From transition crises to macroeconomic stability? Lessons from a crises early warning system for Eastern European and CIS countries.

Kittelmann, Kristina ; Tirpak, Marcel ; Schweickert, Rainer 等

This paper uses a Markov regime-switching model to assess the vulnerability of a series of Central and Eastern European countries (ie Czech Republic, Hungary, Slovak Republic) and two CIS countries (ie, Russia and Ukraine) during the period 1993-2004. For the new EU member states in Central and Eastern Europe, the results of our model show that the majority of crises in those countries can be explained by inconsistencies in the domestic policy mix and by the deterioration of macroeconomic fundamentals, as emphasised by first-generation crises models, while for the CIS countries analysed, financial vulnerability type indicators were the most relevant, that is, indicators connected with the second- and third-generation of crisis model better explain the vulnerability of these countries. Additionally, the set of indicators chosen by our model is rather heterogeneous, supporting the superiority of a country-by-country approach. Comparative Economic Studies (2006) 48, 410-434. doi:10.1057/palgrave.ces.8100162

Keywords: EU, Central and Eastern Europe, CIS, early warning system, currency crisis, Markov switching

3EL Classifications: F47, P20, C22

INTRODUCTION

The concept of early warning systems (EWS) has been connected with various methodologies. The pioneering paper of Kaminsky et al. (1998) and the implementation of a EWS for currency crises by the International Monetary Fund (1) (IMF) set the standard for the early empirical studies. A number of empirical papers dealt with financial crises of the Southeast Asian and Latin American countries. Central and Eastern Europe, and especially the CIS countries, were usually not included in those early studies.

The reasons were obvious: inconsistent and short data series, application of different methodologies for statistical reporting and rapid economic transition. In spite of these difficulties, the first empirical studies dealing with so-called transition countries appeared during the end of the nineties (see Bruggemann and Linne, 1999, 2002) using the traditional 'signals' approach. Recently, new methodologies have been applied to EWS construction. In this paper, we follow the study by Abiad (2003) and estimate a Markov regime-switching model with time-varying transition probabilities. The primary objective of this paper is to estimate crises that occurred in a set of different countries: Czech Republic, Hungary, Russia, Slovak Republic, and Ukraine. (2) Our model correctly identifies most of the crisis periods.

The rest of the paper is organised as follows. Section 2 provides a review of theoretical and empirical literature on currency crises. Section 3 describes the applied methodology, model, and data specification. Section 4 presents the results from the EWS based on a Markov regime-switching model and makes an evaluation of the capabilities of the model, based on a comparative goodness of fit assessment. Section 5 concludes the paper.

THEORETICAL AND EMPIRICAL BACKGROUND

To quantify the potential vulnerability of a speculative attack, the causes of crises must be clearly understood. The following overview of theoretical crises models represents the analytical setting for the choice of possible leading indicators in the early warning model. (3)

According to the literature, one can distinguish between three classes of theoretical models of currency crises. Following the model of Salant and Henderson (1978), Krugman (1979) and Flood and Garber (1984) developed the 'so-called' first-generation currency crises models as response to the currency crises of Mexico and Argentina in the 1970s. In particular, the basic premise in these models is the inconsistency of domestic policies, such as an excessive money-financed fiscal deficit, with the commitment to a fixed exchange rate. Domestic credit expansion in excess of money demand growth leads to a gradual decline in international reserves. In case reserves fall to a critically low level, which is perceived as insufficient by market agents, the currency comes under attack: the attack depletes reserves immediately and the fixed exchange rate regime must be abandoned. To this class belong also models, which suggest that a real appreciation of the currency and a deterioration of the trade or current account balance typically precede speculative attacks. Consequently, a crisis is the unavoidable and predictable outcome in an economy with a constant deterioration of its 'fundamentals'.

To capture the features of the crises in the European Monetary System (EMS) and in Mexico in the 1990s, a second-generation of currency crises models was developed (Obstfeld, 1986, 1994). These models show that the government faces a trade-off between alternative policies to defend or to abandon a fixed exchange rate regime, and that the foreign exchange market could be subject to self-fulfilling expectations if the cost of defending the exchange rate peg rises with the expectations of private agents towards an abandonment of it. This implies the existence of multiple equilibria, since a change in private sector's expectations may lead to a jump from one to another equilibrium. (4) The exact timing of crises is, therefore, unpredictable. However, Krugman (1996) and Obstfeld (1996) emphasise that some weakness in the fundamentals is required before a shift in expectations can push an economy into a crisis. Thus, it is possible to identify zones of vulnerability for 'fundamentals' where a crisis may or may not occur. In principle, any fundamental variable that influences the policymakers' decision whether to defend the fixed exchange rate can be considered in these models. The list of fundamentals originating from first-generation models is, therefore, expanded to output, unemployment, inflation, and domestic and foreign interest rates. An important implication of these models is that anticipating crises may be extremely difficult since a tight link between fundamentals and crises does not necessarily exist.

The Asian crisis of 1997 exhibited particular characteristics that could not be fully explained by first- and second-generation models. This led to the development of third-generation crises models. The focus of these models is to explain the combination of a weak financial sector with currency crises ('twin crises'). (5) Within this new category, two types of currency crises models may be distinguished. On one, explicit or implicit government guarantees for private sector foreign borrowing can create a moral hazard problem. This typically gives rise to an asset price bubble. (6) If the asset price bubble bursts, this causes a severe liquidity problem and a contraction of economic activity as well as a costly fiscal bailout of bad loans. Therefore, expectations of an ensuing excessively expansionary monetary policy rise. Thus, like in first-generation crises models, the speculative attack on the currency originates from inconsistent domestic policies. On the other hand, currency crises can also originate from banking crises if a run on the local banking system encourages panic-stricken foreign investors to flee the country. (7) In these models, high short-term foreign capital inflows (foreign currency loans) lead to currency and maturity mismatches which causes banking system fragility. The collapse of a solvent but illiquid banking system is due to bank runs based on self-fulfilling expectations. In this setting, a currency crisis occurs because the role of the central bank as lender of last resort comes into conflict with the need to defend the fixed exchange rate. In general, third-generation currency crises models emphasise the role played by microeconomic factors implying that the list of potential indicators can be enlarged to banking deposits, short-term foreign debt and M2 to reserves, bank assets, lending deposit rate ratio, portfolio flows, stock price indices, and M2 multipliers. Summary of potential leading indicators with respect to crises symptoms and theoretical models can be seen in Table 1.

