Academic journal article Journal of Money, Credit & Banking

The Great Depression and Output Persistence

Academic journal article Journal of Money, Credit & Banking

The Great Depression and Output Persistence

Article excerpt

THE PERSISTENCE OF SHOCKS to aggregate output has been the subject of extensive and continuing investigation since Nelson and Plosser (1982) presented empirical evidence suggesting that such shocks are largely permanent and account for most of the variation in real output. Methodological innovations such as those of Rudebusch (1992, 1993), and the advent of longer time series due to the work of Balke and Gordon (1989) and Romer (1989), have stimulated renewed interest in measuring the persistence of real shocks. This recent literature, including contributions by Diebold and Senhadji (1996), Cheung and Chinn (1997), Murray and Nelson (2000), Newbold, Leybourne, and Wohar (2001), and Kilian and Ohanian (2002), has produced mixed conclusions, largely due to an implicit disagreement about how to treat the period around the Great Depression. However, none of these papers produce estimates of persistence that reflect that lack of homogeneity in the historical record.

In this paper, we present a formal statistical model that captures the idea that the events surrounding the Great Depression had large, but transitory, effects on output. We construct estimates of output persistence from a parametric bootstrap of a state space model with Markov switching for the annual time series assembled by Maddison (1995) for the period 1870-1994. Contrary to some previous research on long-term annual data, our results suggest a predominant role for permanent shocks.

This paper is organized as follows. Section 1 recaps estimates of persistence under the assumption of homoskedasticity and discusses evidence against that assumption. Section 2 introduces a Markov-switching model to model heteroskedasticity in the historical data, and discusses its estimation. In that model, the economy can switch into a volatile state, such as the Great Depression, and out of it. A parametric bootstrap of that model suggests that historical measures of persistence are consistent with an economy in which shocks persist, but which has experienced at least two regimes of volatility. Section 3 presents two sets of experiments to check on the robustness of our results. Section 4 summarizes and offers concluding remarks.

1. ESTIMATION OF PERSISTENCE IN OUTPUT AND HETEROSKEDASTICITY

We begin by reviewing very briefly the evidence for the persistence of real shocks under the assumption of homoskedasticity. Maddison (1995) has assembled historical data for a large number of countries including an index of U.S. real GDP for 1870-1994. We take the log of that index as our measure of aggregate output and begin with the standard Augmented Dickey-Fuller (AD[F.sub.6]) regression:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

In choosing the lag length k, we employ the "general-to-specific" (GS) approach advocated by Campbell and Perron (1991), Hall (1994), and Ng and Perron (1995) with a maximum lag of 8, noting that this has become standard procedure in the literature. The criterion chooses k=6, and yields a unit root statistic of -3.75, which we call AD[F.sub.6]. Judging by the critical values due to Dickey and Fuller (1979), the hypothesis that the largest AR root is unity is rejected at the 5 percent level. This would imply that shocks to output do not persist indefinitely but rather completely die out over time.

The critical values for the ADF test are based on the maintained hypothesis of homoskedasticity, and a standard diagnostic test is that of White (1980). In this case, White's asymptotic test rejects the null of homoskedastic residuals at the 1 percent level. A plot of the residuals reveals that their variance was much higher around the period of the Great Depression, and generally was lower after World War II. Given the reliance of the ADF test on homoskedasticity, we would like to develop an alternative test within which we can allow for heteroskedasticity. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.