A Regional Early Warning System Prototype for East Asia Regional Economic Monitoring Unit Asian Development Bank 1
A Regional Early Warning System Prototype for East Asia Regional Economic Monitoring Unit Asian Development Bank I. Introduction 1 As part of its contribution to the Kobe Research Project, the Asian Development Bank is implementing a technical assistance for an ASEAN+3 Framework for the Development of Early Warning Systems. The main objective of the technical assistance is to support the ASEAN+3 countries collaborative efforts in regional economic monitoring and surveillance by developing a regional early warning system (EWS) prototype that would help detect emerging macroeconomic, financial, and corporate sector vulnerabilities and prevent financial crises in the future. 1 2 The regional EWS prototype under development comprises four components: (i) a set of macroprudential indicators (MPIs); (ii) a nonparametric EWS model; (iii) a parametric EWS model; and (iv) a set of economic leading indicators of business cycles. ADB has worked closely with other concerned institutions (such as IMF, the Institute of International Economics, and the Korea Center for International Finance) and experts in these fields in developing the four components. 3 The four components of the prototype have different focuses. MPIs are used to assess the health and stability of financial systems in a less formal and more qualitative way, while the other three components make use of statistical techniques and have more specific focuses. The nonparametric and parametric EWS models are designed to assess the probability of a currency crisis within a 24-month time horizon, while leading indicators of business cycles are for predicting turning points of business cycles. The latter three components require high and long time series. Therefore, they can only be applied to countries where such data are available, including Indonesia, Republic of Korea, Malaysia, the Philippines, Singapore, and Thailand. The focuses and applicability of the four components of the EWS prototype are summarized in Table 1. 1 The need for establishing early warning systems was also highlighted in the Ministerial Statement of the ASEAN+3 Finance Ministers at their meeting in Honolulu in May 2001. 2
Table 1 The Regional EWS Prototype and Its Applicability Countries with high Countries without high Macroprudential indicators Macroeconomic indicators Aggregated microprudential indicators Not applicable as these indicators are included in the EWS models Assess the health of banks and nonbank financial institutions Assess the health and stability of financial systems Non-parametric EWS Parametric EWS Leading indicators of business cycles Assess the likelihood of a currency crisis Assess the likelihood of a currency crisis Predict turning points of business cycles Not applicable due to the lack of high Not applicable due to the lack of high Not applicable due to the lack of high II. Components of the Regional EWS Prototype A. Macroprudential Indicators 4 IMF, in collaboration with other international finance institutions, national authorities, and the private sector, initiated work on MPIs a few years ago as part of its efforts to strengthen the surveillance process at IMF. MPIs, broadly defined as indicators of health and stability of financial systems, can be classified into two categories. The first is the aggregated microprudential indicators. These are mostly derived by aggregating indicators of the heath of individual financial institutions, including capital adequacy, asset quality, management soundness, earnings, liquidity and sensitivity to market risk, or the so-called CAMELS. The second category pertains to indicators of macroeconomic developments and external shocks. These cover growth performance, balance of payments positions, monetary and fiscal conditions, interest rates and exchange rates, and asset prices. Worsening of these indicators could affect the stability of financial systems, and lead to currency and banking crises. 5 Key aggregated microprudential indicators we have identified so far include the capital adequacy ratios, nonperforming loan (NPL) ratios, rates of return on equity and assets of the banking sector, growth of real commercial bank deposits, corporate debtequity ratios, corporate profitability, credit ratings, sovereign yield spreads, and price indices of banking equities. 3
6 Key macroeconomic indicators we have identified can be classified into six groups by sources of financial vulnerability. These are the current account, capital account, financial sector, fiscal account, real sector, and global economy. The selection of indicators was based on economic rationale as well as recent findings of empirical studies on currency and banking crises involving various countries and regions, including East Asia. Another major consideration was data availability. The proposed indicators for each group are as follows: Current account indicators: real exchange rates, export growth, import growth, ratios of the trade account balance to GDP, and ratios of the current account balance to GDP; Capital account indicators: ratios of short-term debt to foreign reserves, ratios of M2 to foreign reserves, ratios of banks foreign liabilities to foreign assets, ratios of residents deposits in the Bank of International Settlement banks to foreign reserves, and growth of foreign reserves; Financial sector indicators: ratios of domestic credit to GDP, M2 money multipliers, excess real M1 balances, domestic real interest rates, and lendingdeposit rate spreads; Fiscal account indicators: ratios of the fiscal balance to GDP, ratios of government consumption to GDP, and ratios of public debt to GDP; Real sector indicators: growth of industrial production, changes in stock prices; and Global economy indicators: growth of world oil prices, the US real interest rate, US GDP growth, and the US dollar/yen real exchange rate. 7 Further work in this area is to develop more aggregated microprudential indicators reflecting the health of banks and non-bank financial institutions, and to identify a set of core MPIs. B. A Nonparametric EWS Model 8 The second component of the regional EWS prototype is a nonparametric EWS model. EWS models offer a systematic, objective, and consistent method of assessing financial vulnerability that avoids analysts biases. In such models, composite leading indices provide summary measures of financial vulnerability for a particular country at a particular point in time, and individual indicators reveal sources of financial vulnerability. 