Why Did Farm Product Prices Fall?

Our focus is on the consequences of the farm product price decline, not its causes. Still, the causes of the farm product price decline are both of independent interest and may matter for the interpretation of our results. The most basic explanation for the price decline is the interaction of a decrease in demand (foreign and domestic) combined with a price inelastic demand curve for farm products and a close to completely price inelastic supply curve. The demand curve for farm products shifted in as the United States and the world fell into recession. For farm products overall, the Depression in the United States certainly played the dominant role; much of farm output was nontraded, and even for traded crops, it mattered that the U.S. output decline in 1930 was more severe than that abroad. The decline in demand had large price effects because in the short term, supply was determined by past planting decisions. And even in the medium term, U.S. farmers facing price declines for their products may have maintained production; a price decline for a farmer has a substitution effect pushing a farmer to plant less but an income effect pushing a farmer to plant more. These forces were combined with an influx of workers into agriculture during the Great Depression, as unemployed urban workers moved to rural areas. Throughout the 1920s, there was net migration from farms/rural areas to cities; this pattern reversed in 1930 with net migration to farms from cities each year from 1930–33 (U.S. Department of Agriculture, 1936, table 445, p. 339). The combination of a growing farm population and an income effect encouraging more production meant that the farm product supply curve was close to vertical during the Depression. As Ezekiel and Bean (Reference Ezekiel and Bean1933, p. 21) put it: “Left to themselves, farmers as a group have been unable to readjust their total production in line with the reduced demands.”Footnote ⁹ Despite a collapse in wheat prices, for instance, world wheat production rose between the 1929/30 season and the 1931/32 season.Footnote ¹⁰

The effect of inelastic supply on the farm product price response was compounded by inelastic demand. Food expenditure is insensitive to price (Taylor and Houthakker Reference Taylor and H. S.2009), likely leading to an inelastic demand curve for farm products primarily used for food. Of course, inelastic supply and demand for farm products were not unique to the Great Depression. Bordo (Reference Bordo1980) notes that at least since Cairnes (Reference Cairnes1873), economists have known that inelastic supply may make commodity prices more volatile than the prices of manufactured goods.Footnote ¹¹ Thus farm product prices often swing dramatically in response to shocks. Beyond these general factors, idiosyncratic shocks drove large price declines of certain farm products. Table 1 shows the prices of 12 major farm products early in the Depression. It illustrates that while the prices of all major farm products fell in 1930, the price decline was far from uniform. It is beyond the scope of the present paper to present a full description and explanation of the behavior of different farm product prices in 1930. Rather we consider one natural distinction, that between more and less internationally traded crops.

In 1930, the prices of wheat, cotton, and tobacco fell more than those of other crops. Cotton, tobacco, and wheat are the three crops in Table 1 that were traded most internationally; Figure 3 illustrates the much larger decline in traded than in nontraded farm product prices after the beginning of the Depression. This distinction was noticed at the time; the U.S. Department of Agriculture (1933, p. 94) wrote “The prices of major cash crops, being more subject to international influence, at first suffered more than did the prices of livestock and livestock products, that are consumed almost entirely in the domestic markets.” To illustrate how international factors mattered, we briefly explore the wheat and cotton price decline.

Figure 3

TRADED AND NONTRADED FARM PRODUCT PRICES

Notes: The figure shows seasonally adjusted traded crop and nontraded farm product prices. Traded crops are wheat, cotton, and wool, major internationally traded crops for which monthly price data are available. Tobacco is excluded because only annual prices are available. Nontraded farm products are corn, potatoes, hay, cattle, hogs, milk, eggs, and chickens; these products were traded little internationally. See Online Appendix B for details on the construction of these price indices.

Source: See Online Appendix B.

