Investment during the great depression: uncertainty and the role of the Smoot-Hawley tariff.

Feldman, David H.

Industry and commerce are now in many instances unable to conclude their plans and many programs of development are waiting or threatened with abandonment, while party politics, bloc politics, personal politics and sectional politics are maneuvering for this or that advantage or to avenge this or that defeat or embarrass this or that political personality.

The House of Representatives has performed its duty. Why cannot the Senate act or frankly acknowledge its incapacity to legislate and let the country know the conditions to which our slowing economic machine must adjust itself to move forward?

John E. Edgerton President, National Association of Manufacturers New York Times, November 11, 1929, Section 1, page 2

1. Introduction

The political process set in motion in early 1929 that led ultimately to the Smoot-Hawley tariff likely generated a climate of business uncertainty. We test whether this uncertainty affected business fixed investment during 1929 and the early 1930s. The theory of irreversible choice under uncertainty supports the idea that a climate of increased uncertainty can create an investment cycle. Bernanke (1983a) argues that irreversibility of investment decisions means that firms must make timing decisions that trade off extra gains from early commitment against the benefits of waiting for more information. Events with unclear long-run consequences can depress investment by increasing the returns to waiting. Pindyck (1991) likens an irreversible investment opportunity to a financial call option. The standard net present value rule fails since the value of a unit of capital must exceed its purchase and installation price by an amount equal to the value of keeping the investment option alive. Thus, changing economic conditions that affect the perceived riskiness of future cash flows may have an effect that dominates a change in, for instance, real interest rates.

The effect on investment of trade regime uncertainty should vary based on the international exposure of the industry. If an industry has a large export market, competes with foreign firms in the domestic market, or relies on imported inputs, threatened changes in tariff policies will be very important. This description of exposure likely comprises a sizeable fraction of the economy, including most traded-goods industries. Plus, if the capital/output ratio were higher in trade-exposed industries than in the rest of the economy (which includes labor-intensive services), then the share of an investment decline attributable to trade regime uncertainty could be quite large.

This paper brings together a number of data sources to evaluate the behavior of investment during the depression. We use cross-sectional net investment data from 1927-1936 at the two-digit standard industrial classification level to build a neoclassical investment function augmented by two measures of international exposure. The exposure variables permit us to test the hypothesis that trade regime uncertainty was related to declines in investment during the downturn that became the Great Depression. Our measures of exposure prove important in explaining investment behavior at the start of the downturn (1929) but not in later years (1930-1933). Thus, the tariff was likely one of the triggers that precipitated the downturn that other factors, such as adherence to the gold standard, then exacerbated into the Great Depression.

2. Lessons from the Great Depression

Recent years have seen a significant evolution in our understanding of the causal processes at work during the Great Depression. In the traditional view, as Meltzer (1976) has phrased it, the Smoot-Hawley tariff converted a "sizeable recession into a severe depression" (p. 469). This view has roots back to the Depression itself. In the election of 1932, Roosevelt charged Hoover with complete responsibility for the Depression. Among its domestic causes, he argued, was the new tariff regime that inhibited foreign debt repayment and elicited strong trade retaliation. Lewis (1949) provides a postwar expression of this position. Trade policy could have imparted a contractionary shock to the U.S. through a number of channels. Gordon and Wilcox (1981) posit three. Directly, without any retaliation, the resulting increase in the price of U.S. imports and close substitutes altered the division of the nominal income decline between output and prices in 1930-1932, so output fell more than otherwise and prices fell less. Retaliation could also lower aggregate demand through the trade multiplier. Lastly, foreign retaliation against U.S. food exports could have aggravated the decline in farm prices, worsening the banking crisis and in turn the decline in the supply of money due to currency hoarding.(1)

Much recent research is corrosive to this orthodox view. Eichengreen (1992) argues that the gold standard itself was central in transmitting contractionary shocks from nation to nation. He concludes that nations that abandoned gold fared significantly better in the 1930s than did those who clung to the mechanism longer. Eichengreen (1994) also argues that the collapse of world trade in the 1930s was likely caused by the Depression and not a cause of it.(2) In his view, the tariff was either macroeconomically insignificant or marginally stimulative. "By applying modest inflationary pressure in an environment of low and falling prices, tariffs actually worked to close that Okun's Gap" (Eichengreen 1994, p. 45). He has also shown (Eichengreen 1989) that one can construct a static macro model in which worldwide increases in barriers to trade redirect spending without reducing it. His simulations suggest modest positive output and employment effects of tariffs and retaliation in the 1930s. Romer (1993) also rejects Smoot-Hawley as a causal factor in the Depression. She suggests that the onset and early course of the Great Depression in the U.S., while part of a worldwide phenomenon, can be explained can essentially in a closed economy model. She discounts the role of international trade since U.S. net exports played a smaller role in the decline in output in 1929-1931 than in previous downturns in the 1920s.

Perhaps the most succinct dismissal of the tariff comes from the first chapter of Temin's (1989) Lessons from the Great Depression. He argues that a tariff, like a devaluation, is an expansionary policy since it diverts demand from foreign to domestic goods. Furthermore, he says (see p. 46)

It may thereby create inefficiencies, but this is a second-order effect. The Smoot-Hawley tariff also may have hurt countries that exported to the United States. The popular argument, however, is that the tariff caused the American Depression. The argument has to be that the tariff reduced the demand for American exports by inducing retaliatory foreign tariffs. (Eichengreen 1989)

Exports were 7 percent of GNP in 1929. They fell by 1.5 percent of 1929 GNP in the next two years. Given the fall in world demand in these years from the causes described here, not all of this fall can be ascribed to retaliation from the Smoot-Hawley tariff. Even if it is, real GNP fell over 15 percent in these same years. With any reasonable multiplier, the fall in export demand can only be a small part of the story. And it needs to be offset by the rise in domestic demand from the tariff. Any net contractionary effect of the tariff was small. (Dornbusch and Fischer 1986, pp. 466-470)

