Projecting Monthly Natural Gas Sales for Space Heating Using a Monthly Updated Model and Degree-days from Monthly Outlooks

Abstract

The problem of projecting monthly residential natural gas sales and evaluating interannual changes in demand is investigated using a linear regression model adjusted monthly. with lagged monthly heating degree-days as the independent variable. The relationship between sales and degree-day data for customers of Columbia Gas Company (serving the Columbus, Ohio, area) is studied for a 20-yr period ending in June 1990. Analysis of the phases of the monthly billed sales and the degree-day data indicated that monthly sales reports lagged degree-days and gas consumption by 15 days on average. Running 12-month regressions of Columbia Gas sales on 15-day-lagged degree-days show that lagged degree-days explain, on average, 97% of the variability in the monthly sales reports for the study years. Annualized trends in the regression coefficients indicate changes in consumption due to conservation and changes in price. Since 1974?75 the trends indicate declines of 50% in non-weather- sensitive sales per customer, and 35% in monthly sales per degree-day per customer, with most of the changes occurring prior to 1985. The mode is adapted by using a regression equation based on historical data through the prior 12 months with degree-days as the independent variable. Estimates for sales in the coming period are based on official National Oceanic and Atmospheric Administration (NOAA) monthly temperature outlooks (outlooks) for the Columbus region. For comparison purposes, four lagged monthly degree-day sets are used in a model: 1) a set of degree-day normals, 2) a set of 100% projected degree-day values obtained by use of NOAA outlooks, 3) a set in which the first half of the degree-days in each monthly period are observations and the second half are projected, and 4) a set that is 100% observed (the perfect case). The skill of the degree-day sets for projecting monthly sales is evaluated by a statistical analysis of the projection errors (differences between projected and reported sales). Errors from the sales projection models using the four different degree-day sets are compared with errors from two sets of baseline sales. The first set of baseline sales is estimated with and the second set without foreknowledge of monthly sales norms and annual total sales. The models using partially and fully projected degree-days are found to have measurable skill over models using climatology in projecting monthly gas sales during the heating season.