the 1 percent level and their signs are correct as well, which can be seen by comparing them with the expected signs outlined in paragraph 3. All coefficients of model 3 are significant at the 1 percent level as well, except for the coefficient of the money growth rate which is significant at the 5 percent level, and the signs are also as expected. As these models are more or less equally suitable with respect to the signs and significance of the coefficients, model 3 that will be chosen as the most preferred model as it includes more valuable information. The graphs with predicted probabilities produced by this model will be used to assess the predictive power.
Appendix 1: Estimation results
Dependent variable: Binary crisis index based on occurrence banking and/or
currency crisis (1) (2) (3) (4) Observations 904 993 1476 1581 Domestic Credit .0250668 .0172054 .0225419 .0227662 (0.000)* (0.000)* (0.000)* (0.000)* Inflation -.0008371 (0.423) Growth rate money supply .0038885 .0029727 .0031661 (0.124) (0.036)* (0.025)* Short-term
external debt 6.57e-12 1.12e-11 (0.699) (0.482)
Current account
balance .000014 8.11e-06 .0000116 9.59e-06 (0.252) (0.482) (0.009)* (0.006)* Foreign exchange -.0000172 -.0000175 -.0000142 -.0000156 reserves (0.000)* (0.000)* (0.000)* (0.000)* Summary statistics AIC 823.9804 904.653 1323.294 1411.08 LR chi2(K-1) 82.36 86.68 157.56 154.56
Before assessing the predictive power of the model, a number of things should be noted about the models presented in Appendix 1. Firstly, it is suspected that the models presented in Appendix 1 suffer from the multicollinearity problem, which inflates the standard errors of the estimated coefficients upwards and therefore lowers the Student t test statistics. This creates problematic situations in terms of evaluating the regression model. Appendix 2 displays the correlations between the independent variables. As can be observed from this table a number of variables are highly correlated. For example, inflation and money growth are highly positively correlated, which explains the sudden significance of money growth at the 5 percent level when inflation is dropped from model 1. This positive relation is as expected, because more money circulating in an economy means higher demand for goods which raises the prices of those goods. Another example is the high positive correlation between short-term external debt, the current account balance and foreign exchange reserves, which explains the sudden significance of the current account balance when short-term external debt is removed from model 2. This positive correlation is also as expected, because these variables are connected with each other via the balance of payment which consists of the current account, the capital account and foreign exchange reserves. Short-term capital inflows, for example, need to be bought by the central bank in order to defend the exchange rate peg, which raises the level of foreign exchange reserves.
Secondly, the estimated models presented in Appendix 1 assumed fixed effects, because it is reasonable to suspect that these effects are present in this macroeconomic model. For example, the unobserved expectations of economic agents about the inflation rate are correlated with the growth rate of GDP according to the Phillips curve theory, which in turn could influence the lending and borrowing decisions of economic agents. However, in order to be confident that treating αi as a fixed effect is indeed the correct assumption a Hausman test is performed. If there are fixed effects then the random effect estimators will be inconsistent, so the probability limit of the fixed effect estimators and random effects estimators will be different. If there are no fixed effects both estimators are consistent, so the two estimators have the same probability limit. This suggests that testing for the presence of fixed effects can be done by testing for the difference between the random effect estimator and fixed effect estimator. More generally, consider the vector of random effect estimators
B and the vector of fixed effect estimators b, then the following hypothesis must be tested
H0 : plim(b-B) = 0, no fixed effects H1 : plim(b-B)≠0, fixed effects
B is the efficient estimator and under H0 √N(b-B) N[0, Vh], so the Hausman test statistic can be expressed as
H = (b-B)’ (V[b] – V[B])-1 (b-B)
which is asymptotically chi2(q) distributed under H0, where q is the number of regressors in the model. The results of this test are shown in Appendix 3, where H0 is rejected if H > chi2(q), which is obviously the case for model 3 of Appendix 1.
Appendix 2: Cross-correlations Dom. Credit Inflation Growth rate money supply ST external debt CA balance FER Dom. Credit 1 Inflation 0.0074 1 Growth rate money supply 0.0587 0.8926 1 ST external debt 0.3076 ‐0.0023 0.0082 1 CA balance 0.1588 ‐0.0111 ‐0.0049 0.6282 1
Appendix 3: Hausman test
Coefficients Difference
Fixed b Random B b-B S.E. Domestic Credit .0225419 .0153317 .0072102 .0018098
Growth rate Money
Supply .0031661 .0042305 -.0010644 .
Current account
balance .0000116 8.03e-06 3.56e-06 2.55e-06
Foreign exchange
reserves -.0000142 -.0000122 -2.01e-06 1.10e-06
chi2(4) 17.87 (0.0013)
Having made the above notifications the predictive power of model 3 can now be assessed by constructing a graph for each country that presents the in-sample predicted probabilities that a currency and/or banking crisis occurs in the period 1988-2012. Appendix 4 presents a table for a number of countries that contains the years in which a crisis actually erupted in these countries. A year in which a country experienced a currency crisis (CC) or a banking crisis (BC) is indicated with a one in the tables. Appendix 5 shows the graphs with predicted probabilities generated by model 3 (chart 1) for some South-American crisis countries, Mexico and the Asian crisis countries. Comparing the movement of the graphs with the actual crisis years for each country could gives an impression of the accurateness of the models in predicting crisis episodes. In this comparison peaks in the graphs are taken to be a sign that a crisis must have occurred.
For Argentina the model predicts that a banking and/or currency crisis must have occurred in 1989, 1995 and 2002. Considering the years in which a banking crisis occurred in Argentina and neglecting the fact that the latter crisis actually
started in 2001, these forecasts are fairly good. The only shortcoming of the model for Argentina, besides the fact that it predicts the onset of the 2001 banking crisis with a one-year lack, is its inability to predict the duration of each crisis. This is even more the case when considering the years in which Argentina actually experienced a currency crisis as the table in Appendix 4 shows that the currency crisis that erupted in 1988 lasted until 1992. However, when considering currency crisis episodes the model would have predicted the onset and the duration of the currency crisis in 2002 perfectly.
For Brazil the model does a fairly good job when considering the years in which it actually experienced a currency crisis. Neglecting the downward peak between 1990-1991 and the fact that the currency crisis actually lasted until 1994, the model is fairly accurate as it generates high probabilities in 1988-1989 and 1992-1993. The model does also foresee the currency crisis that erupted in 1999 and 2001-2002, but not very convincingly as the peaks are rather small. However, the model also has a small peak in 2005 while no currency crisis actually took place in this year and the model also misses out on the currency crisis Brazil experienced in 2008. The model is very inaccurate when the attempt is to predict episodes of banking crisis that actually occurred in Brazil.
The probabilities predicted by the model for Mexico are much less convincing than those for the South-American crisis countries. The model completely misses out on the currency crisis that erupted in 1989. Furthermore, the model foresees a crisis in 1995, while in fact a banking and currency crisis actually erupted in 1994. In addition, the duration of the banking crisis that started in 1994 was much longer than predicted by the model. Finally, the currency crisis that occurred in 2008 is predicted by the model with a one-year lack and the currency crisis that erupted in 1998 is not foreseen at all by the model.
For the Asian crisis countries the model predicts the start of a crisis in 1996/1997 fairly well for all countries. Roughly for each country the graphs begin to rise gradually in the period 1994-1996 and eventually show a very abrupt and
all Asian crisis countries, except for Indonesia, a crisis is predicted in 2009, while in reality a crisis did not take place in this year.
In short, based on the above description of the predictive power of the model for each country it can be concluded that the model does a reasonable job in predicting the onset of each crisis, although the onset is sometimes predicted with a one-year lack, and that it is completely unable to foresee the duration of each crisis. Adapting monetary policy to this model will certainly not produce the optimal result as the model sometimes foresees the onset of a crisis with a one-year lack. However, the fact that the model points to a crisis one year later than the crisis actually occurred implies that the regressors are informative for the purpose of crisis prediction. In order to improve the predictive power of the model it might be wise to include the expected value of each regressor into the model.