This focus on net exports as the primary channel for international factors to affect domestic demand misses another potentially important consequence of changes in the international business climate. In particular, macro models that do not incorporate uncertainty may understate the effect on investment of cash flow uncertainty from doubts about the future structure of the trade regime.(3) Since internationally exposed industries comprise a large fraction of the economy, neither the ratio of exports to GNP nor changes in net exports is a sufficient indicator of the likely consequences of trade regime change. If a feedback existed from trade regime change to domestic investment, then the tariff may have been a macroeconomically relevant component of the early years of the Depression.(4)

[TABULAR DATA FOR TABLE 1 OMITTED]

3. The Smoot-Hawley Tariff as a Source of Business Uncertainty

The Smoot-Hawley Tariff legislation passed the House of Representatives on June 14, 1930, and on the following day, the New York Times published a "Chronology of the Tariff Bill From Jan., 1929 to June, 1930" (section 1, p. 26). Table 1 gives this chronology. If anything, it understates the legislative turmoil surrounding this bill. The initial reason for reconsidering the tariff code was to provide more protection for agricultural products. However, as Taussig (1931) explains in The Tariff History of the United States, log-rolling in the House Ways and Means Committee (chaired by Representative Willis C. Hawley of Oregon) and the Senate Finance Committee (chaired by Senator Reed Smoot of Utah) soon involved duties on many manufactured products as well as agricultural products. The process of log-rolling was enhanced by the organization of the House Ways and Means Committee, in which the 15 members of the Republican majority were each given subcommittees to chair dealing with a particular tariff schedule. Each subcommittee had three members, a subcommittee chair, and two other members who were themselves chairs of other subcommittees. The proceedings were behind closed doors, so there was ample opportunity for vote trading.(5)

The bill that eventually passed the House represented a considerably larger alteration of the tariff schedules than had been planned originally, and the bill was changed considerably by the Senate Finance Committee. While the bill was before the full Senate, the leadership lost control of the debate to a group of Democrats and progressive Republicans.(6) Taussig explains the Senate amendment process that followed.

As the individual items were taken up in the Senate and became subject to amendment from the floor, the changes were sometimes in one direction, sometimes in another. There was no rhyme or reason in it all; a deviation from the agreement here, a return to it there; duties shoved up on one motion, then shoved down on the next. The situation toward the close of the Senate's proceedings was nothing less than chaotic (Taussig 1931, pp. 498-499)

The length of the debate, the closeness of the final vote in the Senate (44 to 42), and the fact that several leaders of the Senate made serious threats to kill the bill altogether made it unclear that any tariff change would occur.(7) Also, as the bill progressed through the legislative process, its provisions faced changes at every step. The history of the duties on hides, which were free under the existing tariff, illustrates this nicely.(8) In the bill as passed by the House, there was a 15% duty on hides; in the bill as presented to the Senate by the Finance Committee, there was a 17 1/2% duty; in the bill as fixed by the Senate in the Committee of the Whole, there was no duty; in the bill as passed by the Senate, there was no duty; and in the bill as enacted, there was a 10% duty on hides.

Determining the fate of any part of the legislation that became the Smoot-Hawley tariff would have been very difficult for any observer, either inside the Congress or as an interested party outside the Congress. The cloud of uncertainty about the final outcome of the legislative process could well have influenced business decisions as asserted in the quotation by John Edgerton, the President of the National Association of Manufacturers. Political uncertainty may have caused firms to delay investment projects, thus slowing the economy.

President Hoover's signature on the tariff bill did not eliminate all uncertainties associated with the tariff. After U.S. approval of the tariff, the uncertainty shifted to foreign reactions. Firms that relied on export markets would be concerned about how rapidly and how completely other countries would retaliate. Firms that previously relied on imported inputs would have been concerned about how effectively and at what price domestic suppliers would be able to replace imports. A major change in tariff schedules such as the Smoot-Hawley tariff changed many relative prices. Furthermore, since foreign retaliation has to be accomplished by governments, some of which might be as indecisive as our Congress, the uncertainty surrounding the eventual shape of the economic landscape after the signing of the tariff bill could have lasted for quite some time. Cooper (1992, p. 2125) argues that Smoot-Hawley was no ordinary tariff. It was instead a major shift of the trade regime that "above all foster(ed) a climate of political recrimination and unpredictable reaction that must have greatly increased doubts in the minds of businessmen about investing in any activity for which foreign markets were important."

4. Investment Model

Our investment model is based on the neoclassical investment model pioneered by Jorgenson and several collaborators.(9) The starting place for this model is the assumption that a firm has a desired capital stock in any period, [Mathematical Expression Omitted], which is a function of its output level [Q.sub.t] and the user cost of capital [R.sub.t]. Net investment is then given by the flexible accelerator as a function of the past changes in the desired capital stock, that is,

[Mathematical Expression Omitted]. (1)

For the estimations below, we use what Chirinko (1993) terms the modified Neoclassical Model, in which changes in output and the changes in the user cost of capital, the determinants of changes in the desired capital stock, enter the equation separately, that is,

[I.sub.t] = [summation of] [[Alpha].sub.j][Delta][Q.sub.t-j] where j = 0 to Jq + [summation of] [[Gamma].sub.j][Delta][R.sub.t-j] where j = 0 to Jr. (2)

The coefficients [Alpha] and [Gamma] capture several effects, including the [Beta]'s from Equation 1 and the coefficients in the function that defines how [Mathematical Expression Omitted] depends on [Q.sub.t] and [R.sub.t].

We add two international exposure variables to the investment function. One captures the importance of export markets and the other measures the importance of imported inputs. Ideally, we would also like an openness measure for import competition. Leontief's 1929 input-output table for the U.S., from which our openness measures are derived, does not contain information that would allow us to determine the extent to which a given industry's output competes with imports.

Uncertainty about the tariff structure would increase the variance of a firm's cash flow, working through both of the exposure variables. Tariffs were raised on many products and inputs while many others were added to the free list. Even for products whose tariffs moved little, significant changes were often proposed at some point in the legislative process. For firms engaged in export activities as well as for those who used imported inputs, these changes (and proposed changes) would have increased the confidence interval around any industry's expected effective rate of protection.(10) With the possible exception of agriculture, forecasting the direction of change in an industry's effective rate of protection would have been daunting indeed.

