Stock portfolio construction in the Korean stock market applying XGBoost Classifier

Abstract

The stock selection is identifying the stocks suitable for investors to construct a portfolio. The process of stock selection generally compares the value, quality, and soundness of the firms listed in the stock market. It is known that the analysis of the financial ratios recorded in each company’s financial statements is useful. In addition, several studies have shown that financial statements such as PER, PBR, PSR, F-score, and size can produce excess returns in stock investment. However, when investing in a company selected through financial statement analysis, it is reasonable to invest in referring to previous price movements and technical indicators recorded on the stock chart. Therefore, we use the excess return financial indicators and technical indicators encoded by classical trading strategies as the features of machine learning. This paper analyzes all the stocks in the KOSPI and KOSDAQ market and selects the candidates to construct an optimal portfolio by using XGBoost classifier. Especially, we apply two different methods to improve the predictability of XGBoost classifier and use them to construct our models. If the number of selected stocks remains reasonable, adjusting the threshold on the prediction probability of XGBoost classifier is meaningful to increase the precision. We confirm this relationship and use the new prediction thresholds. In addition, we select four encoded technical indicators based on feature importance and use it to get the final candidates. This is to decide which trading strategies should be mainly used in each trading period. It is confirmed that with this filtering, the precision is mostly higher than that of the simple XGBoost classifier. Finally, when monthly investing with the constructed portfolio by our models, it mostly outperforms our benchmark KOSPI index returns and gets sufficiently good cumulative profits to invest. In conclusion, our models can effectively help investors to make a rational stock investment portfolio.