1890 and 1907: Contagion from Europe

In the post-World War II era, “sudden stops” of foreign investment triggered many financial crises in capital-importing countries of Latin America and Asia (Calvo and Reinhart 1996; Kaminsky, Reinhart, and Vegh 2003). According to many pre-1914 contemporaries and modern economists (e.g., Sprague 1910, p. 132; Fels 1959, p. 167; Friedman and Schwartz 1963, p. 104, 156; Eichengreen 1992, p. 51; Bordo 2006; Reinhart and Rogoff 2009, p. 243), the American crises of 1890 and 1907 were caused by sudden stops of foreign investment from Europe. This is plausible because an exogenous decline in foreign demand for American assets would tend to depress American stock prices and real activity. A drop in stock prices threatened New York banks as it could bring defaults on their loans to stock market “speculators.” A recession could bring defaults on business loans and threatened banks all over the country, draining their interbank deposits from New York.

Through most of the pre-1914 era, the United States imported capital from Europe. Europeans held large fractions of outstanding American securities (Wilkins 1989). At short maturities, Americans borrowed in London by selling “finance bills,” 1–3 month paper “drawn” on a London bank like a trade bill but collateralized by American stocks and bonds like New York time money (Margraff 1912, pp. 34–42). Through international arbitrage, a hike in European interest rates or a European sell-off of securities in general would tend to raise American interest rates and lower prices of American securities.

The 1890 episode was the most notorious European financial disturbance of the pre-1914 era, the “Baring Crisis.” Events in South America had caused European investors to fear defaults on Argentine bonds, depressing those bonds’ prices and endangering the solvency of London financial houses exposed to them, including Barings’. In July 1890 Barings’ and others began fire sales of good assets, including American securities, to raise liquid funds (Noyes 1898, p. 157; Lauck 1907, p. 64; The Economist (Supplement, Feb. 21, 1891); Wilkins 1989, pp. 194, 222, 471). Meanwhile, a monetary tightening by the Bank of England hiked rates on London bills, including American-issued finance bills (Sprague 1910, p. 133).

The year 1907 saw sharp hikes in European bill rates, due to tightening by the Bank of England and other European central banks in response to a persistent international gold drain. According to Kerry A. Odell and Marc D. Weidenmier (2004), the gold drain reflected payments by European insurance companies for damage from the San Francisco earthquake of 1906. The Bank took further actions to choke off American finance bill borrowing, specifically (Sprague 1910, p. 241; Noyes 1898, p. 358; Sayers 1976, pp. 54–56).

An increase in American interest rates due to foreign events, hence exogenous to American conditions, would tend to depress American real activity and could precipitate a business cycle downturn. Calomiris and Gorton (1991) argue that cyclical downturns were the proximate cause of all of the crises listed in Table 1: when people perceived a downturn was underway, they feared a wave of defaults on banks’ business loans; panics in the hinterland due to fears for local banks were transmitted to New York as out-of-town banks withdrew bankers’ balances to build up their own cash reserves. To support their argument, Calomiris and Gorton point out that all crises occurred in the midst of cyclical downturns according to the NBER business cycle chronology, and all crises were preceded by big drops in stock prices. They argue these stock price declines indicated the public expected a recession, which would hit corporate earnings.

Aside from its role as an indicator of expected recession, a drop in stock prices could be a more direct result of foreign financial disturbance. Stock prices reflect required returns to long-term securities as well as expected corporate earnings. An increase in required returns due to international factors would depress American stock prices for any given expectations of future earnings. A drop in stock prices was also a direct threat to the solvency of New York banks, especially. Some New York banks lent to brokerages and other stock “speculators” who could not repay if stock prices fell sharply. The problem was not ordinary callmoney or time money loans. These were collateralized with large “haircuts”—margins between collateral value and loan amount—to protect the lender against drops in collateral value.Footnote ³ We have found no evidence of runs due to fears of losses on such loans. The problem was rather that banks made large unsecured loans to speculators, sometimes illicitly, and depositors did not know exactly which banks had done so. Some of the most conspicuous New York runs in 1884, 1890, and 1907 occurred for this reason (Hanes and Rhode 2012).

