Deep recessions, fast recoveries, and financial crises: evidence from the American record.

Bordo, Michael D. ; Haubrich, Joseph G.

I. INTRODUCTION

The recovery from the recent recession has now been proceeding for over 7 years. Many argue that this recovery is unusually sluggish and that this reflects the severity of the financial crisis of 2007-2008 (Reinhart and Rogoff 2009; Roubini 2009). Yet if this is the case, it appears to fly in the face of the record of U.S. business cycles in the past century and a half. Indeed, Milton Friedman noted as far back as 1964 that in the American historical record "A large contraction in output tends to be followed on the average by a large business expansion; a mild contraction, by a mild expansion" (Friedman 1969, 273). Much work since then has confirmed this stylized fact but has also begun to make distinctions between cycles, particularly between those that include a financial crisis. Zarnowitz (1992) documented that pre-World War II recessions accompanied by banking panics tended to be more severe than average recessions and that they tended to be followed by rapid recoveries.

In this article, we revisit the issue of whether business cycles with financial crises are different. We use the evidence we gather to shed some light on the recent recovery. A full exploration of this question benefits from an historical perspective, not only to provide a statistically valid number of crises but also to gain perspective from the differing regulatory and monetary regimes in place. We restrict ourselves to the United States, where we have a better grasp and can take a closer focus on the monetary regimes and institutional environment. This also avoids comparability problems with other countries, which often have very different stochastic structures of output (Cogley 1990). but of course this means that our data contains many fewer crises and business cycles than is the case with studies based on multicountry panels. We look at 27 cycles starting in 1882 and use several measures of financial crises. We compare the change in real output (real gross domestic product [RGDPJ) over the contraction with the growth in real output in the recovery, and test for differences between cycles with and without a financial crisis. We check for robustness to various definitions of financial crises, output, and business cycles. Finally, we suggest residential investment as a possible explanation for a slow recovery after a recession that involves a housing bust, as the United States is currently experiencing.

Our analysis of the data shows that steep expansions tend to follow deep contractions, though this depends heavily on how the recovery is measured. In contrast to much conventional wisdom, the stylized fact that deep contractions breed strong recoveries is particularly true when there is a financial crisis. In fact, on average, it is cycles without a financial crisis that show the weakest relation between contraction depth and recovery strength. This is our key finding. For many configurations, the evidence for a robust bounce-back is stronger for cycles with financial crises than those without. The results depend somewhat on the time period, with cycles before the Federal Reserve looking different from cycles after the Second World War.

Until quite recently, the extensive literature on business cycle dynamics rarely combined a long data series with a focus on financial crises. Friedman (1969, 1993) has a long series but does not consider the effect of financial crises, in addition to using data and empirical techniques that are somewhat different than ours. In contrast to most subsequent work, Friedman looks at growth over the entire expansion. Wynne and Balke (1992, 1993) include only cycles since 1919 and do not consider the effect of financial crises. They measure growth four quarters into the expansion. Kahn and Rich (2007) have a neoclassical growth model with regime shifts in pemanent and transitory components, but their focus is on identifying post-World War II productivity trends, not business cycles or financial crises. Among more recent papers, the first major split is between papers like ours that focus on one country, and those that use international data. Lopez-Salido and Nelson (2010) explicitly look at the connection between financial crises and recovery strength but look only at post-World War II cycles in the United States. Bordo and Haubrich (2010) find that contractions associated with a financial crisis tend to be more severe but do not gauge the speed of the resulting recovery.

Stock and Watson (2012) use a 198-variable dynamic factor model on data since World War II, attributing recent slow recoveries to demographic factors in the labor market. Gali, Smets, and Wouters (2012) estimate a structural new Keynesian model and attribute the current slow recovery to adverse demand shocks stemming from the zero lower bound and wage markups. Flail (2011) defines a related concept of slump, when employment is below 95.5% of the labor force. On the international side, Howard, Martin, and Wilson (2011) look at the relationship between recoveries and crises for 59 countries since 1970 and reach conclusions similar to ours. Benati (2012), in a paper that originated as a conference discussion of our paper, looks at impulse response functions for five countries and the euro area since 1970 and does not find different dynamics after crises. A recent paper by Romer and Romer (2015) looks at the impact of financial distress on advanced countries in the post-World War II period and find results similar to ours.

The second major split is over the definition of recovery. Most of the literature follows Friedman and considers a recovery or expansion as starting at the cyclical trough. Several important papers, however, measure a recovery from peak to peak, that is, how long it takes the economy to return to the cyclical peak. These are important differences, and seemingly divergent results often stem from differing formulations of the question. Reinhart and Rogoff (2009) concentrate on major international financial crises and document long and severe recessions but make few direct comparisons of the recovery speed with noncrisis cycles. They also measure speed by how long it takes to get back to the previous business cycle peak and not how fast the economy grows once the recovery has started. Jorda, Schularick, and Taylor (2011) find slow growth after a credit boom but say little about the patterns after a recovery has started, again making peak-to-peak comparisons. Cerra and Saxena (2008) look at data for 190 countries and find that output losses in disasters are in general not recovered, in the sense of returning to the pre-crisis trend line. Gourio (2008) finds strong recoveries after disasters, in the sense of exceptionally high growth rates.

The remainder of the article is as follows: Section II presents an historical narrative on U.S. recoveries; Section III examines the amplitude, duration, and shape of business cycles since 1882, testing whether strong recoveries follow deep contractions, and whether financial crises alter that pattern; and Section IV concludes.

II. NARRATIVE

The relation between a contraction, a recovery, and an associated financial crisis can in principle depend on the monetary regime, banking system, and macroeconomic environment. To provide some context for the econometric results, and offer some justification (and skepticism) for treating over a century of financial crises and business cycles in a coherent fashion, we present some descriptive evidence and historical narratives on economic recoveries following recessions associated with a financial crisis in the United States, from 1880 to the present. For example, financial crises before the advent of the Federal Deposit Insurance Corporation (FDIC) in 1934 were banking panics reflecting a scramble for liquidity, fast-moving events that lasted at most a matter of months. Since 1934, financial crises have been mostly insolvency events, not liquidity events, resolved by fiscal rescues by the FDIC and Treasury. They often lasted for years. We do not aim to establish causality, particularly about the extent to which contractions and crises influence each other, but rather to foster a broader appreciation of the idiosyncracies of the timing, policy, and environment that surround each cycle and crisis.

Table 1 shows some metrics on the salient characteristics of the recessions and recoveries. Columns 1 and 2 report the date of the cyclical peak and trough, as determined by the National Bureau of Economic Research (NBER). Column 3 lists the date of the onset of the financial crisis, if any, associated with the cycle. Column 4 shows the total change (usually a drop) in GDP during the contraction, and column 5 shows the total change (usually an increase) during the recovery going out the same number of quarters as the contraction lasted (so if the contraction lasted five quarters, five quarters of the recovery will be counted). Columns 6 and 7 show the percentage changes.

