Haavelmo's Probability Approach and the Cointegrated VAR
34 Pages Posted: 22 Mar 2012
Date Written: March 22, 2012
Abstract
Some key econometric concepts and problems addressed by Trygve Haavelmo and Ragnar Frisch are discussed within the general framework of a cointegrated VAR. The focus is on problems typical of time series data such as multicollinearity, spurious correlation and regression results, time dependent residuals, normalization, reduced rank, model selection, missing variables, simultaneity, autonomy and identification. Specifically the paper discusses the conditions under which the VAR model represents a full probability formulation of a sample of time-series observations, the plausibility of the multivariate normality assumption underlying the VAR, cointegration as a solution to the problem of spurious correlation and multicollinearity when data contain deterministic and stochastic trends, the existence of a universe, the association between Frisch's confluence analysis and cointegrated VAR analysis, simultaneity and identification when data are nonstationary, conditions under which identified cointegration relations can be considered structural or autonomous, and finally a formulation of a design of experiment for passive observations based on theory consistent CVAR scenarios illustrated with a monetary model for inflation.
Keywords: Haavelmo, CVAR, autonomy, identification, passive observations
JEL Classification: B16, B31, B41, C32, C82
Suggested Citation: Suggested Citation