Predicting the unpredictable: a counterfactual causal inference framework for financial market collapse during black swan events

Traditional financial models are often put to test during complex financial crisis, creating a need for the decision-makers for tools that go beyond prediction to provide causal explanations and generate counterfactual insights. This study develops a counterfactual causal inference framework for crisis analysis by using high-frequency daily data during two ‘rare, extreme, and unpredictable’ Black Swan events–the Global Financial Crisis (GFC) and the COVID-19 crash. This study models S&P 500 daily log-returns as functions of changes in implied volatility (CBOE VIX) and the 10-year U.S Treasury Yields, along with credit-spread, funding-stress, MIDAS-aggregated macroeconomic, and global-volatility-factor controls to generate factual and counterfactual trajectories. The results show that volatility shock was the dominant transmission channel during both the crisis (contributing –13.4 percentage points of the cumulative crash in COVID and −7.9 in GFC), while yield shock only acted as a secondary predictor (−1.9 and −3.5 percentage points respectively). The yield-channel coefficient is 3.11 × larger in the high-volatility regime in COVID and 4.76 × larger in GFC, indicating economically meaningful state-dependence. In both crises, the actual crash factual-counterfactual gap is outside the whole range of fifty randomly selected non-crisis placebo windows (empirical p < 0.001). The resilience to theory-informed regularization is demonstrated by the dYield coefficient changing by less than 6 % following Heston-augmented training. The study demonstrates that counterfactual reasoning can generate an interpretable causal narrative in complex systems that can be used to develop new learning systems.