Abstract:
Drawing on the scientific literature on automated stock trading systems based on machine learning techniques, this thesis addresses to evaluate the ability of random forest algorithms to predict whether the expected return of an investment in stocks will be positive or negative at the end of a hypothetical five-day trading window. Specifically, the final aim is to build an effective trading tool calibrated for the Italian utility sector. Therefore, the stocks taken into consideration includes all the utility industries listed in the FTSE MIB index providing oil, gas, and electricity. The time frame from which the data were extracted ranges from January 1st, 2016, to December 31st, 2021. One of the most widely used approaches is to take advantage of technical analysis to build an exhaustive set of predictors and enhance the predictive capabilities of the model. In the first instance, several technical indicators are generated based on the historical data of stock trading price and volume, and the overall performances of the model, and consequently of the trading system, are evaluated. Finally, further steps were taken in the direction of providing the algorithm with not only market information, but also general Italian economy health indicators release dates, holidays, specific weekdays, and climate change awareness events dates. After including the above-mentioned additional information, the model has been tested against the simplest model. Early results suggest that the inclusion of this information boosts the overall performance of the random forest classifier, improving the final profits.
