Asset Pricing Theory And A Comparison Of Machine-learning Techniques
Author: Muhammad Salman Shah

The main rationale of asset pricing theory is to identify the underlying pattern of the drivers and establish their relationship with the financial performance of a firm. The proliferation of hundred of drivers often called the curse of dimensionality in the candidate factor pool is a result of continuous research to achieve higher returns than market. Thus, the fundamental task facing the asset pricing theory today is to bring discipline to the zoo of factors. The leverage to utilize machine learning techniques is enhanced due to their innate ability of handling and extracting valid signals from such complex data structures. The most notable techniques are the neural networks, tree based models, and penalized regressions. The neural networks on the training sample of US market performed the best with the MSE of 3% and hit ratio of 55%. The most prevalent factors include the market capitalization confirming the existence of size anomaly, momentum indicators, and the capital expenditure to sales cash flow ratio among others. The MSE and the hit ratio for the data of Pakistan, where the best performing candidate model is random forest is 0.3% and 84% respectively. Most significant contributors for the data of Pakistan includes momentum, price volatility, and dividend yield etc. The results are encouraging but still warrants further research especially to the formulation of ensembles that may beat the naive equal weighted ensembles. Supervisor: Dr. Mehmood Khalid Co-supervisor: Dr. Hafsa Hina

Meta Data

Supervisor: Mahmood Khalid
Cosupervisor: Hafsa Hina

Related Thesis​