We start by comparing the price that was received by U.S. wheat and cotton farmers to measures of wheat and cotton prices abroad. Figure 4(a) compares the price of wheat received by U.S. producers to the wholesale price of wheat in Winnipeg, Canada and the wholesale price of imported wheat in Liverpool, England. The figure shows that all three prices followed a roughly similar downward path from fall 1929 to fall 1930. Figure 4(b) compares the price of cotton received by U.S. producers to the prices of U.S. and Indian cotton in Liverpool, England. After the United States, India was the largest exporter of cotton in the late 1920s (U.S. Department of Agriculture 1936, table 112, p. 86). Even more than for wheat, for cotton, there is a close correlation between the U.S. producer price and the two foreign prices. Figures 4(a) and 4(b) suggest, as one might expect with an easily transportable commodity, that world developments determined the prices received by U.S. wheat and cotton farmers. While determining causality is difficult, and quantitatively disentangling the influence of different shocks even more so, it is possible to describe the broad forces lowering world wheat and cotton prices in 1930.

Figure 4

WHEAT AND COTTON PRICES

Notes: All series are non-seasonally adjusted. Figure 4(a) shows wheat prices paid to U.S. producers, the wholesale wheat price of No. 3 Manitoba Northern wheat in Winnipeg, Canada and the price of imported wheat in Liverpool, England. Figure 4(b) shows cotton prices paid to U.S. producers, the wholesale price of American Middling cotton in Liverpool, England, and the wholesale price of Indian Oomra No.1, Fine cotton in Liverpool, England.

Sources: Wheat U.S. Department of Agriculture (1936, table 15, p. 19, table 18, p. 21, and table 19, p. 21). Cotton - U.S. Department of Agriculture (1936, table 106, p. 82 and table 111, p. 85).

World wheat prices fell as the world economy fell into depression, with two more idiosyncratic factors also contributing to the decline. First, world wheat production in 1928/29 was exceptionally large; excluding Russia and China, production rose 9 percent between 1927/28 and 1928/29 (U.S. Department of Agriculture 1936, table 5, p. 11). This led to an accumulation of stocks, which put downward pressure on prices in 1930 ( Wheat Studies, VI:9, Aug. 1930, pp. 387–8; Wheat Studies, VII:2, Dec. 1930, p. 90).Footnote ¹²

A second idiosyncratic factor pushing down wheat and other grain prices was exports from the Soviet Union. In 1913, before WWI and the Bolshevik revolution, Russia had been the world’s largest wheat producer (U.S. Department of Agriculture 1936, table 5, p. 11). The disruption of WWI and the Russian Civil War reduced Russian wheat production from 1.03 billion bushels in 1913 to a low of 205 million bushels in 1921 (U.S. Department of Agriculture 1936, table 5, p. 11). But production roughly recovered to 1913 levels by 1930 (U.S. Department of Agriculture 1936, table 5, p. 11), and Soviet wheat exports grew from 124,000 bushels in 1928/29, to 7.4 million bushels in 1929/30, and 112 million bushels in 1930/31 (U.S. Department of Agriculture 1933, table 17, p. 417). While it is hard to judge the quantitative significance, we believe it likely that Soviet exports and the prospect thereof depressed wheat prices.

The decline in cotton prices in 1930 is a simpler story. Likely even more for cotton than for wheat, the global depression itself depressed prices. In its summary of the decline in cotton prices during the Depression, the U.S. Department of Agriculture (1933) writes: “The outstanding forces that led to the present cotton situation were general deflation in commodity prices and declines in business activity and consumer incomes throughout the world” (p. 97).Footnote ¹³ Unlike wheat, which is primarily used to produce food, cotton was used in industry, most obviously, but not only to produce textiles. Thus the demand for cotton was a causality of the world output decline. To give one example of this mechanism, in 1929, the production of car tires accounted for roughly 10 percent of all cotton consumed in the United States (U.S. Department of Agriculture 1933, p. 105). The production of U.S. tire casings fell by 26 percent between 1929 and 1930 with, one imagines, a corresponding hit to demand for cotton.Footnote ¹⁴ Of course, world manufacturing output also fell in 1930. As world demand fell, so did the volume of U.S. cotton exports as well as the price. Exports fell from 8.4 million bales in the year beginning August 1928 to 7.0 million bales in the year beginning August 1929. As with wheat, demand for U.S. cotton was also affected by idiosyncratic factors. In particular, manufacturers abroad were increasingly using cheaper foreign, in particular Indian, cotton instead of U.S. cotton (U.S. Department of Agriculture 1931, p. 12).

