Forecasting financial markets crashes: a case of Russian Trading system
A significant number of recent studies of the financial markets dynamics focus on the question of the market crashes detection and predictability. However, most of them study major critical events on the largest world markets. In this paper we study predictive power of autoregressive conditional duration models in forecasting durations between sequent significant falls of Russian Trading System Index log returns. Being an emerging market indicator the RTS index experience much higher frequency of relatively strong falls. We find strong autocorrelations in the series durations between significant negative log returns of the index arrivals and calibrate a parametric family of autoregressive conditional duration models to derive the durations’ forecasts. The retrospective analysis of 100 crashes arrived in 09/1998 — 11/2011 shows a significant predictive power of ACD models. However the problem of designing an effective real-time prediction technique remains open, as it is rather hard to identify the specification of the model which is the best for prediction the following crash in realtime forecasting.