Comparative Asset Pricing Models Using Different Machine Learning Tools

Abstract

AbstractThe study tries to estimate the returns using three known models, namely the single-index model, capital asset pricing model (CAPM), and arbitrage pricing theory (APT) model. The data for the study was taken from Nifty. The top 100 large-cap stocks were taken into account. The Nifty index was taken as the proxy for the market return. All data was taken with one period lag to make it stationary. These models are subjected to three software-based machine learning, namely decision tree, neural network, and ordinary least square. The results were iterated from the three best returns, and the area of the curve was estimated. The academic point score for the area was taken into consideration. The value 0.8 was taken as the cutoff for the acceptance of the model. The result shows that CAPM outperforms all the models used under different machine learning mechanisms. In the process, it was observed that the arbitrage pricing theory model too performs well. However, while we use the ordinary least square, the single-index model has the best performance. It is kind of a foregone conclusion since the single-index model is based on the return of stock and is close to paper trading.KeywordsSingle-index modelCapital asset pricing modelArbitrage pricing theoryDecision treeNeural networkOrdinary least square