An experiment on controlling a national economy

Abstract

The work described in this paper was carried out as part of the Programme of Research in Econometric Methods (PREM), a programme for investigating whether modern control theory can be applied to improve the control of a national economy. The development of a simple linear econometric model is described which is specifically motivated towards representing the dynamics of the economic system accurately from a control point of view. This is described in terms of four sectors, each sector having up to six dynamic behavioural equations expressed in rational structural form. A method for simultaneously estimating equation parameters within sectors is indicated and seen to result in simplification of equations which provide the basis for the design of optimal controls.

In order to employ known results from optimal control theory the system has first to be transformed into state vector form and techniques for achieving this are indicated. A quadratic criterion function is adopted leading to much simplification in the work. Explanations are given as to how such a quadratic criterion function may be applied in an unobjectionable manner.

To demonstrate the likely performance of controls derived in this fashion from a control model when employed on the actual economy it is proposed to use them to drive one of the large forecasting models in closed-loop fashion. Methods of matching the controls to the large model are described which avoid the necessity for complicated ‘observers’ in deriving the appropriate feedback.

A novel technique is then described whereby automatic modification of the weights attaching to terms of the quadratic performance function can be obtained directly from the modification to outputs desired by a policy-maker. This new method appears to go a long way towards removing rigidities in the use of the quadratic criterion function to which exception has previously been taken. A simple example of its use is given in the paper.