REVERSE STRESS TESTING CORRELATIONS: A REDUCED-FACTOR MODEL FRAMEWORK FOR ANALYZING TRACKING ERROR

Abstract

This research proposes a reduced-factor model framework for performing Reverse Stress Testing (RST) to identify the most probable asset correlation scenario that pushes a portfolio's tracking error beyond a pre-defined threshold. While the existing stress testing framework links asset correlations to economic drivers, automatically including all directly linked drivers leads to high computational time and model complexity. To address this, this study first constructs a "full model" using all directly linked drivers to serve as a baseline. It then constructs a "reduced model" by evaluating all possible combinations of economic drivers for specific target subset sizes and selecting the combination that minimizes the deviation between the stress scenarios identified by the full and reduced models. The proposed framework is empirically tested using data from the Thai equity market. The results show that retaining a higher number of economic drivers in the reduced model does not inherently result in higher accuracy. Furthermore, the optimal reduced model achieves significant operational computational time savings. Finally, the economic drivers that can be safely excluded in this framework cannot be deduced simply by observing the sector weight differences between the managed portfolio and the benchmark.