1: \begin{abstract}

2: {\footnotesize{}We consider the derivation of data-dependent simultaneous

3: bandwidths for double kernel heteroskedasticity and autocorrelation

4: consistent (DK-HAC) estimators. In addition to the usual smoothing

5: over lagged autocovariances for classical HAC estimators, the DK-HAC

6: estimator also applies smoothing over the time direction. We obtain

7: the optimal bandwidths that jointly minimize the global asymptotic

8: MSE criterion and discuss the trade-off between bias and variance

9: with respect to smoothing over lagged autocovariances and over time.

10: Unlike the MSE results of \citet{andrews:91}, we establish how nonstationarity

11: affects the bias-variance trade-off. We use the plug-in approach to

12: construct data-dependent bandwidths for the DK-HAC estimators and

13: compare them with the DK-HAC estimators from \citet{casini_hac} that

14: use data-dependent bandwidths obtained from a sequential MSE criterion.

15: The former performs better in terms of size control, especially with

16: stationary and close to stationary data. Finally, we consider long-run

17: variance estimation under the assumption that the series is a function

18: of a nonparametric estimator rather than of a semiparametric estimator

19: that enjoys the usual $\sqrt{T}$ rate of convergence. Thus, we also

20: establish the validity of consistent long-run variance estimation

21: in nonparametric parameter estimation settings.}{\footnotesize\par}

22: \end{abstract}