The Uncertain Unit Root in Real GNP

The impulse and propagation mechanisms of business cycles have long been debated; however, until recently, economists were in fairly broad agreement that business fluctuations could be studied separately from the secular growth of the economy. This separation was justified because, to a first approximation, the factors underlying trend growth were assumed to be stable at business-cycle frequencies. Indeed, the common practice of macroeconomists of all theoretical persuasions was to model movements in real GNP as stationary fluctuations around a linear deterministic trend (e.g., Finn Kydland and Edward C. Prescott, 1980; Olivier J. Blanchard, 1981). Such a trendstationary (TS) model of real GNP was the canonical empirical representation of aggregate output until the early 1980's. In contrast to previous work, much of the research of the last ten years has assumed a unit root in the autoregressive representation of real GNP, which is inconsistent with a TS model of output. A model with a unit root, commonly termed a "differencestationary" (DS) model, implies that any stochastic shock to output contains an element that represents a permanent shift in the level of the series. If real GNP is best represented by a DS model, the traditional separation between business cycles and trend growth is incorrect. In the usual empirical versions of the DS model estimated for real GNP, output behaves more like a random walk than like transitory deviations