Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
We propose computing HAC covariance matrix estimators based on one-step-ahead forecasting errors. It is shown that this estimator is consistent and has smaller bias than other HAC estimators. Moreover, the tests that rely on this estimator have more accurate sizes without sacrificing its power.