This thesis compares three different methods for predicting the value of stock indices. Two of the methods use machine learning to forecast values, while one uses a combination of statistical analysis and machine learning. The aim of the thesis was to figure out if it is possible to predict future stock index values and to find which of the methods is the best. The comparison was done on data of three different stock indices in three different markets. The forecasted values were compared using multiple different metrics, where the value of final equity was given the biggest weight. Two different algorithms were developed for the purpose of trading simulation.
The results showed that it is possible to predict future values of stock indices, but not with all methods. They also showed that the results can be positively influenced by trading algorithms. All methods provided predictions good enough to trade better than the buy and hold strategy. One of the methods also ended with a positive final capital even with subtracted inflation. It was found that the standard metrics for assessing the quality of the results are not always suitable, especially in the case of predicting stock index values.
|