Figure 3

STOCK PRICES AND INDUSTRIAL PRODUCTION, 1879–1913

Note: Not all provinces were in existence at all times. Bosnia-Hercegovina was an Austrian protectorate between 1878–1908, after which it was annexed. See Appendix 1 for details.Source: See the Data Appendix.

All the crises in Table 1 were, indeed, preceded by large declines in stock prices. All but one took place during cyclical downturns—the exception is, again, 1890. These patterns are apparent in Figure 3, which plots a monthly index of American stock prices and a seasonally adjusted version of Miron and Christina D. Romer's (1990) monthly industrial production (IP) index. The Miron-Romer index begins with January 1884. For earlier years, we link it to Joseph H. Davis’ (2004) annual IP index, which covers the entire nineteenth century.Footnote ⁴ Right-hand columns of Table 1 give dates of business cycle turning points around crises according to the old NBER chronology and two other chronologies based on more recent research. Romer's (1994) is based on the monthly IP index so it does not cover 1884. She shows that the NBER chronology, which was developed prior to the adoption of the modern definition of business cycles, places some peaks too early and includes some nonexistent downturns (when nominal variables fell but real variables continued to grow). Davis’ (2006) peaks and troughs are simply those apparent in his annual IP index. All the chronologies place a downturn prior to the crises of 1884, 1983, 1896, and 1907. Only the NBER's dates a downturn near 1890.

What About 1884, 1893 and 1896?

Like 1890 and 1907, the crises of 1884, 1893, and 1896 were preceded by cyclical downturns and stock-price declines. But none can be plausibly attributed to a foreign financial crises or central bank tightening.Footnote ⁵ A political factor that contributed to these crises, but did not trigger them, was the perceived danger that the United States would link the dollar to silver and float against gold. “Silver risk” contributed to crises by reducing international demand for American securities.Footnote ⁶ Silver risk appears to have become negligible after the 1896 presidential election, which was taken to signal firm commitment to the gold standard (Friedman and Schwartz 1982, pp. 513–17). The literature does not claim that exogenous hikes in silver risk set off crises but rather that potential silver risk was crystallized by, and amplified, cyclical downturns and financial crises already underway.Footnote ⁷ The more-or-less exogenous political events that created silver risk—the 1878 Bland-Allison Act, the 1890 Sherman Silver Purchase Act—occurred years before the crises in question.

Monetary Shocks from Agriculture

The quantity of a crop that is harvested varies greatly from year to year. Some harvest fluctuations may be endogenous responses to macroeconomic conditions, but agricultural output is also uniquely affected by obviously exogenous factors: weather and pests. Scholars have considered two possible channels from harvest fluctuations to pre-1914 American financial crises. The first runs through high-powered money demand. America's three great staple crops, cotton, corn, and wheat, were planted in or before the spring and harvested within a span of months from late summer through autumn—the harvest season.Footnote ⁸ Farmers were paid for their crops, and farmworkers received most of their wages, during or shortly after the harvest, creating a surge in demand for cash which fell off gradually to a low in early summer. The surge was accommodated by cash shipments from the northeast, funded largely by withdrawals of bankers’ balances from New York (Sprague 1903; Kemmerer 1910). V. V. Chari (1989) speculates that American financial crises occurred in years when agricultural conditions created especially large surges in cash demand and withdrawals of banks’ balances, leaving New York banks vulnerable to further shocks. To test this, Calomiris and Gorton (1991) look for a relation between harvest size and occurrence of crises, on the assumption that rural cash demand was positively related to crop production. They conclude “seasonal money demand shocks originating in the countryside cannot possibly be the cause of panics” (p. 148), as they find “years in which panics occurred in the fall… were not years of unusually large harvests for corn, wheat, and cotton” (p. 137).