In the earlier empirical literature (8) two major approaches for predicting currency crises can be distinguished. The first one is the signals approach, originally presented by Kaminsky and Reinhart (1996) and Kaminsky et al. (1998). This approach chooses thresholds for each indicator variable in order to distinguish their movements in periods preceding a crisis from their usual behaviour in normal, non-crisis periods. The level of the threshold is set so that it minimises the 'noise-to-signal-ratio', that is, the risk of false signals to the risk of missing crises. The contribution of each indicator variable to the vulnerability of a country can then easily be determined. This non-parametric approach is typically univariate (9) and does not allow testing the statistical significance levels of variables. These drawbacks can be overcome by applying multivariate logit or probit regressions. Eichengreen, Rose and Wyplosz (1995) and Frankel and Rose (1996) were the first to apply this method to predicting currency crises. All information about a crisis is contained in the predicted crisis probability. By comparing signals models with probit models, Berg and Pattillo (1999a, b) show that probit models perform slightly better in terms of predicting crises. One explanation may be the transformation of the indicator variables into binary variables in the signals approach, which entails a loss of information.

Both signals and probit/logit models require a priori dating of crises episodes before estimation. This entails many ad hoc assumptions since different methods can be applied which result in different crises dates being identified. A common procedure is to construct an index of speculative pressure and set a certain threshold level such that a crisis is being identified when this threshold is crossed.

On the other hand, applying a Markov switching model allows simultaneously identifying crises episodes and estimating crises forecast probabilities. Furthermore, by employing time-varying transition probabilities, the probability of switching from a tranquil regime to a crisis regime can be modelled as a function of a country's fundamentals. Markov switching models, therefore, acknowledge that periods leading to crises are intrinsically different from tranquil, non-crisis periods, and they allow determinants triggering shifts from one regime to another. The statistical significance of the determinants of crises can also be derived.

Markov switching models have been used in several empirical studies to determine currency crises. (10) However, none of them examines Central and Eastern European or CIS countries. Martinez-Peria (2002) used a Markov switching model with time-varying transition probabilities to model the currency crisis in the EMS in the early 1990s. Her results indicate that the regime-switching approach identifies speculative attacks better vis-a-vis using the common threshold crisis-dating procedure. The study by Abiad (2003), to which this work is most closely related, underlines the good predictive ability of the Markov switching model. He looks at the 1997 Asian-crisis countries, but unlike Martinez-Peria, who pooled the data across countries, estimates the model for each country separately. This takes into account that the economic situation in each country is different and, therefore, different leading indicators may be significant for different countries. By assessing the predictive ability of the model, he finds that the model both correctly anticipates more crises periods in the sample and sends fewer false signals than other models. (11) In a recent study, Arias and Erlandsson (2004) also apply the Markov switching concept with time-varying transition probabilities to the Asian-crisis countries, where they correct for the bias of the estimation procedure, which would result in the selection of short-term predictors of regime switches rather than long-term ones. The predictive ability of their model is comparable to the model of Abiad. (12)

As indicated before, only few studies have looked at transition economies, and even fewer to CIS countries, including Russia. Among those, Brtiggemann and Linne (2002) look at the vulnerability of 16 Central and Eastern European countries to currency crises. They find that exports, foreign exchange reserves, and the lending deposit rate ratio are the best performing indicators in signalling a crisis. The real exchange rate, banking deposits, budget deficit, industrial production, M2 multiplier, domestic credit and interest rate, M2 to reserves, short-term foreign debt, and imports are also useful in predicting a currency crisis.

MODEL SPECIFICATION

The endogenous variable [y.sub.t] in our model is assumed to depend on an unobserved first-order two-state Markov chain [{[s.sub.t]}.sub.t=1.sup.T] as follows:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (1)

where [s.sub.t] = 0 denotes a tranquil, non-crisis state and [s.sub.t] = 1 a crisis state. The mean and variance of [y.sub.t] are allowed to shift with the respective state [s.sub.t]. Hence, the conditional density of [y.sub.t] for [s.sub.t] = 0, 1 is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (2)

The transition probability matrix [P.sub.t] associated with the latent regime-switching variable [s.sub.t] is defined as

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (3)

The transition probability [p.sup.t.sub.ij] gives the probability that state i in period t-1 will be followed by state j in period t. The two transition probabilities are time varying, evolving as the cumulative density function of the logistic distribution F. The constant and the early warning indicators, which affect the state transition probabilities, are contained in vector [x.sub.t-1]. (13) To get maximum likelihood estimates of all parameters in the regime-switching model, we use the likelihood function using the iteration described in Hamilton (1994: 692-93).

To be able to assess currency crises under different exchange rate regimes, the 'crisis' variable to be used as dependent variable [y.sub.t] is an 'Index of Speculative Pressure' (ISP), defined as:

IS[P.sub.t] = 1/[[sigma].sub.ER] (E[R.sub.t] - E[R.sub.t - 1/E[R.sub.t-1) + 1/[[sigma].sub.IR] (I[R.sub.t] - (I[R.sub.t-1]) - 1/[[sigma].sub.R]([R.sub.t] - [R.sub.t-1]/[R.sub.t-1]). (4)

In (4) above, ER denotes the nominal exchange rate (defined as domestic currency to Euro), IR denotes nominal interest rate, R denotes international reserves, and a denotes the respective standard deviation. Every variable is determined as one-month growth rate. An increase of the ISP, therefore, originates from an increase in the nominal exchange rate (depreciation of the domestic currency) and/or a rise in interest rates and/or a decrease in reserves.

The model estimates 1-month ahead forecast probabilities, which are transformed into 12-month ahead forecast probabilities. (14) This reflects a compromise between more accurate forecasts shortly before the crisis date and the fact that the model should signal a crisis as soon as possible (Arias and Erlandsson, 2004). As the estimated crisis probability cannot be compared to the actual crisis probability, it has become standard to determine a cut-off probability so that an alarm signal is being emitted if the forecasted probability is higher than this threshold. We set this cut-off probability to 50%, favouring the risk of missing crises to the risk of having more false signals at a lower cut-off probability. In this case a good signal is defined if the estimated crisis probability is higher than 50 % and a crisis ensued within the next 12 month or no signal was issued and no crisis occurred. In the same way, a signal is said to be false if the forecasted crisis probability is higher than 50% and no crisis occurred during the next 12 month or no signal was issued and a crisis ensued.

EMPIRICAL RESULTS

The model is estimated by using monthly data series for the Czech Republic, Hungary, Russia, the Slovak Republic, and Ukraine, taken from IMF's International Financial Statistic (IFS). This is a very heterogeneous set of countries: As three of these countries entered the EU in May 2004, while two others, Russia, and Ukraine, have no such perspective in the short run, and one, Russia, is largely a resource dependent economy, this set enables one to assess if a credible prospect of EU entry makes any difference in terms of both crises determinants and crises probability.

The series cover the period 1993:12 to 2004:06. The selection of potential early warning indicators is made using as reference theoretical models of currency crises, plus specific features of the analysed economies (for instance, its dependency on oil exports, in the case of Russia, and as an ersatz oil/gas exporter--due to the transit fees on Russian oil and gas--for Ukraine), and data availability. The tested variables are listed in Table 2 below.