Measures of international exposure are clearly not pure measures of domestic tariff-policy risk. They may also include the direct distortionary impact of actual changes in tariff levels as well as uncertainty originating abroad.(11) Despite these difficulties, we chose to use these variables primarily because there are no direct measures of domestic tariff-policy risk. Our investment data are at the two-digit industry level, and we have no way to disentangle how tariff proposals on highly disaggregated products and inputs would affect uncertainty at the industry level. Even if we could conquer this aggregation problem, we face further difficulties. Uncertainty could be affected both by formal proposals in Congress (for which there is a written record) and by rumors of back-room negotiations real and imagined (for which there is no written record). Given the amount of vote trading in committees known to have accompanied the legislative process in this case, the volume and variability of the rumors surrounding the tariff would have been very large. Thus we do not think it is possible to construct a variable that directly measures the uncertainty created by the tariff.

Uncertainty abroad could also affect U.S. investment in a manner related to openness. European price levels and tariff policy, for instance, were quite volatile in the 1920s. Yet the extreme volatility of prices that characterized the early 1920s was not evident in the years that prove most relevant to this study (1928-1930), while tariff changes were sprinkled throughout the decade.(12) And after the passage of the tariff, uncertainty originating abroad is subject to the same measurement problems discussed above.

Dixit and Pindyck (1994) show formally how uncertainty would decrease investment spending. They consider an industry with a large number of firms, each of which acts competitively, is risk neutral, and has rational expectations about underlying stochastic processes. Each firm can produce a flow of one unit of output if it incurs a sunk cost. Suppose, also, that the price for any one firm's output is

P = YD(Q), (3)

where Y is an industry wide shock, Q is output and D(Q) is a decreasing function that captures the nonstochastic part of the firm's inverse demand function. Each firm knows that a positive shock to Y makes entry equally attractive to all firms. Entry by other firms shifts the industry supply curve to the right, so the price rises less than proportionately with Y. Therefore, domestic price, and hence profit flow, is a concave function of Y. Increased uncertainty in Y thus reduces the expected value of investing relative to not investing.

Average firm size is an important control variable in our investment function. The debt-deflation literature developed by Mishkin (1978) and Bernanke (1983b) predicts a negative relationship between investment and average firm size during the banking crises of the early 1930s. They link changes in the real economy to financial-sector disturbances using asymmetric information models. In their view, debt-deflation-induced debtor insolvency and public loss of confidence in financial institutions interacted to reduce investment. The waves of domestic bank failures destroyed much local knowledge about potential borrowers, thus raising the cost of credit intermediation and lowering investment. This work implies that the banking crises would have had a disproportionate impact on small firms. Their borrowing costs are presumed equal to the observed rate on safe loans (to large firms and the government) plus a credit intermediation premium that increased during a bank crisis. Larger firms had greater access to securities markets and most had sufficient reserves of cash and liquid assets.

Temin (1989) criticizes the debt-deflation view because he asserts that this major prediction of the hypothesis fails a simple cross-section test. Temin uses production data in two-digit industries and two measures of firm size (concentration ratios and the incidence of identifiable large firms) to show that the presence of large firms is positively related to the fall in production. Since we have information on average firm size by industry we can provide an additional test. Our net investment model controls for other industry characteristics that might be correlated with average firm size.(13)

5. Data

The starting point for the model is the data on net investment in plant and equipment in 1972 dollars by 16 two-digit industries calculated by Bernstein (1987) in The Great Depression: Delayed Recovery and Economic Change in America, 1929-1939. These data are from 1927 to 1940, but unfortunately they are not quite complete. Net investment in 1937 for Industry 27, Printing and Publishing, is missing. Because cross-section information is more important to us than time-series detail, we decided to include Printing and Publishing and not use the data from 1937 through 1940.(14) Given this, our task is to match data for output and the user cost of capital to the industries in Bernstein for 1927 through 1936. We used the indexes of industrial production from the Federal Reserve Board as our measures of output. Unfortunately, we were unable to find such an index for Industry 34, Fabricated Metal Products. This cut the size of our cross section to 15 industries. Calculations and data sources involved in constructing the user cost of capital are discussed in detail in Appendix A. In this section, we focus on the measures of international exposure and average firm size.

International Exposure

Leontief's (1941) classic The Structure of the American Economy, 1919-1929 provides the information we need to calculate our measures of international exposure. That book's Table 6, which is contained in a pouch attached to the back cover of the book, is the detailed input-output table for the American economy in 1929. Rows of the table give the disposition of output of an industry, and columns of the table give the sources of inputs used by an industry. The entries in the body of the table, [X.sub.i,j], then give the outputs of industry i used as inputs by industry j. Our first task was to compress the input-output matrix so that the industry definitions corresponded to the industries in the 1972 SIC. Appendix B is a concordance, which shows how we matched the industry definitions in Leontief's input-output table with the 1972 SIC.

Export exposure EX is the share of the industry's gross output that was exported. Leontief's table lists exports as one possible disposition of each industry's output. The variable IM was a bit more difficult to construct. For each industry, we first determined [d.sub.j], the percentage of inputs that were directly imported. Imports were listed as one of the sources of inputs of each industry, and [d.sub.j] was constructed as imports as a percentage of gross outlays (G[O.sub.j]) of the industry. The next step in the construction of IM was to account for indirect imports. Indirect import exposure is import exposure derived from the fact that the suppliers to an industry rely on imports. The concept is probably best explained by an example. In the automobile industry import exposure comes from two sources, direct exposure (the use of imported steel in the construction of auto bodies) and indirectly through the fact that some inputs, such as tires, are outputs of other industries that rely on imported inputs. To capture this indirect import exposure, we multiply the direct import exposure of each supplying industry by the amount of inputs supplied, sum [TABULAR DATA FOR TABLE 2 OMITTED] across industries, and divide the result by gross outlays of the industry. Mathematically, IM is constructed as follows:

I[M.sub.j] = [d.sub.j] + [[summation over j] [d.sub.i][X.sub.ij]]/G[O.sub.j], (4)

where the first term on the right-hand side is the direct import exposure and the final term is the indirect exposure.

