Analysis of HSVAR identification and inference with an unknown break date

Analyze the consequences of an unknown break date in heteroskedastic Structural Vector Autoregressions that rely on a variance break for identification, and develop methods to conduct identification and inference for structural parameters and impulse responses when the break date is not exogenously given.

Background

The proposed methodology assumes a known (exogenously determined) break date separating two volatility regimes, a common assumption in the heteroskedastic SVAR literature that simplifies identification and estimation. The paper’s identification theory, algorithms, and empirical illustration all rely on this known break.

In many applications the timing of volatility breaks is uncertain. Treating the break date as unknown raises new challenges for identification through heteroskedasticity, for constructing admissible rotations under zero/sign restrictions, and for robust Bayesian inference. Addressing these issues requires adapting both the theoretical identification analysis and the computational algorithms.