A. 1880-1913: The Pre-Federal Reserve Period

During this era, the United States was on the gold standard and did not have a central bank. The NBER demarcates 11 business cycles, of which three, 1882(1) to 1887(11), 1893(1) to 1894(11), and 1907(11) to 1908(11), had associated financial crises. All of the recessions associated with financial crises were followed by recoveries at least as rapid as the downturns.

The recession of 1882-1885 featured a banking panic in May 1884 following the collapse of the brokerage firm, Grant and Ward, and a stock market crash (Wicker 2000, 35). The banking panic ended with the issuance of clearinghouse loan certificates and U.S. Treasury quasicentral banking operations. The recovery of 1885(11) to 1887(11) was driven by capital and gold inflows. The recovery was interrupted by a brief 1-year mild contraction. (1)

The decade of the 1890s was shadowed by silver uncertainty and falling global gold prices, which produced persistent deflationary pressure. The recovery from the 1890-1891 recession ended early in the first quarter of 1893. A stock market crash on May 3, 1893 was followed by a banking panic which spread from New York to the interior and back. The panic, which ran in waves from May to July led to many bank failures across the country (Friedman and Schwartz 1963, chapter 3; Wicker 2000, chapter 4). The issuance of clearinghouse loan certificates and the suspension of convertibility of deposits into currency eventually ended the panic by the fall of 1893. The subsequent recovery from 1894(11) to 1895(111) was aided by the Belmont Morgan syndicate, which was created in early 1895 to rescue the U.S. Treasury's gold reserves from a silver-induced run.

In 1906, the Bank of England reacted to declining gold reserves by raising its discount rate and rationed lending based on U.S. securities. This created a serious shock to U.S. financial markets, triggering a stock market crash and a major banking panic in October 1907. The banking panic led to many bank failures, a steep drop in the money supply, and a serious recession, which ended in May 1908. The recession of 1907-1908 was followed by a vigorous recovery from 1908(11) to 1910(1). Friedman and Schwartz attribute this to gold inflows reflecting a decline in U.S. prices relative to those in Britain stemming from the crisis. The onset of World War I in 1914 led to a recession and a global banking crisis (Silber 2007). The recession was then followed by a major boom, driven by the demand for U.S. goods by the European belligerents and then by the United States as it prepared for war.

B. The World Wars and Between: 1914-1945

The Federal Reserve was established in 1914 in part to solve the problem of the absence of a lender of last resort in the crises of the pre1914 national banking era. In the Fed's first 31 years, there were three very severe business cycle downturns and several minor cycles. Most of the recoveries in this period were at least as rapid as the downturns that preceded them with one important exception: the recovery from the Great Contraction of 1929 to 1933.

Recovery 1933(1) to 1937(11). A recession began in August 1929 with a major stock market crash in October. A series of banking panics beginning in October 1930 ensued. The Fed did little to offset them, turning a recession into the Great Contraction. The recovery began after Roosevelt's inauguration in March 1933 with the Banking Ffoliday. The recovery, although rapid (output grew by 33%), was not sufficient to completely reverse the preceding downturn. The recovery may have been impeded somewhat by New Deal cartelization policies like the National Industrial Recovery Act, which, in an attempt to raise wages and prices, artificially reduced labor supply and aggregate supply (Cole and Ohanian 2004).

C. Post-World War II: 1945-2014

In the post-World War II era, with only two exceptions, recoveries were at least as rapid as the downturns. In general, recessions were shorter and recoveries longer than before World War II (Zarnowitz 1992). There also were fewer stock market crashes. There were only four events demarcated as financial crises in the period. This may have reflected the presence of deposit insurance and the financial safety net.

1975(1) to 1980(1). The recession that began in November was one of the worst in the postwar period. RGDP fell by 3.4% and unemployment increased to 8.6%. The recession was greatly aggravated by the first oil price shock, which quadrupled the price of oil, and by wage price controls which prevented the necessary adjustment. The United States experienced a minor banking crisis with the failure of Franklin National in October 1974 and other significant banks, as market conditions were made more stressful by the payments problems stemming from the Herstatt failure in June 1974 (Lopez-Salido and Nelson 2010). Faced with inflation and rising unemployment the Federal Open Market Committee (FOMC) sought moderate growth in monetary aggregates (Roesch 1975). The recovery began in April 1975. As in most of the postwar recessions, the pace of the recovery exceeded the pace of the downturn.

1982(IV) to 1990(111). Fed policy began to tighten in May 1981 in the face of a jump in inflation. It raised the federal funds rate from 14.7% in March to 19.1% in June. This second and more durable round of tightening induced by Chairman Volcker succeeded in reducing the inflation rate from 10% in early 1981 to 4% in 1983 but at the cost of a sharp and very prolonged recession. RGDP fell by close to 3% and unemployment increased from 7.2% to 10.8%. During this period there were two banking crises. The first, between 1982 and 1984, involved the failure and bail-out of Penn Square Bank in 1982 and Continental Illinois in 1984. Both banks were hit hard by the collapse in global oil prices. The money center banks were hard hit by the Latin American debt crisis in 1982, and the Volcker Fed eased policy as part of a complicated plan to avoid extensive defaults (Meltzer 2010. book 2). The second was the Savings and Loan Crisis from 1988 to 1991: Thrift resolutions spiked from 47 in 1987 to 205 in 1988 as the loan charge-off ratio for commercial banks increased and in 1991 reached a postwar peak not equaled until 2009 (Lopez-Salido and Nelson 2010).

1991(1) to 2001(1). The FOMC only began cutting the funds rate in November 1990 because its primary concern was to reduce inflation, which had reached 6.1% in the first half of 1990 (Hetzel 2008, chapter 15). The recovery from the trough in March 1991 was considered tepid, and it was referred to as a jobless recovery. Unemployment peaked at 7.75% in June 1992. The recession was also viewed as a credit crunch (Bordo and Haubrich 2010), and real housing prices declined by 13%, suggesting a minor housing bust. This is the first recovery in the postwar era where the pace of expansion was less than that in the downturn. The recovery continued into a lengthy expansion, however.

2009(11)-?. The last recession which began in December 2007 was preceded by a crisis in the shadow banking sector after the collapse of the housing market reduced the value of mortgage-backed securities. This later affected the balance sheets of the banking system. The fact that the Fed had supported the insolvent investment bank Bear Stearns in March 2008, but to prevent moral hazard did not do so for Lehman Brothers in September, even though it was in similar shape, many believe triggered a global liquidity panic (Bordo 2014). According to Hetzel (2014), the recession was not caused by the financial crisis but by the Fed following tight monetary policy in the fall of 2007 and early 2008. The recession was the most severe in the postwar period (RGDP fell by more than 5% and unemployment increased to 10.8%). The financial crisis in the fall of 2008 was without doubt the most serious event since the Great Contraction.

