A Window on Econometrics: Computational Issues

Jurgen A. Doornik and David F. Hendry
Nuffield College
Doornik@cole.nuff.ox.ac.uk

Abstract

In a companion paper, the authors illustrated how, using primarily graphical tools, econometricians practice their subject. That paper was mostly aimed at a non-technical audience. The current paper discusses the technical issues underlying the new environments developed by the authors, and demonstrates how econometric researchers could benefit.

First we discuss a new object-oriented matrix language, illustrate some syntax issues, and show how creating various classes for Monte Carlo experiments reduce the required effort in setting up such experiments. It is argued that the system is completely open: one can (1) write programs in the new language, called Ox; (2) tap into the matrix functions directly from a low level language such as C and C++, thus completely ignoring the Ox language; (3) develop front-end programs, which formulate their computations in Ox code (this could be either invisible or visible to the user); (4) with the appropriate compilers, advanced users can extend Ox with pre-existing Fortran or C/C++ code. We argue that user-friendly programs could best be built by generating the interactive components separately using the most modern tools, and linking this to the new system.

Next, we show how the modularization of the new system is achieved using Dynamic Link Libraries and OLE Automation. For example, output from an Ox run (including graphics) could either appear on the console or be piped to the common front end, called GiveWin. Such modularization attempts to reduce the development cycle of econometrics software in a world of ever more demanding users. We introduce some modules which depend on GiveWin for their data input and graphics output. Development of one such module, for Monte Carlo experimentation, is discussed and demonstrated.