We draw three conclusions from the wheat and cotton experience. First—and unsurprising—U.S. traded crop prices moved with world prices, meaning that any explanation for the 1930 price decline likely needs to include international shocks. Second, a crop could be tradeable in the sense that its price tracked world prices even if the share of production actually leaving the United States was modest. Exports of wheat in 1930 accounted for less than 20 percent of U.S. wheat production, yet U.S. wheat prices tracked foreign prices closely. Finally, it is difficult to quantify the effect of idiosyncratic versus aggregate shocks and the exact extent to which agricultural price declines were driven by the U.S. Depression and the Depression abroad. This unfortunately means it is difficult to make causal or quantitative statements about the role of foreign versus U.S. shocks in the U.S. farm product price decline. Thus we interpret the results that follow as an exploration of the impact not of an exogenous shock, but rather of a propagation mechanism.

Farm Incomes

The effect of a large commodity price movement depends on its incidence; it was the U.S. economy’s misfortune in 1930 that the burden of lower farm product prices fell on indebted farm households. An individual farmer’s income roughly equals the price of their product times the quantity produced, so the large decline in farm product prices produced a large decline in the income of farmers. Farm incomes are shown in Figure 5. Between July 1929 and October 1930, real, cpi-deflated farm income fell 29 percent; income from crops fell 42 percent. While individuals who became unemployed may have seen larger income declines, the decline in income for the typical farmer was far larger than that for the typical nonfarm worker. Annual nominal income data from the BEA (table SA04) deflated by the CPIFootnote ¹⁵ show that real nonfarm personal income fell 6.3 percent between 1929–1930; over the same period, real farm income fell 25.1 percent.

Figure 5

FARM INCOME

Note: The figure shows seasonally adjusted total farm income and income from crop production.

Sources: Pre-1932 – Survey of Current Business, May 1934, p. 19. 1932–33: 1936 Supplement – Survey of Current Business, p. 9. These nominal values are deflated by the CPI (not seasonally adjusted), taken from FRED series CPIAUCNS.

To understand whether the fall in farm income in 1930 was driven by lower farm product prices, Figure 6 shows farm income, the quantity of marketed farm products, and farm product prices from 1927 to 1930. Unlike the series shown in Figure 5, these series are not seasonally adjusted; thus they show the regular seasonal peak in fall of farm products sold and farm income received. Income and marketing track each other closely until mid-1930. This is a bit mysterious given the steady fall in farm product prices. In any case, beginning in mid-1930, farm marketing rises in its typical seasonal fashion while farm income increases little relative to its normal seasonal increase. The unusually small increase in farm income reflects falling farm product prices. The behavior of these series shows that before mid-1930 falling seasonally adjusted farm incomes (Figure 5) may have largely reflected less marketing of farm products. Reduced marketing may in part have been a result of drought, which shrank the 1929 harvest. (In our cross-state regression, we control for drought conditions.) In any case, beginning in summer 1930 lower farm product prices are the likely driver of lower farm incomes.

Figure 6

FARM INCOME, MARKETING, AND PRICES

Note: The figure shows nominal farm income, the quantity of farm products marketed, and farm product prices. No series is seasonally adjusted.

Sources: Farm income – Survey of Current Business, May 1934, p. 19. Farm marketing – Survey of Current Business, March 1933, p. 20. Farm product prices – FRED series M04058USM350NNBR, originally from NBER series m04058, which was collected from BLS publications.

Evidence from the U.S. States

Monthly data on new passenger car registrations come from the 1934 Automotive Daily News Review and Reference Book.Footnote ¹⁷ These data closely approximate sales and for conciseness, we will generally refer to auto “sales” rather than “new registrations.” States required the registration of new cars, so new registrations were a direct measure of sales. As discussed further in Online Appendix C, at times the measure could be inexact, but only to a limited degree.

Figure 8 shows a scatter plot of the percent change in new registrations (sales) between the second and third quarter of 1929 and the second and third quarter of 1930 and the share of a state’s population living on farms. Here and in our regressions later we compare car sales between these six-month averages (1929:Q2–Q3 and 1930:Q2–Q3), since doing so filters out idiosyncratic noise in the monthly data, and since using a 12-month sample window allows us to avoid the uncertainty associated with seasonal adjustment. Figure 8 shows a clear negative relationship between car sales growth and farm share of the population during the first year of the Great Depression.

