Investment funds results analysis and forecasting using neural networks

The aim of this bachelor thesis is to create a recurrent neural network, which uses investment fund data and macroeconomic parameters to forecast future returns of investment funds and compare the network’s results to that of regression analysis. Relevant literature of investment funds and their usage in machine learning as well as deep learning problems is reviewed. Literature of most commonly used technologies in the problem area is also reviewed. Monthly historical data of 79 funds and several economic variables, spanning from December of 2006 to March of 2020, is collected. This data is used to perform linear regression, Lasso regression and Ridge regression analyses. The data is also used to train a recurrent neural network. The described methods were used to predict future monthly returns of investment funds. After comparing the results of applied methods by using 3 different error measures, it is apparent that linear regression performed the weakest, while the neural network outperformed all three regression algorithms