The first three columns of Table 2 give the calculated values of EX, [d.sub.j], and IM for our 15 industries. The effect of calculating indirect import exposure is clearly the largest in Industry 23, Apparel and Other Textiles. The value of IM is 0.120, while the value of d is 0.060. This result follows since a very important input in Apparel and Other Textiles comes from Industry 22, Textile Mill Products, which has a value of 0.143 for d, a value for direct import exposure that is only exceeded by that of one other industry, Industry 30, Rubber and Miscellaneous Products. In the results reported below, we use IM as the measure of import exposure.(15) The two measures of international exposure are not strongly correlated with one another. The correlation coefficient is -0.2671.(16)

Average Firm Size

Our measure of average firm size, S, was constructed from data in the 1929 volume of Statistics of Income (U.S. Treasury Department 1931). The major task here was to construct a concordance between the industry categories in Statistics of Income and the 1972 SIC. This concordance is included in Appendix B. The measure of average firm size in the industry was calculated as the average gross income per firm in the industry. Gross income is defined as [TABULAR DATA FOR TABLE 3 OMITTED] follows: "Gross income corresponds to total income as reported on the face of the returns, plus the cost of goods sold" (p. 293). The values of S are included as the final column of Table 2. The industries with the largest average firm size, 29, Petroleum Refining; 36, Transportation Equipment; and 21, Tobacco Manufactures, are not surprises. Interestingly, the measure of average firm size is not correlated with either measure of international exposure described above. The correlation of S with EX is -0.0698, and the correlation of S with IM is 0.0090.

Table 3 summarizes the data we will use in the regression analysis below.

6. Results

We first estimated cross-section regressions for each year. With only 15 cross-section observations, we had to economize on the number of regressors. From the modified neoclassical investment model described above, we included the current value and one lag of the change in output, [Delta]Q, and [Delta][Q.sub.t-1], and the current value and one lag of the change in the user cost of capital, [Delta][R.sub.t], and [Delta][R.sub.t-1]. We also included the measures of international exposure, EX and IM, to capture the effects of uncertainty created by the Smoot-Hawley tariff and average firm size S to control for the possible differential effects of debt deflation. In these regressions, we also included the previous year's investment [I.sub.t-1] as a control for the size of the industries. The basic model we are using was derived for use with time series data and, controlling for industry size, we think it is a reasonable model for cross section purposes.

Table 4. Cross Section Investment Regressions

 Year
 1928 1929 1930

Constant -153,965 -53,091 800,587
 (0.97) (0.13) (0.63)
[I.sub.t-1] 0.63487 1.18143 0.89306
 (5.31) (6.64) (3.36)
[Delta]Q 483.676 -87.627 52.328
 (1.74) (0.35) (0.14)
[Delta][Q.sub.t-1] -137.871 -209.661 -70.298
 (0.20) (0.55) (0.13)
[Delta]R -8,674,539 5,000,936 19,204,336
 (0.41) (0.81) (1.02)
[Delta][R.sub.t-1] -2,020,220 -901,111 -4,114,001
 (0.13) (0.12) (0.35)
EX -1,519,129 -1,629,551 667,872
 (0.99) (1.17) (0.25)
IM 595,414 -2,022,502 1,722,012
 (0.48) (2.63) (0.98)
S 0.036435 0.000038 -0.052713
 (2.03) (0.00) (1.47)
[R.sup.2] 0.91 0.95 0.74
F 7.68 13.02 2.16

In parentheses below the estimated coefficients are the t
statistics.

Table 4 presents the regressions for 1928 through 1934.(17) Not surprisingly, the early 1930s are a difficult time during which to predict investment behavior, and our efforts met with mixed success. Many coefficients have incorrect signs and several equations are totally insignificant. Nevertheless, we find some interesting results. Regarding international exposure, the 1929 equation has a statistically significant negative coefficient for IM, which is consistent with our hypothesis that uncertainty surrounding the Smoot-Hawley tariff would reduce investment disproportionately in industries with large values of international exposure. Regarding the effect of average firm size, the coefficient on S in the 1933 equation is positive and statistically significant, suggesting that the large number of bank failures in 1933 would have made obtaining financing for investment more difficult for small firms.

Before we interpret these results as support for the hypotheses in question, we must recognize that EX, IM, and S measure industry characteristics, and there may well be other industry characteristics omitted from our equations that correlate with these characteristics. To check for this type of problem, we ran a pooled regression for the 1928-1936 period using a fixed effects [TABULAR DATA FOR TABLE 5 OMITTED] model omitting EX, IM, and S.(18) The industry dummy variables in this equation will capture any industry effects on investment. The results of this estimation are included as Appendix C. All of the signs are correct in this equation, but few of the coefficients are statistically significant. As a second test of the hypotheses of interest we used EX, IM, and S as independent variables in regression on the year-by-year residuals from the pooled regression. The dependent variable in these regressions should be purged of the effects of other industry characteristics on investment.

Table 5 contains the regressions of EX, IM, and S on the residuals from the fixed effects equation. The result for international exposure survives this test. In 1929, the coefficients for EX and IM are reduced a bit in absolute value, from -1,629,551 to -1,486,033 for EX and from -2,022,502 to - 1,624,990 for IM, and the coefficient on IM continues to be statistically significant. These coefficients in the 1929 equation for both EX and IM are negative and dramatically larger in absolute value than in any other year. This result suggests that a large international exposure impeded investment in 1929, the year most confounded by uncertainty concerning the outcome of the tariff deliberations in Congress.

The results regarding size tell a different story. The coefficient on S for 1928 is the only one that is statistically significant. Most importantly, the coefficient of S for 1933 is positive, as it was in the regression in Table 2, but its value is dramatically smaller, 0.009588 compared to 0.077959, and it is no longer statistically different from zero. These results give no support to the notion that the average size of firms was related to investment during the waves of bank failures that occurred in the early 1930s.

We performed two additional robustness tests of our result, though with only 15 industries, we did not have many degrees of freedom to exploit. First, we eliminated each industry in turn from the regressions. The basic result from the first cross section and the regression on the residuals from the fixed effects model survived this test. The effects of EX and IM are strongly negative in 1929. As an additional robustness test, we changed the lag structure of the cross-section equation and found again that IM was statistically significant in the 1929 equation.(19)

Our results focus on 1929, the year before the passage of the Smoot-Hawley tariff. As we stated previously, our measures of international exposure (IM and EX) could capture effects of tariff changes and uncertainty abroad as well as the effects of domestic tariff-policy uncertainty. From this perspective, it is interesting that neither IM nor EX are statistically significant for 1930, the year in which the tariff changed, or thereafter.

