Mining for critical stock price movements using temporal power laws and integrated autoregressive models

This paper investigates the practical applicability of the log-periodic power law model to forecast large drawdowns of stock prices and compares its performance with the performance of the classical integrated autoregressive time series model. Both models are fitted to the daily closing prices of the Dow Jones index. In the case of the log-periodic power law model an alarm is issued if any fit conforming to theoretically motivated parameter restrictions can be found. In the case of the integrated autoregressive model an alarm is issued if structural breaks are observed at the end of the fit interval. It is shown that both models are successful in predicting upcoming stock market crises. The log-periodic power law model is superior in filtering out extreme drawdowns. However, its performance is highly dependent on the fit procedure.