Working Paper: CEPR ID: DP775
Authors: Marco Lippi; Lucrezia Reichlin
Abstract: In this paper we argue that modelling the trend component in real GNP as a random walk is inconsistent with its interpretation as productivity growth. As an alternative we specify the trend as an Auto Regressive Integrated Moving Average (ARIMA) process, whose impulse response function follows an S-shaped pattern reflecting the process of diffusion of technical change. Such an ARIMA process is employed to build and estimate an Unobserved Components ARIMA (UCARIMA) model using USA post-war quarterly data. We find that our model, although more parsimonious, fits the data as well as the standard random walk plus AR(2) cycle. Moreover, the cycle has a very low variance relative to the variance of the trend in our model.
Keywords: Nonstationarity; Productivity Growth; Random Walk
JEL Codes: E32; E37; O49
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Technological change (O33) | Decomposition of output into trend and cycle (E32) |
| Modeling the trend component of real GNP as a random walk (C51) | Inconsistency with interpreting it as productivity growth (O49) |
| ARIMA process with S-shaped impulse response function (C22) | More accurate representation of technological changes diffusion (O33) |
| Technological innovations absorbed within approximately 15 months (O39) | Contrast with immediate absorption implied by random walk model (G19) |
| Cycle variance lower than trend variance (C22) | Most variability in GNP due to permanent changes rather than transitory shocks (E39) |