walt_sak@proper.UUCP (Walt Sakai) (10/08/85)
> From dual!lll-crg!seismo!harvard!h-sc1!quah > Subject: Re: FORECASTING CAPITAL MARKETS > > For forecasting in general, a representation known as bayesian > vector autoregressions seems to have been relatively successful > in certain macroeconomic applications. This is work most closely > associated with the ideas of Bob Litterman and Chris Sims at > Minnesota. They seem to indicate that BVAR also outperforms the expensive commercial forecasting services! References are: Thomas Doan / Robert Litterman / Christopher Sims 1984 "Forecasting and Conditional Projection Using Realistic Prior Distributions" Econometric Rev. 3:1-100. William Lupoletti and Roy Webb 1984 "Defining and Improving the Accuracy of Economic Forecasts: Contributions from a VAR Model" Federal Reserve Bank of Richmond, Working Paper 84-6. > I have not myself seen these ideas applied in the more "finance"- > oriented areas. Everyone I know who uses these things is kind of > macro-academic. Litterman and Doan also have a package called > RATS (Regression Analysis of Time Series) that implements many of > these ideas, available on both mainframes and IBM-type PC's. > (Alas no unix (tm) implementation yet). I know that certain > financial firms have found this product useful although I haven't > seen any of their empirical research and success published as of > yet :-). There is some work work going on in the University of Chicago using RATS for testing bias in forex forward pricing. Preliminary results indicate that the usual random walk dominates. Walt Sakai uucp: {ucbvax, hplabs, ihnp4, cbosgd, unisoft, decwrl, fortune, sun}!dual!proper!walt_sak {intelca, qantel}!proper!walt_sak ARPA(new): dual!proper!walt_sak@UCBVAX.BERKELEY.EDU ARPA(old): dual!proper!walt_sak@BERKELEY.ARPA