Prediction Of Stock Performance in The U.S. Stock Market Based on Logistic Regression Model

Abstract

How to select stocks as investment targets is a hot issue of concern in the investment market. Some researchers have found that the quintile ranking of payout, valuation, profitability, and historical growth in the market can be used as a screening basis to help investors make initial selections. However, the differences in ranking within the same quintile are still obvious, which does not allow for a fine and clear selection. Therefore, in this paper, by collecting the data profile of more than five thousand US stocks from 2015 to 2021 and fitting a logistic model to the data, the author examines price to book ratio, sales growth rate, price to earnings ratio, gross profit to asset ratio, return on equity, dividend yield and earnings per share to predict whether the stocks will outperform in the future, and further derive the result of whether it is worth investing or not under different decision thresholds to help differently risk-averse investors to screen stocks. The results show that for the prediction of whether stocks would “outperform or not”, the final model’s prediction was 71.28 percent accurate. This excellent result shows that this logistic model can be applied to the real market to help general investors and investment institutions to select the right stocks.