|
Abstract:
|
Previous research over many decades has investigated to what extent seasonal variations in U.S. monthly regional housing starts can be explained using time series models with weather variables as covariates, beyond the fairly regular seasonal patterns captured by models without such covariates. See Pang (2022) for a detailed review and for recent results. Trimbur and Bell (2012) accounted for such variation by adding to conventional seasonal time series models a heteroskedastic seasonal noise term affecting just the winter months. Here we show that both model features – regression effects using weather covariates taken from Pang (2022), as well as seasonal noise terms – are important to the modeling of Northeast and Midwest regional housing starts, i.e., one cannot get by with either one alone. Seasonal adjustment results from such a model are illustrated.
|
