TWO-COMPONENT EXTREME VALUE DISTRIBUTION FOR ASIA-PACIFIC STOCK INDEX RETURNS

Abstract

Financial risk management typically deals with low-probability events in the tails of asset return distributions. To better capture the behavior of these tails, several studies have clearly highlighted that one should rely on a methodology that directly focuses on the tails of the distribution rather than getting the tails as an outcome of modelling the entire density function. Traditional Extreme Value Theory (EVT) distributions, however, provide a good fit for the bulk of the extreme data but usually underestimate a small amount of observations considered as "outliers". Since the main objective of risk management analysis is to estimate the size and probability of very large price movements, these "outliers" are by definition the very events we need to investigate. In this paper we suggest the use of a Two-Component Extreme Value (TCEV) distribution where a 'basic distribution' generates ordinary extremes (more frequent and less severe in the mean) while an "outlying distribution" generates rarer but severe extremes. Goodness-of-fit tests show the superiority of this distribution to capture the extremes of eleven MSCI Indices of the Pacific-Basin region relative to traditional EVT densities. Measures of accuracy and efficiency used to assess the performance of VaR forecasts also indicate that the additional flexibility brought by the TCEV model provides strong improvements for risk management.