Prediction, Volatility and Complexity of Financial Markets:a move-based approach

Abstract

We consider financial time series by representing the movements in the data rather than the prices at fixed time steps. More precisely, we first identify a set of discrete levels and then record the first time the process crosses a level. This information can be simplified still further by simply recording the direction of the movements, creating a binary string: 1 for “up” and 0 for “down”. We show that it is possible to use this representation in order to create effective forecasting models based on the distance between the different sequences. Their parameters are determined by cross validation, the maximum likelihood and the data complexity. Afterwards, we construct several trading strategies based on the predictions given by the model, then, we measure the trading rules performances and compare them to a Buy and Hold strategy.