Abstract
Abstract
Financial return dynamics are inherently unstable, reflecting both recurring latent market regimes and structural changes in the underlying data-generating process. This study examines whether causal structural-break information improves regime-dependent financial forecasting across heterogeneous Asia-Pacific equity markets. The empirical analysis covers five major indices: the S&P BSE SENSEX, Nikkei 225, Hang Seng, KOSPI, and S&P/ASX 200. The framework combines retrospective Gaussian quasi-maximum-likelihood multiple-break segmentation, Bayesian Online Change-Point Detection (BOCPD), and a Markov-switching AR(1) model with regime-dependent parameters. The number of regimes is selected separately for each market using only pre-out-of-sample information, after which performance is evaluated through a strict 1,000-observation one-step-ahead walk-forward experiment against historical-mean, AR, ARMA, EWMA, GARCH, and conventional Markov-switching benchmarks. Controlled simulation with known regimes and break locations provides additional identification validation. The results reveal substantial cross-market heterogeneity in both structural-break evidence and selected regime structure. Break-aware adaptation does not consistently improve conditional-return or average-volatility forecasts, while conventional Markov switching without memory resetting remains highly competitive and, in some markets, significantly superior under QLIKE loss. At the 5% Value-at-Risk level, break-aware forecasts provide acceptable or improved calibration in several markets, although the benefit does not generalize uniformly across all indices or tail levels. Overall, the evidence indicates that structural-break information is market- and forecast-target dependent, offering selective value for adaptive downside-risk calibration rather than universal improvements in financial forecasting.
JEL classification C22; C53; C58; G17