Using machine learning for quantitative trading strategies

V diplomski nalogi smo s pomočjo zgodovinskih podatkov delnice BAC, menjalnega tečaja BTC-USD in tehnične analize trgovanja naučili različne modele strojnega učenja trgovalnih strategij. Podatke iz tehnične analize smo pridobili s pomočjo različnih indikatorjev. Podatke smo označili po strategiji, ki temelji na povratku k srednji vrednosti. Strategija se lahko spreminja s pomočjo parametrov in tako izkoristi nihanja lokalnih ekstremov v različnih dolžinah časovnega okna. Pozorni smo na to, da je strategija donosna v različnih trendih (naraščajočem, padajočem ter stranskem). Podrobno smo opisali uporabljene metode, torej modele nadzorovanega strojnega učenja. Strategije smo preizkusili na omenjenih realnih podatkih in analizirali rezultate. Dobro strategijo, ki je donosna tako v naraščajočem kot tudi padajočem trendu, je prikazal le eden od modelov. To je naključni gozd z 10 drevesi.In this thesis, we trained different machine learning models the trading strategies with the help of the historical data of BAC shares, the exchange rate of BTC-USD and the technical analysis of trading. The data from the technical analysis were obtained by using different indicators. We marked the data according to the strategy based on the mean reversion. The strategy can be changed with the help of parameters and thus exploiting the fluctuations of the local extremes in different lengths of the time window. We are mindful of the fact that the strategy should be profitable in various trends (the rising, the decreasing, and the lateral). We described in detail the methods used, i.e. the models of supervised machine learning. The strategies were tested on aforementioned real data and the results were analysed. A good strategy that is profitable, both in rising and decreasing trends, was only achieved by one of the models. That is a random forest with 10 trees