Estimate industry cost of equity from machine learning : Master of Business Studies (Finance), Massey University
Loading...
Date
2025
DOI
Open Access Location
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Massey University
Rights
The author
Abstract
This thesis aims to contribute to the existing literature in several specific ways. To the best of our research, there has been relatively limited research applying advanced machine learning (ML) techniques specifically to the estimation of industry-level cost of equity (ICoE). While existing studies have generally focused on firm-level return forecasting or relied heavily on traditional asset pricing models such as CAPM, FF3, or FF5, our study extends this scope by systematically exploring how ML methods perform in the broader context of industry-level equity cost estimation. Through this investigation, we provide empirical insights that help clarify the robustness and applicability of ML techniques at the industry rather than the firm level.
We propose a hybrid modelling framework designed to integrate factors selected by machine learning methods into the traditional FF5 asset pricing framework. Specifically, we utilize the extensive JKP factor library, which comprises 153 factors categorized into 13 economic themes, to investigate whether ML-selected factors can provide incremental explanatory power relative to the original, fixed FF5 factors. By adopting this approach, we seek not only to enhance theoretical understandings of factor models but also to offer practical improvements that investors and financial analysts could potentially adopt in their valuation processes.
We conduct a detailed comparative analysis between traditional models (especially FF5) and selected ML methods, such as LASSO, Gradient Boosting Machines (GBM), and Light Gradient Boosting Machines (Light GBM). We evaluate the predictive accuracy and stability of these methodologies using a range of commonly accepted error metrics, including Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Adjusted R². This structured comparison may illustrate the conditions and industry features. Under these conditions and features, ML methods might outperform traditional asset pricing models. To this extent, it may provide useful practical insights for both academic researchers and financial practitioners.
Finally, we attempt to bridge existing gaps between traditional asset pricing theories and modern machine learning methodologies with our empirical framework. We validated the potentiality of integrating ML-selected factors into traditional models, and provide a basic approach for future researchers and practitioners interested in improving industry-level equity cost estimation. Overall, our research seeks to provide a novel perspective about factor selection in asset pricing models. Also we seek to suggest methodological improvements that could enhance investment decision-making under complex and volatile market conditions.