We present the results for the five countries mentioned before: the Czech Republic, Hungary, Russia, the Slovak Republic, and Ukraine. Our country sample is restricted for two reasons. First, for most of the remaining new EU member states the EWS was estimated, (15) but it just does not captures any significant 'crisis event', even when using the broad definition of an ISP. This is specially, but not only, the case for currency board or quasi-currency board regimes (Estonia, Latvia, and Lithuania). Of course, this non-result is in itself an interesting one, as it may be seen as an indication of the robustness of such an extreme type of hard currency regime. Second, for other new EU member states, the EU accession countries Bulgaria and Romania, and other CIS countries, either the shortness of the usable sample, the questionable quality of the date series, or statistical problems with the data hinders interpretation of the results and puts some doubts on their overall robustness. For these five countries for which data is available and which reveal crisis events daring the data availability period our EWS show clear, robust and interpretable results, which we will detail below.

The final selection of explanatory variables for the model is made by using the approach of Abiad (2003). A bivariate model is estimated with only one indicator and a constant at a time to get t-statistics for the coefficients of the indicators and log-likelihood values for the corresponding model. (16) The different signs and degrees of statistical significance of various indicators confirm the country-by-country assessment, since the chosen countries exhibit different sources of vulnerability. (17) Only some indicators are significant per country: for the Czech Republic, deviation of real exchange rate from trend, current account balance as a GDP share, the growth rate of the industrial production and the FDI-current account deficit on GDP, in case of Hungary the lending deposit rate ratio and the gap between foreign direct investment and current account deficit, in the case of the Slovak Republic the gap between foreign direct investment and current account deficit, while for Russia the deviation of real exchange rate from trend, the LIBOR, changes in the ratio of deposits to M2 and ratio of loans to deposits, and for Ukraine only banking deposits/M2, M2/reserves and the real interest rate are significant.

However, considering that there is correlation among the indicators, the t-statistics may be misleading for the significance assessment of indicators. Therefore, the selection criterion for the multivariate models is the log-likelihood value of each bivariate model. Based on this final criterion, significant indicators have been chosen for each country. Performing a likelihood-ratio test for joint significance of indicators showed them to be significant.

The final results of the multivariate regression are shown in Table 3. As can be seen, rather traditional indicators of crises are chosen for the Czech Republic and the Slovak Republic, mostly related to external imbalances, and in the case of Hungary a mix of financial sector and external imbalances indicators is relevant. For Russia and Ukraine, the list mainly includes financial sector indicators. Furthermore, the expected conditions [[mu].sub.o] < [[mu].sub.1] and [[sigma].sub.o] < [[sigma].sub.1] hold, so that state 0 is identified as low-mean, low-volatility regime, and state 1 as high-mean, high-volatility regime.

Czech Republic

Following the results from the bivariate regression, we chose four indicators for the Czech Republic--the deviation of the real effective exchange rate from trend, current account balance to GDP, industrial production, and the gap between foreign direct investment and current account deficit to GDP. According to the selection, we conclude that balance of payment indicators play a substantial role in explaining speculative attacks in this case. Moreover, the deviation of the real effective exchange rate from trend, considered as the most complex indicator of speculative pressures, appears to be important in evaluating incoming problems.

Three out of four chosen indicators refer to external imbalances in the economy. The first-generation of theoretical models describes external imbalances as symptom of crisis. The inclusion of industrial production as an indicator shows that speculative attacks in the Czech Republic could be predicted also by indicators described in the second-generation theoretical models.

In Figure 1, we can observe crisis periods along with the 12-month forecast probabilities and alarm signals based on 50% cut-off probability. Currency speculators in the first months after the break-up of Czechoslovakia directed their attention to the appointed monetary authorities of the Czech Republic and the Slovak Republic. Although the result of this speculative attack was not a currency crisis, international reserves of both central banks decreased significantly. The first speculative attacks in a not-fully liberalised environment, a common feature of the experiences of the Czech and Slovak Republics, were not analysed, due to data availability limitations.

[FIGURE 1 OMITTED]

In the case of the Czech Republic one major currency crisis occurred in May 1997. Our model sends alarm signals from September 1996, 8 months prior to the crisis. The principal trigger of the speculative attack against the Czech crown was excessive external imbalance. High real wage growth exceeding productivity growth induced a substantial current account deficit. (18) Through rapid appreciation of the real exchange rate, the domestic corporate sector lost competitiveness. Consequently, increasing domestic absorption required an upswing in imports.

During the turbulent years of the late 1990s, our model is sending alarm signals up to the beginning of 1999. Forecast probabilities are oscillating noticeably above the 50 % threshold. One can observe contagion effect on the behaviour of the real exchange rate during 1998. Excessive appreciation pressures at the beginning of the summer of 1998 were replaced by flight of short-term foreign capital right after the Russian currency crisis erupted. Although we did not include an indicator for measuring contagion in our model, depreciation pressures against the Czech crown resulted in persistent signalling up to the end of 1998. Gibson and Tsakalaatos (2004) highlight the possible effect of the Asian crisis in 1997 and the Russian crisis in 1998 on accession countries. They found '... the strong effect from the Russian crisis, providing the evidence that contagion is an important factor in determining the probability of speculative attacks'. (19) Throughout 2002-2003, our model sends alarm signals sporadically.

Hungary

We ran the model for Hungary with four indicators--the current account balance to GDP, real domestic credit growth, the gap between FDI and current account deficit to GDP, and the ratio of banking deposits to M2. The results are shown in Figure 2.

[FIGURE 2 OMITTED]

The development of the Hungarian economy was disturbed by several devaluations of forint. During the summer of 1993, the Hungarian forint experienced three minor devaluations (June, July, and September), totalling 9.4%. The first major devaluation, by 8%, took place in August 1994 and served as a prologue to the introduction of government stabilisation measures at the end of 1994. Stabilisation measures took place with a supportive effect of a 9 % devaluation and a switch to more flexible crawling band regime on March 1995.

Our estimations correctly marked all the cases of speculative pressures in Hungary with sufficient time in advance. After the change of exchange rate regime in March 1995 and the initial devaluation of 9%, our model keeps sending signals of anticipated speculative pressures. The persistence of signals in that case could be interpreted as the result of the adopted crawling band exchange rate regime with its gradual devaluations. (20)

The signals during the year 1998 can be considered the outcome of the Russian crisis. Although there were no major movements in Hungarian financial markets, the negative sentiment concerning emerging markets after the Russian crisis put the Hungarian forint under temporary pressure. Worries about fiscal stability after the general elections in 1998 strengthened this behaviour of the markets. Nevertheless, the signals issued by the model during the end of 2000 and January 2001 are not linked to any open crisis.

The most recent speculative attack Hungary experienced took place during the first months of 2003, resulting in a 2.26% shift of the forint's central parity. Our model signals the possible currency crisis 8 months in advance of the first speculative attack against the Hungarian forint in January 2003. A continual increase in the current account deficit, together with low fiscal discipline, expressed in a growing budget deficit, undermined the defence of the Hungarian forint by the central bank.

