|
Article Excerpt
Prompted by the recent surge in light oil product consumption, this paper analyses the demand for non-commercial motor fuel and proposes a longrun forecasting model. In doing so, our aim is to be able to reproduce a few key stylized facts observed in secular evolutions of the motor fuel intensity of GDP and related notably to the derived nature of oil demand. Using a database covering 77 countries over the 1986-1998 period, we explain sequentially the stock of private vehicles per capita and fuel consumption per vehicle. The former is expressed as an S-shaped function of real per-capita income, which takes into account the dynamics specific to the dissemination of a durable good in a population. By explicitly considering the distinct phases of the development of the automobile market, our approach enables us to propose an explanation to the space-time variability in long-run income elasticities reported in the literature - especially its decline as per-capita income increases and the resulting gap between elasticities in emerging countries compared to developed countries. Our two-equation model also enables us to reproduce the "bell" shaped curve of the motor fuel intensity of GDP as a function of per-capita income, as well as the other principal properties of resource intensity-of-use linked to the process of dematerialization which, for any country, follows the industrialization period.
1. INTRODUCTION
The first two oil shocks led to a rationalization of energy consumption, and oil is increasingly used in the transport sector. Its use is growing steadily, propelled by the greater need for mobility in the developed countries, and the progressive dissemination in the emerging countries of a model of society inspired by the western paradigm.
In this context, the modeling of motor fuel demand has been the subject of many studies, aimed at identifying its fundamentals. Economists mainly attempt to evaluate demand elasticities in relation to price and income. Many models have been proposed for this purpose. Most of them use log-linear specifications but various models arise depending on the kind of data used (time series, cross-section or pooled) and the type of equation estimated (static or dynamic). Regardless of the approach taken, these models imply a constancy of the elasticities estimated.
However, it is well-known that the relationship between GDP and energy or oil product demand evolves as the economy passes through successive stages of development and faces various needs in terms of industrial products, mobility and households appliances. Moreover, energy demand is mostly a derived demand and the quantity consumed therefore is strongly conditioned by the stock of energyconsuming durable goods, whose diffusion ultimately reaches a saturation level. Thus the energy and motor fuel intensities of GDP first increase and then decrease as real per-capita income rises.
According to Galli (1998), any long-run forecasting model of motor fuel demand should account for such a turning point and allow the income elasticity to vary with income. Otherwise its results could present a significant misspecification bias.
At a rather aggregated level, those concerned with general tendencies of fuel demand can deal with this problem using a specification that lets the energy intensity evolve non-monotonically like the log-quadratic specification used by Galli or Medlock and Soligo (2001).
Another approach, with a stronger explanatory power, would be to consider the problem as one of omitted variables. One particular feature of the demand for gasoline is its derived nature and its close link to the stock of durable goods that uses it, i.e. the stock of automobiles. Consequently, many empirical studies of fuel demand introduce the stock of vehicles in use as an explanatory variable, which should lead to more efficient estimates. However, most of them consider the vehicle stock as exogenous (e.g. McRae, 1994). Thus the income elasticities estimated correspond only to partial elasticities because they fail to evaluate the influence of income variations that takes place through the variations in vehicle ownership. Other studies explain the vehicle stock with a log-log specification (e.g. Drollas, 1984) and do not enable the total income elasticity to vary, which is not consistent with various contributions in the literature. Indeed, a large body of research has been done to explain the vehicle stock and model its diffusion process. It has become common to use a nonlinear S-shaped relationship to express vehicle ownership rates as a function of real per-capita income (Greenman, 1996, Dargay and Gately, 1997, 1999). This indirect influence is a major source of variability for the income elasticity of gasoline demand.
Using time series cross-section data describing annual evolutions in 77 countries over the 1986-1998 period, we estimate sequentially a two-equation system which enables us to model explicitly the nonlinear relationship between motor fuel consumption and real income. Inspired by previous work, we break down per-capita fuel consumption into two terms: the stock of private vehicles per-capita and consumption per private vehicle. The first equation explains passenger car ownership as an S-shaped function of the real per-capita income and demographic factors. The second is a panel equation expressing the fuel consumption per vehicle as a function of real per-capita income, the real price of gasoline and the car ownership rate. We use the estimated parameters of our model to derive out-of-sample properties of the relationships between the real per-capita income and both the income elasticity of motor fuel demand and the GDP motor fuel intensity. We also use past out-of-sample data to asses the ability of our model to forecast trends in the car ownership rate. This enables us to evaluate our model against some alternatives by comparing how they are able to reproduce long-run tendencies. We interpret the fact that our two-equation system performs better than a number of other specifications to explain secular evolutions highlighted by many studies as evidence that it is better suited to longrun analyses and forecasts. In particular, considering energy demand, we are able to replicate 1) the decrease of the long-run elasticity of motor fuel demand with respect to per-capita income as per-capita income rises, 2) the bell-shaped pattern of the motor fuel intensity of GDP as a function of per-capita income, 3) the stability of the level of per-capita GDP at which the peak of intensity is reached and 4) the lowering of this peak as time passes. As for the car ownership rate, we can reproduce the long-run trends, but the short-run adjustment could be improved for developed countries.
In the next part, we comment briefly on the long-run influence of economic development on gasoline consumption through the diffusion of the automobile in the population. We then present, in the third section, our two-equation econometric model. The empirical results are reported in the fourth section as well as the long-run evolutions of the income elasticities of motor fuel demand and the motor fuel intensity of GDP implied by the parameter estimates. In the fifth section, we compare our results with those of previous studies and perform out-of-sample forecasts with past data in order to benchmark our model. Finally,...
|