Both the crisis and the recession were dealt with by vigorous policy responses: on the monetary policy side, the Federal Reserve cut the funds rate from 5.25% in early fall 2007 to close to zero by January 2009 and created a variety of programs that tripled the Federal Reserve's balance sheet, on top of a massive fiscal stimulus package. The recovery since 2009 has been tepid, with real growth expanding at slightly above 2%. The pace of recovery is well below the pace of the decline in output.

III. THE RELATIONSHIP BETWEEN RECESSIONS AND RECOVERIES

In this section, we take a more statistical view of the relationship between the depth of the contraction and the strength of the following expansion. We make no claims about causality: we do not consider here whether financial crises contribute to recessions or recessions create financial crises. That is, we examine recoveries that occur against the backdrop of a recent financial crisis; the analysis does not require attributing the preceding recession to the crisis.

Our data is based on Bordo and Haubrich (2010), where we provide a more detailed description. Business cycle turning points (in quarters) come from the NBER. RGDP, again at a quarterly frequency, is based on Balke and Gordon (1986) and Gordon and Krenn (2010), extended via the NIPA accounts. This gives us quarterly RGDP for 27 business cycles, starting with the peak in 1882 and ending with the recovery from the 2007 recession.

[FIGURE 1 OMITTED]

We measure the amplitude of the contraction by the percentage drop (from the peak to the trough) of quarterly RGDP. We measure the recovery strength as the percentage change from the trough at two horizons: four quarters after the cyclical trough and after a time equal to the duration of the contraction. Figure 1 plots the strength of the recovery against the depth of the contraction, for both recovery conventions. Going out the length of the contraction, while it appeals to symmetry, appears to be new, as most papers restrict their attention to four quarters (or 12 months if they use monthly data). Friedman (1993) is the exception, looking at growth to the next cyclical peak. Morley and Piger (2012) in their discussion of bounceback models, consider recoveries of up to six quarters after a trough, while Ploward, Martin, and Wilson (2011) look out 3 years after a trough. Papell and Prodan (2012) look at when growth rates return to trend and when both levels and growth return to trend.

Exactly what constitutes a financial crisis depends on how it is defined, and the question has been answered several ways. In this section, for the pre-World War II years we use the chronology from Eichengreen and Bordo (2002) and also add 1914, a year in which the bond markets closed. Eichengreen and Bordo describe their definition as follows (15-16):

"For an episode to qualify as a banking crisis, we must observe either bank runs, widespread bank failures and the suspension of convertibility of deposits into currency such that the latter circulates at a premium relative to deposits (a banking panic), or significant banking sector problems (including but not limited to bank failures) resulting in the erosion of most or all of banking system collateral that are resolved by a fiscally underwritten bank restructuring." (2)

For the postwar period, we use the chronology of Lopez-Salido and Nelson. This gives us crisis periods of 1884-1885, 1892-1893, 1907, 1914, 1930-1933, 1973-1975, 1982-1984, 1988-1991, and 2007 (Table 1).

Consequently, the recessions we associate with a financial crisis are those that start in 1882, 1893, 1907, 1913, 1929, 1973, 1981, 1990, and 2007. (We drop the 1945 recession from our sample. This is reasonable, but it matters, as it was the deepest recession of the century outside the Great Depression, with an extremely weak recovery.) It is perhaps interesting to note that the five GDP "disasters" picked out by Barro and Jin (2011) correspond to the cycles with troughs in 1914, 1921, 1933, and 1947, with two of those five being associated with a financial crisis. (With our data, we would add 1894 as a disaster where RGDP fell by 10% or more.) Interestingly, of our 27 business cycles, only 4 did not have some form of financial crisis, according to the reckoning of Reinhart and Rogoff (3) (1899, 1923, 1953, and 1960).

A. Are Deep Recessions Followed by Steep Recoveries?

The visual evidence (Figure 1) definitely suggests that deep recessions are followed by strong recoveries, though a few outliers, particularly the Great Depression, may have a disproportionate impact. Regressing growth four quarters after the trough against contraction amplitude shows a positive but small and statistically insignificant relationship. The relationship is tighter, and stronger, if we examine more of the recovery, measuring growth out to the duration of the contraction after the trough. Looking out only four quarters can give a misleading picture, particularly for longer recessions. Much of the difference is in fact driven by the Great Depression, and it should not be surprising that the drop in output from 1929 to 1933 was not fully reversed by 1934. since monetary policy (using M2) only became expansionary in early 1934 with the devaluation of the dollar and resulting gold inflows (Friedman and Schwartz 1963, chapter 8). By 1936 though, output was much closer to its pre-crisis level.

Do financial crises affect the bounceback? Or more precisely, since we cannot assess causality without a carefully identified model, do the recoveries from recessions with a financial crisis look different than those without? Figure 2 plots the strength of the recovery against the depth of the recession, separately identifying crisis and noncrisis cycles. The scatterplot suggests a difference, but for reasons contrary to the conventional wisdom. In crisis times, strong recoveries follow deep recessions, but outside of a crisis, they do not. The relation is in fact negative, though not statistically significant.

To get a more formal view of the difference, we run a set of regressions based on the following specification:

(1) %[DELTA][Y.sub.T+k] = [[alpha].sub.1] + [[alpha].sub.2][D.sub.F] + [[beta].sub.1] [%[DELTA][Y.sub.P-T]] + [[beta].sub.2][D.sub.F] [%[DELTA][Y.sub.P-T].

Here, % [DELTA][Y.sub.T+k] is the percentage increase in RGDP from the cyclical trough (T) to quarter T + k; as mentioned above, k is either 4 or equals the number of quarters in the recession. [D.sub.F] is a dummy for financial crises, % [DELTA] [Y.sub.P-T] is the percentage change in RGDP from peak to trough (peak [RGDP] minus trough [RGDP] over trough [RGDP], so in typical contractions this will be a positive number). Thus the strength of the expansion is regressed against a constant, a dummy for financial crises, a measure of the depth of the contraction and an interaction between the financial crisis dummy and the depth of the contraction, in effect a slope dummy. An F-test then determines the significance of excluding the dummy and the interaction term. This specification is meant to capture two separate ways that a crisis may affect the bounceback. On average, such contractions may have a slower recovery, but the relationship between contraction depth and recovery strength may also be affected.

[FIGURE 2 OMITTED]

It may not be proper to lump all crises and all cycles together, given the very different monetary standards and regulatory regimes in place over time. We split the data several ways, dropping the Great Depression, and separately examine the years after the founding of the Federal Reserve and after World War II. Table 2 reports the results out to the first four quarters of the expansion, and Table 3 looks at the expansion going out the duration of the contraction.

Looking only at four quarters of the recovery, there is little evidence that recoveries following financial crises are much different. At a formal level, the F-test does not show a significant difference. Informally, the coefficient on the financial dummy is uniformly negative, indicating that recoveries following financial crises are on average somewhat smaller, though the coefficient is significant in only one case. The interaction term, also insignificant, is split between negative and positive values: in the one significant case the depth of the recession has a greater impact on the strength of recoveries when there is a crisis.