9 The nonparametric EWS model we have developed follows the signaling approach pioneered by Kaminsky and Reinhart. 2 Technically, the model can generate, for each country and on a monthly basis, estimates of the likelihood of a currency crisis within the next 24 months. The model was initially estimated using monthly data from 1970 to 1995 for Indonesia, Korea, Malaysia, Philippines, Singapore, and Thailand. After 2 See Goldstein, Kaminsky and Reinhart, 2000, Assessing Financial Vulnerability: An Early Warning System for Emerging Markets. Institute for International Economics, Washington, DC. 4
demonstrating that the model is capable of predicting the 1997 Asian financial crisis outof-sample, it was reestimated and updated to include data up to 2000. 10 In the model, a crisis episode is considered to occur in a particular month if the month-over-month percentage change in a bilateral nominal exchange rate (e.g., local currency/the US dollar) exceeds its sample mean by two standard deviations. Thirtyeight economic and financial indicators are used to predict such depreciation episodes. Most of these indicators are among the MPIs of the first component described earlier. Some MPIs, such as nonperforming loan ratios, capital adequacy ratios, ratios of government debt to GDP, although important, are not part of the model, because they are available only for more recent years and do not have a long enough time series. In many cases, the model looks at a particular variable not only in its level form, but also its change (or percentage change) over 12 months or deviations from its trend. 11 In the model, each individual indicator is examined to determine whether it has crossed its threshold value that has historically been associated with heightened probability of currency crises. When an observed outcome of an individual leading indicator crosses its threshold value, it is considered as issuing a warning signal. A warning signal is associated with a certain conditional probability. For example, if a particular leading indicator has a conditional probability of 50 percent, this means that, on average, in the six countries under consideration, 50 percent of the months in which this indicator crossed its threshold during 1970-2000 were actually associated with increased financial vulnerability leading to currency crises within 24 months. The conditional probability differs across different leading indicators. The higher the conditional probability, the more reliable a leading indicator would be in detecting vulnerability. 12 The threshold value of a leading indicator is set such that the so-called noise-tosignal ratio that is the ratio of the likelihood of the indicator signaling during times with low financial vulnerability (or tranquil period) to the likelihood of it signaling during times with high vulnerability (or crisis period) is minimized. To take into account countryspecific characteristics that are not necessarily related to financial vulnerability but due to factors such as differences in policies, institutional structures, and variable definitions, thresholds of leading indicators are all country-specific. 13 Estimation results show that eight out of the 38 leading indicators have a conditional probability greater than 50 percent. These are, in order of probability value, the deviation of the real exchange rate against the US dollar from its trend, the deviation of the real effective exchange rate from its trend, the ratio of short-term debt to foreign reserves, the ratio of M2 to foreign reserves, the ratio of banks foreign liabilities to foreign assets, the ratio of residents deposits in Bank for International Settlements (BIS) banks to foreign reserves, the deviation of the real dollar/yen exchange rate from its trend, and the deviation of the foreign liabilities/foreign assets ratio from its trend. The conditional probability for the rest of the individual indicators ranges from 30 to 49 percent. Across the six indicator categories, on average, the current account has the highest conditional probability, which is followed by the capital account, the global economy, the financial sector, the real sector, and the fiscal sector. These suggest that the current account and capital account indicators are on average more reliable than other types of indicators, in particular, the real sector and fiscal sector indicators, in assessing the probability of currency crises. 5