Passage of the tariff act in 1930 also resolved some of the uncertainty in ways that might have affected investment. Hayford and Pasurka (1991) provide estimates of industry-level changes in effective rates of protection caused by Smoot-Hawley.(20) For 1930 and 1931, we added the actual change in the effective rate of protection ([Delta]ERP) to our investment model. This allows us to separate, especially for 1930, any certainty effect from our uncertainty effect.(21)

This experiment produced mixed results. In cross-section regressions (Table 6a) with and without the openness variables, the coefficient on [Delta]ERP is statistically significant and positive in 1930 and 1931. Industries that experienced increases in the effective rate of protection invested more. In Table 6b, we include [Delta]ERP in a regression on the residuals from a fixed effects model (the procedure used to generate Table 5). Only the result for 1930 remains significant. This result, however, is not robust when we remove outliers. Table 6c reruns the regression on residuals without Industry 1, Food and Kindred Products. This industry has a large value for [Delta]ERP, which is driven in large part by increases in protection for input-output sector 5 (Sugar, Glucose, and Starch) and input-output sector 9 (Butter, Cheese, etc.), both of which had exceedingly high effective rates of protection prior to Smoot-Hawley and even higher rates after Smoot-Hawley. The calculated ERP in sector 5, for instance, rose from 1800% to 2950%. Many of these agricultural tariffs were redundant in practice, so the [Delta]ERP calculations likely overstate the effect of the nominal tariff change. Dropping this industry from our group makes the coefficient on [Delta]ERP in the 1930 regression statistically insignificant.

The statistical unreliability of the results for 1930 should not be surprising. The tariff was passed in June of 1930. At this point, domestic political uncertainty about the tariff code was resolved, and this reduction in uncertainty should have stimulated investment, particularly in industries that were internationally exposed. Industries whose ERP increased dramatically would have been more inclined to invest than industries whose ERP increased only marginally or decreased. The rebound caused by resolved uncertainty and any effect of actual changes in the ERP occurred in the second half of the year. Political uncertainty over the tariff reigned in the first half of the year. If quarterly data were available, we would expect to see quite different investment behavior in the two halves of the year. With annual data, there is some evidence that the effects from the second half of 1930 are stronger, for example, positive coefficients on [TABULAR DATA FOR TABLE 6 OMITTED] EX and IM suggest some investment rebound from the resolution of the uncertainty and a positive coefficient on [Delta]ERP indicates more investment in industries that benefitted from the tariff, but these coefficients are not statistically reliable.

At this point, we want to return to our positive conclusion, the statistical significance of import exposure in 1929. Does this result represent an economically important event in the year 1929? Using the fixed effects model, the coefficient for IM in 1929 is - 1,624,990. This value implies that an extra percentage point of import exposure yields a reduction of investment that is 4.88% of the average net investment for the industries in the cross section. Given that the average value of IM is 0.082, and if we assume that the coefficient on this variable would have been zero in the absence of political uncertainty, the effect of eliminating the uncertainty in 1929 would be to increase investment in 1929 by 40% in our 15 industries.

As we have indicated earlier, the industries in our sample are all from the traded-goods sector, and they are likely to be among the more capital-intensive industries. These factors suggest that the estimate of a 40% drop in investment is a considerable overestimate for the entire economy. Also, our counterfactual exercise (setting IM to zero) likely generates an upper bound on the fall in investment spending. Nevertheless, the 15 industries in our sample are large. Net investment in these industries represented 42.8% of nonresidential fixed investment in 1929.(22) A 20% percent increase in investment in this large a portion of the total economy would have had a sizeable impact (8.5%) on aggregate investment.

7. Summary and Conclusions

Uncertainty surrounding the Smoot-Hawley tariff may have slowed investment spending. This uncertainty is conceivably of two types: domestic political indecision surrounding the lengthy tariff debate and uncertainty about reactions to the tariff. The political uncertainty should have started with the discussion of possible tariff changes in 1928, intensified in 1929 as the House and Senate debated various versions of the bill, and continued in the first half of 1930 until the bill eventually was passed and signed. The uncertainty about reactions to the tariff would then intensify and continue until it became clear how much, if any, retaliation the Smoot-Hawley tariff would generate.

Unfortunately, we do not have a direct measure of either of these uncertainties. We account for uncertainty indirectly by including variables measuring the international exposure of industries in cross-section regressions on net investment by industry. We argue that firms with higher international exposure (they export and/or use imported inputs) are more likely to slow investment spending because of uncertainty about tariff changes, so the coefficients on our international exposure variables provide a test of this hypothesis. Our results support the notion that political uncertainty had an effect on investment. The year 1929, a year of great political turmoil over the tariff, is the only year in which either of the variables measuring international exposure is statistically significant. Intensive use of imported inputs had a strong negative impact on investment. This is a clear positive conclusion of our statistical work. The tariff was far from macroeconomically irrelevant.

The fact that international exposure is not important in any of the other years suggests important negative conclusions as well. First, uncertainty about reactions to the Smoot-Hawley tariff do not appear to be important. Second, our measures of international exposure might pick up other effects of the tariff, retaliation to the tariff, or the general decline in international trade in the early 1930s, but they do not. This finding is consistent with the views of Eichengreen, Romer, and Temin, discussed in the Introduction, that the Smoot-Hawley tariff was not an important factor in the deepening of the Great Depression once it had begun.

Last, we discover no relationship between average firm size and investment behavior. This finding is inconsistent both with Temin's claim that the economic consequences of the banking crises were concentrated in larger firms and with Bernanke's hypothesis that increases in the cost of credit intermediation would disproportionately affect smaller firms.