The persistence of the twin deficits phenomenon in Hungary during 2003, amid recurrent speculative pressures on the forint, played a dominant role in the decision process of the Hungarian central bank. In an attempt to counteract the discrepancies produced by an inconsistent policy mix, the central bank responded by two rapid hikes of its base rate by 300 basis points. (21)

Slovak Republic

In our estimated model for the Slovak Republic, we chose as indicators the current account balance to GDP, real domestic credit growth, the gap between foreign direct investment and current account deficit to GDP, and the import-export ratio. Figure 3 displays the results.

[FIGURE 3 OMITTED]

The results of the bivariate regression highlight the indicators described in the first-generation of currency crises models. In line with results for the Czech Republic and Hungary, most of the indicators for the Slovak Republic focus on external imbalances. The growth of domestic credit also signals overborrowing cycles, which according to the theoretical models can precede both currency and banking crisis. In the late 1990s, the Slovak banking sector was near a collapse. Nevertheless, the rapid restructuring and the gradual removal of non-performing loans (22) to consolidation agencies averted an open banking crisis in the Slovak Republic.

The first speculative attack in a not-fully liberalised environment started right after the break-up of Czechoslovakia and the monetary separation in 1993. After depletion of international reserves, the National Bank of Slovakia came up with a 10% devaluation of the Slovak crown against a currency basket without widening the oscillation bands. (23) The main incentive for the devaluation was the defence of an initially low level of international reserves. Since our model starts to evaluate the vulnerability periods from early 1994, this speculative attack could not be detected.

The Slovak Republic experienced several speculative attacks during this period. Although the inconsistent policy mix and unfavourable macroeconomic developments increased the vulnerability of the Slovakian economy to speculative attacks, currency speculators succeeded only once in causing an open crisis, namely in the fall 1998. The model identifies almost persistent crisis signs, where the forecasted crisis probability is above the 50% threshold, from 1997 up to spring of 1999.

In May 1997, a speculative attack against the Slovak crown was led by currency speculators, a few days after they attacked the Czech crown. Although the strength of the Slovak economy in terms of international reserves was smaller when compared to the Czech economy, the first speculative attack against its currency was unsuccessful. However, devaluation pressures on the currency peg forced the central bank to increase interest rates substantially. According to Arvai and Vincze (2000), the main reason, why the speculative attack against the Slovak crown was not successful is that speculative capital inflow had been relatively low in the preceding years.

The successful speculative attack in the fall of 1998 ended with the abandonment of the pegged exchange rate regime. Besides large external imbalances and political uncertainty before general elections, contagion from the Russian crisis played an important role. The abandonment of the pegged regime was followed by a depreciation of about 20 %, after the exchange rate regime of the Slovak crown changed to a managed float. The model estimations are in line with the observed speculative pressures against the Slovak crown, since alarm signals are emitted before the excessive volatility periods of May 1997 and October 1998.

During 2001 and 2002 a high current account deficit and uncertainty about the general election in 2002 increased depreciation pressures against the Slovak crown. Politically motivated increases in public sector wages and pensions put pressures on the fiscal side and, through increased domestic absorption, on the external balance. The response of the National Bank of Slovakia was a massive intervention aiming at halting the fall of value of the Slovak crown. (24)

Russia

For Russia the deviation of real exchange rate from trend, the LIBOR, changes in the ratio of deposits to M2, and the ratio of loans to deposits have been chosen as indicators for the multivariate model. Those indicators are mostly linked to the so-called second- and third-generation crises models, and they highlight Russia's financial vulnerabilities.

Figure 4 shows that the issued signals are very much in line with the stylised description of the Russian crisis during the 1990s.

[FIGURE 4 OMITTED]

The dissolution of the Soviet Union at the end of 1991 was followed by the usual sharp GDP downturn (the so-called 'transitional recession', see Bakanova et al., 2004). A certain macroeconomic stabilisation around the mid 1990s was followed by the introduction of a pegged exchange rate regime with a crawling band against the US dollar, from July 1995 onwards, replacing the previous 'dirty float'.

However, the start of the Asian crisis in 1997 spread a negative shock throughout emerging markets. This external shock decreased investment confidence in Russia and caused capital outflows, forcing the Bank of Russia to defend the band. Although during the exchange market interventions in November 1997 the Bank of Russia lost over USD 6 billion of its liquid reserves, which was equal to two-thirds of total reserves at that time, the exchange band was successfully defended in that occasion.

Nevertheless, after renewed attacks in the run up to August 1998, the government was forced to default its domestic debt obligations: this is the onset of the famous Russian 1998 crisis, which also had substantial regional implications, including crises in some of countries covered by this paper.

The Russian rouble was devalued and the exchange rate band was abandoned, leading to the adoption of a 'dirty floating' regime (effectively, still a nominal exchange rate targeting, see Esanov et al., 2005). One consequence of the sharp depreciation was a rapid initial acceleration in inflation. However, the GDP fall was much less severe and prolonged than expected, given first, the gains in competitiveness from the devaluation in industrial sector with plenty of excess capacity, and the still ongoing increase in energy commodities prices (oil and gas), which represent almost 50% of Russia's exports. Those two factors--plus the undeniably more sustainable monetary and fiscal policy mix pursued since 1999, which is also related to the previous factors, given the importance of the energy sector in terms of fiscal revenues in Russia--have underpinned a GDP growth of almost 7% per year since 1999 (see Vinhas de Souza and Havrylyshyn, forthcoming 2006).

Particularly, one can see that the 1998 crisis is clearly forewarned by our model.

Ukraine

For Ukraine only banking deposits/M2, M2/reserves and the real interest rate variables were used in the final multivariate model, again reflecting second- and third-generation type of crisis determinants, that is, financial sector vulnerabilities.

Figure 5 reveals that this small multivariate EWS reflects the crisis period and vulnerability of the Ukraine quite well (see Vinhas de Souza et al., 2005). The first years of independence resulted in substantial adjustment costs for Ukraine. This was partly due to unfavourable initial conditions: Ukraine had one of the highest shares of large-scale intermediate goods industrial enterprises of the former Soviet Union, highly integrated and dependent on the rest of the USSR economy. As a result, Ukraine suffered one of the largest declines in output among the CIS, with manufacturing output declining by over 60% in the first 5 years of 'transition'. Monetary and fiscal policies were clearly on an unsustainable path during this period: budget deficits were close to 10% of GDP (a substantial part of which was linked to para-fiscal operations to support the energy sector). As these deficits were largely monetised, they also resulted in inflation, which reached almost 5,000% in 1993.

