Interpreting Cointegrated Models

Abstract

Error-correction models for cointegrated economic variables are commonly interpreted as reflecting partial adjustment of one variable to another. We show that error-correction models may also arise because one variable forecasts another. Reduced-form estimates of error-correction models cannot be used to distinguish these interpretations. In an application, we show that the estimated coefficients in the Marsh-Merton (1987) error-correction model of dividend behavior in the stock market are roughly implied by a near-rational expectations model wherein dividends are persistent and prices are disturbed by some persistent random noise. Their results thus do not demonstrate partial adjustment or 'smoothing' by managers, but may reflect little more than the persistence of dividends and the noiseness of prices.