The effects of moderate levels of serial correlation on one-sided and ordinary cross-validation in the context of local linear and kernel smoothing is investigated. It is shown both theoretically and by simulation that one-sided cross-validation is much less adversely affected by correlation than is ordinary cross-validation. The former method is a reliable means of window width selection in the presence of moderate levels of serial correlation, while the latter is not. It is also shown that ordinary cross-validation is less robust to correlation when applied to Gasser-Müller kernel estimators than to local linear ones. © 2003 Elsevier Inc. All rights reserved.
Hart, J. D., & Lee, C. L. (2005). Robustness of one-sided cross-validation to autocorrelation. Journal of Multivariate Analysis, 92(1), 77–96. https://doi.org/10.1016/j.jmva.2003.08.005