[FIGURE 5 OMITTED]

In 1994 an initial stabilisation programme was finally attempted. Similarly to other adjustment programmes in Eastern Europe, it included price and import/exports liberalisation, the unification of the exchange rate, some limited fiscal consolidation and the introduction in 1996 of a national currency, the hryvnia, which was linked to the US Dollar via an exchange rate band of 1.7-1.9 hryvnia/USD. These measures were successful in bringing down inflation from 400% in 1994 to 10% in 1997. Nevertheless, the persistent fiscal deficits were incompatible with a fixed exchange rate regime. The situation came to a head with contagion from the Russian crisis in August 1998. Foreign reserves fell to just over a week of imports, forcing the authorities to devalue the hryvnia (by more than 50 %) and to introduce strict restrictions on foreign exchange transactions. Inflation briefly increased, but returned to a downward trend by the early 2000s.

In December 1999, Viktor Yushchenko, a former Central Bank Governor who had built a solid reformist reputation during and after the 1998 crisis, was appointed Prime Minister. He moved fast to introduce reforms during its brief period in power (he was voted out of office in April 2001 by a coalition of 'oligarch' and Communist parties, after only 16 months in power). The strong growth resumption in Ukraine (interrupted in late 2004-2005 by the policy disorganisation linked to the change in power in the country) is considered by most analysts to be linked to the fiscal and tax reforms initiated during this period, and to the devaluations of hryvnia in 1998-1999 and its posterior linking to the USD (given that most of Ukraine's external markets are in the euro area, this implied a further depreciation of the hryvnia from 2003 onwards: a real cumulative depreciation of 40% happened since 1998) and the resumption of growth in major CIS markets.

During subsequent years the government' continued its efforts towards hardening budget constraints and making the tax system more transparent. Also, in 2000 a nominally free-floating exchange rate regime was introduced (de facto the hryvnia has been kept at almost constant rate with respect to US dollar, by means of foreign exchange market interventions). Since 2000 the trade and current accounts show surpluses, which lead to an increase of the money supply, as often the monetary authorities refrained from sterilising these inflows. The main reason behind the lack of effective sterilisation was lack of sterilisation instruments and ineffectiveness of NBU rates as a monetary policy tool. Also, due to the success of the stabilisation policy, the demand for financial assets increased, leading to high growth rates of money supply and a credit boom.

Therefore, one might see above that there was a concentration of crises episodes until 1996, when the Ukrainian national currency was introduced after the first stabilisation programme. This was followed by a tranquil period, which ended with the spillover of the Russian crisis in 1998. It is rather surprising that this crisis period was almost predicted by the 12-month forecast, according to which the crisis probability increased to 38% from virtually zero in a very short period of time. Owing to our sample, the instability associated with the Yushchenko's 2004 election is not registered. The figure also reveals that the introduction of a de-jure floating exchange rate regime in 2000 was followed by warnings, which where not related to actual crises events.

Forecast assessment

For assessing the predictive ability of our model, we constructed several goodness-of-fit measures which have become standard for EWS. This allows us also to compare our model to different EWS, although one has to take into account that this comparison is a more indicative one because of different country samples, time periods and definitions of crises. The results are shown in Table 4.

On average, our model correctly assesses 78% of the observations. Forecasting the pre-crisis periods is also very impressive (71% on average), the correct assessment of tranquil periods reached 84% of observations, and only 19% of alarms where false. (25)

Comparing our model to the similar model implemented for Asian countries (Abiad, 2003) and a signals approach implemented for Central and Eastern European countries (Bruggemann and Linne, 2002) highlights the overall good performance of our model. In both cases less pre-crisis periods could be predicted correctly but the percent of tranquil periods predicted correctly was higher. The comparison also suggests that the Markov switch approach is especially good in predicting crisis while the signals approach clearly estimates a much smaller share of false alarms. The relatively weak performance in the case of Bruggemann and Linne (2002) may also be due to the pooling of data across countries and the longer forecasting horizon of 24 month. This supports our assumption on the superiority of using a country-by-country approach with a medium-term time horizon.

The good performance of our model may be due to the fact that crises in our country sample were mainly (but not exclusively) caused by deteriorating fundamentals and, thus, according to first-generation of crises models, are clearly predictable. Also, the definition of currency crises may contribute to this result, since we set up our objective to assess not only devaluation periods, but unsuccessful speculative attacks as well.

One may also point out that, while roughly one-third of the periods in our sample were estimated as having crisis warnings in the covered new EU member states, for Russia and Ukraine almost half of all the periods in the sample had crises warnings. Albeit some of these may be linked to structural questions (ie, a higher dependency on more cyclical commodities in the case of Russia), this may also be seen as an indication that the EU accession process could have decreased the vulnerability of some of those countries to crises. The importance of regional (and subregional, for the Czech-Slovak case) contagion should also be stressed, given that most of the open crises periods observed in our sample are clustered around the 1998 Russian crisis.

SUMMARY AND POLICY CONCLUSION

This paper examined vulnerability periods in a series of Central and Eastern European countries during the period 1993-2004. For three new EU member states (Czech Republic, Hungary, and Slovak Republic), the results of our model have shown that the majority of currency crises in those can be explained by inconsistencies in the domestic policy mix, and by the deterioration of macroeconomic fundamentals with consequent effects in terms of external imbalances, that is, mostly traditional, first-generation type of crises, which means that crises in these countries were clearly predictable.

Opposed to that, and beyond an apparently greater overall vulnerability to crises than for the new EU member states (which may be linked to the EU accession process), for the CIS countries analysed here (Russia and Ukraine), second- and third-generation, financial vulnerability type indicators were the most relevant ones. A corollary of this is that crises may not be as clearly predictable in these countries, since those sorts of crises can also be subject to self-fulfilling expectations and multiple equilibria.

This study represents, to the best of our knowledge, the first attempt to apply a Markov regime-switching model to assess vulnerability periods of these countries. Although it is clear that EWS are far from perfect, and that the results do not represent a mechanical tool to avert potential crises, the surprisingly robust performance of this model leads one to conclude that the regime-switching approach may be quite useful to assess vulnerability periods in the chosen countries.

The different sets of vulnerabilities indicate different types of policy prescriptions. Given that the importance of external vulnerabilities is expected to decrease substantially for the new EU member states (especially after an eventual euro adoption), one can expect the importance of those external sustainability indicators to be reduce, and, therefore, the crises related to them. For Russia and Ukraine, the (ongoing) strengthening of their financial sectors could arguably be the priority task.

REFERENCES

Abiad, A. 2003: Early warning systems: A survey and a regime-switching approach. IMF Working Paper 03/32.

Alvarez-Plata, P and Schrooten, M. 2003: The Argentinean currency crisis: A Markov-switching model estimation. DIW Discussion Paper 348.

Arias, G and Erlandsson, UG. 2004: Regime switching as an alternative early warning system of currency crises--an application to South-East Asia. Department of Economics, Lund University, Scandinavian Working Papers in Economics 11.