A second possible channel runs through revenues from crop exports and the potential link between net exports and the money supply. Under the gold standard, for most countries the sum of net exports, international capital inflow and foreign investment income was balanced by an inflow of monetary gold from other countries. In the United States, the supply of high-powered money to banks and the public consisted of monetary gold plus currency (greenbacks, silver notes, national banknotes, silver coins) minus money held in U.S. Treasury vaults (Friedman and Schwartz 1963, pp. 124–34). Currency growth was unresponsive to economic conditions and mostly out of the Treasury's control.Footnote ⁹ On occasion, Treasury officials deliberately managed their reserves of vault cash to affect the money supply (Timberlake 1978) but these operations were limited by the degree of slack between expenditures, tax receipts, and bond operations. The Treasury did not generally sterilize international gold flows. Thus, the change in the American high-powered money supply was linked to export revenue, as an element in the balance of payments.

Cotton and wheat (raw wheat and wheat flour) were important American exports. For both crops, export revenue was positively related to the size of the harvest, and export-revenue fluctuations were large relative to the high-powered money supply (Davis, Hanes, and Rhode 2009). Thus, fluctuations in cotton or wheat harvests could produce substantial exogenous shocks to potential high-powered money supply growth. Contemporaries believed cotton and wheat harvests affected financial conditions through this channel (Sprague 1903, p. 50; Andrew 1906). Note it implies unusually large harvests of cotton or wheat would tend to reduce, not raise the probability of a financial crisis.

Milton Friedman and Anna Jacobson Schwartz (1963) argue that the crisis of 1893 was triggered by fluctuations in wheat harvests through this money-supply channel and the classic specie-flow mechanism. Wheat and wheat flour export revenue was extraordinarily high following the harvest of autumn 1891, partly because of a large harvest, an “accident of weather” (p. 107). The resulting gold inflow “fostered a spurt in the stock of money from 1891 to 1892” that interrupted a longer-term deflationary trend. A reversion to deflation in 1893 caused bank runs because “[l]oans that would have been good and banks that would have been solvent if prices had been stable or rising became bad loans and insolvent banks under the pressure of price deflation” (p. 108). Wheat export revenues remained low until “another one of those fortuitous combinations of good harvests at home and poor harvests abroad” in 1897, which contributed to recovery from the 1896 crisis (p. 140).Footnote ¹⁰

Davis, Christopher Hanes, and Paul W. Rhode (DHR) (2009) examine the relation between pre-1914 American business cycles and harvests of all three of America's staple crops, cotton, wheat, and corn, constructing instruments for harvest fluctuations from weather data. DHR find that, within 1879–1913, the size of the cotton harvest had positive effects on the following year's industrial production, gold inflows, and high-powered money supply growth, and negative effects on commercial paper interest rates. Wheat harvests had positive effects on gold inflow and money growth but no apparent effects on industrial production or interest rates. Corn harvests affected none of these variables. With respect to cotton, DHR find no systematic differences between OLS and IV coefficient estimates, and no evidence of reverse causality from business cycles to harvest fluctuations. Thus, it appears cotton harvest fluctuations in general, not just those due to weather, were exogenous to the macroeconomy.

DHR argue that a crop's macroeconomic effects reflected the relative intensity of its more immediate effects on export revenue and rural cash demand. To the degree that a bumper harvest boosted export revenue, it drew in gold, lowered interest rates, and stimulated real activity. To the degree that a harvest boosted cash demand, it drew in gold but raised interest rates and depressed real activity. Corn harvests had no apparent macroeconomic effects because they had little effect on export revenue or cash demand: little corn (or corn meal) was exported; relatively little was even sold off the farm for cash income. Wheat harvests had strong positive effects on export revenue and cash demand, so wheat affected gold flows but not, on net, interest rates and real activity. Cotton harvests had strong effects on export revenue, but, unlike wheat, relatively weak effects on cash demand. Thus, a big cotton harvest lowered interest rates and raised industrial production. Cotton's export-revenue effect was relatively strong partly because cotton exports were a large share of domestic production: export revenue was very large relative to farmers’ income from sales of the crop. For wheat, exports (in the form of wheat and wheat flour) were a smaller share of production, so export revenue was smaller relative to crop sales. It is also possible, DHR argue, that cotton's cash-demand effect was relatively weak because Southern agricultural institutions tended to substitute merchants’ and landlords’ credit for cash balances (pp. 1682–86, 1708–11).

DHR do not examine the relation between harvests and financial crises. In this article, we take on that task.

Figures 4 and 5 plot fluctuations in wheat and cotton harvests and export revenues. The top line in each figure is the deviation from a quadratic time trend (over 1869–1913 to match results we will present below) in the log of a year's autumn harvest, in bushels or bales. The lower line is the deviation from a similar quadratic time trend in “real” revenue from exports of each crop (cotton or wheat and wheat flour).Footnote ¹¹

Figure 4

WHEAT HARVEST AND DEFLATED EXPORT* REVENUE, 1879–1913 LOGS, DEVIATIONS FROM TREND

*Wheat and wheat flour.Sources: See the Data Appendix

From the figures, it is obvious that harvest fluctuations were just one factor affecting crop export revenue. Wheat revenue was high in 1907 despite a poor harvest. Cotton revenue was low in 1898 and 1908 despite big harvests. Some of the other factors affecting export revenue must have been endogenous to macroeconomic conditions: e.g., industrial demand for raw materials. Thus, we do not argue that export-revenue fluctuations in general were exogenous shocks to the macroeconomy. Our focus is on export-revenue fluctuations due to harvest fluctuations.

Figure 5

COTTON HARVESTS AND DEFLATED EXPORT REVENUE, DEVIATIONS FROM TREND, 1879–1913

Sources: See the Data Appendix

Recall Friedman and Schwartz attribute the crisis of 1893 to a drop-off in wheat export revenue after 1891. That drop-off is apparent in Figure 4. It is clearly due, at least partly, to a decrease in the size of the American harvest. But Figure 5 shows something Friedman and Schwartz missed. The lower line of Figure 5 indicates that the cotton harvest prior to 1893, that is the harvest of autumn 1892, generated extremely low export revenue. The harvests prior to 1884 and 1896 also generated extremely low export revenue. The two crises that can be accounted for as the result of sudden stops, 1890 and 1907, were different: neither was preceded by a cotton harvest that generated low export revenue. The crisis of November 1890 was coincident with a cotton harvest that generated extremely high export revenue. The upper line of Figure 5 shows that the low cotton revenues of 1883, 1893, and 1896 were all associated with poor harvests. The high revenue of 1890 was associated with a bumper crop. Importantly, according to contemporaries the poor harvests of 1883, 1892, and 1895 and the good harvest of 1890 were all due to the exogenous factors of pests and weather.Footnote ¹²

To check whether the relation between financial crises and cotton harvest fluctuations apparent in Figure 5 holds for wheat as well, we ran a probit regression with the left-hand side variable equal to one in years with crises, and the previous years’ cotton and wheat harvest fluctuations (as in the figures) on the right-hand side, over 1879–1913 (harvests of 1878–1912). To account for the foreign financial shocks of 1890 and 1907, we removed those years from the sample.Footnote ¹³ The coefficient on cotton is negative and significantly different from zero at the 5 percent level; the wheat coefficient is much smaller and far from significance:

Money Supply Growth, the Balance of Payments, and the Current Account

In the gold standard United States, growth in the high-powered money supply from one period to the next was determined by the net inflow of monetary gold, the change in the supply of non-gold currency, and the change in the quantity of high-powered money sequestered in Treasury vaults. Thus, we describe the change in the log of the real money supply across a span of time as Δ(m – p) = g + z − Δ p where m is the high-powered money supply, p is the American price level, g is the inflow of monetary gold, and z represents the other money supply factors (expressed as a fraction of the previous period's money supply).

The balance of payments is the sum of the current account and the capital account. The current account—net exports less the net outflow of factor payments and transfers—might be affected by the level of real nonagricultural activity, denoted y, through American demand for imports. Net exports might also be affected by the relative price of foreign goods, as determined by the domestic price level p, the foreign price level p*, and exchange rates. Variations in the key exchange rate, against sterling, were large enough to affect short-term returns to foreign assets but too small to matter for trade.Footnote ¹⁴ Thus, we describe the current account (in real value, scaled by the average magnitude of its fluctuations relative to trend money growth) as a function CA(y⁻,(p –p*)⁻) + ɛ^CA. Superscripts in the function give the sign of a partial derivative: hence CA_y < 0, CA_{(p – p*)} < 0. The disturbance term ɛ^CA represents exogenous shocks to the current account, such as variations in export revenue due to exogenous fluctuations in an export crop's harvest.

The Capital Account and Imperfect Capital Mobility

The capital account reflects the change in net international demand for a country's assets. International demand for American assets must have been sensitive to returns on American versus European assets. Europeans traded American stocks and bonds in European markets as well as New York, with rapid arbitrage across the Atlantic. There was also arbitrage at the shorter maturity of bills. London was the central bill market of Europe.Footnote ¹⁵ Americans financing security portfolios shifted between London finance bills and New York time money depending on relative interest rates and expected exchange rates (Morgenstern 1959, p. 527). Some firms borrowed in London through finance bills to lend time money in New York (Margraff 1912, p. 41). Arbitrage between London bills and matching maturities in New York also operated through the mechanics of foreign exchange (Goodhart 1969). Except at times when perceived silver risk was high, expected future changes in the exchange rate were small relative to variations in interest rates. Thus, international arbitrage tended to equalize both bill maturity interest rates and expected returns to long-term securities between New York and European financial centers, especially London.

Within New York, commercial paper was a very close substitute for time money, from a lender's point of view. Unlike today, there was no active secondary commercial paper market, so commercial paper was about as illiquid as time money loans. Thus, changes in New York commercial paper rates were highly correlated with changes in time money rates.Footnote ¹⁶ Commercial paper was distributed throughout the United States by national dealers. Outside New York, it was the only available short-term open market investment. For a hinterland bank, commercial paper rates were the opportunity cost of short-term loans to local borrowers. Through commercial paper, therefore, short-term interest rates across the United States were linked to New York rates, and therefore European rates.

However, there is no reason to believe international arbitrage was sufficient to equalize expected returns in the United States and Europe, to the extreme of perfect capital mobility. European investors did not view American and European assets as perfect substitutes: in London, realized risk-adjusted rates of return were consistently higher for American securities even after the diminution of silver risk in 1896 (Edelstein 1982, pp. 157–58). Many empirical studies have rejected the hypothesis that uncovered interest rate parity (UIP) held between London bills and New York time money or commercial paper (Giovannini 1993, pp. 133–36; Foster 1994).Footnote ¹⁷ Demand for American assets was as depicted in models of “imperfect” international capital mobility (e.g., Kouri and Porter 1974). We describe the capital account as a function KA(Δ (i − i* − ɛ^Ee)⁺) + ɛ^KA, where the benchmark foreign return i* would be best indicated by the open market bill rate in London. The corresponding American return i corresponds to interest rates for time-money or commercial paper, or to expected short-term returns to American stocks and bonds. ɛ^Ee is the expected increase in the New York price of London sterling, which would be affected by perceived silver risk. ɛ^KA accounts for any other factors that might affect demand for American assets.

Demand for High-Powered Money

As the expected asset return relevant for capital flow i was linked to commercial paper rates, it would be related to demand for high-powered money. For the public and out-of-town banks, commercial paper rates represented the opportunity cost of holding cash. Aggregates of deposits and cash held by the public for the pre-1914 era show a normal money-demand relation to interest rates and national income or wealth (Friedman and Schwartz 1982). Thus, we describe demand for real high-powered money balances as L(y⁺, i⁻) + ɛ^md. The disturbance term ɛ^md represents exogenous shocks to high-powered money demand, such as variations in rural demand for cash due to exogenous fluctuations in a crop's harvest.

Interest Rates, Output, and the Price Level

Relations between interest rates, the price level, and real activity were governed by a constraint on long-run trend inflation imposed by the gold standard. For all gold standard countries, trend inflation was largely determined by the balance between world gold production, growth in real output and financial innovations in the use of monetary gold. There was little, if any, serial correlation in inflation from year to year. Judging from the behavior of interest rates, expected inflation at both long and short horizons was always close to zero and remarkably unresponsive to economic events (Barsky 1987; Barsky and Delong 1991; Bordo and Dewald 2001).

Under stable trend inflation, an expectations-augmented Phillips curve of either the Old-Keynesian or New Keynesian variety implies a positive relation between inflation and output along the lines of the original Phillips curve, rather than the “accelerationist” Phillips curve of the postwar era (Alogoskoufis and Smith 1991; Hanes and James 2012). Thus, we describe the change in the price level Δp as a function of nonagricultural real activity π(y⁺)+ ɛ^p. (The last term accounts for any extraordinary changes in inflation expectations, as well as “supply shocks.”)

Also, a positive shock to nominal interest rates would raise real interest rates and reduce real spending. Empirical studies of postwar data generally find that an exogenous shock to interest rates affects real spending and output after about six months, with the strongest effect appearing after about a year (e.g., Christiano, Eichenbaum, and Evans 2005). We are aware of no evidence that the lag was shorter in pre-1914 America. Thus, we describe y in a period as determined by a function of the interest rate in the prior period E(i₋₁⁺, (p − p*)⁻)+ ɛ^IS. In this context, a period must be about six months. We allow for an effect of the relative American price level on the volume of spending directed to domestic output (and foreign demand for exports). The last term accounts for exceptions to the usual pattern for expected inflation (which would affect the relation between nominal and real interest rates). It also accounts for extraordinary disturbances to the operation of financial intermediaries which would affect the supply of credit to potential borrowers (as in Woodford 2010).

Effects of Harvest Fluctuations

An exogenous fluctuation in the harvest of cotton or wheat meant a shock to export revenue reflected in ɛ^CA, and a shock to high-powered money demand reflected in ɛ^md. The timing of the shocks was important. Foreign payments for cotton and wheat exports began within the autumn harvest season and continued through the early months of the following year (Kemmerer 1910, pp. 136–40). Taking a period to be about six months and letting h denote exogenous fluctuations in a crop's harvest, we describe the export-revenue effect of a harvest fluctuation as a disturbance ∂ɛ^CA/∂h > 0 that occurs within one period—period 1. Another period—period 2—passes before the next harvest season. The harvest's money-demand disturbance ∂ɛ^md/∂h occurs within period 1 and persists through period 2.Footnote ¹⁸ Let (x₁ − x₀) denote the change in a variable to the harvest period 1 from period 0 prior to the harvest. Within the harvest period, the size of the harvest cannot affect real activity y but it may affect the required asset return i₁ depending on the value of ∂ɛ^CA/∂h relative to ∂ɛ^md/∂h

If ∂ɛ^CA/∂h > ∂ɛ^md/∂h, the size of the crop's harvest has an immediate negative effect on i. If ∂ɛ^CA/∂h = ∂ɛ^md/∂h, the crop's harvest does not affect i, even though it affects export revenue. In the post-harvest period 2, real activity y₂ responds to the crop's effect on i in the harvest period, while i₂ reflects the response of y₂ as well as the initial disturbances to export revenue and money demand

Note that the harvest's effect on i may diminish from period 1 to period 2: that is because an increase in output tends to reduce net exports and boost high-powered money demand. Whether or not an export crop's harvest affects i and y, the size of the harvest will have a positive effect on gold inflow

where (g¹ + g²) is cumulative gold inflow since the pre-harvest period.

Our hypothesis is that, for wheat, (∂ɛ^CA/∂h − ∂ɛ^md/∂h) was about equal to zero. Thus, the wheat harvest affected gold flows, but not financial markets or real activity. For cotton, (∂ɛ^CA/∂h − ∂ɛ^md/∂h) was positive. Thus, the cotton harvest affected required returns to American assets, which is to say interest rates for commercial paper or time money, within the harvest period, and real nonagricultural activity in the post-harvest period.

As cotton harvests affected commercial paper rates within the harvest period, they could have affected the volume of deposits in New York City banks at the same time. Out-of-town banks held bankers’ balances partly to satisfy regulatory reserve requirements, but bankers’ balances were usually far in excess of requirements and varied substantially from quarter to quarter (Myers 1931, pp. 237–39; Goodhart 1969, pp. 76–77). For a bank outside New York, the commercial paper rate was the opportunity cost of holding New York bankers’ balances. New York banks paid interest on bankers’ balances, but they held the rate nearly fixed across the pre-1914 era (James 1978, pp. 176, 265–67). Thus, increases (decreases) in commercial paper rates due to cotton harvest fluctuations should have been associated with decreases (increases) in New York deposits.

As i represents required returns to securities as well as interest rates, our hypothesis implies that cotton harvests must have affected prices of American stocks and bonds, within the harvest period. For any given expectations of future real activity and corporate earnings, an increase (decrease) in required returns would depress (boost) securities’ prices. Additionally, financial market participants might forecast the upcoming effect of the cotton harvest on real activity. That would reinforce the effect of a change in required returns: on stock prices, by changing expectations of future corporate earnings; on bond prices, by changing probabilities of future defaults. In any case, in the post-harvest period the realized change in real activity would affect stock prices.

Through these channels, a poor cotton harvest could create all the conditions for a financial crisis by the post-harvest period. Thus, a poor cotton harvest could trigger a financial crisis. The occurrence of a financial crisis could be associated with many more disturbances, such as a decrease in ɛ^IS or an increase in ɛ^Ee (due to heightened silver risk). Many of these disturbances which would tend to reinforce the effects directly due to the harvest's export-revenue shock. But the effects described by equations 1–5 are independent of the occurrence of a financial crisis.

With respect to observable variables, our hypothesis implies that the size of the cotton harvest should have negative effects on commercial paper rates, and positive effects on New York deposits, stock, and bond prices, within the harvest season. The cotton harvest's effect on industrial production should appear with a lag, about six months after the appearance of effects on financial variables. All of these effects should appear prior to the occurrence of a financial crisis and whether or not a crisis follows. The wheat harvest should affect international gold inflow, but not financial variables. The corn harvest should not affect financial variables or gold inflow.

Call Money Rates

Our hypothesis does not have clear implications for call money rates. Call money was not subject to direct international arbitrage. There are many contemporary references to American borrowing from London at the maturity of time money, but we have found no references to trans-Atlantic lending at the overnight maturity.

A link between call money rates and i would come through time money or commercial paper. Generally, one would expect rates for these bill maturity assets to equal the expected return to call money lending over the appropriate horizon plus a liquidity premium. Descriptions of the call money market by contemporaries (e.g., Woodlock 1908; White 1910) and later observers (Myers 1930, pp. 265–87; Goodhart 1969) suggest the premium varied greatly and depended on a distinct supply and demand for call money funds. Borrowers attempted to finance a fairly stable fraction of their portfolios’ value with call money. They were reluctant to substitute call-money for time money even when call money rates were especially low, for fear they would be unable to roll over the loans (Woodlock 1908, p. 29). Thus, an increase in American securities’ prices raised the demand for call money loans as it raised the value of borrowers’ portfolios and collateral. On the supply side, the bulk of call money lending was by New York banks, which usually tried to match their holdings of this low return but uniquely liquid asset to the volume of bankers’ balances, a volatile liability. The supply of call money increased with the volume of bankers’ balances. According to Margaret G. Myers (1930, p. 275):

A study of call money liquidity premiums with proper accounting for the details of interest rate definitions, timing and causality is beyond the scope of this article. But the relation between call money rates, securities’ prices, and New York deposits posited by Myers is obvious in our data (Hanes and Rhode 2012). Thus, a poor cotton harvest could tend to lower call money rates by reducing time money rates, but it could also tend to raise call money rates by depressing securities’ prices and draining bankers’ balances.

Interaction Between Cotton Harvests and Foreign Financial Conditions

An increase in foreign interest rates due to tightening by foreign central banks corresponds to an increase in i*. A decrease in foreign demand for American assets due to fire sales by European investors in an event such as the Baring Crisis corresponds to a decrease in ɛKA. How would these shocks interact with the cotton harvest's effects on export revenue and money demand? Within the harvest period, the net effect on i would be

with the corresponding effect on real activity in the following period. Thus, as a big cotton harvest means a large value of (∂ɛ^CA − ∂ɛ^md), it would tend to counteract the financial and macroeconomic effects of interest rate hikes or other financial disturbances in Europe.

What If America Had a Central Bank?

If America had a central bank, its open market operations and discounting would affect z as they affected the supply of non-gold high-powered money—not only currency, but also financial institutions’ balances at the central bank. Thus, a central bank could adjust z in response to cotton harvests. Within the harvest period, the net effect on i would be

An American central bank could have counteracted the financial effects of cotton harvest fluctuations by ensuring that ∂z/∂h = − ∂ɛ^CA/∂h + ∂ɛ^md/∂h. A central bank could achieve this by buying (selling) domestic securities in response to a poor (abundant) harvest, or simply by setting its discount rate slightly above the normal market rate for the season. Cotton harvests would then, like wheat harvests, affect international gold flow but not financial markets or real activity.

Statistical Specifications

Crop production trends were defined to be quadratic in time over 1869–1913 to match DHR (2009). (HP trends or quadratic trends over 1878–1913 gave similar results.) We want to pin down the seasonal timing of harvests’ effects but the timing of harvests themselves varied from year to year. The cotton harvest, for example, could peak in September one year, October the next. To allow for some slack, we average monthly series up to a quarterly frequency.

Our main dependent variables are changes in stock prices, bank deposits, and so on from a pre-harvest value to Q3 or Q4 of the harvest year, and to each of the four quarters of the calendar year following the harvest. The pre-harvest value is a four-quarter average up through Q2 of the harvest year, for variables known to be subject to strong seasonal influences in the pre-1914 era (industrial production, commercial paper, NYCH bank deposits, call money, London bills, New York stock volume [on stock volume seasonals see Owens and Hardy 1925]). For other variables, the pre-harvest value is the average for harvest-year Q2. Immediate effects of a harvest, within the harvest season, should appear in regressions with changes to harvest-year Q3 or Q4 of the harvest year on the left-hand side. Delayed effects appear in results for quarters of the following year. To look for “effects” of harvests on financial variables prior to the harvest season, we run regressions where the dependent variables are changes to Q2 of the harvest year from harvest-year Q1 or a four-quarter average up through harvest-year Q1.

Our samples begin with the harvest of autumn 1879, the first after the United States resumed gold payments, and stop at the end of 1913. For most variables all-harvest samples have 35 observations for harvest-year Q3 and Q4, 34 observations for following-year Q1-Q4. In the tables, coefficients and standard errors are scaled to the standard deviation of a cotton harvest fluctuation (0.12 for harvests 1878–1912); standard deviations for wheat (0.13) and corn (0.14) were similar. Thus, the tables give estimated coefficients and SE's multiplied by 0.12. On a causal interpretation, they give the effect of a one-standard-deviation harvest fluctuation.