Abstract
This paper develops a pre-registered framework for evaluating a potential intraday structural break in the S&P/TSX Composite without turning an exploratory signal into a premature statistical claim. The protocol fixes the significance threshold, testing windows, boundary search, multiplicity correction, and separate mean and volatility diagnostics before confirmatory analysis. A simulation stress test shows that even a seemingly large volatility shift can remain difficult to confirm once dependence, heavy tails, intraday seasonality, boundary uncertainty, and multiple testing are taken into account. The current S&P/TSX observation therefore remains a research candidate rather than a confirmed regime break.
Research overview
This paper formalizes a pre-registered protocol for testing an exploratory local regime-break candidate in the S&P/TSX Composite. The screening snapshot contains 240 path points, a reported MeanShiftSigma of 0.064, a VolatilityRatio of 3.292, and a 24-candidate screening family. Because the full timestamped path was unavailable at the time of analysis, no empirical p-value or confirmed structural break is reported. The protocol fixes the familywise significance level at 5%, uses 60 validated returns on each side of the candidate boundary, applies a Welch-based mean statistic, separates mean and variance null constructions, and calibrates boundary-specific statistics while preventing nuisance AR estimation across the suspected break. A simulation stress test incorporating AR(1) dependence, U-shaped intraday volatility, Student-t innovations, local boundary scanning, and 24-candidate multiplicity produces an empirical familywise size of 3.98%. Under the same design, a planted variance ratio of 3.292 is detected in only 9.78% of simulations, emphasizing the distinction between screening evidence and confirmatory inference.
Reference identifier: 10.5281/zenodo.22706824