An analysis of the efficacy of regression techniques in forecasting stock prices

Author: Joe Carman
Major: Physics and Math
Approved: Spring 2020
Status: In progress

This project aims to look at least squares regression, ridge regression, lasso regression, elastic net regression, and seasonal decomposition techniques as methods for forecasting times series stock price data. The study will aim to quantify the efficacy of these methods for several stocks in different sectors of the market to identify if regression is a better forecasting method in some cases, and if so, which regression technique outperforms the others. Identifying the conditions for which certain forecasting methods work and those for which they don’t is important, because accurately predicting time series data and understanding when these predictions are invalid can yield financial gain and prevent losses. In addition, a study that seeks to find for which stocks are various forecasting methods more effective and if certain methods are more effective within certain groups of stocks will bring new knowledge to the field. The way in which this well be done is as follows. The stock price data used will be obtained from yahoo finance and additional predictors will be obtained from Old School Value’s database. In order to carry out these regression techniques, the way the prices cluster as a result of the other data is examined first. Using this information, the data inputs with the greatest effect on the price will be found by evaluating the effect of these inputs on the clusters observed. These variables will then serve as the inputs for the regression models.