Robust Portfolio Optimization using GOPALS: Geospatial Optimization and Portfolio Allocation using Landscape Segmentation

Code

• github.com/JatinPatni/GOPALS
  none★ 4

Abstract

Portfolio Optimization is an important area of research in Financial Investments. The traditional mean-variance framework for portfolio optimization is very sensitive to the estimation errors in the expectations of returns and covariances, and leads to suboptimal port folios. Even after using techniques such as shrinkage methods, robust optimization, and resampling methods, we are unable to address these limitations comprehensively. This study introduces GOPALS: Geospatial Optimization and Portfolio Allocation using Landscape Segmentation, a simulation-based portfolio optimization framework designed to overcome these limitations and improve the robustness of the portfolio optimization process. GOPALS methodology relies on simulating market scenarios and identifying ”stable regions” in the high-dimensional portfolio weight space (i.e. regions where portfolio allocations consistently perform favorably across varying market conditions) using clustering methods. By restricting the choice of portfolio allocations from such stable regions, the portfolio performance can be shown to be consistently favorable and robust to estimation errors. GOPALS is applied to a handpicked universe of Indian Equity Mutual Funds, with practical constraints-such as long-only positions and concentration limits-in order to reflect real-world constraints of investing. The robustness of GOPALS is shown via backtests per formed over multiple historical regimes. GOPALS offers a resilient approach to portfolio construction, hence it has considerable promise, not only in academic research, but also in the real world of financial investments. It offers investors and portfolio managers a new tool for navigating uncertain markets across different asset classes and global investment universes.

Tasks

Reproductions