November 2007

The article, Using base-year data with empirical scenario models, was published in the August edition Journal of Technological Forecasting and Social Change.

Dr. Eric Kemp-Benedict, whose research focuses on cross-disciplinary policy analysis for sustainable development strategies, provides in his paper a theoretical rule for estimating residuals in scenarios.

This rule can be implemented in a straightforward and reproducible way, Kemp-Benedict says.

Filling the gap between base-year data and model
In his article, Kemp-Benedict makes no assumption about the source, use or design of scenario models, only that a scenario model is being developed to represent variables for which there are base-year data.

He presents specific examples used to illustrate the method employing linear regression models.

- By making use of an empirical formula we are often forced to come up with a solution to what to do with the gap between the obsverved base-year values and the model – the residuals or shift parameters.

These are known to be zero in the base year, raising the suspicion that they should not be set to zero in scenario years. Despite this, there have been no standard rules for how the residuals should evolve.

"This is an unglamorous problem, but it is also unavoidable," Kemp-Benedict says.

Using base-year residuals to estimate scenario residuals
His paper presents a solution to this technical problem. By developing a quantitative scenario for a multi-country region accompanied by an empirical model, Kemp-Benedict shows how known base-year residuals can be used to estimate residuals in the given scenario.

Furthermore, by adding the estimated residual to the empirical model, the resulting estimate will typically outperform the unadjusted empirical model as measured by the mean squared error (MSE) of the estimate.

While the estimate for a particular country may not improve by adding the estimated residual, the average performance will typically improve.