Univariate vs. Multivariate Forecasts of GNP Growth and Stock Returns: Evidence and Implications for the Persistence of Shocks, Detrending Methods

Lagged GNP growth rates are poor forecasts of future GNP growth rates in postwar US data, leading to the impression that GNP is nearly a random walk. However, other variables, and especially the lagged consumption/GNP ratio, do forecast long-horizon GNP growth, and show that GNP has temporary components. Labor income and stock prices (using the dividend/price ratio) display the same behavior. This paper documents these facts and examines their implications for the persistence of shocks to GNP and time-variation in expected stock returns. I find that GNP has an almost entirely transitory response to a GNP shock that holds consumption constant. This is intuitive: if consumption does not change, permanent income did not change, so the change in GNP should be transitory. Similarly, a stock price shock that holds dividends constant suggests a discount rate change, and prices display a large transitory movement in response to this shock. The paper also examines implications of transitory variations in GNP and labor income for methods of extracting stochastic trends or "cyclically adjusting" GNP, and for explaining "excess smoothness" violations of the permanent income hypothesis.