The differences are more striking once more of the recovery is considered, as Table 3 shows (Figure 2 plots the data). The interaction term and the F-test are significant in all but the post-World War II samples. The dummy for financial crises is always negative, but it is significant only half the time. The post-World War II sample shows no significant difference between crisis and noncrisis recoveries (perhaps because crises were less severe, and the intercept dummy dominates the slope dummy), but for the other samples a clear pattern emerges: the dummy for financial crises is negative, but the interaction term is positive. This means that relatively mild recessions with a crisis have slower-than-average recoveries, but the deeper the recession, the stronger the recovery. For the entire sample, the crossover point is about 3.25%, met by most crisis contractions except 1882, 1973, 1981, and 1990, which perhaps contributes to the impression that the crisis recoveries are weaker. The numbers have economic heft: a 1% deeper recession with a crisis will lead to greater than an extra 1.5% of growth in the quarters following the trough.

As a check on the robustness of the results, we look at alternative measures of crises. Bordo and Landon-Lane (2012) find the world has had five global financial crises since 1880 (1890-1891, I907-1908, 1913-1914, 1930-1933, and 2007-2008). Figure 3 plots recovery amplitude (four quarters) against contraction amplitude for these crises and all post-World War II contractions, including 2007-2008. The Appendix conducts further robustness tests with alternate measures of output and of business cycles.

Like the previous figures, Figure 3 shows a positive relationship between contraction amplitude and recovery strength, though the coefficient is relatively small. In part, this arises from the three most recent cycles, which appear different, with both a lower intercept and a lower slope. Some speculation suggests that this results from changes in labor market behavior since the 1980s (Beauchemin 2010: Jaimovich and Siu 2012).

Another classification of financial crises comes from Reinhart and Rogoff (2009), and using their measure also lets us make distinctions between financial crises based on their severity. Replacing the crisis dummy in the interaction term with the Reinhart and Rogoff index of crisis severity (which essentially adds up the number of different crises occuring during particular years, which we sum over contractions), provides yet another robustness check. The results looking out either four quarters from the trough or the duration of the contraction into the expansion were quite similar, so in Table 4 we report only the results for duration. Again the results are somewhat mixed. The financial crisis dummy is usually positive, but insignificant. The interaction term is significant in all four cases, and positive except for the post-World War II period. Thus, for most time periods, a more severe crisis means a stronger recovery.

Yet another approach to measure financial stress uses the spread between CD spreads and 3-month Treasury bills (see Figure 4). Using the spread removes 1990 as a crisis event. As the Appendix shows, the results that recoveries are generally more rapid from financial recessions are robust to this change.

[FIGURE 3 OMITTED]

IV. PROSPECTS FOR THE CURRENT RECOVERY

Recessions that accompany a financial crisis tend to be long and severe (Bordo and Haubrich 2010; Reinhart and Rogoff 2009). What that portends for economic growth once a recovery has started is less certain, however. On the one hand, there is the feeling that "growth is sometimes quite modest in the aftermath as the financial system resets" (Reinhart and Rogoff 2009, 235). On the other hand, there is the stylized fact behind Friedman's plucking model, that "A large contraction in output tends to be followed on the average by a large business expansion" (Friedman 1969, 273). One popular measure of recovery strength, the time required to return output to the pre-crisis level, confounds the depth of the recession with the strength of recovery. For many purposes, it is important to separate the notions of contraction depth and recovery strength. Our results show that if there is a difference between recoveries, it is because the recoveries following a financial crisis tend to be steeper, and far from being an exception to the plucking rule, are the major evidence for it.

[FIGURE 4 OMITTED]

Where does that leave the current recovery? It remains an outlier, as one of the few cases where output did not return to the level of the previous peak after the duration of the recession. In this, it resembled two very different recessions, the Great Depression and 1990. Significantly, both of those combined financial problems and (real) housing price declines, albeit of strikingly different magnitudes. In Bordo and Haubrich (2012), we provide suggestive evidence that supports housing as at least a partial factor. Others point to policy uncertainty as having a role (Baker, Bloom, and Davis 2015), but at this point the evidence does not provide a definite resolution. The unanswered question, of course, relates to causality; tracing out the exact shocks, and their transmission, remains key. Must housing recover for the recovery to take off, or will the economy pull the industry along? Would monetary policy, credit policy, or fiscal policy be an appropriate response? These are questions for another day.

APPENDIX: DATA

Construction of the quarterly bank loan numbers. We started with the annual numbers for all commercial banks, Millennial Statistics, table Cj253 "commercial banks-number and assets," Total Loans. After I9I4Q4 these are made quarterly by interpolation using the RATS disaggregate (linear, arl) procedure using total loans for all member banks, from the Fed's Banking and Monetary Statistics, 1914-1941 No. 18, All Member banks-Principal assets, "Loans" p. 72-74. The data were pushed forward to 1955 Q4 using the data from Banking and Monetary Statistics 1941-1970 table 2.1 All Member Banks A. Total assets and number of bank loans. There were several missing quarters, which we interpolated using No. 48, Weekly Reporting Banks in 101 Leading Cities-Principal Assets and Liabilities, weekly and monthly.

[FIGURE A1 OMITTED]

To go back further, we again began with the Millennial statistics, and interpolated again, this time using national and state bank data. The national bank data came from the NBER series, adding up NBER 14016, Loans and Discounts, national banks, country districts, NBER 14018, Loans and discounts, national banks. Reserve cities other than central, and NBER 14019, Loans and Discounts, National Banks, Central Reserve cities. Some judgment was used to apportion the data into quarters.

Annual State bank data comes from Millennial Statistics, Cj 151, State banks. Loans, and discounts. It was made quarterly by linear interpolation. The state and national bank numbers were added together and used to interpolate the Millennial statistics annual number. This let us interpolate from 1896 to 1913. Then we attached the series for state and national banks, discounting by the ratio for that series to the Millennial series in 1914. This gave us a series from 1882 to 1955. Figure Al shows the series.

FURTHER ROBUSTNESS RESULTS

The next step is to check whether the results depend on using RGDP as an output measure. One approach is to use per capita income (population taken from Millenial Statistics): the cycles with a financial crisis show a positive correlation between contraction depth and recovery strength. Though not reported here, the regression analysis for per capita RGDP follow along the same lines as RGDP; going out only four quarters into the expansion shows no significant difference between crisis and noncrisis cycles. Going out further into the recovery, the difference is highly significant, again with the intercept dummy negative and the slope dummy positive.