Figure 8

PERCENT CHANGE IN CAR SALES AND FARM POPULATION SHARE

Sources: Car sales – see text; farm share of the population – Haines and ICPSR (Reference Haines2010).

Figure 9(a) provides another way to see the relationship between farm share and economic performance. It graphs the average level of auto sales in each of four quartiles of states, where states are grouped by the share of their population living on farms. To construct this graph, we seasonally adjust auto sales using data from 1929 (when the series begins) through 1934, excluding 1933.Footnote ¹⁸ We note that before 1930 there is no consistent ranking of auto sales by farm quartile, which is to say there is no apparent pre-trend. There is a notable upward spike in auto sales in quartile 4 in late 1929. This upward spike is driven by very high auto sales in Alabama and Mississippi, numbers that we suspect may be errors. For this reason, Figure 9(b) repeats the calculation with Alabama and Mississippi excluded.

Figure 9

CAR SALES BY FARM SHARE QUARTILE

Sources: Car sales – see text; farm share of the population – Haines and ICPSR (Reference Haines2010).

The first indication of a divergence between high and low farm share quartiles begins in February or March 1930 depending on whether Alabama and Mississippi are included. In these months we first see the expected pattern: the least farm intensive states have the smallest drop in car sales (relative to 1929), followed by quartiles 2 and 3, with the largest decline in car sales occurring in quartile 4. While this timing accords with the drop in farm product prices in Figure 2, one needs to be cautious in reading too much into short-term movements given the uncertainty associated with the seasonal adjustment. Still, by the second half of 1930, a clear pattern emerges, in which auto sales had fallen most in the highest farm share states (quartile 4) and least in the lowest farm share states (quartile 1).

Table 2 investigates the relationship between farm intensity and auto sales more carefully by estimating regressions of the form:

(1) $$\% \Delta {\rm{Auto}}\;{\rm{sale}}{{\rm{s}}_{i,{\rm{1929}}:{\rm{Q2 - Q3 - 1930}}:{\rm{Q2 - Q3}}}} = {\beta _0} + {\beta _1}{\rm{Agricultural}}\;{\rm{exposur}}{{\rm{e}}_i} + \gamma '{{\rm{X}}_{\rm{i}}}{\rm{ + }}{\varepsilon _{\rm{i}}},$$

where $${\rm{\% }}\Delta {\rm{Auto\;sale}}{{\rm{s}}_{i,{\rm{1929:Q2 - Q3 - 1930:Q2 - Q3}}}}$$ is auto sales growth in state i at the beginning of the Depression, “Agricultural exposure” is a measure of a state i’s exposure to falling farm prices, and X is a set of control variables. Column (1) shows results for the single-variable regression analogous to the scatter plot in Figure 8. The coefficient is both economically and statistically significant (the t-statistic equals 5). The coefficient of –0.32 implies that a one-standard-deviation change in farm population share (17 percentage points) is associated with a 5.5 percentage point decline in auto sales growth. For comparison, nationwide auto sales fell 34 percent over this period.Footnote ¹⁹ Note also that the R² is 0.30: as measured by auto sales, the farm share of the population alone explains 30 percent of the cross-state variation in the severity of the early Great Depression.

Table 2

CROSS-STATE REGRESSIONS

Notes: The dependent variable is the percent change in non-seasonally adjusted car sales from the 1929:Q2–Q3 average to the 1930:Q2–Q3 average; p.c. means per capita. Robust standard errors in parenthesis. *p < 0.1, **p < 0.05, ***p < 0.01.

Sources: New car sales—see text; population and percent of the population on farms—the 1930 Census as reported in Haines and ICPSR (Reference Haines2010); 1929 value of crops sold per capita—the 1940 Census as reported in Haines, Fishback, and Rhode (Reference Haines, Price and Paul2015); 1928 car sales—Automotive Industries, 23 February 1929, p. 271; we construct the drought dummies using data from the National Climate Data Center. Regional fixed effects are dummy variables for the four census regions: northeast, midwest, south, and west.

