[ Agenda | Sessions | Program ]

Toward a Generic Macroeconomic Modeling Environment

Stephen Wright - Union Bank of Switzerland


The rapid development of computer technology is bringing increasingly powerful capabilities within reach. More complex systems can be modeled more quickly, and developments in graphical user interfaces allow the user to interact with this simulation in more intuitive ways. On their own however, developments in computing technology are not enough to achieve the full potential benefit.

There are three computational phases to model building, each with its own critical time lag.

  1. Learning to use new modeling software
  2. The process of building the model
  3. Using the model
At each stage, the more domain specific the structure that can be built into the software, the easier it is to learn to use, the faster models can be built and tested, and the more highly optimized is the final run time code.

This paper briefly surveys some of the ways that domain specific knowledge has been used in commercially available economic modeling packages to date, then describes an innovative approach to building generic multi sectoral or multi regional macroeconomic models.

This approach specialty exploits an Input/Output structure to represent all the system states in a multisectoral model of the economy.

Structuring the problem in this way allows other domain specific information to be built in a way that particularly applies to multi sectoral or multi regional models. It also allows results analysis and presentation tools to be built in which make it practical for the user to work with very large models Examples of such domain specific information include "One mans sale must be another mans purchase", or "volume times price equals value" among many others. Doing this can not only improve the productivity of a user, it can also significantly improve the quality of the data and the consistency of the results matrices (ref. Stone, Champemowne and Meade 1942 and Weal 1992).

Innovative algorithms developed to deal with the specific challenges of this approach are discussed and illustrated with reference to a long standing sectoral model of the UK economy implemented with a mixture of tools. Primarily MODLER created mainly by Charles Renfro combined with other proprietary software developed by analysts employed by the Union Bank of Switzerland.


Scheduled for Session 8.4 Neural Networks

[ Agenda | Sessions | Program ]