Arvai, Z and Vincze, J. 2000: Financial crises in transition countries: Model and facts. National Bank of Hungary, Working Paper 2000/6.

Bakanova, M, Vinhas de Souza, LV, Kolesnikova, I and Abramov, I. 2004: Transition and growth in belarus. In: Ofer, G and Pomfret, R (eds). The economic prospects of the CIS: sources of long-term growth. Edward Elgar Publishing: UK. pp. 57-75.

Berg, A, Borensztein, E, Milesi-Ferretti, GM and Pattillo, C. 1999: Anticipating balance of payments crises: The role of early warning systems. IMF Occasional Paper 186.

Berg, A and Pattillo, C. 1999a: Are currency crises predictable? A test. IMF Staff Papers 46(2): 107-138.

Berg, A and Pattillo, C. 1999b: Predicting currency crises: The indicators approach and an alternative. Journal of International Money and Finance 18: 561-586.

Blackburn, K and Sola, M. 1993: Speculative currency attacks and balance of payment crises. Journal of Economic Surveys 7(2): 119-144.

Bruggemann, A and Linne, T. 1999: How good are leading indicators for currency and banking crises in Central and Eastern Europe? An empirical test. IWH Discussion Paper 95.

Bruggemann, A and Linne, T. 2002: Die Bestimmung des Risikopotenzials von Finanzkrisen anhand eines Fruhwarnindikatorensystems: Eine Untersuchung der EU-Beitrittskandidatenlander und ausgewaihlter Staaten Mittel- und Osteuropas Schriften des Instituts fur Wirtschaftsforschung Halle, Vol. 13, Nomos Verlagsgesellschaft, Baden-Baden.

Chang, R and Velasco, A. 1998: Financial Crises in Emerging Markets: A Canonical Model. NBER Working Paper 6606.

Corsetti, G, Pesenti, P and Roubini, N. 1998: Paper tigers? A model of the Asian crisis. Internet Version November 2004.

Diebold, FX, Lee, J-H and Weinbach, GC. 1993: Regime switching with time-varying transition probabilities. Federal Reserve Bank of Philadelphia, Working Paper 93-12.

Dooley, MP. 1997: A model of crises in emerging markets. NBER Working Paper 6300.

Eichengreen, B, Rose, AK and Wyplosz, C. 1995: Exchange market mayhem: The antecedents and aftermath of speculative attacks. Economic Policy 21: 249-312.

Esanov, A, Merkl, C and Vinhas de Souza, L. 2005: Monetary policy rules for Russia. Journal of Comparative Economics 33 (3): 484-499.

Flood, R and Garber, P. 1984: Collapsing exchange rate regimes: Some linear examples. Journal of International Economics 17: 1-13.

Flood, R and Marion, N. 1999: Perspectives on the recent currency crisis literature. International Journal of Finance and Economics 4(1): 1-26.

Frankel, JA and Rose, AK. 1996: Currency crashes in emerging markets: An empirical treatment. Journal of International Economics 41: 351-366.

Fratzscher, M. 1999: What causes currency crises: sunspots, contagion or fundamentals? EIU Working Paper 99/39.

Gibson, H and Tsakalaatos, E. 2004: Capital flows and speculative attacks in prospective EU member states. Economics of Transition 12(3): 559-586.

Goldfajn, I and Valdes, RO. 1997: Capital flows and the twin crises: The role of liquidity. IMF Working Paper 97/87.

Hamilton, JD. 1989: A new approach to the economic analysis of nonstationary time series and the business cycle. Econometrica 57: 357-384.

Hamilton, JD. 1994: Time series analysis. Princeton University Press: New Jersey.

Jeanne, O. 2000: Currency crises: A perspective on recent theoretical developments. Special Papers in International Economics, Department of Economics, Princeton University, Princeton, New Jersey.

Jeanne, O and Masson, R 2000: Currency crises, sunspots and markov-switching regimes. Journal of International Economics 50(2): 327-350.

Kaminsky, GL. 1998: Currency and banking crises: The early warnings of distress, Board of Governors of the Federal Reserve System, International Finance Discussion Papers 629.

Kaminsky, GL, Lizondo, S and Reinhart, CM. 1998: Leading indicators of currency crises. IMF Staff Papers 45(1): 1-48.

Kaminsky, GL and Reinhart, CM. 1996: The twin crises: The causes of banking and balance-of-payments problems. Board of Governors of the Federal Reserve System, International Finance Discussion Papers 544.

Kaminsky, GL and Reinhart, CM. 1999: The twin crises: The causes of banking and balance-of-payments problems. The American Economic Review 89(3): 473-500.

Krugman, P. 1979: A model of balance-of-payments crises. Journal of Money, Credit and Banking 11(3): 311-325.

Krugman, P. 1996: Are currency crises self-fulfilling? NBER macroeconomics annual. The MIT Press: Cambridge, pp. 345-406.

Krugman, P. 1998: What happened to Asia? November 2004.

Martinez-Peria, MS. 2002: A regime-switching approach to the study of speculative attacks: A Focus on EMS crises. Empirical Economics 27(2): 299-334.

Masson, P. 1999: Multiple equilibria, contagion, and the emerging market crises. IMF Working Paper 99/164.

McKinnon, RI and Pill, H. 1997: Credible economic liberalizations and overborrowing. The American Economic Review 87(2): 189-193.

Obstfeld, M. 1986: Rational and self-fulfilling balance-of-payment crises. The American Economic Review 76(1): 72-81.

Obstfeld, M. 1994: The logic of currency crises. NBER Working Paper 4640.

Obstfeld, M. 1996: Models of currency crises with self-fulfilling features. European Economic Review 40: 1037-1047.

Radelet, S and Sachs, J. 1998: The onset of the east Asian financial crisis Harvard Institute for International Development, November 2004.

Rosenberg, MR. 1998: Currency crises in emerging markets: A guide to speculative-attack models and early warning systems. Merrill Lynch, International Fixed Income Research: New York.

Salant, S and Henderson, D. 1978: Market anticipation of government policy and the price of gold. Journal of Political Economy 86: 627-648.

Vinhas de Souza, LM and Havrylyshyn, O. 2006: Return to growth in CIS countries and the macroeconomic framework. Springer: Germany (forthcoming).

Vinhas de Souza, LM, Schweickert, R, Movchan, V, Bilan, O and Burakovsky, I. 2005: Now so near, and yet still so far: Economic relations between Ukraine and the European Union. Institute for World Economics, Kiel Discussion Paper 419. Kiel.

The views expressed here are exclusively those of the authors and do not necessarily reflect the official views of their institutions. All usual disclaimers apply.

(1) The IMF's EWS is described in Berg et al. (1999).

(2) As explained later in the paper, from the group of Central and Eastern European countries we excluded Poland, because of data availability. Other countries from the region (ie, the Baltics, Belarus, and Bulgaria) were excluded due to the lack of open crisis during the period of data availability.

