15.062 Term Project: Exploring and Creating Asset Pricing Models

Hannah Chiou, May 2023

Created for MIT Sloan Management Course 15.062 Data Mining

Executive Summary

The goal of every investor is to “play the market” — to actively trade and sell stocks and assets in the market in the hopes of making a profit. Investors generally hope to get returns that are higher than what they would expect, given the generally accepted risk factors, such as interest rate risk. In this project, I explored four different well-known asset pricing models and their ability to accurately predict abnormal returns (that is, returns that are higher or lower than expected). I use the Boeing Company daily rate of return from January 2 1980 to August 11 2022 (with rate of return being calculated as ln(Adjusted Close Pricet / Adjusted Close Pricet-1). The Adjusted Close Price has already been adjusted for the impact of time, so no time-series methods had to be applied. The possible predictors were the six variables included in the four original models. After exploring the linear regression models, I use various data mining methods (lasso regression, random forest feature importance selection, and neural nets) as alternative ways to predict stock performance, and compare the performance of these models created from these techniques against the four pre-existing models. To compare each of these models, I applied k-fold cross-validation to see which model yielded the most accurate prediction for Boeing stock’s rate of return. The Fama-French five-factor model was found to be the best model.

For the full report, please see this link