«By Andrew J. Jalil + May 2012 Abstract There are two major problems in identifying the output effects of banking panics of the pre-Great Depression ...»
A NEW HISTORY OF BANKING PANICS IN THE UNITED STATES, 1825-1929:
CONSTRUCTION AND IMPLICATIONS !
Andrew J. Jalil +
There are two major problems in identifying the output effects of banking panics of the pre-Great
Depression era. First, it is not clear when panics occurred because prior panic series—lists of
when panics occurred—combine panics with other developments in financial markets, fail to distinguish among different types of financial panics, and employ unreliable strategies to identify panics. Second, establishing the direction of causality is tricky: are panics causing downturns or are downturns causing panics? This paper addresses these two problems (1) by deriving a new panic series for the 1825-1929 period—one that rectifies many of the problems of earlier series— and (2) by studying the output effects of major banking panics via vector autoregression (VAR) methods. My paper derives four main empirical findings: (1) major banking panics have large and strongly negative effects on both output and prices, (2) panics were a substantial source of economic instability prior to the founding of the Federal Reserve, (3) on average, downturns with major banking panics were more severe than downturns without them and output recoveries were longer for downturns with major banking panics than output recoveries for downturns without them and (4) panics are associated with persistent level and trend effects. Moreover, the new series has important implications for the history of financial panics in the United States.
I thank participants at the Economic History seminars at UC Berkeley, UC Davis, UC Irvine, and Northwestern !
University, the ALL-UC Group in Economic History Conference on Financial Crises and the Real Economy, the International Transmission of Business Cycles session at the 2011 ASSA meetings, and the DAE and Monetary Economics Workshops at the 2010 NBER Summer Institute for helpful comments. I gratefully acknowledge financial support from the Economic History Association. The appendix to the paper can be downloaded at http://academic.reed.edu/economics/jalil.
Department of Economics, Reed College, 3203 SE Woodstock Blvd, Portland, OR 97202 (Email:
The general absence of banking panics in the United States since the Great Depression means that there are few modern-day equivalents to the banking crisis of 2008. Indeed, as the crisis was unfolding, there was a growing sense that the developing recession might more closely resemble the downturns of the 19th and early 20th centuries that accompanied banking crises than the recessions of the postwar period. As a consequence, commentators were speculating at the time, “this may be your great-great-grandfather’s recession.”1 However, the effects of financial panics of this era are not very well understood. Indeed, there are two main problems in identifying the macroeconomic effects of the financial panics of the pre-Great Depression era. First, it is not clear when panics occurred because prior panic series—lists of when panics occurred—differ in their identification of panic episodes. Some series document panics occurring at a rate of more than one per year, whereas other series identify recurring periods of ten to twenty years without a panic. For example, two studies on the output effects of the financial panics of the pre-Federal Reserve era—DeLong and Summers (1986) and Miron (1986)—arrive at contradictory conclusions because they rely on different panic series to identify when panics occurred.
Second, even if the timing of when panics occurred is consistent with panics having real output effects, establishing the direction of causality is inherently difficult. Are panics causing downturns or are downturns causing panics? There are two schools of thought. According to one view, panics are shocks to the real economy that cause downturns, but according to the alternate view, panics are consequences of major recessions. Under this latter specification, even if a correlation between downturns and panics exists, it would be misleading to attribute output declines to panics since panics would be products of downturns rather than causes.
This paper seeks to identify the macroeconomic effects of a main class of financial panics of the preGreat Depression era: banking panics. It accomplishes this by addressing these two major problems. To address the first problem, this paper derives a new series on banking panics for the pre-Great Depression era. To address the second problem, this paper studies the output effects of major banking panics via 1 Krugman, Paul, “Who’ll Stop the Pain?”, New York Times, Feb 19, 2009.
panics occurred—a chronology that has important implications for our understanding of the causes, effects and frequency of financial panics throughout U.S. history.
Section 1 begins by asking: when did the financial panics of the pre-Great Depression era occur? I show that—perhaps, somewhat surprisingly—answering this question is not a straightforward task. I document nine leading panic series and demonstrate that each of these series would arrive at a different answer to this question. Some series document panics occurring at a rate of roughly one per year, whereas other series identify recurring periods of ten to twenty years without a panic. Why do these series differ so dramatically? By studying each of the panic series on a case-by-case basis, I find that methodological problems behind the development of earlier series are the likely source of these extreme variations.
Specifically, I find two common problems. First, many series do not explicitly define a panic, making it unclear exactly what kind of financial disturbance is being recorded. Second, most series do not adopt a systematic rule to identify panics over a specified period, raising the possibility that some panic episodes might be omitted or that non-panic episodes might be mistakenly classified as panics.
Due to the numerous contradictions across these series and in light of these methodological concerns, Section 2 derives a new panic series for the 1825-1929 period—one that rectifies many of the problems of earlier series. I begin by providing a definition of financial panic—one that distinguishes among different types of financial panics. In developing the new series, however, I restrict attention to one key type of financial panic of this era: banking panics. To identify banking panics, I adopt a systematic rule to search through more than one hundred years of contemporary financial and economic newspapers.
Section 2 outlines the algorithm I use to identify banking panics from the historical news record and develops a consistent set of criteria for classifying banking panics as major or minor.
Section 3 presents the results of the new series and its implications. The new series on banking panics identifies seven major banking panics, as well as twenty minor banking panics. The new series has two main initial findings. First, using my new series and the accounts of contemporary observers contained in the historical news record, I discover that earlier series presented flawed lists of when panics
distinguish among different types of financial panics, and a few series even went so far as to mistakenly identify foreign banking panics as domestic ones. Second, contrary to the accepted wisdom in the literature, I find that there is no evidence of a decline in the frequency of financial panics during the first fifteen years of the existence of the Federal Reserve. Previous studies used the Kemmerer series—a series that the analysis in this paper shows to be severely flawed—to document the pre-1914 frequency of panics, leading to spurious conclusions regarding the historical frequency of panics.