(3) For a comprehensive review of the theoretical literature for the first- and second-generation crises models, see Blackburn and Sola (1993), Flood and Marion (1999) and Jeanne (2000).

(4) What triggers the jump between multiple equilibria remains largely unexplained. Possible explanations are contagion effects or herding behaviour in the presence of imperfect information, see Masson (1999).

(5) Kaminsky and Reinhart (1996, 1999) pioneered the empirical work on twin crises. They found empirical evidence that banking crises tend to precede currency crises, hut the causal link is not unidirectional since the currency crisis deepens the banking crisis*

(6) See Corsetti et al. (1998), Dooley (1997), Krugman (1998) and McKinnon and Pill (1997).

(7) See Chang and Velasco (1998), Goldfajn and Valdes (1997) and Radelet and Sachs (1998).

(8) For an extensive survey of the empirical literature, see for example Kaminsky et al. (1998) and Abiad (2003).

(9) Kaminsky (1998) presents a method to combine individual indicators into a composite indicator.

(10) In addition to the studies mentioned, Alvarez-Plata and Schrooten (2003), Jeanne and Masson (2000) and Fratzscher (1999) use Markov-switching models with constant transition probabilities to model the switches between multiple equilibria leading to currency crises.

(11) At a cut-off probability of 50%, the model correctly calls 65% of pre-crisis periods, whereby 27% of total alarm signals are false.

(12) The model correctly calls approximately 70% of pre-crisis periods at a cut-off probability of 40%.

(13) Diebold et al. (1993) extended the baseline Hamilton (1989) regime-switching model to allow for time-varying transition probabilities.

(14) Here, it is assumed that the indicators that influence the crisis probability neither worsen nor improve during this period.

(15) Results not presented here, but available from the authors upon request.

(16) Each indicator is standardised to be zero mean and unit variance.

(17) Results not presented here, but available from the authors upon request.

(18) At the beginning of 1997, the estimations for current account deficit to GDP for the whole of 1997 were around 10%, far exceeding the expected inflows of long-term non-debt capital.

(19) Gibson and Tsakalaatos (2004: 577).

(20) The rate of devaluation in the crawling band regime decreased continuously, from 0.060 % of daily devaluation in March 1995 to 0.00654% of daily devaluation in April 2001.

(21) The first base rate increase took place in May, while the second was at the end of November. In both cases, the increase of the base interest rate was 300 basis points.

(22) The estimated costs of the removal of non-performing loans are about 105 billion Slovak crowns (about 12% of the nominal GDP in 1999).

(23) After the break-up of Czechoslovakia and the following monetary separation, both countries pegged their currencies to baskets with relatively narrow oscillation bands ([+ or -] 0.5% from central parity).

(24) During May 2002, the central bank decided to increase interest rates to cool down excessive demand pressures.

(25) The goodness-of-fit values differ somewhat ranging between 88% in Russia to 69% in Hungary for all observations. The results are most homogeneous for calling the tranquil periods, that is, above 80% success rate for all countries.

KRISTINA KITTELMANN (1), MARCEL TIRPAK (2), RAINER SCHWEICKERT3 & LUCIO VINHAS DE SOUZA (4)

(1) Sovereign Risk Unit, Moody's Deutschland, An der Welle 5, D-60322 Frankfurt/Main, Germany. E-mail: kristina.kittelmann@moodys.com

(2) Faculty of National Economy, Banking and International Finance Department, University of Economics, Bratislava, Slovakia. E-mail: marcel.tirpak@gmail.com

(3) Kiel Institute for World Economics, Duesternbrooker Weg 120, D-24105 Kiel, Germany. E-mail: rainer.schweickert@ifw-kiel.de

(4) Kiel Institute for World Economics and Head, Russia/Belarus Desk, DG-ECFIN, European Commission, Avenue de Beaulieu, 1, B-1160 Brussel, Belgium. E-mail: Lucio-Mauro.Vinhas-de-Souza@cec.eu.int

Table 1: Leading indicators, in terms of crises symptoms and
theoretical models

 Generation
 of Crises
Symptom Indicators Model (a) Sign

Expansionary M1 1 +
Monetary policy
 Foreign exchange 1 -
 reserves

 Domestic credit 1 +

Expansionary Budget Deficit/GDP 1 +
fiscal policy
 Public debt/GDP 1 +

Bank runs Banking deposits/M2 3 -

Overborrowing Domestic credit 1 +
cycles
 M2 multiplier 3 +

Current account Exports 1 -
problems
 Imports 1 +

 Real exchange rate 1 -

 Current account 1 +
 deficit/ GDP

 Terms of Trade 1 -

 FDI-current account 3 +
 deficit

Capital account Foreign exchange 1 -
problems reserves

 Interest rate 2 +/-
 differential

 M2/reserves 3 +

 Short-term foreign 3 +
 debt/reserves

 Portfolio flows/ 3 +
 Total capital flows

 Bank assets/GDP 3 +

Growth Real interest rate 1 +
slowdown
 Industrial 2 -
 production

 Output 2 -

 Unemployment 2 +

 Inflation 2 +

 Lending/deposit 3 +
 Rate ratio

 Stock price index 3 -

Symptom Description

Expansionary Loose monetary policy can lead to currency
Monetary policy crises if the central bank cannot guarantee
 the fixed peg anymore.

Expansionary Loose fiscal policy can be starting point for
fiscal policy a currency crisis if the government wants to
 overcome the problem by inflation.

Bank runs Bank runs can proceed (banking and) currency
 crises.

Overborrowing Currency (and banking) crises can be the
cycles consequence of rapid credit growth after
 liberalization of the domestic financial
 system and the elimination of capital
 account controls.

Current account External imbalances and a real exchange rate
problems overvaluation are part of a currency crisis.
 The loss of competitiveness can lead to
 recessions, business failures and a decline in
 the quality of loans. Therefore, large negative
 shocks to the terms of trade, exports, the real
 exchange rate and positive shocks to imports are
 crises symptoms.

Capital account High foreign interest rates lead to capital
problems outflows and may therefore anticipate currency
 crises. Large capital inflows usually fuel a
 lending boom. If the country's foreign debt is
 large and capital flight increases capital
 account problems become more severe since this
 raises issues of debt sustainability. High
 short-term foreign debt increases the
 vulnerability of a country to external shocks.

Growth Currency (and banking) crises are preceded by
slowdown recessions and the burst of asset price bubbles.
 High real interest rates could signal a
 liquidity crunch, which leads to a slowdown
 and banking fragility. A decline in loan quality
 can be shown by an increase in the lending deposit
 rate ratio.

Source: See Kaminsky (1998) and Rosenberg (1998).

(a) This column is meant to indicate from which generation of currency
crises the indicator originates. Therefore, indicators originating from
first- and/or second-generation crisis models are also important in
explaining third-generation currency crises.

