ABSTRACT
This chapter deals with dynamic portfolio selection. The main methodological approaches are the machine learning (ML) method random forest and the grey relational analysis (GRA) within grey system theory. Although many approaches to forecasting and portfolio selection exist today, this chapter combines approaches, which have been proven in previous literature as being efficient and robust. Previous research deals with one or another approach, but here we use the forecasts from fandom forests to rank stock indices via the grey approach in order to make the decision on the structure of the portfolio. The empirical analysis includes selected stock market indices, for the period 2015–2020. Firstly, daily data are used for the ML part and forecasting. Next, the out-of-sample forecasts are used to construct weekly data for the portfolio return and risk. Trading strategies are simulated so that comparisons can be made between those that are based on the approaches of this chapter and those that are usual benchmarks from portfolio theory. Since the results are promising, future research and applications could take into consideration such an approach.