While considerable debate persists about what caused the recession that started in 1929 to deepen into the Great Depression, no one doubts that a recession started in 1929. The National Bureau of Economic Research puts the business cycle peak in August of 1929. Between August and December, industrial production declined by almost 20% (Bernstein 1987). Almost half of this decline occurred before the October stock market crash. We have annual investment data and therefore can say nothing about the timing of events during 1929. Nonetheless, we find that investment in an important sector of the economy was substantially less than it otherwise would have been. This suggests that the political uncertainty surrounding the creation of the Smoot-Hawley tariff may have been one of the causal factors in the recession of 1929. Our work thus offers another channel through which the Smoot-Hawley tariff likely exerted a macroeconomic influence. While our results are consistent with the notion that Smoot-Hawley did not contribute to the deepening of the Great Depression through an uncertainty channel, they do suggest that Smoot-Hawley played an important role in the recession that later became the Great Depression.

Appendix A. User Cost of Capital

Chirinko (1993) gives the following definition of the user cost of capital:

[R.sub.t] = [PI.sub.t] ([r.sub.t] + [Delta]) (1 - [m.sub.t] - [z.sub.t])/(1 - [t.sub.t])

where [R.sub.t] is the user cost (or rental price) of capital, [PI.sub.t] is the purchase price of new capital (relative to the price of output), [r.sub.t] is the real financial cost of capital net of taxes, [Delta] is the geometric rate of capital depreciation, [m.sub.t] is the rate of the investment tax credit, [z.sub.t] is the discounted value of the tax depreciation allowances, and [t.sub.t] is the rate of business income taxation. For our project, we add an industry subscript to PI, [Delta], and z since these variables differ by industry.

Data Sources

[PI.sub.t]. To construct the relative purchase price of new capital goods we needed four pieces of information, the price of new equipment, the price of new plant, the price of output, and the relative importance of equipment and plant in overall investment. The first two price indexes are found in Chawner (1941). Chawner gives separate estimates for plant and equipment expenditures for 1915-1940 in current prices and in 1939 prices. From these two estimates, we can obtain separate implicit price deflators for expenditures on equipment and expenditures on plant. For 10 of the 15 industries, the price of output was taken from Wholesale Price Index (WPI) for Major Product Groups. The industries for which this was not possible were 21, Tobacco; 23, Apparel; 26, Pulp and Paper; 27, Printing and Publishing; and 35, Non-electrical Machinery. In several cases, we were able to find series that were highly correlated with the series we needed in an out-of-sample time frame, and we were able to estimate the price series we needed for our sample years. These estimates will be described in detail in the final section of this appendix. For the relative importance of plant and equipment expenditure, we are using industry level data from the 1939 Census of Manufacturers, the earliest census of manufacturers that provides the data in that way by industry.

[r.sub.t]. We are using the ex ante real interest rate calculated from data provided by Cecchetti (1992).

[Delta], Equipment. The basic source for this technique is the estimates in Hall and Jorgenson (1967). For manufacturing equipment, they use a depreciation rate of 0.1471. They say that [Delta] is "taken to be 2.5 times the inverse of the Bulletin F lifetime." Given this, the lifetime they must have used is 17 years (i.e., [Delta] = 2.5 (1/F) and thus F = 2.5/[Delta], given a [Delta] = 0.1471, F = 17).

We then theorized that an exponential depreciation scheme has to satisfy the following equation:

0 [e.sup.-[Delta]T] - X,

where T is the length of time the piece of capital is in use and X is the percentage of the piece of capital's original value left when it is scrapped. Using Hall and Jorgenson's numbers yields X = 0.08203. We take this value for X and assume it does not vary across industries. We have estimates of the average length of life for machinery and equipment from Creamer, Dobrovolsky, and Borenstein (1960). We use these values for T along with the calculated value for X to find depreciation rates for machinery in the industries.

[Delta], Structures. For structures, Hall and Jorgenson set [Delta] = 0.0625. This implies a life for structures of 40 years. Since we have no data that suggests the length of life for structures differs across industries, we will use [Delta] = 0.0625 for structures for every industry.

[m.sub.t]. There was no investment tax credit during this time period.

[z.sub.t]. This variable is calculated following a technique for straight-line depreciation found in Hall and Jorgenson given the ages found in Creamer, Dobrovolsky, and Borenstein.

[t.sub.t]. The tax rate was the corporation income tax rate for the relevant year.

Details on the Prices of Output

Industry Source of Price Information

20. Food and Kindred Products Wholesale Price Index (WPI)
 by Major Product Group-Food
 (1925-1940).

21. Tobacco Manufactures No price data were available
 for Tobacco until a CPI
 (consumer price index)
series
 that started in 1940. To
 estimate the price of
tobacco
 for the earlier period, we
 ran a regression for 1940
 through 1970 using the price
 of raw tobacco and the
 average compensation per
 employee in the tobacco
 industry as independent
 variables. Both of the
 independent variables had
 statistically significant
 coefficients, and the
 equation had an [R.sup.2] =
 0.989. Predicted values from
 this regression were then
 used for 1929 through 1940.
 The remaining four years
were
 added by backcasting from a
 regression in which the
 dependent variable is the
 price index in question and
 the independent variables
 were all of the other price
 indexes for which values for
 the complete period were
 available.

22. Textile and Textile Products WPI, Textile Products
 (1925-1940).

23. Apparel and Other Textiles Consumer Price Index (CPI),
 Apparel (1925-1940).

24. Lumber and Wood Products WPI, Lumber and Wood
Products
 (1926-1940).

26. Paper and Allied Products No price data for Paper and
 Allied Products were
 available until 1947 when a
 WPI for Pulp, Paper, and
 Allied Products became
 available. To estimate the
 price of paper in the
earlier
 period, we ran a regression
 from 1947 to 1970 using the
 price indexes of Lumber and
 Wood Products and Chemicals
 and Allied Products as
 independent variables. Both
 of the independent variables
 had statistically
significant
 coefficients, and the
 equation had an [R.sup.2] =
 0.831.
Predicted values from
 this regression were then
 used for 1926 to 1940.

