An introduction to entropy estimation of parameters in economic models

The results from quantitative economic modelling are highly dependent on the parameter values that are adopted. A common practice is to use elasticity values from previous modelling which in some cases are simply assumed or stylised values or can be traced back to some econometric study. While borrowing elasticities is a sensible starting point in any modelling exercise, users are left in doubt as to whether the elasticities from other times and places, that use different aggregations or are based on longer or shorter periods of adjustment are applicable for the current exercise. It therefore puts the robustness of results in doubt. Furthermore, using conventional econometric methods to estimate parameters is not an option when data are limited, as is often the case with the economic variables required for CGE models. Entropy estimation, developed by Golan, Judge and Miller (1996), is an approach that allows economic modellers to use data to improve the assumptions they make about parameters in economic models. It works by using prior information — a combination of prior beliefs, educated guesses and theoretical constraints — and limited data to inform estimates. Importantly, entropy estimation places more weight on the data (and less on the priors) as the number of observations increase. A further attraction is that the resulting entropy parameter estimates must satisfy the underlying economic model equations since those equations are constraints in the entropy estimation. The purpose of this short paper is to provide an introductory guide to entropy estimation for economic modellers with a particular emphasis on estimating elasticities from limited time series. The objective is to provide all the information that researchers need (how it works, the importance of the assumptions and when and how it should be used) to be able to use the technique confidently.