Robustness of the split-trend stationary hypothesis for the U.S.

• Author: Davis, George K.

1. Introduction

   Since Nelson and Plosser's[5] seminal contribution in 1982, macroeconomic empirical studies have generally proceeded on the presumption that time series possess unit roots. However, Perron [6] has recently challenged this view. He argues that real GNP has neither a unit root nor a simple trend, but instead is stationary around a linear trend that changes slope in 1973. This result is important since finding that macro time-series do not have unit roots would make it necessary to reconsider much of the previous empirical work in macroeconomics. In this paper we investigate the robustness of Perron's finding for other measures of economic activity using different methods of determining lag lengths. More specifically, we extend the analysis to industrial production (IP), employment (EMP), and the components of real GNP. We study the breakdown of real GNP based on the type of good produced: manufactured goods (GM), structures (GC), and services (GS). We then reexamine Perron's split-trend stationarity (STS) result for real GNP.(1)

   This exercise is not only useful because it checks on the robustness of the STS, but also because these measures of economic activity are many times treated as substitutes.(2) The choice of output measure is often made on the basis of the frequency of the other variables included in the analysis. It is therefore important for researchers to know the extent to which series meant to measure real economic activity are similar.

   The results differ across series. There is no evidence in support of STS for structures or employment. For longer lag lengths, which are advocated by Perron, there is also little evidence of STS for industrial production or manufacturing. Services do appear to be stationary around a split trend. Perron's finding for real GNP is reversed with longer lag lengths.

   Our work complements other recent work on changing trends.(3) This recent work does not assume a priori knowledge of the timing of the split, but instead searches the data for the break in the trend. Banerjee, Lumsdaine, and Stock[1] find no evidence of a split trend for real GNP in the U.S. We reach the same conclusion using Perron's technique and his assumption on the timing of the break. Our results thus reconciles Perron's finding with this recent work.

2. The Perron Test

   Perron [6] suggests the following procedure to test for a unit root against the altemative of a trend with a split at a known date. First, run the...