Picking the Newey-West lag for autocorrelated intraday errors

Question

You regress intraday returns on a signal, and market microstructure (bid-ask bounce, order-flow persistence) leaves the residuals autocorrelated over a short but non-trivial horizon.

Explain how the Newey-West lag length is chosen, what happens if it is set too short or too long, and how the choice connects to the residual autocorrelation you observe.

Accepted Answer

What the lag controls. The Newey-West estimator sums weighted residual covariances from lag 1 up to a truncation lag L. That L says how far apart two errors can still be treated as correlated. Beyond L, covariances are assumed negligible and dropped. Choosing it. Two common approaches: - Rule of thumb. A widely used automatic choice is L \approx 4 (T/100)^{2/9}, where T is the sample size, rounded up. It grows slowly with the sample. - Data-driven. Look at the residual autocorrelation function and set L past the lag where the autocorrelations fade into the noise band. If microstructure effect…

Picking the Newey-West lag for autocorrelated intraday errors

Your answer

Discussion

Learn the concepts

Related questions