Stock Prediction via Machine Learning and Factor Analysis

Since the application of quantitative strategy in the process of stock selection, many researchers in securities institutions or investment companies have been trying to use quantitative models rather than experienced analysts to select stocks with more stable and higher excess returns. Because the human brain cannot simultaneously process and analyze large amounts of data, the application of machine learning could enable this process more efficient and precise. Our framework is to combine simple machine learning models with traditional factor analysis to quantitatively invest. We consider the stock prediction process as a classification problem and use the random forest classifier to make the factors as input features and the prediction of good stocks as target values. Then create a monthly-running trading strategy according to model prediction to gain excess profits. To optimize our model, we can analyze factors to replace model training inputs and select the hyperparameters through grid-search to improve the performance of prediction. Additionally, we explored the feature importance of factors and the order of searching hyperparameters on model optimization, which can lead us to upgrade the model itself. In the end, this paper delivers a full practice on the utilization of machine learning with factor analysis in quantitative strategy.