Industrial production is a less compehensive measure of output, but it is also available monthly and so makes another natural comparison point. We use two data sets (both taken from Mark Watson's website) used in Watson (1994). For January 1884 to December 1940, we use the Miron and Romer (1990) IP numbers. For January 1947 to December 1990, we use Watson's post-war approximation of the Miron-Romer numbers, which re-weights the post-war index to approximate the Miron-Romer pre-war index. The gap means we cannot calculate statistics for the 1945 contraction, but as discussed above, we drop that from the results for other reasons. With these data, strong recoveries follow steep recessions in both crisis and non-crisis cycles. The response looks similar, and indeed is not signficantly different in a statistical sense, but we wish to stress that it certainly is not the case that crisis recoveries look weaker.

A third check for robustness is with different definitions of the business cycle. This is perhaps particularly important in our case because the early NBER cycles were determined before the development of the GDP concept, much less its implementation. As a result, cyclical troughs do not always coincide with troughs of RGDP. We use the Bry and Boschan (1971) approach to finding cyclical turning points, as implemented by the RATS program bryboschan.src. This results in a smaller number of business cycles (18 as oppoosed to 27) and a somewhat different distribution of crisis cycles--10 instead of 9. Table A1 reports the regression results. The striking change is that Bry-Boschan cycles show that a deep recession is followed by a strong recovery, regardless of whether or not there is a financial crisis. It remains true, however, that the response of the two groups is (statistically and economically) different: as before, crisis cycles have a lower intercept but a higher slope. Again the bounce back is stronger for a good sized crisis cycle.

Another possible relationship between contractions and expansions concerns their shape. One obvious measure of shape is steepness, or change in output divided by duration. Tables A2 and A3 report the results of regressing the steepness of the recovery against the steepness of the contraction, again dropping 1929 and looking at post-Fed and post-World War II periods.

Steepness does pick out differences between recoveries with and without financial crises. The financial crisis dummy is significantly negative in six of eight cases, turning positive for the post-World War II period both times. So everything else equal, crisis recoveries are smaller. But everything else is not equal, for the interaction term is significantly positive in the same six cases, while the coefficient on contraction depth is insignificant, small, and generally negative. The conclusion is that crisis recoveries show a strong relationship between contraction depth and recovery strength, but the noncrisis recoveries do not. Far from overturning the stylized fact, crisis recoveries account for it!

Steepness is not the only measure of shape. For example, the contraction might be "L" shaped, that is, dropping quickly at first, but then only slowly reaching a trough (Macroeconomic Advisers 2009). The recovery might be slow, with a "U" shape as opposed to a quicker recovery yielding a "V" shape. Some even described the 2007 cycle as having a "square root sign" shape. Quantifying shape appears a daunting task, but Harding and Pagan (2002) have a measure of how much the change in output deviates from a straight line. The specifications we have considered, however, have not uncovered any meaningful relationship between the shape of the contraction and the shape of the recovery.

Another approach preserves the interaction between the simple dummy for financial crises but adds controls for financial conditions during the recovery. To capture both the market-based and the intermediary-based aspects of finance, we include stock prices and bank loans. While certainly the price of credit should matter, some authors (Owens and Schreft 1995) define a "credit crunch" as nonprice rationing of credit, and thus observable mostly from the quantity side. Schularick and Taylor (2012) use (annual) bank lending as a measure of credit conditions. The stock price index for 1875-1917 is the Cowles commission index, releveled to match the Standard and Poor index which begins in 1917. For bank lending, we construct a new quarterly series from 1882 to 2010 for all commercial banks, detailed in the Appendix.

Table A4 reports the results of regressing the expansion strength against contraction depth, the change in real loans over the expansion and the change in the stock index over the expansion. It is meant to uncover whether financing problems held back the recovery. Because adding dependent variables reduces the degrees of freedom, we only report results for the entire sample and for a subsample that excludes the 1929 cycle. The interaction term is positive in the entire sample. both at four quarters and duration, and positive, statistically significant and quantitatively large in three of the four cases considered. Again it appears that if there is a difference between cycles with and without a crisis, the rebound from a financial crisis is particularly strong if the recession was deep.

Looking at the controls, however, shows that financing matters for the recovery, though the results do depend on the horizon, particularly for bank lending. Loans are quantitatively and statistically significant except when 1929 is dropped from the four-quarter specification. Higher bank lending is associated with a stronger recovery. The effect on the stock index is more consistent across horizons. There is a positive relationship between changes in the stock index and the strength of recovery, both with and without the Great Depression. Of course, particularly with the stock market, causality is hard to determine, and the results may only be telling us that strong recoveries have bull markets.

