Dynamic Bayesian Hierarchical Models for Causal Attribution in Multi-Period Financial Intervention

Abstract

 

This study examines how multi-period financial interventions, specifically repeated central-bank liquidity operations, shaped bank-level lending growth, liquidity stress, market volatility, credit spreads, and non-performing asset movement in the Indian banking sector. The problem is empirically difficult because liquidity interventions rarely operate as isolated one-period events; they are sequenced, repeated, and absorbed differently by banks with different capital strength, ownership structures, liquidity positions, and market exposure. To address this issue, the study uses a bank-month panel covering January 2018 to December 2025, including 42 banks, 4,032 monthly observations, five intervention phases, a 24-month pre-intervention window, and a 12-month post-intervention window. A dynamic Bayesian hierarchical causal-attribution model was estimated in Stan/PyMC to capture time-varying intervention effects, bank-level heterogeneity, posterior uncertainty, and counterfactual response paths. The results show that liquidity interventions reduced bank-level liquidity stress, increased lending growth, lowered market volatility, and reduced credit-spread pressure. The decline in liquidity stress was most pronounced within the first three months, whereas lending growth reached its strongest response at the six-month horizon. Effects were stronger among well-capitalized, large, and private-sector banks. The study contributes a policy-ready causal attribution framework for evaluating when interventions work, how long effects persist, and which banks transmit liquidity support most effectively under uncertainty.