Selecting time-series hyperparameters with the artificial jackknife
This work addresses hyperparameter tuning in time series analysis, which is incremental as it adapts an existing jackknife method to handle temporal dependencies.
The authors tackled hyperparameter selection for time series by proposing an artificial delete-d jackknife method that replaces data removal with artificial missing values, showing finite-sample advantages in simulations and applying it to regulate forecasting models for foreign exchange rates.
This article proposes a generalisation of the delete-$d$ jackknife to solve hyperparameter selection problems for time series. I call it artificial delete-$d$ jackknife to stress that this approach substitutes the classic removal step with a fictitious deletion, wherein observed datapoints are replaced with artificial missing values. This procedure keeps the data order intact and allows plain compatibility with time series. This manuscript justifies the use of this approach asymptotically and shows its finite-sample advantages through simulation studies. Besides, this article describes its real-world advantages by regulating forecasting models for foreign exchange rates.