Forecasting New Jersey's Electricity Demand Using Auto-Regressive Models

Abstract

Forecasting is an important tool in planning and policy making. Electricity load forecasting is necessary for power systems planning, efficient dispatching of electricity in the grid and to forecast other macro-economic trends. This paper summarizes and presents auto-regressive techniques/processes as a practical tool in forecasting electricity demand. This paper attempts to model the long-term electricity demand for New Jersey using three different auto-regression models: ARMAX (autoregressive moving average with exogenous variables) model, Vector auto-regressions (VAR) and Bayesian VAR (BVAR). The application of VAR/BVAR to electricity demand forecasting is relatively new and untested. The forecasting performance of each model is assessed using different forecast error metrics. For the given case study, the VAR model produced the best forecast.

Keywords: electricity demand, Vector Autoregressions (VAR), Bayesian Autoregressions (BVAR)

JEL Classification: Q41, Q43