Timing or Aggregation? Weekly Energy Prices and Inflation Nowcasts

Abstract

Weekly energy prices provide timely information about monthly inflation, but their usefulness need not depend on estimating a separate coefficient for each week. We assess the value of this flexibility for United States Consumer Price Index (CPI) energy inflation using weekly West Texas Intermediate (WTI) crude-oil and retail gasoline prices over 2005-2026. Matched comparisons hold the weekly observations, autoregressive moving-average (ARMA) errors, estimation window, and target history fixed. A separate applied comparison evaluates monthly-average regressors that use every available endpoint. Free coefficients on target-month WTI changes give no reliable accuracy gain over their block mean, although moderate gains remain unresolved. Monthly-average retail gasoline reduces nowcast root mean squared error by 43-60% relative to the ARMA benchmark in the main windows and schemes; adding WTI gives no reliable gain. Free gasoline coefficients improve on matched block means by 14.8% in contemporaneous comparisons but perform worse than the monthly-average regressor. With predictors lagged one month, the matched gain is an imprecise 2.7%; the larger 16.7% gain against the monthly average also changes endpoint coverage. Thus the predictor and the comparison benchmark both matter when assessing weekly flexibility. The retrospective evidence supports monthly averaging as a parsimonious benchmark, with prospective and historical release-time validation still needed.

Dae-Jin Lee
Dae-Jin Lee
Assistant Professor in Statistical Sciences

P-splines, mixed models, computational statistics, flexible survival, and latent memory for athlete monitoring.