It is perhaps more common to measure financial conditions via some sort of credit spread, and in our earlier work we followed that tradition and looked at the spread between Baa and safe bonds, or between different grades of railroad bonds for the nineteenth century. As mentioned in the main text, using a risk spread to define banking crisis makes 1990 look less compelling. Table A5 drops 1990 as a crisis, and results are similar to the steepness results. TABLE A1 Bry-Boschan Cycles: Expansion Growth on Contraction Amplitude Variable All No 29 Post-Fed Constant 4.841 ** 4.841 ** 6.139 *** (2.151) (0.215) (2.204) FinDum -5.536 ** -2.546 --6.924 *** (2.665) (0.300) (2.649) [C.sub.stp] 0.286 *** 0.286 ** 0.225 ** (0.104) (0.104) (0.104) FinDum * [C.sub.stp] 0.950 *** 0.432 * 1.081 *** (0.158) (0.2.30) (0.115) [R.sup.2] 0.85 0.34 0.95 Chow Test [chi square](2) 176.3 *** 1126 *** 13.52 *** Sig level 0.000 0.000 0.000 N 17 16 11 Note: HAC robust standard errors in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%. TABLE A2 Regressions of Four Quarter Expansion Steepness on Contraction Steepness Variable All No 29 Post-Fed Constant 1.833 *** 1.833 *** 1.713 *** (0.252) (0.252) (0.363) FinDum -0.906 *** -0.916 *** -0.923 * (0.298) (0.300) (0.473) [C.sub.stp] -0.776 ** -0.776 ** -0.771 ** (0.346) (0.346) (0.390) FinDum * [C.sub.stp] 1.585 *** 1.617 *** 1.487 *** [R.sup.2] (0.368) (0.366) (0.405) 0.42 0.41 0.39 Chow Test [chi square](2) 22.77 *** 23.48 *** 13.52 *** Sig level 0.000 0.000 0.001 N 27 26 17 Variable Post-WWII Constant 1.430 *** (0.310) FinDum 2.096 *** (0.593) [C.sub.stp] 0.366 (0.453) FinDum * [C.sub.stp] -3.787 *** [R.sup.2] (0.764) 0.30 Chow Test [chi square](2) 25.61 Sig level 0.000 N 11 Note: HAC robust standard errors in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%. TABLE A3 Regressions of Expansion Steepness on Contraction Steepness Variable All No 29 Post-Fed Constant 1.965 *** 1.965 *** 2.068 *** (0.361) (0.361) (0.487) FinDum -1.3741 *** -1.371 *** -1.513 *** (0.381) (0.382) (0.540) [C.sub.stp] -1.475 ** -1.475 ** -1.652 ** (0.627) (0.627) (0.641) FinDum * [C.sub.stp] 2 475 *** 2.468 *** 2.616 *** (0.639) (0.643) (0.649) [R.sup.2] 0.48 0.47 0.54 Chow Test [chi square](2) 123.45 *** 22.972 *** 17.615 *** Sig level 0.000 0.000 0.000 N 27 26 17 Variable Post-wwn Constant 1.467 *** (0.269) FinDum 2.126 *** (0.717) [C.sub.stp] 0.655 ** (0.3050) FinDum * [C.sub.stp] -4.253 *** (1.037) [R.sup.2] 0.56 Chow Test [chi square](2) 38.90 Sig level 0.000 N 11 Note: HAC robust standard errors in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%. TABLE A4 Regressions of Expansion on Contraction Depth. Controlling for Change in Stock Price, and Bank Loans, 1887-2007 Variable All(4Q) No 29 (4Q) All (Dur) Constant 3.206 3.563 *** 3.723 *** (1.084) (1.168) (1.206) FinDum -0.400 -2.908 -5.158 *** (1.112) (1.793) (1.746) [C.sub.Amp] -0.535 -0.535 -0.719 (0.567) (0.557) (0.588) FinDum * [C.sub.Amp] 0.853 1.219 ** 1.909 *** (0.558) (0.573) (0.560) Loan 0.307 *** 0.200 0.199 *** (0.093) (0.131) (0.071) Stock 0.127 *** 0.143 *** 0.146 *** (0.041) (0.042) (0.044) [R.sup.2] 0.32 0.33 0.68 Chow Test [chi square](2) 2.35 5.45 * 18.48 *** Sig level 0.309 0.066 0.000 N 26 25 26 Variable No 29 (dur) Constant 3.031 ** (1.252) FinDum -1.025 (1.762) [C.sub.Amp] -0.652 (0.573) FinDum * [C.sub.Amp] 1.243 ** (0.532) Loan 0.361 *** (0.110) Stock 0.108 *** (0.041) [R.sup.2] 0.41 Chow Test [chi square](2) 5.62 * Sig level 0.060 N 25 Note: HAC robust standard errors in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%. TABLE A5 Regressions of Expansion Growth on Contraction Amplitude, Excluding 1990 Variable All No 29 Post-Fed Post-WWII Constant 7.229 *** 7.229 *** 6.281 *** 5.159 *** (1.089) (1.089) (1.025) (1.251) FinDum -2.825 -0.860 -3.351 8.798 *** (1.988) (1.482) (2.314) (1.787) [C.sub.Amp] -0.790 -0.790 -0.677 0.415 (0.573) (0.573) (0.737) (0.488) FinDum * [C.sub.Amp] 1.847 1.425 ** 1.854 ** -2.290 *** (0.550) (0.587) (0.735) (0.531) [R.sup.2] 0.58 0.23 0.66 0.39 F-Test [chi square](2) 13.00 *** 6.23 ** 7.87 ** 24.31 *** Sig level 0.001 0.04 0.02 0.00 N 26 25 16 10 Note: HAC robust standard errors in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%. ABBREVIATIONS FDIC: Federal Deposit Insurance Corporation FOMC: Federal Open Market Committee NBER: National Bureau of Economic Research RGDP: Real Gross Domestic Product

doi: 10.1111/ecin.12374

Online Early publication July 19,2016

REFERENCES

Baker, S. R., N. Bloom, and S. J. Davis. "Measuring Policy Uncertainty." NBER Working Paper No. 21633, 2015.

Balke, N. S., and R. J. Gordon. "Appendix B: Historical Data," in The American Business Cycle: Continuity and Change, edited by R. J. Gordon. NBER Studies in Business Cycles, Vol. 25. Chicago: University of Chicago Press, 1986, 781-850.

Barro, R. J., and T. Jin. "On the Size Distribution of Macroeconomic Disasters." Econometrica, 79(5), 2011, 1567-89.

Beauchemin, K. R. "Not Your Father's Recovery?" FRB Cleveland Economic Commentary, September 9, 2010.

Benati, L. "Why are Recession Associated with Financial Crises Different?" University of Bern Working Paper, 2012.

Bernanke, B. S. "Nonmonetary Effects of the Financial Crisis in the Propagation of the Great Depression." American Economic Review, 73, 1983, 257-76.

Bordo, M. D. "The Federal Reserve's Role: Actions Before, During, and After the Panic in the Historical Context of the Great Contraction," in Across the Great Divide: New Perspectives on the Financial Crisis, Chapter 6, edited by M. N. Bailey and J. Taylor. Stanford. CA: Hoover Institution Press, 2014.

Bordo, M. D., and J. G. Haubrich. "Credit Crises, Money and Contractions: An Historical View." Journal of Monetary Economics, 57(1), 2010. 1-18.

Bordo, M. D., and J. G. Haubrich. "Deep Recessions, Fast Recoveries, and Financial Crises: Evidence from the American Record." NBER Working Paper No. 18194, 2012.

Bordo, M. D., and J. S. Landon-Lane. "The Global Financial Crisis of 2007-2008: Is It Unprecedented?" in Global Economic Crisis: Impact, Transmission and Recovery, Chapter 2, edited by M. Obsfeld, D. Cho, and A. Mason. Cheltenham, UK: Edward Elgar, 2012, 19-56.

Bordo, M. D., and C. M. Meissner. "Fiscal and Financial Crises." NBER Working Paper No. 22059, 2016.

Bry, G., and C. Boschan. "Cyclical Analysis of Time Series: Selected Procedures and Computer Programs." NBER Technical Paper No. 20, National Bureau of Economic Research, Columbia University Press, New York, 1971.

Cerra, V., and S. C. Saxena. "Growth Dynamics: The Myth of Recovery. "American Economic Review, 98(1), 2008, 439-57.

Cogley, T. "International Evidence on the Size of the Random Walk in Output." Journal of Political Economy, 98(3), 1990, 501-18.

Cole, H. L., and L. E. Ohanian. "New Deal Policies and the Persistence of the Great Depression: A General Equilibrium Analysis." Journal of Political Economy, 112(4), 2004, 779-816.

Eichengreen, B., and M. D. Bordo. "Crises Now and Then: What Lessons from the Last Era of Financial Globalization?" NBER Working Paper No. 8716, 2002.

Friedman, M. "The Monetary Studies of the National Bureau," in The Optimum Quantity of Money and Other Essays, Chapter 12, by M. Friedman. Chicago: Aldine Publishing Company, 1969, 262-85 (Originally from "The National Bureau Enters Its 45th Year," 44th Annual Report. New York, 1964).

Friedman, M. "The 'Plucking Model' of Business Fluctuations Revisited." Economic Inquiry, 31, 1993, 171-77.

