Estimating underlying energy demand trends using UK annual data

Abstract

Employing the Structural Time Series Model (STSM) approach suggested by Harvey (1997), and based on annual data for the UK from 1967–2002, this paper reiterates the importance of using a stochastic rather than a linear deterministic trend formulation when estimating energy demand models, a practice originally established by Hunt et al. (2003a, 2003b) using quarterly UK data. The findings confirm that important non-linear and stochastic trends are present as a result of technical change and other exogenous factors driving demand, and that a failure to account for these trends will lead to biased estimates of the long-run price and income elasticities. The study also establishes that, provided these effects are allowed for, the estimated long-run elasticities are robust to the different data frequencies used in the modelling.

Notes

A full list of all possible set of restrictions is given in Hunt et al. (2003b).

In addition impulse dummies are included where there is some evidence of non-normality of the auxiliary residuals following Harvey and Koopman (1992) (see Hunt et al., 2003b for details). However, in all cases the inclusion of the dummies has no discernable effect on the estimated elasticities or the shape of the estimated UEDTs.

The activity variables, being GDP for the whole economy and the transportation sector, Real Household Income for the residential sector and the Index of Industrial Output for the manufacturing sector, are all based on 2000.

Given this result, further experimentation is undertaken for the residential sector by starting with the general model with a deterministic trend and testing down accordingly. This produces very similar elasticities to those in Table 1 but the problem of predictive failure persists. Therefore, the model given in Table 1, with the mild stochastic trend, is retained. All additional estimates are available from the authors on request.

Notably, 86% of the housing stock in the UK was built before 1984, with almost half of that stock built before 1940. However, improvements in building regulations and insulation standards, especially since 1994, have considerably reduced energy losses (DTI, 2001, p. 26).

Dargay (1992) amongst others have attempted to estimate possible asymmetries of the price elasticities. The estimated UEDT may in fact be picking up these effects and future research will attempt to combine the UEDT approach with asymmetric price and income elasticities.

This is included since the coefficients on the fourth and fifth lags of the dependent variable are of almost equal size (in absolute terms) but of opposite signs. Therefore, the two variables (e_{t  − 4} and e_{t  − 5}) are replaced by their difference (Δe _{t  − 4}) that is significant.

Diesel engines are approximately 54% more efficient and also the number of diesel engine vehicles has quadrupled during the 1990s (DTI, 2000, p. 21).