ACV – package for optimal out-of-sample forecast evaluation andtesting under stationarity
Package ACV (short for Affine Cross-Validation) offers an improvedtime-series cross-validation loss estimator which utilizes bothin-sample and out-of-sample forecasting performance via a carefullyconstructed affine weighting scheme. Under the assumption ofstationarity, the estimator can be shown to be the best linear unbiasedestimator of the out-of-sample loss. Besides that, the package alsooffers improved versions of Diebold-Mariano and Ibragimov-Muller testsof equal predictive ability which deliver more power relative to theirconventional counterparts. For more information, see the accompanyingarticle “Optimal Out-of-Sample Forecast Evaluation Under Stationarity”by Filip Staněk.