Online detection of financial time series peaks and troughs: A probability-based approach

The problem related to the identification of a change in time series trajectories plays\r\na crucial role in many contexts. In this paper, we propose a flexible and computationally\r\nefficient procedure for turning point identification based on hypothesis testing\r\napplied to the difference between two consecutive slopes in a rolling regression\r\nframework. Along with the description of the methodology, to measure the performance\r\nof the method we have applied it to the S&P 500 Stock Index and its subsector\r\nindices. By using an in-sample/out-of-sample approach we compare results with the\r\nprofit/losses we could obtain by using themoving average crossover strategy. Results\r\nshow that the operating signals obtained by our proposal may better enable financial\r\nanalysts to make profitable decisions. Finally we present an extensive simulation\r\nstudy to show the weaknesses and strengths of the proposal under different expected\r\nreturns and volatility scenarios.