Models, markets, polls and pundits: a case study of information efficiency

This paper develops several simple multivariate statistical models and applies them to explain fluctuations in the aggregate vote for the United States House of Representatives, over the period 1896-1964. The basic hypothesis underlying these models is that voters are rational in at least the limited sense that their decisions as to whether to vote for an incumbent administration depend on whether its performance has been “satisfactory” according to some simple standard. Because of data limitations, the analysis focuses on measures of economic performance, treating other aspects of an incumbent's performance, such as its handling of foreign affairs, as stochastic perturbations of the underlying relationship to be estimated. (Examination of residuals suggests this assumption is not unreasonable, at least during peacetime.) Possible effects of coattails from presidential races, of incumbency, and of secular trends in the underlying partisanship of the electorate are also taken into account. The models, estimated by maximum-likelihood methods, are found to be successful. Close to two-thirds of the variance in the vote series is accounted for, and the structural coefficients of the models are of the correct signs and of quite reasonable magnitudes. Economic growth, as measured by the changes in real per capita income, is the major economic variable; unemployment or inflation have little independent effect. Presidential coattails are also found to be of some importance.