In Appendix 6 the results are shown of 3 models, which uses the expected value of the regressors as explanatory variables. The models differ from each other by the way the expected values of the regressors are calculated. The expected values of the regressors in each model are calculated as follows:
Model 1: E(Xt) = Var + 0.5*(Xt + Xt-1) Model 2: E(Xt) = Xt+1
Appendix 6: Estimation results using expected values
Dependent variable: Binary crisis index based on occurrence banking and/or
currency crisis (1) (2) (3) Observations 1407 1448 1407 E(Domestic Credit) .0123731 .0152812 .0104824 (0.000)* (0.000)* (0.000)* E(Inflation) E(Growth rate money supply) .0042486 .0021779 .001999 (0.004)* (0.041)* (0.028)* E(Short-term external debt) E(Current account
balance) 5.56e-06 .0000152 6.35e-06 (0.021)* (0.001)* (0.011)*
E(Foreign exchange -7.04e-06 -.0000112 -6.20e-06 reserves) (0.000)* (0.000)* (0.000)*
Summary statistics
LR chi2(K-1) 165.51 112.08 136.07 (0.000) (0.000) (0.000)
As can be observed from Appendix 6 exactly the same regressors are significant and have the correct sign as when the current values of the regressors were used. Appendix 7 displays the graphs with predicted probabilities for a number of countries that are generated by each model presented in Appendix 6, along with the predicted probabilities generated by model 3 of Appendix 1 for comparison purposes. A quick glance at the graphs directly shows that the peaks of the graphs generated by model 3 of Appendix 6 are observed one year earlier in general than the peaks of the original model that used the current values of the regressors. This is an obvious result as it is assumed that economic agents know the exact value of each regressor one year in advance. This assumption of perfect information is, of course, a very unrealistic one, because there are always unexpected shocks that could not have been foreseen by the market. Therefore, two other models are considered which make more realistic assumptions about the forecasting power of economic agents. Model 1 is very realistic because the expectations are formed on the basis of information that is available at the time economic agents form their forecast. Model 3 is less realistic than model 1, but more plausible than model 2 as it does not fully account on the future value of the regressor. However, as can be observed from the graphs these models do not improve the predictive power as their graphs roughly display similar behaviour as the graph generated by model 3 of Appendix 1.
By closer inspection of the description of the predictive power of the model 3 (Appendix 1) it could be noticed that the performance of the model differed depending on which type of crisis was considered. In the case of Argentina the model performed better when years of banking crisis were considered, while for most of the other countries the actual years in which it experienced a currency crisis was more accurately described by the model. The observation that the predictive power of the model could possibly vary when currency crisis episodes or banking crisis episodes are considered separately brings us to the point where we re-estimate the model for each of these types of crisis. The estimation results for both cases are presented in Appendix 8.
Appendix 8: Estimation results using different types of crisis
Table 1
Dependent variable: Binary crisis index based on occurrence banking crisis (1) (2) (3) (4) Observations 853 1303 1376 1481 Domestic Credit .0213104 .0279009 .0217886 .0236379 (0.000)* (0.000)* (0.000)* (0.000)* Inflation .00251 .0028679 (0.007)* (0.004)* Growth rate money supply -.0023509 -.0027694 .000467 (0.021)* (0.010)* (0.049)* Short-term
external debt 2.80e-11 (0.131) Current account balance .0000298 .0000138 .0000117 .0000128 (0.038)* (0.006)* (0.015)* (0.003)* Foreign exchange -.0000181 -.0000112 -.0000105 -.0000113 reserves (0.001)* (0.000)* (0.000)* (0.000)* Summary statistics AIC 660.5194 1014.519 1099.265 1152.854 LR chi2(K-1) 61.73 137.69 112.09 126.31 (0.000) (0.000) (0.000) (0.000)
Table 2
Dependent variable: Binary crisis index based on occurrence currency crisis (1) (2) Observations 1264 1410 Domestic Credit .0108365 .008696 (0.001)* (0.003)* Inflation Growth rate money supply Short-term external Current account balance 7.89e-07 (0.863) Foreign exchange -.0000245 -.0000242 reserves (0.000)* (0.000)* Summary statistics AIC 868.5365 925.464 LR chi2(K-1) 58.14 56.62 (0.000) (0.000)
Note: P-values are presented between parentheses. * indicates that a variable is significant at the 5 percent significance level. ** indicates that a variable is significant at the 10 percent significance level.