Friedman, M., and A. J. Schwartz. A Monetary History of the United States, 1867-1960. Princeton, NJ: NBER and Princeton University Press, 1963.

Gali, J., F. Smets, and R. Wouters. "Slow Recoveries: A Structural Interpretation." NBER Working Paper No. 18085, 2012.

Gordon, R. J., and R. Krenn. "The End of the Great Depression 1939-41: Policy Contributions and Fiscal Multipliers." NBER Working Paper No. 16380. 2010.

Gorton, G. B., and E. Tallman. "How Did Pre-Fed Banking Panics End?" Federal Reserve Bank of Cleveland Working Paper No. 16-03, 2016.

Gourio. F. "Disaster and Recoveries." American Economic Review, 98(2), 2008, 68-73.

Hall, R. E. "The Long Slump." American Economic Review, 101(2), 2011, 431-69.

Harding, D., and A. Pagan. "Dissecting the Cycle: A Methodological Investigation." Journal of Monetary Economics, 29(2), 2002, 365-81.

Hetzel, R. L. The Monetary Policy of the Federal Resen'e: A History. Cambridge: Cambridge University Press, 2008.

--. The Great Recession Market Failure or Policy Failure? Cambridge: Cambridge University Press, 2014.

Howard, G., R. F. Martin, and B. A. Wilson. "Are Recoveries from Banking and Financial Crises Really So Different?" International Finance Discussion Papers No. 1037, Board of Governors of the Federal Reserve System, 2011.

Jaimovich, N., and H. E. Siu. "The Trend Is the Cycle: Job Polarization and Jobless Recoveries." NBER Working Paper No. 18334, 2012.

Jalil, A. J. "A New History of Banking Panics in the United States, 1825-1929: Construction and Implications." American Economic Journal: Macroeconomics, 7(3), 2015, 295-330.

Jorda, O., M. Schularick, and A. M. Taylor. "When Credit Bites Back: Leverage, Business Cycles, and Crises." FRBSf Working Paper No. 2011-27, 2011.

Kahn, J. A., and R. W. Rich. "Tracking the New Economy: Using Growth Theory to Detect Changes in Trend Productivity." Journal of Monetary Economics, 54, 2007, 1670-1701.

Laeven, L., and F. Valencia "Systemic Banking Crises Database: An Update." IMF Working Paper No. 12/163, 2012.

Lopez-Salido, D., and E. Nelson, "Postwar Financial Crises and Economic Recoveries in the United States." Working Paper, May 21, 2010.

Macroeconomic Advisers. "The Shape of Things to Come." MA Macro Focus, 4(6), 2009. 1-11.

Meltzer, A. H. A History of the Federal Reserve, Volume 2, Book 1: 1951-1969. Chicago: University of Chicago Press, 2010.

Miron, J. A., and C. D. Romer. "A New Monthly Index of Industrial Production, 1884-1940." Journal of Economic History, 50(2), 1990, 321-37.

Morley, J., and J. Piger. "The Asymmetric Business Cycle." Review of Economics and Statistics, 94(1), 2012, 208-21.

Owens, R. E., and S. L. Schreft. "Identifying Credit Crunches." Contemporary Economic Policy, 13(2), 1995, 63-76.

Papell, D. H., and R. Prodan. "The Statistical Behavior of GDP after Financial Crises and Severe Recessions." The B.E. Journal of Macroeconomics, Special Issue: Long-Term Effects of the Great Recession, Article 2, 2012, 1-29.

Reinhart, C. S., and K. S. Rogoff. This Time Is Different: Eight Centuries of Financial Folly. Princeton. NJ: Princeton University Press, 2009.

Roesch, S. R. "The FOMC in 1974: Monetary Policy during Economic Uncertainty." Federal Resene Bank of St. Louis Review, 57(4), 1975, 2-13.

Romer, C. D., and D. H. Romer. "New Evidence on the Impact of Financial Crises in Advanced Countries." NBER Working Paper No. 21021, 2015.

Roubini, N. "A Phantom Recovery?" Project Syndicate No. 14, August 2009.

Schularick, M., and A. M. Taylor. "Credit Booms Gone Bust: Monetary Policy, Leverage Cycles, and Financial Crises, 1870-2008." American Economic Review, 102(2), 2012. 1029-61.

Silber, W. L. When Washington Shut Down Wall Street: The Great Financial Crisis of 1914 and the Origin of America's Monetary Supremacy. Princeton, NJ: Princeton University Press, 2007.

Stock, J. H., and M. W. Watson. "Disentangling the Channels of the 2007-2009 Recession," NBER Working Paper No. 18094, 2012.

Watson, M. W. "Business-Cycle Durations and Postwar Stabilization of the U.S. Economy." American Economic Review, 84(1), 1994. 24-46.

Wicker. E. R. Banking Panics of the Gilded Age. Cambridge: Cambridge University Press, 2000.

Wynne, M. A., and N. S. Balke. "Are Deep Recessions Followed by Strong Recoveries?" Economics Letters, 39, 1992, 183-89. "

--. "Recessions and Recoveries." Economic Review, Q1, 1993, 1-17.

Zamowitz, V. Business Cycles: Theory, Histoiy, Indicators, and Forecasting. Chicago: University of Chicago Press, 1992.

(1.) The November 1890 Baring crisis in London led to a sudden stop of capital flows to all emerging markets, including the United States. The issuance of clearinghouse loan certificates (Wicker 2000. chapter 3) quickly stemmed a local banking panic in New York in November. Because it was averted we do not include it in the narrative.

(2.) The demarcation of banking panics in Eichengreen and Bordo (2002) is based on Sprague (1912) and Friedman and Schwartz (1963). Kemmerer (1912), Wicker (1991). and more recently Jalil (2015) consider a broader range including 1890, 1896, and 1902.

(3.) Reinhart and Rogoff have a more liberal definition of financial crisis than do Eichengreen and Bordo (2002), whose chronology we follow. See Bordo and Meissner (2016) for a comparison of the dating of financial crises across databases. There is likewise considerable debate over measuring crisis severity. On the one hand, the Friedman and Schwartz view is that the banking panics pre-1934 had serious consequences because they led to bank failures and declines in the money supply. On the other hand, others measure the capitalized losses of the banking sector in the earlier period, adding bailout costs in the later period. The IMF (Laeven and Valencia 2012) measures losses by the decline in real output from the crisis to the recovery. And of course there is the tradition stemming from Bemanke (1983) which looks at increases in the cost of credit intermediation.

MICHAEL D. BORDO and JOSEPH G. HAUBRICH *

* The views expressed here are solely those of the authors and not necessarily those of the Federal Reserve Bank of Cleveland or the Board of Governors of the Federal Reserve System. We wish to thank David Altig, Luca Benati, John Cochrane, and Ricardo Reis for helpful comments, and participants at the Swiss National Bank conference on Policy Challenges and Developments in Monetary Economics, the New York Area Monetary Policy Workshop, the Julis-Rabinowitz Conference, at the Federal Reserve Bank of Dallas, Stanford, Claremont, UCLA, Santa Clara, UC Santa Cruz, Notre Dame, and the Reserve Bank of New Zealand. Patricia Waiwood provided excellent research assistance.

