Angela-Frances Osehiemen Ibhade, a Master of Science in Statistics candidate, will defend the thesis titled Markov-Switching Models: Applications in Credit Risk Analysis on Wednesday, September 2, 2026 from 1:00 PM– 3:30 PM in Plaza 600F.
The examination committee includes Yifeng Li, Chair; Walid Ben Omrane (Goodman School of Business), External Examiner; William Marshall, Supervisor; Xiaojian Xu and Jan Vrbik, Supervisory Committee Members.
Abstract:
This thesis investigates the macroeconomic determinants of Canadian mortgage delinquency using a progression of time series models, culminating in a proposed mixed-lag Markov-switching vector autoregressive (MS-VAR) framework that allows autoregressive lag structure to vary across economic regimes. The analysis draws on 120 quarterly observations (1995 Q2-2024 Q4) covering mortgage delinquency, unemployment, GDP growth, inflation, and household debt servicing ratios.
A vector error correction (VEC) model identifies four cointegrating relationships among the five variables, with significant error correction terms confirming that delinquency rates adjust toward a long-run macroeconomic equilibrium. A standard two-regime MS-VAR identifies a persistent baseline state and a transient stress state aligning with the 2008-2009 financial crisis and the COVID-19 pandemic of 2020, but near-zero stress-regime variance estimates reveal a degeneracy driven by the rarity of stress episodes in the sample. A proposed mixed-lag extension, allowing regime-specific lag orders via BIC grid search, selects an economically intuitive asymmetric structure but exhibits the same degeneracy, indicating the instability stems from limited stress-period data rather than lag misspecification alone. The VEC framework is therefore recommended as the preferred specification, producing interpretable dynamics and stable forecasts over a 50-quarter horizon. Both the standard and mixed-lag MS-VAR frameworks are implemented in the MSMVAR R package,1 publicly released as open-source software on GitHub, providing a foundation for future applications with larger samples or clearer regime separation.
Keywords: Markov-switching VAR, mixed-lag models, mortgage delinquency, cointegration, EM algorithm, Canadian credit risk, regime-switching, BIC model selection, open-source software