In interpreting the results in Table 2, it is worth emphasizing that Specification (1) is not directly measuring the change in purchases of cars by farmers themselves; the difference between the change in car sales in more and less farm intensive states is due not only to the purchasing behavior of farmers but also to the purchasing behavior of segments of the population whose livelihood was linked to that of farmers. When farmers’ spending fell, the owner of the local general store may have also foregone an auto purchase.

Column (2) adds control variables to address omitted variable bias concerns. We control for population to assure that the percent of a state’s population living on farms is not simply proxying for a small versus large state effect;Footnote ²⁰ we control for the per-capita number of cars sold in 1928 to assure that estimates are not biased by greater propensities to purchase cars in some states; and we control for region fixed effects to isolate the effects of farm intensity within regions. These control variables have essentially no effect on the coefficient. Column (3) controls for drought with two dummy variables equal to 1 in states that suffered a moderate drought or worse in the second and third quarters of 1930 and 1929.Footnote ²¹ With the controls for drought, the coefficient on farm share of the population is again little changed.

Column (4) of Table 2 uses an alternative indicator of a state’s agricultural exposure: the value of crops sold per capita in 1929. The coefficient is again economically and statistically significant (t-statistic equal to 4). The coefficient of –0.12 implies that a one-standard-deviation increase in crops sold per capita in a state ($39.70) results in a 4.8 percentage point larger decline in auto sales in the first year of the Depression. This is very similar to the decline in auto sales associated with a one-standard-deviation change in the farm share of the population. Column (5) adds control variables and region-fixed effects. The coefficient shrinks by one-third, but remains economically and statistically significant. Column (6) adds controls for drought. This results in an increase in the coefficient on crops sold per capita.

The results in Table 2 show that agriculture-intensive states suffered more during the first year of the Great Depression. This is consistent with a story in which lower agricultural prices depressed farm incomes and farm consumption and investment. We would also like to know what sort of farm areas did worst in 1930. If spending declined most in areas growing crops whose prices declined most, this would support our argument that lower farm product prices and income lowered farm area spending. To investigate this, we look at how state performance varied with the type of agricultural product produced.

RESULTS BY CROP

Table 3 shows the relationship between the auto sales growth in the first year of the Depression and the production of two categories of farm products: internationally traded crops and nontraded farm products. In distinguishing between traded and nontraded farm products, our goal is to distinguish between those farm products whose prices were likely to be strongly influenced by world demand (e.g., cotton) from those less influenced by world demand (e.g., milk). Whether it reflects the impact of world demand or not, the division between traded and nontraded roughly captures the division between farm products whose prices collapsed in 1930 and those whose prices fell more modestly (Table 1 and Figure 3 ). We define traded crops to be the value of cotton, tobacco, cereals, and wool production. While not all cereals were traded, their substitutability meant that their prices often moved together (Hausman, Rhode, and Wieland Reference Hausman, Rhode and Wieland2019). Nontraded farm production equals nontraded crop production plus the value of dairy and livestock sold. Nontraded crop production is equal to the value of total crop production minus the value of traded crops. We also include corn in the nontraded crops category. Corn is a cereal, and corn prices may have moved with other cereals prices; but movements in corn prices did not necessarily directly impact farm incomes, because corn was often grown to feed hogs. An increase in the market price of corn had little or no effect on many corn farmers’ incomes since the same farmers were using the corn to feed their hogs.Footnote ²²

Table 3

CROSS-STATE REGRESSIONS BY FARM PRODUCT TYPE

Notes: The dependent variable is the percent change in non-seasonally adjusted car sales from the 1929:Q2–Q3 average to the 1930:Q2–Q3 average. Corn is included in nontraded crops because of its use in hog production; p.c. means per capita. Robust standard errors in parenthesis.

*p < 0.1, **p < 0.05, ***p < 0.01.

Sources: Per capita farm product production—the 1940 Census as reported in Haines, Fishback, and Rhode (Reference Haines, Price and Paul2015); all other variables—see Table 2.

The first column of Table 3 shows that traded crop production was much more correlated with auto sales than was nontraded farm product production. The coefficient on traded crop production of –0.15 implies that a one-standard-deviation increase in traded crop production ($39) would have resulted in 5.7 percentage points lower auto sales growth. The coefficient is estimated precisely, with a t-statistic above 4 and a 95 percent confidence interval of [–0.21, –0.08]. Thus we can be confident that there was an economically significant relationship between traded crop production and economic performance.