27. Printing and Publishing No price data for Printing
 and Publishing were
available
 until 1947 when an implicit
 price deflator for this
 industry could be
calculated.
 To estimate a price index
for
 the earlier period, we ran
a
 regression from 1947 to 1970
 using the price indexes of
 Paper and Allied Products
and
 the average compensation per
 worker in Printing and
 Publishing. Both of the
 independent variables had
 statistically significant
 coefficients, and the
 equation had an [R.sup.2] =
 0.793. Predicted values from
 this regression were used
for
 1929 to 1940.

28. Chemicals and Allied Products WPI, Chemicals and Allied
 Products (1925-1940).

29. Petroleum and Coal Products WPI, Fuel and Lighting
 (1925-1940).

30. Rubber and Plastic WPI, Rubber and Plastic
 Products (1926-1940).

31. Leather and Leather Products WPI, Hides and Leather
 Products (1925-1940).

32. Stone, Clay, and Glass WPI, Nonmetallic Mineral
 Products (1926-1940).

33. Primary Metals Industries WPI, Metals and Metal
 Products (1925-1940).

35. Nonelectrical Machinery WPI, Metals and Metal
 Products (1925-1940). A WPI
 for Machinery and Equipment
 was available starting
 in 1939. A regression of
this
 price index on the price
 index for Metals and Metal
 Products for 1939 to 1970
 had a very good fit, an
 [R.sup.2] = 0.989, so we
felt
 safe in using the price
index
 Metals and Metal Products
for
 this industry.

37. Transportation Equipment WPI, Motor Vehicles and
 Equipment (1926-1940).

[TABULAR DATA FOR APPENDIX B OMITTED]

Appendix C. Full Fixed Effects Model

Constant 31,324 Petroleum 21,484
 (0.31) (0.21)
[I.sub.t-1] 0.50794 Rubber 22,617
 (6.56) (0.21)
[Delta]Q 4.9839 Leather 17,621
 (0.07) (0.17)
[Delta][Q.sub.t-1] 23.8539 Stone, clay and glass 60,534
 (0.33) (0.59)
[Delta]R -615,162 Metals 48,606
 (0.51) (0.46)
[Delta][R.sub.t-1] -8690.11 Machinery 23,883
 (0.01) (0.23)
Food 197,778 1928 91,273
 (1.85) (0.83)
Tobacco 15,407 1929 162,284
 (0.15) (1.48)
Textiles -94,924 1930 -90,901
 (0.92) (0.68)
Apparel 23,646 1931 -106,578
 (0.23) (0.81)
Lumber -42,233 1932 -147,596
 (0.41) (1.41)
Paper 17,709 1933 -210,045
 (0.17) (0.72)
Printing 11,121 1934 -65,332
 (0.11) (0.22)
Chemicals 56,310 1935 -78,527
 (0.55) (0.97)
[R.sup.2] 0.62
F 6.37

We would like to thank Mario Crucini, Barry Eichengreen, Lynne Kiesling, Robert Margo, Carl Moody, Ed Tower, and seminar participants at Duke University and at William and Mary for their comments. Jeffrey Previdi provided valuable research assistance. We also thank the anonymous referees for insights that have substantially improved the final product.

1 See Eichengreen (1989) for a good critique of these channels.

2 Irwin's (1996) partial and general equilibrium modeling indicates that the tariff itself likely reduced imports by 4-8% (ceteris paribus), and that nearly one quarter of the observed 40% decline in imports can be attributed to the increase in the effective tariff (Smoot-Hawley plus deflation).

3 Our argument thus complements Romer's (1990) work in which the stock market crash explains the onset of the downturn. She shows how the crash could have reduced the demand for durable goods by generating significant income uncertainty among households. Industrial production in the last half of 1929 was declining almost as rapidly before the crash as afterward. This leaves room for other explanatory variables such as cash-flow uncertainty induced by the tariff debate.

4 Crucini and Kahn (1996) offer another dissenting view. They use a multisector dynamic-equilibrium model with imported inputs to show how tariffs could significantly affect GDP even when trade is a small share of output. They argue that tariffs were at least as important as monetary and nonmonetary factors in contributing to the volatility of interwar output.

5 Taussig's account suggests that economic interests were more important than party discipline in determining the shape of the final tariff. A recent paper by Irwin and Kroszner (1995) finds empirical support for the hypothesis that the economic interests of Senate constituencies shaped the log-rolling coalitions that determined tariffs on specific goods.

6 The Republicans who broke ranks with the party had to weather the wrath of the party leadership. Senator George H. Moses of New Hampshire, the president pro tempore of the Senate and the Chair of the Senate Campaign Committee, went so far at to call the insurgent Republican senators "sons of the wild jackass" on the floor of the Senate (New York Times 1929b).

7 For example, on August 25, 1929, the New York Times reported, "From high Republican councils there was issued today a warning that should a filibuster develop, endangering passage of the proposed legislation at the special session or early in the regular session, a move would be made to defer action of the tariff for another year."

8 Taussig (1931, p. 508) gives this example.

9 See Jorgenson (1971) and Chirinko {1993) for review articles on empirical investment models.

10 This would also be true of the missing import-competition exposure measure.

11 Passage of Smoot-Hawley in 1930 does allow one to calculate changes in effective rates of protection that could affect investment in 1930 and beyond. See our Results section for a full discussion of this issue. We note also that, since tariffs were often specific duties (as opposed to ad valorem), changes in the price level mean that ERPs were changing every year in ways we cannot quantify at our level of disaggregation (see Crucini, 1994).

12 Actual tariff changes in our major trading partners may have affected investment through the export exposure term, but we have no way to quantify this issue at our level of disaggregation.

13 See Calomiris (1993) for another critique of Temin's test.

14 Net investment is also missing for Industry 29, Petroleum, in 1940, so we would lose another industry if we attempted to do a full time series.

15 We recognize that indirect import exposure could be calculated for additional rounds. Firms may know not only about the direct exposure of their suppliers but also about their suppliers' indirect exposure. Inclusion of additional rounds, however, does not materially change the story. The correlation between IM and a measure that includes the next round is 0.995.

16 Exposure is presumed constant over the time period. This assumption is more problematic for the out years of our sample (1933-1934).

17 Using the White (1980) test, our data show no evidence of heteroskedasticity.

18 We cannot include EX, IM, or S in a pooled regression because they have no time-series variation. The resulting model would not be of full rank.

