Introduction

Portfolio allocation models assign percentages to asset classes. Clients want to understand: "If markets were different (VIX higher, GDP growth lower), how would allocation change?" Counterfactual explanations—showing how predictions change given hypothetical input changes—provide intuitive explanations of model sensitivity and help clients understand portfolio logic.

Counterfactual Generation

For a given portfolio allocation prediction, generate counterfactuals: alter specific market inputs (e.g., "if VIX were 30 instead of 15"), recompute allocation, report difference. Systematically vary each input to measure sensitivity. Generate plausible counterfactuals (realistic market scenarios) vs arbitrary ones.

Client Communication

Counterfactuals enable intuitive explanation: "In current market (low volatility), model recommends 60% equities. If volatility spiked (VIX 40), model would recommend 40% equities instead." Clients understand the logic and can contextualize allocation to their risk tolerance.

Conclusion

Counterfactual explanations improve client understanding of AI-driven portfolio recommendations, building trust and acceptance.