By contrast, conditional on traded crop production, there is no evidence of a negative association between nontraded farm product production and car sales growth in the first year of the Depression. The conclusion remains the same in Columns (2) and (3) when we add control variables. In sum, the cross-state results show that it was traded crop production rather than agricultural production as a whole that was associated with a more severe beginning of the Depression. This is consistent with the much larger declines in traded crop prices than in the prices of farm products as a whole.

Evidence from Ohio Counties

To obtain further evidence on the relationship between agriculture and the severity of the early Great Depression, we collected data on new car registrations in Ohio counties. To our knowledge, these are the only available county-level data on new car registrations at an annual or higher frequency in 1929/30. The data come from the Bulletin of Business Research prepared by the College of Commerce and Administration of the Ohio State University. The data are monthly and are presented as “Registrations of New Automobile Bills of Sales in Ohio Counties” with the source specified as “Clerks of Courts of Listed Counties.”Footnote ²³ Unfortunately, the data do not cover all counties: we have data for 50 of the 88 counties in Ohio. But these counties accounted for most car sales; in 1928, more than 80 percent of all new car sales in Ohio occurred in these 50 counties.Footnote ²⁴

The Bulletin of Business Research presents data on both new passenger car sales and new truck sales. Unfortunately, however, there are too few counties with substantial truck sales to make the truck sales data useful for understanding the early Great Depression. Thus we confine ourselves to an analysis of the new passenger car sales. This has the added advantage of easy comparability with our cross-state results, which are only for passenger cars.

Figure 10 presents a cross-county scatter plot analogous to the cross-state scatter plot in Figure 8. Across Ohio counties, there is no clear relationship between farm population share and new car sales growth. Close inspection of the data, however, reveals that the null result is driven by a few counties, in particular, Gallia, Geauga, and Union. These three counties had large portions of their population living on farms but relatively low values of crop production. As discussed previously, it is crop-producing areas that we expect to have suffered most at the beginning of the Depression, because crop prices fell more than farm product prices as a whole. Figure 11 thus shows a similar scatter plot, but with the value of crops sold per capita rather than the proportion of the population living on farms on the x-axis. As expected, here there is a more obvious negative relationship. This results in part from the shift of Gallia, Geauga, and Union Counties from the far right of the graph to near the middle, reflecting their mid-range crop production despite large farm population shares.

Figure 10

OHIO COUNTIES: PERCENT CHANGE IN NEW CAR SALES AND FARM POPULATION SHARE

Sources: Car sales – see text; farm share of the population – Haines and ICPSR (Reference Haines2010).

Figure 11

OHIO COUNTIES: PERCENT CHANGE IN NEW CAR SALES AND VALUE OF CROPS SOLD PER CAPITA

Sources: Auto sales – see text. Crops sold per capita – Haines, Fishback, and Rhode (Reference Haines, Price and Paul2015).

To more formally investigate the relationship in Ohio between agricultural intensity and performance early in the Depression, we run regressions across counties like those estimated across states in the previous section. Specifically, we estimate

(2) $$\% \Delta {\rm{Auto}}\;{\rm{sale}}{{\rm{s}}_{j,{\rm{1929}}:{\rm{Q2 - Q3 - 1930}}:{\rm{Q2 - Q3}}}} = {\beta _0} + {\beta _1}{\rm{Agricultural}}\;{\rm{exposur}}{{\rm{e}}_j} + \gamma '{{\rm{X}}_{\rm{j}}}{\rm{ + }}{\varepsilon _{\rm{j}}},$$

where $$\% \Delta {\rm{Auto}}\;{\rm{sale}}{{\rm{s}}_{j,{\rm{1929}}:{\rm{Q2 - Q3 - 1930}}:{\rm{Q2 - Q3}}}}$$ is new auto sales growth in county j at the beginning of the Depression, “Agricultural exposure” is a measure of exposure to falling farm product prices, and X is a set of control variables.

