Stochastic Seasonality, Contemporaneous Inference, and Forecasting in the Presence of Extreme Weather

Abstract

Contemporaneous inference from economic data releases for policy and business decisions has become increasingly relevant in the high pace of the information age. The released data are typically filtered to eliminate seasonal patterns to reveal underlying trends and cycles. The nature of economic seasonal behavior is such that average seasonality, not actual seasonality, is filtered from the data. First, the paper suggests adjustments of the inference accounts for the stochastic seasonality. We formalize the issue and present a simple method to the informal inferential practice. Second, we provide a data-based method that allows for temperature adjustment to improve forecasting outcomes. With the assumption of climate change taking place, these methods are particularly important as weather patterns become more extreme.