numeraire.core.stats.clark_west_test#

numeraire.core.stats.clark_west_test(realized: NDArray[float64], forecast: NDArray[float64], benchmark: NDArray[float64], *, nw_lags: int = 0) → ClarkWestResult[source]#

Clark-West (2007) MSPE-adjusted test for nested models.

Per-period adjusted loss difference f_t = e_b^2 - (e_m^2 - (bench - model)^2); the t-statistic of its mean (HAC long-run variance with nw_lags; use horizon - 1 for multi-step forecasts) is compared to one-sided standard-normal critical values. Degenerate when the model equals the benchmark (t = nan).