Bordo: Board of Governors Professor of Economics, NBER and Hoover Institution, Rutgers University, New Brunswick, NJ 08901. Phone 732 932 7069. Fax 732 932 7416, E-mail bordo@econ.rutgers.edu

Haubrich: Vice President and Economist, Federal Reserve Bank of Cleveland, Cleveland, OH 44101-1387. Phone 216 579 2802. Fax 216 579 3050, E-mail jhaubrich@clev.frb.org TABLE 1 Output Change in Business Cycles, 1880--2011 RGDP Change Peak Trough Crisis Date Contraction Recovery March 1882 May 1885 June 1884 2.59 5.14 March 1887 April 1888 No -3.36 2.32 July 1890 May 1891 No -1.03 13.27 January 1893 June 1894 May 1893 -12.38 17.21 December 1895 June 1897 October 1896 1.67 4.86 June 1899 December 1900 No 2.92 15.39 September 1902 August 1904 No 3.34 31.37 May 1907 June 1908 October 1907 -24.13 23.54 January 1910 January 1912 No 7.71 11.55 January 1913 December 1914 August 1914 -19.38 24.28 August 1918 March 1919 No -23.86 -43.67 January 1920 July 1921 No -7.66 33.28 May 1923 July 1924 No 13.99 12.32 October 1926 November 1927 No -3.78 26.84 August 1929 March 1933 October -99.95 87.64 1930 (a) May 1937 June 1938 No -21.96 27.52 February 1945 October 1945 No -122.94 -49.97 November 1948 October 1949 No 3.86 40.02 July 1953 May 1954 No -15.96 48.43 August 1957 April 1958 No -21.95 46.31 April I960 February 1961 No -4.04 42.22 December 1969 November 1970 No -1.80 50.50 November 1973 March 1975 1973-1975 -42.00 88.90 January 1980 July 1980 No -35.10 61.00 July 1981 November 1982 1982-1984 -42.40 153.42 July 1990 March 1991 1988-1991 -29.09 23.25 March 2001 November 2001 No 21.92 57.71 December 2007 June 2009 September -182.49 153.21 2008 Percent Change Peak Contraction Recovery March 1882 3.15 6.25 March 1887 -3.75 2.70 July 1890 -1.05 13.72 January 1893 -11.09 17.33 December 1895 1.41 4.04 June 1899 2.10 10.88 September 1902 2.09 19.23 May 1907 -11.82 13.08 January 1910 3.61 5.22 January 1913 -8.33 11.39 August 1918 -8.34 -16.66 January 1920 -3.42 15.39 May 1923 5.43 4.54 October 1926 -1.26 9.08 August 1929 -31.60 40.51 May 1937 -6.96 9.37 February 1945 -19.25 -9.69 November 1948 0.81 8.36 July 1953 -2.53 7.88 August 1957 -3.15 6.86 April I960 -0.54 5.62 December 1969 -0.16 4.45 November 1973 -3.18 6.96 January 1980 -2.23 3.96 July 1981 -2.64 9.81 July 1990 -1.36 1.10 March 2001 0.73 1.90 December 2007 -5.14 4.55 Notes: Recoveries are measured by the duration of contraction after the trough; RGDP in 1972 dollars. (a) Additional crises in March to June 1931 and January to March 1933. Source: Authors' calculations based on Bordo and JJaubrich (2010), Balke and Gordon (1986), and NIPA accounts. Crisis dates from Eichengreen and Bordo (2002) and Lopez-Salido and Nelson (2010) with monthly dating from Gorton and Tallman (2016). TABLE 2 Regressions of Four Quarter Expansion Growth on Contraction Amplitude Variable All No 29 Post-Fed Post-WWII Constant 6.434 *** 6.434 *** 5.658 *** 5 539 *** (1.044) (1.044) (1.301) (1.242) FinDum -0.5864 -2.360 * -1.254 -0.404 (1.293) (1.381) (1.536) (1.917) [C.sub.Amp] -0.565 -0.565 -0.510 0.756 (0.556) (0.556) (0.705) (0.541) FinDum * [C.sub.Amp] 0.796 1.219 ** 0.692 -0.816 (0.593) (0.559) (0.704) (0.740) [R.sup.2] 0.16 0.20 0.12 0.20 F-Test [chi square](2) 2.18 8.13 ** 1.574 2.605 Sig level 0.34 0.02 0.46 0.27 N 27 26 17 11 Note: HAC robust standard errors in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%. TABLE 3 Regressions of Expansion Growth on Contraction Amplitude Variable All No 29 Post-Fed Constant 1.229 *** 7.229 *** 6.281 *** (1.875) (1.090) (1.026) FinDum -3.663 * -2.251 -4.376 ** (2.246) (1.820) (1.776) [C.sub.Amp] -0.790 -0.790 -0.677 (0.573) (0.573) (0.738) FinDum * [C.sub.Amp] 1.886 *** 1.549 ** 1.893 ** (0.552) (0.603) (0.735) [R.sup.2] 0.51 0.23 0.67 Chow Test [chi square](2) 15.51 *** 7.48 ** 11.95 *** Sig level 0.004 0.02 0.003 N 27 26 17 Variable Post-WWII Constant 5 159 **** (1.251) FinDum -1.168 (2.538) [C.sub.Amp] 0.415 (0.488) FinDum * [C.sub.Amp] 0.109 (0.849) [R.sup.2] 0.06 Chow Test [chi square](2) 0.303 Sig level 0.86 N 11 Note: HAC robust standard errors in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%. TABLE 4 Regressions of Expansion Growth on Contraction Amplitude and RNR Crises Severity Variable All No 29 Post-Fed Constant 6.839 *** 7.207 *** 6.215 *** (1.144) (1.098) (1.036) FinDum 1.708 -0.755 0.111 (2.703) (1.884) (2.305) [C.sub.Amp -0.382 -0.767 -0.640 (0.462) (0.566) (0.727) RnRDum * [C.sub.Amp] 0.103 *** 0.316 *** 0.123 ** (0.031) (0.122) (0.048) R2 0.55 0.23 0.66 Chow Test [chi square](2) 18.23 *** 7.86 ** 13.70 *** Sig level 0.000 0.020 0.001 N 27 26 17 Variable Post-WWII Constant 4.811 *** (1.388) FinDum 0.764 (2.109) [C.sub.Amp 0.761 (0.60605) RnRDum * [C.sub.Amp] -0.227 ** (0.124) R2 0.14 Chow Test [chi square](2) 4.43 Sig level 0.109 N 11 Note: HAC robust standard errors in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%.