19 The results of these robustness tests are available on request from the authors.

20 The industries used in Hayford and Pasurka (1991) come from Leontief's input-output table, so the concordance in Appendix B was used to combine these industries into the 1972 SIC. The effective rates of protection found in Hayford and Pasurka (1991) contained errors, which Pasurka has corrected in collaboration with us. The revised ERPs are available from us on request.

21 The ERP is an implicit subsidy to domestic value added coupled with a consumption tax equal to the tariff rate. If this value added subsidy raises the return to capital or residual profits, we can plausibly link increases in ERPs to higher investment. Yet this relationship should be interpreted with caution. The ERP measures the partial equilibrium incidence of the tariff structure instead of the general equilibrium incidence. For example, import taxes have the same general equilibrium incidence as export taxes, yet the measured ERP does not capture this. Other complications include induced price changes for other primary inputs (wages, e.g.) together with the existence of nontradeable inputs.

22 From the Economic Report of the President, nonresidential fixed net investment was $26.2 billion in 1982 dollars. Convening this figure to 1972 dollars using the implicit price deflator for nonresidential fixed investment yields a figure of $11.63 billion. Net investment in our 15 industries in 1929 was $4.98 billion 1972 dollars.

References

Bernanke, Ben S. 1983a. Irreversibility, uncertainty, and cyclical investment. Quarterly Journal of Economics 98:85-106.

Bernanke, Ben S. 1983b. Nonmonetary effects of the financial crisis in the propagation of the Great Depression. American Economic Review 73:257-76.

Bernstein, Michael A. 1987. The Great Depression: Delayed Recovery and Economic Change in America, 1929-1939. Cambridge: Cambridge University Press.

Calomiris, Charles W. 1993. Financial factors in the Great Depression. Journal of Economic Perspectives 7:61-86.

Cecchetti, Stephen G. 1992. Prices during the Great Depression: Was the deflation of 1930-1932 really anticipated? American Economic Review 82:141-56.

Chawner, Lowell J. 1941. Capital expenditures in selected manufacturing industries. Survey of Current Business 21: 19-25.

Chirinko, Robert S. 1993. Business fixed investment spending: A critical survey of modeling strategies, empirical results, and policy implications. Journal of Economic Literature 31:1875-911.

Cooper, Richard N. 1992. Fettered to gold? Economic policy in the interwar period. Journal of Economic Literature 30: 2120-8.

Creamer, Daniel, Sergei Dobrovolsky, and Israel Borenstein. 1960. Capital in manufacturing and mining: Its formation and financing. Princeton: Princeton University Press.

Crucini, Mario J. 1994. Sources of variation in real tariff rates, 1900-1940. American Economic Review 84:732-43.

Crucini, Mario J. and James Kahn. 1996. Tariffs and aggregate economic activity: Lessons from the Great Depression. Journal of Monetary Economics 38:427-67.

Dixit, Avinash K., and Robert S. Pindyck, 1994. Investment under uncertainty. Princeton: Princeton University Press.

Dornbusch, Rudiger, and Stanley Fischer. 1986. The open economy: Implications for monetary and fiscal policy. In The American business cycle: Continuity and change, edited by Robert J. Gordon. Chicago: University of Chicago Press, pp. 459-518.

Eichengreen, Barry. 1989. The political economy of the Smoot-Hawley Tariff. Research in Economic History 12:1-43.

Eichengreen, Barry. 1992. Golden fetters: The gold standard and the Great Depression, 1919-1939. New York: Oxford University Press.

Eichengreen, Barry. 1994. The international economy: Historical perspectives and future prospects. In The challenge of European integration: Internal and external problems of trade and money, edited by Berhanu Abegaz, Patricia Dillon, David H. Feldman, and Paul F. Whiteley. Boulder, CO: Westview Press, pp. 23-62.

Gordon, Robert J., and James A. Wilcox. 1981. Monetarist interpretations of the Great Depression: An evaluation and critique. In The Great Depression revisited, edited by Karl Brunner. Boston, MA: Kluwer-Nijhoff.

Hall, Robert E., and Dale Jorgenson. 1967. Tax policy and investment behavior. American Economic Review 57:391-414.

Hayford, Marc D., and Carl A. Pasurka, Jr. 1991. Effective rates of protection and the Fordney-McCumber and Smoot-Hawley Tariff Acts. Applied Economics 23:1385-92.

Irwin, Douglas A. 1996. The Smoot-Hawley Tariff: A quantitative assessment. NBER Working Paper No. 5509.

Irwin, Douglas A., and Randall S. Kroszner. 1995. Log-rolling, partisanship, and economic interests in the passage of the Hawley-Smoot Tariff. Unpublished paper, Graduate School of Business, University of Chicago.

Jorgenson, Dale W. 1971. Econometric studies of investment behavior: A survey. Journal of Economic Literature 9: 1111-47.

Leontief, Wassily W. 1941. The structure of the American economy, 1919-1929. Cambridge: Harvard University Press.

Lewis, W. A. 1949. Economic survey, 1919-1939. London: George Allen and Unwin.

Meltzer, Allen H. 1976. Monetary and other explanations of the start of the Great Depression. Journal of Monetary Economics 2:455-71.

Mishkin, Frederick. 1978. The household balance sheet and the Great Depression" Journal of Economic History 38: 918-37.

New York Times, 25 August 1929a.

New York Times, 9 November 1929b. p. 9.

New York Times, 11 November 1929c.

New York Times, 15 June 1930.

Pindyck, Robert S. 1991. Irreversibility, uncertainty, and investment. Journal of Economic Literature 29:1110-48.

Romer, Christina D. 1990. The Great Crash and the onset of the Great Depression. Quarterly Journal of Economics 105: 597-624.

Romer, Christina D. 1993. The nation in Depression. Journal of Economic Perspectives 7:19-40.

Taussig, E W. 1931. The tariff history of the United States. New York: G.P. Putnam's Sons.

Temin, Peter. 1989. Lessons from the Great Depression. Cambridge: MIT Press.

U.S. Treasury Department. 1931. Statistics of Income for 1929. Washington, DC: Government Printing Office.

White, Halbert. 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48:817-38.