|
|
Comparison of LSTM and random forest on forecasting small-cap stocks from different regions
Michalski, Jakub ; Šíla, Jan (advisor) ; Červinka, Michal (referee)
The following thesis focuses on using machine and deep learning on predicting small-cap stock index returns. Namely, Random forest and LSTM models are tested on indices from different regions, which are Russell 2000, FTSE Smallcap, Nifty Smallcap 100, S&P/AXS Small Ordinaries, and B3 Smallcap Index. By writing this thesis I want to emphasise possible benefits that come with machine learning implementation in small-cap stocks indices analysis, and show which method is better. The R2 , which was used as the main metric, indicates that LSTM performs better than Random forest for every index. Indexwise the best results was achieved by the FTSE Smallcap with 61.1% R2 . We can also see some possible improvements in results by optimizing each index separately, or by including more features that are not that easy to get. JEL Classification C01, C45, C49, C51, C53 C67 C88 Keywords excess returns, LSTM, random forest, machine learning, deep learning, forecasting, small-cap stocks, small-cap index, financial market, neural network Title Comparison of LSTM and random forest on forecasting Small-cap stocks from different regions
|
guest ::
