A Dynamic Modeling and Validation Framework for the Market Direction Prediction

There are many research papers talking about building various machine learning models to predict the market index. However, very few attention has been paid to effectively validating or calibrating the prediction results. The focus of this paper is to present a dynamic modeling and validation framework for the market direction prediction. The central idea is to calibrate the probabilistic prediction by estimating two conditional probabilities of correct forecast from the dynamic validation data set. The calibration method can be combined with any predictive model that generates probabilistic prediction of the market direction.