Multidimensional Forecasting and Pattern Matching
Arnold Polanski · UEA Digital Repository (University of East Anglia) · 2014
We apply a pattern matching algorithm to multidimensional forecasting. The algorithm searches for occurrences of patterns in multidimensional time series and computes their predictive accuracy. A genetic algorithm breeds then patterns that maximize this accuracy evolving ever better predictors. In an application to financial data, we show that the most successful patterns in training samples can retain a statistically and economically significant predictive power out-of-sample.