2013年-IMF国际货币组织全球_Comparing_Parametric_and_Non_29页_918kb

Summary of "Comparing Parametric and Non-parametric Early Warning Systems for Currency Crises in Emerging Market Economies"

Core Content

This working paper compares the in-sample and out-of-sample performances of three early warning systems (EWS) for currency crises in emerging market economies (EMs). The EWS include both parametric and non-parametric models, and the study evaluates how the performance of these models is affected by the policymaker's risk preferences.

Main Points

1. Early Warning Systems Overview

• Parametric EWS: Uses a fixed effects logit model to estimate the probability of a currency crisis based on macroeconomic indicators and a political risk variable.
• Non-parametric EWS: Calculates crisis probability as a weighted average of crisis signals from selected indicators, with optimal thresholds chosen to minimize total misclassification error.

2. Crisis Definition

• A currency crisis is defined as a large depreciation of the nominal exchange rate and/or extensive losses of foreign exchange reserves over a 24-month forecast horizon.
• The exchange rate pressure index (ERPI) is used to identify crisis episodes. A crisis is detected when the ERPI exceeds the country-specific mean by three standard deviations.

3. Key Indicators and Their Significance

• Parametric EWS:
  • Real GDP growth is significant and negatively related to crisis incidence.
  • Growth rate in the stock of foreign exchange reserves is also significant and negatively related.
  • Current account balance (CAB/Y) is significant and negatively related.
  • Reserves to short-term external debt (FXR/STED) is significant and negatively related.
  • Political risk variable (government instability) is significant and positively related.
  • M2 to foreign exchange reserves (M2/FXR) is significant and positively related.
• Non-parametric EWS:
  • Current account balance and FXR/STED are the most reliable indicators.
  • Real GDP growth, ΔFXR, M2/FXR, and government instability are less reliable.

4. Performance of EWS

• The parametric EWS outperforms the non-parametric EWS in out-of-sample results due to lower total misclassification error.
• The performance of both EWS does not improve when the policymaker becomes more cautious, as this increases the number of false alarms without reducing the number of missed crises.

5. Impact of Policy Preferences

• When the policymaker assigns equal weights to type 1 and type 2 errors, the EWS performance remains the same.
• A more cautious policymaker (assigning higher weight to missing a crisis) results in more false alarms but does not improve crisis detection accuracy.

6. Role of Government Instability

• Government instability is a significant factor in the parametric EWS, but not in the non-parametric EWS.
• This suggests that the parametric model is better at capturing the influence of political factors on currency crises.

7. Empirical Findings

• The parametric EWS consistently identifies key indicators as significant across different estimation periods.
• The non-parametric EWS performs less reliably, especially in capturing the effects of government instability and other indicators.
• The performance of EWS improves after 2008, likely due to increased focus on macroeconomic indicators by international investors during the global financial crisis.

Key Information

• Data Period: January 1995 to December 2011, with analysis of different sub-periods (up to 2006, 2007, and 2008).
• Sample Size: 28 EMs with monthly data.
• Methodology:
  • Parametric: Fixed effects logit model with crisis incidence as the dependent variable.
  • Non-parametric: Crisis probability is a weighted average of signals from indicators, with optimal thresholds derived to minimize misclassification errors.
• Crisis Signals:
  • Type 1 error: Missing a crisis due to a high threshold.
  • Type 2 error: Issuing a false alarm due to a low threshold.
• Policy Trade-off: Policymakers face a trade-off between correctly identifying crises and minimizing false alarms.

Conclusion

The study concludes that the parametric EWS is more effective in predicting currency crises in EMs compared to the non-parametric EWS. While the non-parametric model is based on signal extraction from indicators, the parametric model provides more accurate forecasts by incorporating macroeconomic and political risk variables. Additionally, the performance of EWS does not improve with increased prudence, as it leads to more false alarms without better crisis detection. The findings support the idea that government instability and macroeconomic factors like credit growth and reserves are important in predicting currency crises, particularly in the parametric model.