Classification research / edition 2026.10
Evaluate conditional behaviour
Examine sample size, exposures, costs and testing choices before attributing strategy results to a regime.
Conditional Returns and the Sample Behind Them
Read regime-conditioned return summaries without confusing a large row count, a favorable average, or repeated exposure to one episode with independent evidence.
Building an Auditable Regime-Conditioned Backtest
Separate retrospective state descriptions from a chronological test of decisions, with explicit vintages, execution timing, training boundaries, and comparison rules.
Exposure Effects Are Not Automatically Regime Alpha
Distinguish changing market participation, benchmark-relative returns, and model-dependent alpha before attributing performance to regime information.
Turnover and the Cost of Changing States
Translate label changes into actual allocation changes, define turnover consistently, and keep invented cost assumptions separate from measured implementation evidence.
Benchmark and Universe Changes in Regime Research
Keep comparator changes, constituent eligibility, and return-series revisions visible so that a new research definition does not masquerade as new performance.