958. Data-Driven Approach to Robust Portfolio Optimization Using Deep Neural Networks
Invited abstract in session MB-31: Machine Learning for Optimization under uncertainty 1, stream Stochastic and Robust optimization.
Monday, 10:30-12:00Room: Maurice Keyworth 1.06
Authors (first author is the speaker)
| 1. | Divyanee Garg
|
| Mathematics, Indian Institute of Technology, Delhi | |
| 2. | Aparna Mehra
|
| Department of Mathematics, Indian Institute of Technology Delhi |
Abstract
The construction of optimal portfolios under uncertainty, stemming from data imperfections and market volatility, has received significant attention from scholars and practitioners. Traditional portfolio optimization (PO) models inadequately account for these uncertainties and often produce suboptimal solutions. This study integrates deep learning techniques with robust optimization to enhance portfolio performance and improve decision-making under uncertainty. We propose a robust PO model with deviation expectile value at risk (DEVaR) considering the uncertainty in expected asset returns. DEVaR is selected as the risk measure due to its growing prominence in recent literature. It is attributed to its theoretical strengths of coherence and elicitability and its computational efficiency enabled by its linear formulation. This article uses a deep neural network to learn hidden structures from data, leading to a non-convex uncertainty set. The trained neural network is integrated into a robust PO model by formulating the adversarial problem as a mixed integer linear program. We find the robust portfolio through the iterative constraint generalization algorithm. The effectiveness of the proposed model is demonstrated by constructing optimal portfolios on eight global stock indices. The out-of-sample statistics obtained from the proposed model are compared with techniques available in the literature. The proposed model outperforms all other models across most datasets.
Keywords
- Robust Optimization
- Machine Learning
- Risk Analysis and Management
Status: accepted
Back to the list of papers