ANZJS-11-10-008 Final EVALUATION OF RECURSIVE DETECTION METHODS FOR TURNING POINTS IN FINANCIAL TIME SERIES

Timely identification of turning points in economic time series is important for plan-ning control actions and achieving profitability. This paper compares sequential methods for detecting peaks and troughs in stock values and deciding the time to trade. Three semi-parametric methods are considered: double exponential smooth-ing, time-varying parameters and prediction errors statistics. These methods are widely used in monitoring, forecasting and control, and their common features are recursive computation and exponential weighting of observations. The novelty of this paper is to select smoothing and alarm coefficients on the maximization of the gain (the difference in level between subsequent peaks and troughs) of sample data. The methods are compared on applications to leading financial series and with simulation experiments. Key words: adaptive estimation; exponential smoothing; gain maximization; pre-diction errors; time-varying parameters