Monday 22 December 2008

How can theory and empirics best serve policy?

Economics is a subject which aims for real world application. Clarifying the relation between theoretical analysis, empirical analysis, and policy may help to produce more useful research. Of interest here will be how understanding improves control of or response to economic events, and what the relative roles and potential are for theory and empirics.

We borrow the graphical representation from Classification and Regression Tree (CART) models to present a model for a real world relation between two economic quantities. Two nodes represent each of the quantities, and all their possible causal interactions are represented by paths connecting them. By interaction we mean a mechanism of covariation that is readily conceptually distinguished from other mechanisms. For example, if the first node represents domestic technology usage and the second node represents foreign technology usage, then one path may represent change in technology due to foreign direct investment and a second path may represent change in technology due to trade exposure. The paths may be split into sub-paths, with sub-nodes breaking up the path and representing quantities which causally vary as one of the end node quantities varies and whose variation leads to variation in the other end node quantity. Thus, the sub-paths represent chains of causality connecting the two variables. In the technology spread example, the sub-nodes along the foreign direct investment path may be {transfer to other companies in the economy beyond the immediate recipient of technology} then {skilled usage by all domestic companies} then {profitability of the operation} then {technology usage throughout the economy}. Further sub-nodes may be introduced if the mechanism of transfer is further deconstructed, such as introducing {the movement of people between foreign direct investing companies and the rest of the economy} as a sub-node between {the original transfer} and {the spread to other companies}.

The last three steps in the FDI path (skilled usage, profitability, and wide technology usage) are shared with the other path, capturing the effect of technology exposure through trade. Thus we have a graph showing the modelled interactions:



The upper path represents the FDI route for technology transfer. The lower path represents the trade path, and is less fully modelled.

In analysing such modelled graphs, a theory proposes the existence of a new path between two nodes, or the introduction of new nodes in an existing path. In the proposed technology graph, suggesting that licensing is also a means of technology transfer is a theory. Another theory is that trade exposure’s action on growth requires foreign importers to demand high production standards and a domestic knowledge base to adapt to the standards, introducing two new nodes in the lower path. The theories here would be more properly called hypotheses since they do not have to be correct, in which case the nodes would have no real world connection and the node quantity covariation is zero.

An empirical study tests the relation between quantities on two nodes, and may be informed by and test for the graph connections suggested by theory. It may test the relative strengths of two different paths between the nodes. If the empirical study tests the relation between quantities on two nodes separated by other sub-nodes, then an empirical test controlling for the sub-node quantities is removing the effect of the first node quantity acting along the sub-node paths, so that any robust remaining effect indicates the presence of other sub-paths between the two nodes.

Policy usually looks to control the variation in one node’s quantity in response to changes in other node quantities. It will often be indifferent to the precise paths taken from the control quantities to the response quantities. In the technology graph, a policymaker may wish to maximise the use of a technology in their economy by transferring it from abroad, and will use any means necessary to obtain the maximum throughput in the graph. Theory sometimes returns at this stage to specify decision rules by which the policymaker, having precisely stated outcome preferences, optimises the changes in control quantities. The decision rule is required for optimisation if all paths cannot simultaneously operate with their maximum response. In the technology graph, for example, increases in intellectual property rights protection may raise the level of investment but lower the ability of local companies to copy the technology.

Under these definitions, we can propose some best possible outcomes for policy from theoretical and empirical analysis. Theory could find new paths for a modelled situation and emphasise the most important ones. Empirics could show overall node covariation along multiple paths, show covariation along specified paths, compare covariation along different paths, and demonstrate how path covariation changes in response to path choices in the whole graph. Theory could also connect the entire analysis with policy by presenting means of exploiting the paths and proposing decision rules for changing node quantities that reflect policymaker objectives. A decision rule may be specific to the objective and not readily adaptable to alternative objectives, which in a competitive setting – for example in competition for the receipt of a particular technology – may give the policymaker an advantage in achieving their objectives over rivals with alternative objectives, at least until they can devise their own decision rule and compete on the basis of economic fundamentals.

No comments: