The Effect of Cash Injections: Evidence from the 1980s Farm Debt Crisis

Abstract

A large theoretical literature shows that in the presence of financial frictions, weak firm balance sheets detrimentally affect economic activity (see e.g., Bernanke and Gertler 1989, 1995; Shleifer and Vishny 1992; Gertler and Gilchrist 1994; Kiyotaki and Moore 1997). Strengthening firm balance sheets during a financial crisis is thus a much discussed and debated question.¹ By reducing financial frictions, such interventions may increase investment, support lending, raise employment, and mitigate the severity of a financial crisis.

Estimating the economic effect of interventions meant to strengthen firms’ balance sheets during a financial crisis is a difficult task, because the timing and strength of such interventions are likely to be endogenous and driven by the severity of the crisis itself. Interventions generally occur in response to severe crises, so a simple correlation would suggest, likely erroneously, that such interventions have detrimental effects on the economy.

To understand the effect of interventions that strengthen firm balance sheets during financial crises, we focus on the farm debt crisis of the 1980s. Assembling a yearly, county-level data set of weather and farm data in Iowa, our identification strategy relies on exploiting variation arising from random weather shocks. As a large literature in agronomics shows, weather shocks affect crop yields and hence farm cash flow (see, e.g., Deschênes and Greenstone 2007; Schlenker and Roberts 2009). Geographic variation in weather realizations thus provide plausibly exogenous variation in local cash flow and are akin to cash injections of differing magnitudes to farms operating across different counties. In this paper, we analyze and measure the effect of such exogenous cash injections during a financial crisis on both the real and financial sectors.

Spanning the period 1981-1987, the Farm Debt Crisis resembled in many ways the financial crisis of 2008-2009, with agricultural land prices and farm debt—much of it collateralized by land—substantially increasing prior to the crisis onset. Subsequently, during the crisis itself, land prices in the U.S. corn belt plummeted by nearly 50%. The farming sector saw severe deleveraging—total agricultural debt declining by 29% from 1984 to 1988—and experienced substantial disruptions with numerous farm bankruptcies as well as agricultural bank failures throughout the period.²

As a first step in our empirical analysis, we confirm that county-level weather variation is related to farm yields in our data. Focusing on corn production in Iowa, we measure how temporary shocks in weather during the corn growing season affect yields. Consistent with a large literature in agronomics, we find that corn is highly sensitive to small changes in temperature, with even a few additional hot days during the growing season reducing annual corn yields substantially.

Local weather affects yields, so weather shocks provide exogenous variation in local cash flows and farms’ net worth during the debt crisis. We exploit this variation in our empirical strategy by relating weather-driven cash flow shocks to a host of real and financial variables, both during the farm debt crisis and outside of it. Our analysis focuses on land markets, the propagation of shocks into the financial sector, and on labor markets. During normal times, firms should be able to smooth temporary shocks, but, in financial crises and other periods of large financial frictions, such smoothing is difficult, because external finance is often prohibitively costly or unavailable.³ An inability to smooth shocks during a crisis is, therefore, predicted to translate to a host of market outcomes, both real and financial. Consistent with financial accelerator theories, our results show that weather-driven cash injections during the crisis positively affect land markets, the financial sector, and labor markets in an economically meaningful manner, and ultimately translate into increases in county per capita income.

We start by examining the effect of weather shocks on agricultural land prices. We expect that during a financial crisis, increases in cash available to firms will support asset prices by mitigating fire sales. When financial frictions are high, asset prices will be affected by cash-in-the-market pricing, as economic agents cannot raise external finance to bring prices to fundamental value (Allen and Gale 1994, 1998; Shleifer and Vishny 1992; Kiyotaki and Moore 1997).⁴

Consistent with cash-in-the-market pricing, we show that weather-induced cash flow injections do indeed increase land prices. Because our specifications include both year and county fixed effects, our identification strategy is driven by comparing, within a given year, counties that received differential weather shocks, as compared to their sample mean. We find that counties that received a positive cash flow injection, which is driven by relatively good weather, exhibit higher land values than do counties that receive a negative cash flow shock, which is driven by a few additional days of high temperature weather during the growing season. Instrumenting for county-level crop yields with our weather shock variable, we find an elasticity of land prices to yields during the farm debt crisis of 0.34. As a placebo test, we rerun our analysis on the period outside of the farm debt crisis and find no statistically significant relation between land values and weather shocks. To our knowledge, this is the first study that provides causal evidence of cash-in-the-market pricing.

Because weather shocks affect both farm income and land values, our empirical methodology cannot separately isolate the impact of variation in income and land values on the additional economic outcomes analyzed below. Indeed, one of the key insights of Kiyotaki and Moore (1997) is that the income channel and the asset value channel (i.e., land values in the present context) are inherently and endogenously intertwined, with exogenous cash flow shocks affecting equilibrium asset values and asset values further affecting cash flows through their impact on collateral constraints. In discussing the impact of exogenous weather variation, we thus refer to the impact of weather-driven cash flow shocks, which should be understood as affecting farm net worth, both directly through the income channel (through the impact on yields) as well as indirectly through the associated change in land values.

We continue our analysis by examining how weather-driven cash flow shocks propagate into the financial sector. We first show that during the crisis, counties that experience reduced crop yields due to bad weather shocks exhibit higher agricultural loan delinquencies: as would be predicted, farms in these counties find it more difficult to repay their obligations. We then show that these county level cash injections reduce the probability of local bank failure during the crisis. The effect is economically significant, with a 10% drop in crop yields increasing the probability of a county bank failure by 3.2%.⁵ Thus, weather-driven cash flow shocks during the crisis create long lasting effects in the financial sector.

We then turn to the effect of cash flow injections during crises on labor markets. We begin by focusing on the labor market in the agricultural sector and then examine spillovers into labor markets in other sectors. The results show that counties that experience a negative cash flow shock during the crisis exhibit lower agricultural employment rates as well as a reduction in average county agricultural wages, consistent with a reduction in farms’ labor demand.⁶ During the debt crisis when financial frictions are high, lower firm net worth thus translate into labor market disruptions and decreased employment. We rerun the analysis outside of the farm debt crisis, and show that weather-driven cash flow shocks do not affect employment and wages during this time period, consistent with firms’ greater ability to smooth shocks when financial frictions and the cost of external finance are lower.

Next, we examine labor market spillover effects of cash flow shocks in the agriculture sector into the service sector.⁷ We hypothesize that during the debt crisis disruptions in agricultural labor markets following cash flow shocks will spill over into other labor markets. This is indeed what the data show. During the debt crisis, county-level weather-driven negative cash flow shocks in the agricultural sector are related to employment increases, as well as average wage decreases, in the local service sector. During the crisis, a negative cash flow shock in agriculture appears, therefore, to increase labor supply in the service sector, with workers reallocating from agriculture to services.

We continue by examining whether these labor market spillovers in the service sector depend on the share of agriculture in the local economy. Our hypothesis is that when the agriculture sector is large within a given county, reductions in agricultural employment following a negative cash flow shock will reduce local demand, and hence employment, in the service sector.⁸ Running interaction specifications conditioning on the share of agricultural income within the county, we find results consistent with this demand channel: in counties where farming is more dominant, during the farm debt crisis negative cash flow shocks in agriculture have a negative effect on employment within the service sector. Firms’ inability to smooth cash flow shocks during the debt crisis is thus transmitted to other industries located within the same area, as employees are dislocated within the economy.

Overall, our results regarding the impact of external shocks during the debt crisis are consistent with the amplifying effect of financial frictions and models of the financial accelerator. As further evidence in support of this financial-friction channel, we also exploit cross-sectional variation in local banking market characteristics. In particular, we show that during the debt crisis, the relation between weather shocks and the various outcome variables described above is stronger in counties with higher funding constraints in the banking sector as measured by loan-to-deposit ratios. Thus, for example, we find that the elasticity of land values to yields is approximately 40% greater in high loan-to-deposit counties as compared to low loan-to-deposit counties.

We conclude by analyzing whether and to what extent exogenous cash flow shocks ultimately affected county-level income during the debt crisis. To this end, we calculate the local-level cash-flow to income multiplier, that is, the increase in income associated with a dollar injection of cash flow. The results show that positive cash flow shocks during the debt crisis did indeed increase local income levels, with our estimates pointing to a multiplier of approximately 1.63. In periods outside of the debt crisis, we do not find a statistically significant relation between cash flow shocks and county income levels. The size of the cash flow to income multiplier is thus state dependent, and larger during crises.⁹

Taken together, our results show how cash injections into an economy during a debt crisis can have important effects on a host of real and financial outcomes. When firm net worth is reduced, asset prices decline, delinquency rates rise, banks are more likely to fail, labor market disruptions ensue, and income levels decline. Conversely, increased net worth during the crisis improve conditions in local land markets, financial markets, and labor markets and ultimately raise income levels. From a policy perspective, our results thus point to the potential value of cash injections during a financial crisis that serve to strengthen firm balance sheets, thereby aiding firms in overcoming frictions in financial markets.

1. Empirical Methodology and Data

1.1 Empirical strategy

Our empirical strategy employs idiosyncratic weather shocks and their attendant effect on agricultural growing productivity as a source of variation in local cash flow. An extensive body of literature has shown that variation in weather has a strong effect on agricultural productivity (see, e.g., Dell, Jones, and Olken 2014 for a review). This variation is plausibly exogenous to farm-level activity, certainly within the frequency we study.

The analysis focuses on the state of Iowa, because it provides an ideal setting for examining the effects of weather on agricultural outcomes. Agricultural production is significant in Iowa and constitutes a large portion of economic activity in the state.¹⁰ Iowa also ranks first of all states in the production of corn, an important U.S. crop whose response to temperature fluctuations is well understood. Finally, agricultural data for Iowa are available at a more detailed level and for a longer time period compared to other states, allowing for a more complete time series of our empirical tests.

Our main empirical strategy uses an instrumental variable approach to relate exogenous weather-driven changes in crop yields to economic outcomes in various markets of interest: the market for land, the local financial sector, and labor markets. In doing so, we rely on an extensive prior literature in agricultural economics showing that corn is highly sensitive to variation in temperature during the growing season—the months from April through September—with even a few additional days of hot weather significantly reducing annual corn yields (see, e.g., Schlenker and Roberts 2006, 2009). Thus, the variation exploited in our identification strategy is not periods of drought or extreme heat throughout the growing season, but rather relatively small variation in temperature across counties within a given year which generate strong negative shocks to yields. Counties that did not experience the negative weather shock exogenously have greater cash flows, on average, than counties which did. It is this variation—that is, the relative increase in cash flows between counties which did and did not experience a negative weather shock—that drives our identification strategy.

It is important to note that our empirical methodology cannot differentiate whether the effect of weather shocks on the various real and financial outcomes analyzed is solely driven by the impact on farm income (stemming from the impact on yields) or, alternatively, is driven, at least in part, by the impact weather shocks have on farmland values. Indeed, weather shocks are found in our analysis to shift both income and land values, and so the relative importance of each channel in affecting the economic variables of interests is not identifiable. In line with this, one of the key insights of Kiyotaki and Moore (1997) is that the income and asset value channels are endogenously intertwined, with exogenous cash flow shocks affecting equilibrium asset values, and changes in asset values further influencing cash flows via their impact on collateral constraints. As previously noted, when discussing the impact of weather variation, we therefore refer to the impact of weather-driven cash flow shocks, which should be understood as affecting farm net worth, both through the income channel (via the impact on yields) as well as through the associated change in land values.

One potential concern in the interpretation of our results stems from farmers’ ability to hedge weather shocks by purchasing crop insurance. However, because of the relatively late development of crop insurance in the United States, hedging is of limited concern in our empirical setting. Indeed, while crop insurance markets have been available since the 1930s, they operated on a limited basis until the 1990s, during which time the U.S. government passed a number of laws that greatly expanded the insurance market (see, e.g., Cornaggia 2013).¹⁴ Further, the presence of hedging would bias our effects downward, as any crop insurance or weather-related governmental transfers to the agricultural sector would make farm net worth less sensitive to the effect of weather shocks.

1.2 Data sources

We construct a novel data set of county-level outcome variables in Iowa using a variety of sources. For our temperature data, we collect daily weather station data for Iowa from the National Oceanic and Atmospheric Administration (NOAA) from 1950 to 2010. Using this daily data, we calculate for each weather station the number of days in the corn growing season (from April 1st to September 30th) with an average daily temperature above 83°F.¹⁵ We then construct county-level estimates of this temperature measure for Iowa following the procedure in Deschênes and Greenstone (2012): Using geographical data for each county in Iowa from the U.S. Census Bureau, we construct a county-level estimate of the annual number of hot days (i.e., above 83°F) in the growing season by using a weighted average of all weather station estimates within a 50-km radius of the geographical center of each county. The weights used are the inverse of the squared distance from each weather station to the geographical center of the county. This process yields a total of 6,032 county-year temperature observations across Iowa’s 99 counties for the sample period 1950–2010 and 693 observations for the crisis period 1981–1987.

Our measure of corn yields comes from the USDA’s National Agricultural Statistics Service (NASS) yearly crop surveys. The NASS provides yearly county-level data of average corn yields from 1950 to 2010, measured in bushels per acre harvested. Our measure of farmland values comes from the Iowa State University Farmland Value Survey, which provides yearly county-level estimates of the average value per acre of Iowa farmland from 1950 to 2010.¹⁶ Studies have shown that these survey values closely track actual land sales prices (see, e.g., Stinn and Duffy 2012; Kuethe and Ifft 2013).¹⁷

We use two different data sources to examine the effect of weather-driven shocks on banks. The first source is data on agricultural loan delinquencies from the Federal Reserve’s Commercial Bank Data Call Reports. Delinquent loans are defined at the bank level as the outstanding balance of agricultural loans that are 90 days or more past due and upon which the bank continues to accrue interest (these data are available from 1984 to 2000). For each county in Iowa, we aggregate delinquent balances of all banks headquartered in that county, to obtain a county-level measure of delinquent agricultural loans.¹⁸ In addition, we use data on bank failures for each county taken from the Federal Deposit Insurance Corporation (FDIC). These data run from 1950 to 2010. To properly attribute the effects of temperature shocks during the growing season to subsequent bank failures, we mark a bank failure as occurring in year |$t $| if it occurred within the period from the end of the growing season in year |$t$| (October and onward) through the growing season of year |$t$|+1.

Our final data source is the Quarterly Census of Employment and Wages taken from the Bureau of Labor Statistics. We collect data on county-level employment, average annual wages, and total county-level wages. The data are available for the period 1975–2000. The agricultural crop production sector is defined as SIC code 01. In addition, we use the services sector (SIC division 0I) and manufacturing sector (SIC division 0D). A caveat with our agricultural wage and employment data is that the Quarterly Census of Employment and Wages only covers larger farms; it does not cover most agricultural workers in small farms or self-employed agricultural workers.¹⁹

1.3 The Farm Debt Crisis

The period preceding the 1980s farm debt crisis exhibited sharp increases in debt levels and land values, as is common in many financial crises.²⁰ During the 1970s, increasing commodity prices along with an expansion in demand for U.S. exports of agricultural commodities led to increased farm production and greater investment in farmland. Between 1971 and 1980, agricultural exports roughly doubled, the real price of commodities such as corn increased by over 35%, whereas farmland values rose by 88% (see Calomiris, Hubbard, and Stock 1986). During this period of land price appreciation, leverage played an increased role in the financing of agricultural land purchases.²¹ For example, whereas in 1950, 42% of all agricultural land transactions occurred with no debt financing, by 1978 only 11% of transactions occurred without relying on debt capital. The increased reliance on debt, coupled with rising farmland prices, led to a 66% increase in aggregate farm debt levels over the period 1971–1980 (Calomiris, Hubbard, and Stock 1986).

The farm debt crisis is generally thought to have been triggered in the early 1980s by the combination of several factors. The first was a tightening of monetary policy undertaken by the Federal Reserve in 1979 under Paul Volcker, which increased interest rates and raised the burden of farmer debt repayment. The interest rate increase also contributed to a strengthening of the U.S. dollar, making U.S. agricultural exports less competitive in the global market. Finally, the U.S. implemented a ban on grain exports to the Soviet Union in 1980, which contributed to a further decline in exports. As a result of these factors, many farmers who had invested heavily in production over the previous decade, often increasing their leverage in the process, faced a sudden reduction in demand for agricultural commodities coupled with a large increase in the cost of borrowing. The result was a period of severe financial distress and deleveraging in the agricultural sector with significant drops in farm income, sharp declines in farmland values, impaired farm balance sheets, and an erosion in farm credit conditions.

From 1981 to 1987, the average value of farmland dropped by 50% across corn belt states (Barnett 2000). Nationwide, nonperforming loans at agricultural banks rose from 2.8% of total loans in 1982 to 6.7% in 1986, and 100 small agricultural banks failed in 1984 and 1985 alone (see Calomiris, Hubbard, and Stock 1986; FDIC Division of Research and Statistics 1997). The farming sector saw significant deleveraging with total real agricultural debt declining by 37% from 1981 to 1987. In Iowa, real farmland prices dropped by an average of 67% across all counties between 1981 and 1987, with 39 commercial bank failures in that period. Furthermore, the majority of bank failures—34 of the 39—occurred between 1984 to 1987, considered to be the peak of the crisis.

1.4 Summary statistics

Table 1 presents summary statistics of the main variables.²² Panel A provides summary statistics for the debt crisis period defined from 1981 to 1987, and panel B provides summary statistics for the height of the crisis from 1984 to 1987. For comparison, panel C provides summary statistics for the noncrisis period.

During the crisis years of 1981 to 1987, the average number of days in the growing season with an average temperature exceeding 83°F is 3.25, whereas the average number of days above 83°F outside of the crisis is 2.51. As would be expected, the average annual number of days above 83°F during the growing season does not differ substantially in the crisis period as compared to the noncrisis period.²³Figure 1 reports density plots of the distribution of days above 83°F over our entire sample. As can be seen from the figure, roughly 45% of the county-year observations have one or fewer days with an average temperature above 83°F, whereas the density function is monotonically decreasing. As our main specifications include county and year fixed effects, Figure 1 exhibits variation that we do not exploit in our identification strategy. Figure 2, therefore, presents density plots of temperature variation demeaned with year and county fixed effects. The distribution of demeaned days above 83°F appears symmetric around zero but also exhibits substantial variation. Density plots for individual years (Figure 3) in the crisis and noncrisis period indicate substantial variability across counties for any given year, with some years exhibiting a significantly higher number of days above 83°F.

Figure 1

Distribution of temperature shocks

This figure shows the distribution of temperature shocks during the growing season for the entire sample from 1950 to 2010. The vertical axis depicts the density, and the horizontal axis depicts the number of days in the growing season for a given county-year that were above 83°F.

Open in new tabDownload slide

Figure 2

Distribution of temperature shocks in excess of averages

This figure shows the distribution of temperature shocks during the growing season in excess of county and yearly averages for the entire sample from 1950 to 2010. The vertical axis depicts the density, and the horizontal axis depicts the demeaned number of days in the growing season for a given county-year that were above 83°F.

Open in new tabDownload slide

Figure 3

Distribution of temperature shocks in different years

This figure shows the distribution of temperature shocks during the growing season for various years. In each panel, the vertical axis depicts the density, and the horizontal axis depicts the number of days in the growing season for a given county in the indicated year that were above 83°F.

Open in new tabDownload slide

Table 1 further shows that mean corn yields range from 105.7 to 123.8 bushels per acre. The mean land value during the 1981–1987 period is |${\$}$|2,948 per acre and is |${\$}$|1,978 per acre during the 1984–1987 period (all in real 2010 dollars). Finally, agricultural loan delinquencies are higher during the crisis, as would be expected. Figure 4 depicts the evolution of average corn yield, land value, and agricultural debt across all counties during the sample. Average corn yields increase over the sample period, while land values gradually increase from 1950 to 1970, and then substantially from 1970 to 1980. In the early 1980s, corresponding to the period of the farm debt crisis, land values drop precipitously. By contrast, corn yields do not exhibit such a trend during the debt crisis. Finally, agricultural debt steadily increased from 1960 to 1980 but significantly dropped during the farm debt crisis, as would be expected by a deleveraging process common in financial crises.

Figure 4

Corn yields, farmland values, and agricultural debt over time

This figure depicts average corn yields, land values, and agricultural debt over time. Each data point is an average across all counties in Iowa. Corn yield is defined as bushels of corn produced per acre of harvested land. Land value is the dollar value of farmland per acre, in real (2010) dollars. Total agricultural debt is the sum of agricultural loans to finance production and real estate debt secured by farmland, in real (2010) dollars.

Open in new tabDownload slide

A potential concern is that our results are driven by confounding differences between counties that are correlated with the weather shocks. While placebo tests (reported below) establish that our outcome variables are not correlated with weather-driven shocks in the noncrisis period, we further examine a host of other observable county characteristics, comparing county-year observations with below-median number of hot days (as measured by the weather shock variable Days above 83) to county-year observations with above-median number of hot days.²⁴ Panel D of Table 1 provides the comparison of means. As can be seen, across all variables examined, we do not find a statistically significant difference between counties with higher versus lower hot days prior to the crisis.

2. Empirical Results

We begin our analysis by confirming that county-level weather variation is indeed related to farm yields in our data.

2.1 Weather shocks and crop yields

As described above, to measure weather shocks we construct a variable, Days above 83, defined at the county-year level, which equals the number of days during the growing season where the average daily temperature within the county was above 83°F. This temperature threshold is taken from Schlenker and Roberts (2009), although our results are robust to alternate definitions of high temperature values.

|$Y_{i,t}$| is corn yields (bushels of corn produced per acre) in county |$i$| in year |$t$|⁠, and Days above 83 is the weather shock variable capturing hot average-temperature years. Regressions include a vector of year fixed effects |$\delta_{t}$|⁠) and a vector of county fixed effects |$\gamma _i )$|⁠. A potential concern with the regression above is that the error terms may be correlated across nearby geographical regions. In particular, temperature shocks to counties geographically close to one another are likely to be correlated, which will bias the standard errors in a typical ordinary least squares (OLS) regression. To address this, we follow the literature in agronomics (e.g., Deschênes and Greenstone 2007; Schlenker and Roberts 2006, 2009) and calculate standard errors correcting for spatial correlation as in Conley (2008).

Table 2 reports the results of regression (3) over the farm debt crisis sample period of 1981 to 1987. Employing year, but not county, fixed effects, Column 1 shows that high temperature is indeed detrimental to corn yields. Column 2 shows that adding county fixed effects does not substantially change the results. As the coefficient on the weather shock variable, Days above 83, shows, adding an extra day during the growing season with an average temperature above 83°F reduces annual corn yields by 3.3%. While seemingly high, this result is consistent with much prior work in the agronomics literature, such as Schlenker and Roberts (2009). Corn is extremely sensitive to high temperature values during the growing season, an established fact that lies at the heart of our identification strategy. The size of the cash flow shock caused by Days above 83 could also plausibly depend on the number of corn acres in the county. To account for this possibility, we rerun our main regressions weighting by the number of corn acres planted. As can be seen in Table A1 in the Online Appendix, we find nearly identical results to our main findings.

To understand the dollar magnitude implication of weather shocks on farm profit, one could consider the following back-of-the-envelope calculation of the impact of a 10% reduction in yields due to bad weather. This 10% reduction in yields is approximately equal to the effect of three additional hot days during the crisis, representing a weather shock smaller in magnitude than the standard deviation of Days above 83 (equal to 4.63). Given that the average yield during the crisis was 123.8 bushels/acre and the (real) price of corn was |${\$}$|4.10/bushel, a 10% decline in yields is associated with a drop in sales of |${\$}$|50.76 per acre (in real 2010 dollars). With 122,854 acres of corn grown at the county-level on average, this implies that a 10% reduction in yields is associated with an aggregate county revenue drop of |${\$}$|6.24 million. Assuming that costs are unaffected by the bad weather shock (in fact, costs tend to rise following such a shock), and considering a mean profit margin for farmers of 5.3% (Hoppe and Banker 2006), the 10% shock in yields ultimately translates to a shift in county-level profits from a positive |${\$}$|3.31 million to a loss of |${\$}$|2.94 million.²⁵

In Column 3 of Table 2 we report the results for the period of 1984 to 1987—the peak of the farm debt crisis—and find similar results to Column 2. In Column 4 we estimate the results for the noncrisis period. The effect of weather on yields is biological and hence, as expected, the estimated coefficient on the weather shock variable during the noncrisis period (Column 4) is similar to those during the crisis period (Columns 1–3).

Here, we note that an additional channel through which financial constraints can amplify the effect of weather shocks stems from their impact on the ability of farms to adapt to detrimental shocks. In particular, farmers may respond to adverse weather shocks with various strategies meant to mitigate the effect of hot weather on crop yields. Such strategies include extra fertilizer, more watering, tillage, and seed varieties, among others (Reilly 1999), and naturally require additional financing. However, when external financing is costly, as during a debt crisis, farm adaptation may thereby be constrained, and a given external shock will thus have a larger impact on farm yields and income.²⁶ Some suggestive evidence for this additional channel may be seen in the fact that the impact of weather shocks on yields is somewhat larger during the crisis years, when the crisis is defined over the period 1981-1987.

2.2 Cash flow shocks and asset prices

Having confirmed the effect of temperature on yields, we analyze how temperature shocks, and the variation they induce in farm cash flows, affect local land prices. Following cash-in-the-market pricing theories (Shleifer and Vishny 1992; Allen and Gale 1994; Kiyotaki and Moore 1997), we hypothesize that during debt crises, when financial frictions and the cost of external finance are high, counties that receive negative cash flow shocks stemming from weather variation will exhibit lower agricultural land prices: negative weather shocks reduce the net worth of local buyers—that is, nearby farmers—who will thus have less funds to purchase land

As in all models of cash-in-the-market pricing, an implicit assumption required for land prices to be affected by local liquidity conditions is that the market for land is at least partially segmented in that capital cannot flow seamlessly from afar (see, e.g., Shleifer and Vishny 1992; Allen and Gale 1994, 1998; Duffie 2010). This assumption is likely satisfied in the market for agricultural land, which is generally thought to be highly localized. However, to confirm this assumption, we hand-collect a micro-level data set on land transactions within one county in Iowa—Hamilton county—between the years 1970 and 1988.²⁷ For each of the 1,971 sales of agricultural land in Hamilton county, we mark the county of the buyer and calculate the monthly fraction of out-of-county buyers.²⁸Figure 5 shows that the data confirm that agricultural land sales are highly localized: only 9.4% of transactions occur with an out-of-county buyer. Interestingly, the fraction of out of county buyers increases substantially during the financial crisis, reaching 25% in 1985. This spike in out-of-county purchases is very much consistent with, and in fact would be predicted by, the existence of fire sales, in which capital from afar flows into the market to buy liquidated assets.

Figure 5

Cross-county land purchases in Hamilton County

This figure depicts cross-county land purchases in Hamilton County, purchases where the buyer is located outside of the county. The red horizontal line represents the mean over the sample period.

Open in new tabDownload slide

Having confirmed that agricultural land markets are localized, we examine the effect of exogenous county-level weather-induced cash flow shocks on the price of land during the crisis. Table 3 reruns the reduced-form specification in regression (3) but employs log(Land value), the average county-level price per acre of farmland (in 2010 dollars), as the dependent variable. Consistent with the prediction of cash-in-the-market pricing, we find that counties that received a positive cash flow injection, which is driven by relatively good weather, exhibit higher land values than counties that receive a negative cash flow shock, which is driven by a few additional days of high temperature weather during the growing season. Focusing on Column 2, which includes county fixed effects, an additional day during the growing season with an average temperature greater than 83°F reduces average price per acre by 0.4%. In Column 3, which reports the results for the period during the peak of the farm debt crisis, 1984–1987, the estimated magnitudes are even larger: during the growing season, an additional day with average temperature exceeding 83°F reduces land values by 0.8%. To our knowledge, this is the first study that provides causal evidence of cash-in-the-market pricing.

The results in Table 3 regarding the relation between weather shocks and land prices focuses on the farm debt crisis period. At the center of the theoretical argument behind this result is the assumption that financial frictions prevent firms from raising external financing to smooth shocks or make it prohibitively costly for them to do so. According to this argument, we thus expect that outside of the crisis, the effect of weather shocks on land prices is greatly diminished (or nonexistent), even while these shocks continue to affect yields and hence cash flows. Column 4 conducts this test by considering the impact of exogenous weather shocks outside of the 1980s farm debt crisis. As can be seen, in contrast to the results in Columns 1–3 of Table 3, weather variation outside of the crisis years has no statistically significant relationship with land prices, consistent with an increased ability of firms to smooth cash flow shocks. Thus, even though negative weather shocks continue to detrimentally affect yields outside of the crisis (see Table 2, Column 4), they have no effect on land values outside the crisis.²⁹

Table 4 shows the results. Column 1 of the table provides the first-stage estimation. Column 2 of the table exhibits the results of the second stage, finding an elasticity of land values to yields of 0.12; a 10% exogenous increase in county yields is thus associated with a 1.2% increase in land values during the debt crisis. Columns 3 and 4 conduct the instrumental variable strategy starting from 1984—the height of the crisis years—and up to its end in 1987. Consistent with higher financial constraints during the height of the crisis, the second-stage elasticity of land prices to yields is 0.34, or roughly 3 times larger than the effect during the full crisis period.

It is instructive to consider the magnitude of these results in light of the impact of weather shocks on aggregate county revenue. As noted in Section 2.1, a 10% change in yields during the crisis is associated with a change in aggregate county farm revenue of |${\$}$|6.24 million. Given the elasticity estimates in Table 4, a 10% drop in yields causes a 1.2% decline in land values, which given the average county land value per acre of |${\$}$|2,948, implies an aggregate decline in county land values of |${\$}$|4.35 million. Thus, a |${\$}$|6.24 million shift in county level revenue is associated with a |${\$}$|4.35 million shift in aggregate county land values. Consistent with cash-in-the-market pricing, the results thus imply an economically significant dollar sensitivity of land values to cash flow shocks of 0.70 (⁠|$=4.35/6.24$|⁠).³⁰

2.3 Cash flow shocks and the financial sector: Delinquencies and bank failures

Having shown how weather shocks affect yields and land prices, in this section we analyze how temporary shocks to cash flow during the debt crisis propagated into the financial sector. In the presence of financial frictions, temporary—that is, short-lived—weather shocks during a crisis affect farmers’ repayment ability, in turn leading to defaults. Thus, farmer inability to smooth temporary weather shocks in a crisis is predicted to create long lasting effects in the financial sector in the form of borrower defaults and bank failures.³¹

Panel A of Table 5 presents the results. As can be seen in Column 1 of the table, delinquency levels vary negatively with yields. The results imply an elasticity of 3 between county aggregate delinquent loans and county average yields: during the crisis, counties which experience a 10% weather-driven exogenous increase in yields (as compared to their mean) exhibit a 30% decrease in aggregate delinquency levels.³⁴ As would be predicted, exogenous positive cash injections translated into reduced delinquencies among borrowers.

Column 2 of Table 5A presents the results. As the table shows, a 10% decrease in yields leads to an approximately 3.2-percentage-point increase in the probability of bank failure. The effect is economically sizeable, as 28% of the county-year observations during the period of 1984 to 1987 exhibit a bank failure.³⁵ Consistent with the hypothesis, temporary cash flow variation driven by exogenous weather shocks did indeed lead to spillovers into the financial sector in the form of bank failures.

Column 3 of the table repeats the analysis but allows a lag in the time to bank failure. Specifically, we define an indicator variable, Bank failure crisis, that takes on the value of one if there was a bank failure from the given year until the end of the crisis (i.e., to 1987), and zero otherwise. As can be seen, the effect of predicted yields on bank failures rises when a time lag to failure is accounted for, with a coefficient in the level-log specification that is approximately -0.4.

As a placebo test, panel B of Table 5 examines the effect of temporary cash flow shocks outside of the debt crisis. Lower financial frictions and stronger balance sheets during this period would predict muted effects. This is indeed what the results indicate. As can be seen in Columns 1 and 2 of Table 5B, cash flow shocks outside of the crisis are not related to delinquencies or bank failure rates in a statistically significant manner.

2.4 Cash flow shocks and labor markets: Employment and wages

Column 1 of Table 6A exhibits results using total county-level employment as the dependent variable.³⁶ We find a positive relation between yields and total county employment. Estimating the economic magnitude of the effect, a 1% drop in yields reduces agricultural employment by 4.1% of the sample mean. Thus, during the crisis, farms in counties that received a positive cash flow injection (driven by relatively good weather) reduce their total agriculture employment by less than those that received a negative cash flow shock. Consistent with increased financial frictions during the crisis, temporary shocks to firm balance sheets affect employment rates. When financial constraints bind and external capital is costly, labor demand can be influenced by firm net worth.

Continuing with labor market outcomes, Column 2 of Table 6A replaces employment with average county wages per employee as the dependent variable. As can be seen, predicted crop yields are positively related to average wages per employee. Counties that experienced a negative weather-induced cash flow shock exhibit a relative decline in average wages per employee, consistent with a drop in labor demand stemming from reduced ability to finance employee wages out of internal capital. The elasticity of yields to average county pay is 2.9: a 1% reduction in yields is associated with approximately a 3% relative reduction in average wage per employee. Column 3 analyzes total county wages, which combines variation in total county employment as well as variation in county average wage per employee. Unsurprisingly, given the results in the prior two columns, we find that weather-driven cash flow injections are positively related to total county wages.

Panel B of Table 6 repeats the analysis but focuses on the period outside of the farm debt crisis. Outside of financial crises, firms’ ability to smooth temporary cash flow shocks is greatly enhanced, and so we expect the relation between employment and predicted yields to be dampened. Consistent with this, the results show that outside of the debt crisis, county level employment, average wage per employee, and total wages are unrelated to exogenous weather-driven variation in yields. While the strength of a firm’s balance sheet, and variation in it, plays a role in determining labor market outcomes during periods of high financial constraints, they play no role outside of the crisis. The results thus suggest that cash injections into the real sector affect labor market outcomes during a debt crisis, but not outside of it.

Table 7 continues by analyzing how cash flow shocks spill over into other labor markets during the debt crisis. Specifically, we use the instrumental variable strategy from above to relate variation in predicted yields to local level employment and wages in the service sector, a natural sector where employees dislocated from farming might seek employment.³⁷ Column 1 of Table 7A shows that total county-level employment in the service sector is negatively related to weather-driven cash flow shocks in the agricultural sector: when a county is hit with a negative cash flow shock in the agricultural sector, the data show that agricultural employment declines while employment in services rises (compared to the mean county level).³⁸ Following a negative cash flow shock, workers thus appear to shift from the adversely affected agricultural sector toward other industries. The results in Column 1 of the table show that a 1% reduction in county predicted yields is associated with a 0.34% increase in county-level service sector employment relative to the sample mean.

Still focusing on the debt crisis period, Column 2 of Table 7 examines how average wages in the service sector relate to cash flow shocks in the agricultural sector. Consistent with an outward shift in the supply of workers in services, the coefficient on predicted yields shows that counties that experienced an exogenous negative weather-driven cash flow shock in agriculture exhibit a relative decline in wages in the service sector. As workers shift from agriculture to services, labor supply rises and, correspondingly, wages in the sector decline. The elasticity of average county wages in the service sector to county yields is 0.075; that is, a 10% decline in agricultural yields translates into a 0.75% drop in service sector wage.

Column 3 of the table examines total county wages in the service sector and finds that these are unrelated in a statistically significant manner to yields. This is not altogether surprising, as the effect on wages and employment run in opposite directions following a negative shock to yields: while average wages in the service sector falls, county employment in the sector rises.

Taken together, the results in panel A of Table 7 paint a picture by which, during a debt crisis, firms’ inability to smooth shocks in one sector create externalities in other sectors within the labor market. Workers shift away from firms hit by temporary cash flow shocks, increasing the supply of labor in other sectors. The result is higher employment and lower wages in sectors unrelated to the original cash flow shock.

For completeness, panel B of Table 7 conducts a placebo test and reruns the specifications of panel A focusing on the period outside of the crisis. As was shown in panel B of Table 6, outside of the crisis weather shocks do not affect agricultural employment. Because the agriculture sector smooths cash flow shocks, we expect to find no effect on labor outcomes in the services sector outside of the crisis. This is indeed what the results show. Using the instrumental variable specification outside of the crisis, none of the service sector labor market outcomes are related in a statistically significant manner to (predicted) county level yields.

Next, we test a second channel, which is related to shifts in demand and through which cash flow shocks during a financial crisis can spill over into other sectors. The results in Table 6 show how sectoral cash flow shocks during a financial crisis translate into labor market dislocation within the agriculture sector, as firms find it difficult to utilize capital markets to smooth temporary funding shortages. Accordingly, we test the following demand channel for inter-sector spillovers during financial crises: once a given sector is hit by a cash flow shock, firms in the sector cut employment, causing dislocated employees to reduce overall consumption. The shock to the first sector thus propagates into other sectors, which, faced with a reduction in demand, cut employment in their respective sectors.³⁹

To test this mechanism, we run similar regressions to those in Table 7 relating employment and wages in the service sector to weather-induced cash flow shocks in agriculture, but interact the weather-driven cash flow shocks with a measure of the importance of agriculture within each county. Weather shocks are measured, as usual, using the number of growing season days with average temperature above 83°F, while the importance of agriculture within each county is measured by the ratio of farm crop income to total income within each count.⁴⁰ We predict that in counties with a dominant agriculture sector, negative (weather-driven) cash flow shocks will lead to greater declines in overall demand, which will tend to reduce employment in the service sector. This demand-driven effect goes in the opposite direction to the reallocation effect analyzed above whereby workers from agriculture move into other sectors following a cash flow shock in the agriculture sector.