Equipped with the new panic series, Section 4.1 presents the empirical tests I conduct to identify the real output effects of major banking panics. Part A presents the basic vector autoregression. The basic VAR that I estimate has two variables: a panic dummy series and annual output growth. The impulse response function implied by the VAR of the panic dummy series to an output shock is insignificant, suggesting that the panics are unrelated to past output movements. By contrast, the impulse response function of output to a unit shock in the panic dummy is large and significant, indicating that panics have substantial real output effects. Part B then restricts the VAR using additional evidence from the narrative record to isolate those panics that the reports of contemporary observers suggest were the result of idiosyncratic disturbances, as opposed to declining output conditions. The impulse response functions with these additional restrictions are similar to those from the basic VAR. They indicate that major banking panics have large effects on output. Methodologically, this approach is similar to the work of Ramey (2011), Ramey and Shapiro (1998), and Romer and Romer (1989, 2004, 2010) who use narrative evidence to identify the effects of fiscal and monetary policy. Lastly, Part C estimates a three-variable VAR with monthly data on output and prices. The results indicate that banking panics have rapid, large, and strongly negative effects on both output and prices.
Armed with a reliable listing of when panics occurred, Section 4.2 then investigates three related macroeconomic issues. First, I investigate whether banking panics were a significant source of output volatility prior to the founding of the Federal Reserve. I find that nearly half of all business cycle downturns between 1825 and 1914 contained major banking panics. According to the estimates I derive in
these downturns, they amplified them. These findings suggest that banking panics were a substantial source of economic instability throughout much of U.S. history and that major banking panics either caused or amplified nearly half of all business cycle downturns between 1825 and 1914.
Second, I examine how downturns with major banking panics differed from downturns without major banking panics. I find that in the post-Civil War era, downturns with major banking panics were slightly longer and substantially more severe than downturns without major banking panics on average, and output recoveries for downturns with major banking panics were two-to-three times the length of output recoveries for downturns without them.
Third, I analyze the behavior of output in the aftermath of major banking panics. Following three of the four major banking panics of the post-Jacksonian period, output did not rapidly revert back to its prepanic trend. Moreover, following two of these panics, trend output growth declined substantially. These results support the growing consensus in the literature that banking crises can have highly persistent and lingering effects on output. They also reveal that the sluggish output growth the U.S. economy has experienced in the aftermath of the Great Recession is consistent with the historical record.
Part I. Prior Panic Series
1.1. When did Financial Panics Occur?
When did financial panics occur before the Great Depression? Surprisingly, answering this question is not a simple task. Table 1 presents nine different panic series: the (1) Bordo-Wheelock, (2) Thorp, (3) Reinhart-Rogoff (two versions), (4) Friedman-Schwartz, (5) Gorton, (6) Sprague, (7) Wicker, (8) Kemmerer, and (9) DeLong-Summers series. Each panic series comes from a different study and covers a
series regarding the timing, incidence, and frequency of panics throughout U.S. history.
Consider first the period between 1825 and the end of the Civil War. The three series that span the 1825-1864 period—the Thorp, Bordo-Wheelock, and Reinhart-Rogoff series—identify 1825 and 1857 as panics.3 However, there is substantial disagreement regarding other episodes. The Bordo-Wheelock and Thorp series identify 1833 as a panic, while both versions of the Reinhart-Rogoff series omit 1833. The Bordo-Wheelock and Thorp series identify 1837 and 1839 as two distinct panics, while one version of the Reinhart-Rogoff series classifies 1836 as a banking crisis, whereas the other version identifies a crisis from 1836 to 1838. The Thorp series identifies 1847 as a panic, whereas the other series do not contain this episode. One version of the Reinhart-Rogoff series identifies December 1861 and April 1864 as banking crises, whereas the other series do not contain these episodes.
Similar discrepancies extend into the post-Civil War period. From 1864 to 1929, the nine series concur in classifying 1873 and 1907 as panics; however, they disagree on other episodes. Some series list 1884, 1890, and 1893 as panics, whereas other series either omit these episodes or classify them as incipient panics, financial distress, or financial stringency, as opposed to full-scale panics. Some series also identify panics in 1896 and 1914, whereas other series omit these episodes. The Kemmerer series identifies three additional major panics (Dec 1899, May 1901, and Mar-Aug 1903) and twenty-one minor panics—the vast majority of which are not noted in any of the other series. The DeLong-Summers series also contains additional panic episodes that are not noted in any of the earlier series.
2 The series come from the following studies: (1) Bordo and Wheelock (1988), (2) Thorp (1926), (3) Reinhart and Rogoff (2009), (4) Friedman and Schwartz (1963), (5) Gorton (1988), (6) Sprague (1910), (7) Wicker (2000), (8) Kemmerer (1910), and (9) DeLong and Summers (1986). Reinhart and Rogoff (2009) provide two versions of their series. They list one version in Table A.3.1 and another version in Table A.4.1. The two versions occasionally contradict one another. As a consequence, I provide both versions in Table I.
3 The Reinhart-Rogoff series uses the term crisis rather than panic.
An understanding of when panics occurred is critical to studies on the real output effects and causes of panics. Studies on the real output effects of panics depend on the panic series to identify the timing of panics in relation to changes in output. Studies on the causes of panics depend on the panic series to identify the timing of panics in relation to forces that may have increased the susceptibility of the financial system to a panic.