Evolving fuzzy modelling in risk analysis

Traditionally, forecast methodologies emphasize precise point-forecasts of stationary data. Risk analysis demands forecasts that, in practice, must be developed using imprecise and nonstationary data. Currently, value-at-risk (VaR) is widely employed in risk analysis. VaR requires a form of interval forecasts. Generalized autoregressive conditional heteroskedasticity (GARCH) models are stochastic recursive systems commonly adopted in financial prediction. This paper addresses a new approach to handle imprecise and nonstationary data using evolving fuzzy modelling translated into a recursive, adaptive forecasting procedure. VaR analysis is conducted to compare the performance and robustness of evolving fuzzy forecasting against GARCH using São Paulo Stock Exchange data. Copyright © 2009 John Wiley & Sons, Ltd.