Column 1 of Table 8A presents the results of the interaction specification, analyzing the effect of weather-driven cash flow shocks on service sector employment. As can be seen, the coefficient on the noninteracted weather shock is positive, but the coefficient on the interaction term between the weather shock and the county-level agricultural importance is negative. Thus, as in Table 7, in counties where farming plays a relatively small role, negative cash flow shocks in agriculture tends to increase employment in services via a reallocation channel. However, if agriculture plays a sufficiently large role in a county, cash flow shocks in the agriculture sector reduce employment in services. At the 25th percentile of agricultural importance within the county (captured by the ratio of farm crop income to total income), an additional hot day with temperature above 83°F increases service sector employment by 1.3% of the sample mean, but in contrast, at the 75th percentile of agricultural importance such a weather shock reduces service sector employment by 0.8%. Thus, during a debt crisis, aggregate county-level cash flow shocks in one sector impose employment externalities on other sectors operating within the same geography. The sign of these externalities depends on the relative importance within the economy of the sector receiving the shock.

Column 2 of Table 8A repeats the analysis but analyzes the impact on service sector wages (rather than employment). We predict that detrimental weather-driven cash flow shocks reduce wages and that this effect will be greater when agriculture plays a larger role within a county. However, as can be seen in the table, while the noninteracted coefficient on weather shocks does indeed predict a reduction in wages following a negative cash flow shock, the coefficient on the interaction term with the fraction of county-level farm income is not statistically significant.

Panel B of Table 8 repeats the interaction specification in panel A of the table, but uses the instrumental variable strategy relating labor market outcome variables to predicted log yields (as in Table 7 above). To this end, we separate the sample into two based on median county-level farm importance and rerun instrumental variable specifications for below-median and above-median farm importance counties.⁴¹ The results are consistent with those in Table 7. Employment in services is positively related to predicted yields in counties with above-median farming importance but is negatively related to predicted yields in counties with below-median farming importance. Negative weather-driven cash flow shocks thus decrease employment in services amongst counties where farming plays a large role, consistent with a demand channel effect, but increases employment in counties where farming plays a relatively smaller role, consistent with a reallocation effect. Further, as can be seen from Columns 2 and 4 of Table 8B, wages are positively related to predicted yields, although the effect is not statistically significant in above-median farming importance counties.

We note that the reason that the point estimate on services employment in the pooled sample in Table 7A does not lie in between the point estimates on services employment in the two subsamples in Table 8B is that the pooled regression allows for only one set of fixed effects, while the subsample regressions allow for two different sets of fixed effects (i.e., each subsample has its own set of fixed effects). Indeed, rerunning the pooled IV regression in Table 7 with interaction terms between the year fixed effects and an indicator variable based on the median fraction of agricultural income—that is, the same criteria used to create the sample split—results in a coefficient on predicted yields which lies in between the analogous coefficients in the two subsamples, as would be expected. Table A2 in the Online Appendix provides these results.

2.5 Banking market frictions and the heterogeneous effect of cash flow shocks

In this section we test whether increased financing frictions at the local level affect the impact of weather-driven cash flow shocks on the real and financial outcomes described above. To this end, we exploit cross-sectional variation in local banking market characteristics and the degree of financing frictions therein. We hypothesize that in markets with greater bank financing frictions, weather-driven shocks will have a larger effect on the variables of interest, as banks, and hence the farms which rely on them, are less able to smooth the impact of negative shocks.⁴²

To test this hypothesis, we proxy for local bank financing frictions by calculating the aggregate loan-to-deposit ratio for banks in each county prior to the crisis (following, e.g., Acharya and Mora 2015).⁴³ We then rerun the specifications relating weather shocks to the real and financial outcomes described above, splitting the sample between counties which had high loan-to-deposit ratios and counties with low loan-to-deposit ratios. Table 9 presents the results.

As can be seen in the table, the results show that for all dependent variables but one—services employment—the effect of (instrumented) yields on the dependent variable of interest is indeed greater in counties with greater bank financing frictions. Thus, for example, we find that the relation between the elasticity of land values to instrumented yields is approximately 40% greater in high loan-to-deposit counties as compared to low loan-to-deposit counties. Similarly, the impact of instrumented yields on agricultural employment is more than 3 times larger in counties with high loan-to-deposit ratios as compared to in low loan-to-deposit ratios (with the effect in the latter counties not statistically different from zero.)

As mentioned above, employment in the services sector is the only variable that does not exhibit a stronger sensitivity to weather-driven shocks in counties with high bank financing frictions. Indeed, the results indicate that the impact of weather shocks on employment in the services sector are smaller in counties with larger bank financing frictions (i.e., high loan-to-deposit ratios). We note, though, that the predicted effect of bank financing frictions on the sensitivity of services employment to weather shocks is ambiguous: on the one hand, employment in agriculture is predicted to drop more, but on the other hand, when financing frictions are large, the ability of firms in the services sector to absorb the dislocated employees could be impaired, as hiring employees could strain finances (for financial friction effects on employment, see, e.g., Chodorow-Reich 2013 and Falato and Liang 2016). Our results are consistent with the second effect dominating the first.

2.6 Cash flow shocks and income per capita: The income multiplier during the 1980s Farm Debt Crisis

The results of the prior sections show how exogenous county-level cash flow shocks during the debt crisis had a sizeable effect on a host of real outcomes across a number of markets. These include the market for land, labor markets, and the local banking sector. A natural question to ask, then, is whether and to what extent cash flow shocks ultimately affect county-level income during the debt crisis. To investigate this question, we use our standard instrumental variable approach regressing the log of county income per capita on the log of county-level yields, with yields instrumented by the exogenous weather shock variable Days above 83.

Table 10 presents the results. As can be seen, during the farm debt crisis, instrumented yields are positively related to income per capita, with an elasticity of 0.138. In contrast, the point coefficient on predicted yields outside of the farm debt crisis period is approximately one third as much and not statistically significant.