Evaluation of Fama-French asset pricing models with machine learning methods: An application on international market portfolios

The nonlinear dynamic nature of financial markets complicates the analysis and prediction of these markets. Additionally, the opportunities provided by the markets and the risks taken by investors to access these opportunities involve processes that are both necessary and challenging to manage. In this study, which evaluates the dynamic structure of markets from the perspective of investors, the validity of the Fama-French Asset Pricing Three, Four, Five, and Six Factor Models, which have been supported by numerous academic studies in the literature, was tested using machine learning methods to measure their ability to explain market excess returns. The data used in the research were derived from the monthly return performances of companies included in the US Industry Portfolios and International Market Portfolios, as provided by Fama-French, covering the period from July 1, 1992, to January 31, 2022. The US Industry portfolios consisted of portfolio groups of 5, 10, 12, 17, 30, 38, 48, and 49 companies traded on NYSE, AMEX, and NASDAQ. The International Market Portfolios encompassed groups consisting of portfolios from the Asia-Pacific region excluding Japan, Developed Countries excluding the US, Developed Countries excluding Japan, Japan, and North America excluding the US. The analysis conducted using machine learning algorithms employed Support Vector Machines, Linear Regression, k-Nearest Neighbors, and Multilayer Perceptron. As a result of the analysis, Support Vector Machines emerged as the most successful method, achieving the highest correlation value in 25 out of the 60 models included in the study. Linear Regression yielded relatively lower values across all models, but provided the highest correlation value in a total of 24 models. Multilayer Perceptron yielded high values only in the 11 models applied to the US Industry Portfolios. On the other hand, the KNN method did not perform successfully in any of the models and produced weak results. From a financial perspective, it was observed that in the International Market Portfolios, portfolios with small company size achieved the highest correlation value in 26 out of 28 models. Regarding DD/PD, investments, operating profit, and momentum, DD/PD portfolios achieved the highest correlation value in 16 out of 28 models, investment portfolios in 10 models, and portfolios including operating profit in 2 models. However, portfolios including the momentum factor did not yield successful results in any model.