The billing cycle and weather variables in models of electricity sales

Abstract

Because utilities bill their residential and commercial customers by cycle on each working day of the month, the calculation of weather variables to associate with monthly sales data is complicated. We examined three different methods of calculating weather variables. 1. (1) For a utility that bills monthly, the most appropriate method is to calculate daily weather measures, then take a weighted sum of these daily measures over the current and previous month, with the weights for each day being proportional to the number of customers whose consumption on that day is billed in the current month. When weather variables are calculated in this way, accurate econometric models of electricity sales can be estimated. 2. (2) If data on the number of customers in each cycle are unavailable, the first procedure can be applied under an assumption concerning the number of customers consuming on each day. For the three utilities in the study, using these approximate weights reduced the model accuracy noticeably but not substantially, implying: if data on the number of customers in each cycle can be retrieved, the effort expended in doing so will be rewarded with more accurate models; however, if such data are impossible to obtain, fairly accurate models can still be estimated. 3. (3) The easiest method for calculating weather variables is to ignore the billing cycle phenomenon and take an unweighted sum of daily weather measures over days in the previous or current, or both, months. Our estimation results indicate that these simple measures decrease the accuracy of the models substantially, implying that the additional effort required to calculate weather variables that reflect the billing cycle phenomenon is clearly worthwhile in terms of increased model accuracy.