Incorporating causal notions to forecasting time series: a case study.

Financial time series have been analyzed with a wide variety of models and approaches, some of which can forecast with great accuracy. However, most of these models, especially the machine learning ones, cannot show additional information for the decision maker or the financial analyst. The notion of causality is a concept that provides a more complete understanding of a problem beyond improved forecasts. In this study, we propose integrating the treatment/control concept of causality into a forecasting framework to better predict financial time series. Our results show that the proposed methodology outperforms classic econometric approaches such as ARIMA and Random Walk, as well as machine learning approaches without the proposed methodology. This improvement is statistically significant, as indicated by the Model Confidence Set test in the complete test set and quarterly analysis. [ABSTRACT FROM AUTHOR]