Table 2: Tested variables

1 Deviation of real exchange rate
from trend

2 Current account balance/GDP

3 Real domestic credit, growth rate

4 Portfolio flows

5 Lending deposit rate ratio

6 FDI-current account deficit/GDP

7 Import-export ratio

8 M2 multiplier

9 M2 as share of reserves

10 Changes of M2 as share of reserves

11 GDP growth

12 Exports growth

13 Changes in reserves

14 Stock prices lagged growth

15 Lagged current account balance

16 Real interest rate

17 IPI growth

18 Growth of ratio of loans on deposits

19 LIBOR

20 Growth in bank assets

21 Lagged reserve ratio

22 Monetary authority credit

23 The ratio of non-FDI inflows

24 Ratio of deposits to M2

25 Changes in the ratio of deposits to M2

26 Ratio of Loans to deposits

27 Brent oil price

28 Changes in Brent oil price

29 Budgetary position of the central government
in GDP %

30 Changes of the budgetary position of
the central government in GDP%

Table 3: The multivariate EWS models

 Czech Republic

Early Warning Indicators Coeff. T-stat.

Mean, [s.sub.t] = 0 -0.19 -2.19
Mean, [s.sub.t] = 1 2.07 1.30
Sigma, [s.sub.t] = 0 0.80 12.60
Sigma, [s.sub.t] = 1 2.79 3.30
Deviation of real exchange rate from trend -1.65 -1.03
Current account balance/GDP -1.96 -0.70
Real domestic credit, growth rate
Industrial product-ion, growth rate -0.70 -0.52
FDI-current account deficit/GDP -0.32 -0.28
Import export ratio
Banking deposits/M2
M2/reserves
Real interest rate
LIBOR
Changes in the ratio of deposits to M2
Ratio of loans to deposits
Constant ([[beta].sub.0]) 2.91 1.17
Constant ([[beta].sub.1]) -0.38 -0.48
Number of observations 121
Likelihood ratio test 9.26
P-value 0.05

 Hungary

Early Warning Indicators Coeff. T-stat.

Mean, [s.sub.t] = 0 -0.09 -1.06
Mean, [s.sub.t] = 1 2.43 2.71
Sigma, [s.sub.t] = 0 0.81 10.80
Sigma, [s.sub.t] = 1 1.63 3.15
Deviation of real exchange rate from trend
Current account balance/GDP -0.27 -0.50
Real domestic credit, growth rate -0.84 -1.40
Industrial product-ion, growth rate
FDI-current account deficit/GDP -0.24 -0.21
Import export ratio
Banking deposits/M2 -1.16 -0.87
M2/reserves
Real interest rate
LIBOR
Changes in the ratio of deposits to M2
Ratio of loans to deposits
Constant ([[beta].sub.0]) 1.82 2.53
Constant ([[beta].sub.1]) 0.41 0.49
Number of observations 133
Likelihood ratio test 15.61
P-value 0.00

 Slovak Republic

Early Warning Indicators Coeff. T-stat.

Mean, [s.sub.t] = 0 -0.21 -2.49
Mean, [s.sub.t] = 1 2.77 1.32
Sigma, [s.sub.t] = 0 0.84 13.31
Sigma, [s.sub.t] = 1 2.42 2.79
Deviation of real exchange rate from trend
Current account balance/GDP -0.45 -0.56
Real domestic credit, growth rate -0.60 -1.01
Industrial product-ion, growth rate
FDI-current account deficit/GDP -0.48 -0.59
Import export ratio -0.41 -0.44
Banking deposits/M2
M2/reserves
Real interest rate
LIBOR
Changes in the ratio of deposits to M2
Ratio of loans to deposits
Constant ([[beta].sub.0]) 1.80 2.35
Constant ([[beta].sub.1]) -0.88 -1.28
Number of observations 124
Likelihood ratio test 11.68
P-value 0.02

 Russia

Early Warning Indicators Coeff. T-stat.

Mean, [s.sub.t] = 0 -0.28 0.07
Mean, [s.sub.t] = 1 0.25 0.48
Sigma, [s.sub.t] = 0 0.45 0.05
Sigma, [s.sub.t] = 1 2.92 0.25
Deviation of real exchange rate from trend 0.32 1.08
Current account balance/GDP
Real domestic credit, growth rate
Industrial product-ion, growth rate
FDI-current account deficit/GDP
Import export ratio
Banking deposits/M2
M2/reserves
Real interest rate
LIBOR -1.90 1.24
Changes in the ratio of deposits to M2 -0.45 0.56
Ratio of loans to deposits -0.51 0.50
Constant ([[beta].sub.0]) 1.74 0.64
Constant ([[beta].sub.1]) 1.08 0.38
Number of observations 122
Likelihood ratio test 11.73
P-value 0.02

 Ukraine

Early Warning Indicators Coeff. T-stat.

Mean, [s.sub.t] = 0 0.00 0.01
Mean, [s.sub.t] = 1 -0.11 -0.13
Sigma, [s.sub.t] = 0 0.22 14.05
Sigma, [s.sub.t] = 1 3.75 11.74
Deviation of real exchange rate from trend
Current account balance/GDP
Real domestic credit, growth rate
Industrial product-ion, growth rate
FDI-current account deficit/GDP
Import export ratio
Banking deposits/M2 -0.35 -0.13
M2/reserves -1.94 -1.35
Real interest rate 6.27 0.54
LIBOR
Changes in the ratio of deposits to M2
Ratio of loans to deposits
Constant ([[beta].sub.0]) -20.63 -1.28
Constant ([[beta].sub.1]) 0.76 2.15
Number of observations 125
Likelihood ratio test 7.98
P-value 0.05

Table 4: In-sample forecast assessment: Measures of predictive power

Goodness-of-fit (cut-off probability of 50%) Our Abiad
 model (2003)

(a) Per cent of observation correctly called 78 81
(b) Per cent of pre-crisis periods correctly called 71 65
(c) Per cent of tranquil periods correctly called 84 89
(d) False alarms as per cent of total alarms 19 27

Goodness-of-fit (cut-off probability of 50%) Bruggemann/
 Linne (2002)

(a) Per cent of observation correctly called 74
(b) Per cent of pre-crisis periods correctly called 16
(c) Per cent of tranquil periods correctly called 96
(d) False alarms as per cent of total alarms 5

(a) This is equal to the sum of pre-crisis month correctly called
and tranquil periods correctly called divided by the number of
observations.

(b) This is the number of pre-crisis periods correctly called
(observations for which the estimated probability of crisis is above
the cut-off probability and a crisis ensues within 12 month) as share
of total pre-crisis periods.

(c) This is the number of tranquil periods correctly called
(observations for which the estimated probability of crisis is below
the cut-off probability and no crisis ensues within 12 month) as share
of total tranquil periods.

(d) A false alarm is an observation with an estimated probability of
crisis above the cut-off probability not followed by a crisis within
12 month.