14 Signals by individual indicators (taking the value of one when signaling and zero otherwise) were then aggregated into composite leading indices to provide summary measures of economic and financial vulnerability that could lead to currency crises. Composite leading indices contain more information than individual leading indicators, and are considered to be more reliable. In the model, there are six sector-specific composite leading indices and two overall composite leading indices. The six sector composite indices are, respectively, the Current Account Composite Index, Capital Account Composite Index, Financial Sector Composite Index, Fiscal Account Composite Index, Real Sector Composite Index, and Global Economy Composite Index. Each of these is a weighted average of one/zero signals of individual leading indicators of the concerned sector, with weights being their respective noise-to-signal ratios. The first overall composite leading index is a weighted average of the six sector composite indices, with weights being its noise-to-signal ratio. The second overall composite leading index is a simple sum of all the one/zero signals of individual leading indicators. 15 Estimation results show that, during 1970 to 2000, the weighted composite leading index captures 83 percent of the crisis episodes at a conditional probability of 82 percent, and the unweighted composite leading index captures about 70 percent of crisis episodes at a conditional probability of 65 percent. 16 Limitations of the EWS approach are: (i) at a theoretical level, there are many conceptual and methodological issues to be further studied; (ii) availability and accuracy of data are also major constraints to the usefulness of such models; and (iii) EWS models have so far found to perform reasonably well when they are applied to currency crises, while the track record of EWS models for banking crises is not so good. For these reasons, EWS models should be used to complement, rather than substitute for, sound and balanced judgments on financial weaknesses. C. A Parametric EWS Model 17 A third component of the regional EWS prototype is a parametric EWS model based on the probit regression analysis. Compared to the signaling approach, a major advantage of a regression-based EWS model is that it is multivariate and considers all explanatory variables simultaneously. Hence impacts of a particular indicator on crisis probability are conditional on values of other indicators in the model, and could be more accurately measured. A further advantage is that it allows testing of statistical significance of individual indicators. 18 The parametric EWS model was also estimated using monthly data of Indonesia, Korea, Malaysia, the Philippines, Singapore, and Thailand from 1970 to 1995. In-sample estimation results indicate that the following variables possess the most significant predictive power in identifying currency crises: deviations of the real exchange rate from its trend; changes in the ratio of M2 to foreign reserves; changes in real stock prices (denominated in US dollars); the US real interest rate; and changes in the M2 money multiplier. Employing a cutoff probability of 20 percent, the in-sample test of the probit EWS model predicts 69 percent of all crisis episodes. The out of sample performance of the model is even better, with 78 percent of all crises predicted accurately. This suggests that with data up to 1995, the model is able to predict the 1997 Asian financial crisis with a considerable amount of confidence. 6
19 The regression approach also has its limitations. First, within this approach, it is difficult to assess which of the variables is "out of line." Second, the regression-based EWS models require larger sample size than the signalling-based EWS models for them to work properly. Often these large samples are not available and this becomes a major constraint. Third, the regression-based approach permits only a small number of indicators to be used, in order to maintain a sufficient degree of freedom and to avoid multicolinearity. Because each of the two approaches has its advantages and limitations, they should be considered as complements to, not substitutes for, each other. D. Leading Indicators of Business Cycles 20 The last component of the regional EWS prototype is leading economic indicators of business cycles. Leading indicators are arguably the chief forecasting tool used to predict recessions and recoveries. They can provide advance warning of a sharp economic upturn (overheating) or downturn (recession) and thus enable policymakers to initiate preemptive action. Our work on leading indicators is still ongoing and now covers only Malaysia and the Philippines. The main constraint is data availability. Future work will involve expanding country coverage and working together with government agencies that compile similar indicators. III. Summary and Conclusions 21 ADB s work on the regional EWS prototype is still ongoing. But the progress so far has been very encouraging. Further work include (i) identifying a set of core MPIs; (ii) improving the parametric and nonparametric EWS models; (iii) integrating the parametric EWS model with the nonparametric EWS model; and (iv) extending the analysis of leading indicators of business cycles to more countries where high are available. 22 At the recently concluded ASEAN+3 Deputies meeting in Myanmar in early April, ADB presented the Vulnerability Assessment Report containing the simulation results of the EWS prototype using monthly data from January 2001 to February 2002, and received favorable feedback from the Deputies. ADB will continue to present Vulnerability Assessment Reports to the ASEAN+3 Deputies meetings in the future. 23 We have adopted a participatory approach in developing the EWS prototype. Various workshops have been organized, jointly with the Bank of Thailand, the Ministry of Finance and Economy of Korea, and the ASEAN Secretariat, to discuss the prototype and seek views from representatives of all member countries of the ASEAN+3 group. At the Myanmar ASEAN+3 Deputies meeting, the People s Bank of China announced that it would organize jointly with ADB a workshop later this year to finalize the regional EWS prototype. 24 It is envisaged that after the prototype is finalized, individual countries would use it as a model to set up their own early warning systems, taking into account their own circumstances. ADB is processing another technical assistance for supporting the development and implementation of individual country early warning systems. 7