Columns (1) and (2) of Table 4 show the single variable regressions corresponding to the scatter plots in Figures 10 and 11. Columns (3) and (4) add controls for population and 1928 car sales per capita. Unsurprisingly given the scatter plot (Figure 10 ), with and without controls the coefficient on farm share is small and insignificant. By contrast, with and without controls, the coefficient on crop sales per capita is economically and statistically significant. Its magnitude (–0.13) is somewhat larger than that in the cross-state regression with these controls (Column (5) of Table 2). Thus the results support the cross-state finding of an economically significant relationship between the importance of crops in a county and the depth of the Depression in 1929/30. Columns (5) and (6) explore the relationship between traded and nontraded farm product production and auto sales. As in the cross-state results in Table 3, across Ohio counties, the negative impacts of agriculture are driven by the cultivation of traded crops. But the conclusion is more tentative in the Ohio county data than it is in the state data, since when controls are added (Column (6)) the coefficient on traded crops loses statistical significance.

Table 4

CROSS-COUNTY REGRESSIONS

Notes: The dependent variable is the percent change in non-seasonally adjusted new car sales from the 1929:Q2–Q3 average to the 1930:Q2–Q3 average; p.c. means per capita. While we observe monthly new sales in 1929/30 in 50 counties, there are only 49 observations since 1928 sales were not reported for Morgan County. Robust standard errors in parenthesis. *p < 0.1, **p < 0.05, ***p < 0.01.

Sources: New car sales—see text; population and percent of the population on farms—the 1930 Census as reported in Haines and ICPSR (Reference Haines2010); 1929 value of crops sold per capita and farm product categories—the 1940 Census as reported in Haines, Fishback, and Rhode (Reference Haines, Price and Paul2015); 1928 car sales—calculated from the Industrial and Commercial Ohio Yearbook (1930, table XVI, p. 104), which lists by county both 1929 new car sales and the 1928/29 percent change in new car sales.

Unlike the state results, which change little when weighted by population (Online Appendix Table E.1), some of the cross-county specifications are sensitive to population weighting. Specifically, Online Appendix Table E.2 shows that the univariate results (Specifications (1) and (2)) become stronger, with more evidence of a negative relationship between auto sales and farm share or crops sold per capita. But the results with controls (Specifications (3) and (4)) become weaker. When weighted, there is no longer a negative coefficient on crops sold per capita. Reassuringly, weighting strengthens the finding in Columns (5) and (6) that traded crop production drove worse economic performance while nontraded farm product production did not.

Taken together, the cross-county data are supportive of the findings from the cross-state data. The county sample is too small and noisy for precise, statistically significant conclusions in all specifications, but the results support the robust message from the cross-state data that areas producing traded crops suffered most in the first year of the Depression.

Narrative Evidence

Further evidence for the effect of lower farm product prices on car sales comes from narrative evidence. Narrative evidence itself does not establish the importance of a channel from farm product prices to car sales. But combined with the previous quantitative evidence, it is reassuring. That contemporaries noticed the channel from farm product prices to auto sales suggests that the effect was significant; it supports the mechanism we posit, in which lower farm product prices reduced farmers’ incomes and hence their expenditure.

Narrative evidence comes from the publication Automobile Topics, which reported on car sales conditions around the country. In the summer and fall of 1930, it reports many instances of lower farm product prices depressing sales. The 2 August 1930 edition includes these reports:

• “IOWA - Grain prices are too low. Farmers will not buy” (p. 1028).

• “MICHIGAN - Better prices for agricultural products would help sales” (p. 1028).

• “SOUTH DAKOTA - Prices of farm products hurting our business” (p. 1028).

Similar comments were made in the 4 October 1930 issue:

• “SOUTH CAROLINA - Prices on cotton and tobacco retarding sales” (p. 678).

• “GEORGIA - Low price of cotton hurting” (p. 678).

• “NORTH DAKOTA - Poor crops and low grain prices retarding sales. There can be no prosperity here until grain prices go up” (p. 678).

To be sure—and consistent with our argument—low farm product prices are far from the only factor discussed in Automobile Topics. Many quotes are also to be found on depressed conditions in manufacturing, on complaints about banks, and on idiosyncratic local conditions. Like our quantitative evidence, however, the narrative evidence is consistent with a large role for lower farm product prices in explaining lower auto sales in 1930.