Forecast Errors and Efficiency in the US Electricity Futures Market

In March 1996 the US introduced electricity futures contracts which provided for delivery at the California/Oregon border (COB) and Palo Verde switchyard. These were followed in July 1998 by contracts which provided for delivery of electrical energy into the Cinergy and Entergy systems, and in March 1999 by a contract for the Pennsylvania/New Jersey/Maryland (PJM) area. This paper employs the forecast error approach to study the informational efficiency of these markets (the model prediction and event studies approaches are not feasible). COB and Palo Verde spot and futures prices are stationary, while Cinergy spot and futures prices are I(1) and cointegrated (Entergy and PJM data were not utilised due to insufficient observations). All forecast errors, therefore, are stationary. Estimation is by instrumental variables in the presence of a lagged dependent variable and overlapping observations. For the period 1996 (04) to 1999 (12), with COB and Palo Verde data, the efficient markets hypothesis (EMH) is rejected. This outcome is confirmed by cointegration of the COB and Palo Verde futures prices. A significant M-GARCH term in the forecast error relationship for both contracts supports the view that a time-varying risk premium was present. A significant negative relationship between forecast errors and volume for the COB contract suggests that agents were still learning the true model driving this market, and that Stein’s ‘Bayesian error’ was non-zero. For the period 1998 (07) to 1999 (12), with COB, Palo Verde and Cinergy data, the EMH cannot be rejected. Finally, an attempt is made to relate this study to the recent electricity crisis in California.