A (conservative) test is applied to investigate the optimal lag structure for modeling realized volatility dynamics. The testing
procedure relies on the recent theoretical results that show the ability of the adaptive least absolute shrinkage and selection operator (adaptive lasso) to combine efficient parameter estimation, variable selection, and valid inference for time series processes.
In an application to several constituents of the S&P 500 index it is shown that (i) the optimal significant lag structure is time-varying and subject to drastic regime shifts that seem to happen across assets simultaneously; (ii) in many cases the relevant information for prediction is included in the first 22 lags, corroborating previous results concerning the accuracy and the difficulty of outperforming out-of-sample the heterogeneous autoregressive (HAR) model; and (iii) some common features of the optimal lag structure can
be identified across assets belonging to the same market segment or showing a similar beta with respect to the market index.
