Antidumping and Production-Line Exit: The Case of the US Steel Industry

USITC statistics from July 2012 report that steel mill products accounted for about 39 percent of all current AD/CVD orders in place. Data from Bruce Blonigen (http://​pages.​uoregon.​edu/​bruceb/​adpage.​html) indicate that from 1980 to 1995, steel petitions comprised about 40 percent of the total caseload, while data from World Bank’s Global AD and Countervailing Databases (developed initially by Chad Bown) show that from 1996 to 2010 steel petitions were 48 percent of the total caseload.

Other forms of protection include price supports (1978–1982), quotas against Europe and Japan (1969–1974, 1982–1984), comprehensive quotas (1984–1992), and safeguards (2002–2003).

See Blonigen and Wilson (2010) for an empirical analysis of this issue. Their results fail to find evidence to support this argument.

See, e.g., Bernard and Sjöholm (2003) and Van Beveren (2007).

The closest that the Census data come to the disaggregation of our within-plant production line data is in having some economic output (e.g., revenues) by product code for a plant. But this is different from our data, which contain capacity by production line, as the Census data may aggregate multiple production lines that are producing the same product in a plant.

In some cases, data were verified and/or taken from a variety of secondary sources. These include: Christopher Hall’s 1997 volume, Steel Phoenix—the Fall and Rise of the US Steel Industry; the University of Pittsburgh’s Center for Industry Studies ‘Steel Plant Database, The US International Trade Commission’s 2005 report, “Steel: Evaluation of the Effectiveness of Import Relief”; William T. Hogan’s 1994 volume, Steel in the United States: Restructuring to Compete, and Purchasing Magazine’s 1999 report, “North America Galvanized Steel Capacity.”

We define foreign ownership as having at least ten percent of the US firm owned by non-US investors (generally non-US steel firms), although in most cases, foreign ownership is greater than 30 percent. Also, in some cases, foreign investment takes place only for particular plant within a firm, or even for a particular product at a particular plant. Information for this variable came primarily from Hall (1997).

The minimum efficient scale of an integrated steel firm is generally much larger than that of a minimill, and minimills often carry cost advantages over integrated firms even when the latter are producing at efficient scales of operation. Through the years, the integrated producers held that the steel that they produced was “purer” than that of minimills and hence they could compete through quality differences. Recent technological innovations, however, have allowed minimills to encroach upon downstream markets (such as automobiles) that traditionally required integrated steel products.

AD and countervailing duties are product specific, and vary across our five different products (hot-rolled sheet, cold-rolled sheet, galvanized sheet, plate, and wire rod).

Feinberg and Hartigan (2007) study the inter-temporal link between AD investigations and plant closure. Their study, which includes US petitioners of dumping protection during the period 2003–2005, finds little evidence of any link between recent case outcomes and the timing of exit, regardless of whether the case

outcome was an affirmative or negative determination.

We also use survival analysis to check the robustness of our logit results. The results are qualitatively very similar, regardless of whether we use a survival or an exit model.

The inclusion of firm-specific dummy variables causes the sample size to decrease from 1415 to 1164. The is because some firms have no variation over time in the exit rate of their production lines, i.e. they never exit or they all exit during the same period. For such firms, the inclusion of the firm dummy variables serves as a perfect predictor for exit and is therefore dropped from the regression.

Production lines that underwent modernization were, on average, modernized 11.5 years prior to the observation year. The standard deviation of this lag is 10.2 years. We experiment with different definitions of “recent” modernization, including any modernization in the previous 10 years, previous 20 years, etc., and the results are basically unaffected by these differences.

As a robustness check, we also perform disaggregated tests across our sample’s five different products, and find that the production line capacity variable generates a particularly large coefficient compared to the estimated coefficients for the other four products.

We also test for the impact of the total capacity of the plant that contains the production line, and find that this variable is not statistically significant.

The exception to this result is found when we disaggregate the data across our sample’s five products, and find that the “foreign-owned production line” coefficient is negative and significant for wire rod. This suggests that while foreign ownership has had no overall effect on production line exit, it may have prevented exit for wire rod lines.

Feinberg and Hartigan (2007) note that they include a multinational dummy variable in results not reported in the paper and find that this variable is not statistically significant.

This result is consistent with Feinberg and Hartigan (2007), who find that the timing of AD determinations had no discernible impact of plant closures of petitioning firms, regardless of whether the case resulted in a negative or positive determination.

As an additional robustness test, we remove our final observation period (2000–2007), which contains a spell of safeguard protection (2002–2003) that might interact with our AD variable. Even after removing this period, however, the AD/CVD variable continues to be insignificant. Finally, we separate the dataset by technology type and re-estimate the model for each type (integrated, minimills, and processors). Integrated firms have by the far the largest number of production lines, and are the only type for which the logit model converges when we include company-specific dummy variables in the specification. Results are quite similar to the results from the overall dataset, including lack of significance of the AD/CVD variable. The minimill subset converges once company dummy variables are excluded, and produces similar results to the full dataset, with the exception that the AD/CVD variable is negative and significant at the 5 % level. A similar test for integrated firms, in which company dummy variables are removed, fails to produce a statistically significant AD/CVD variable. Thus, it is tempting to suggest that perhaps AD protection reduces the likelihood of production line exit for minimills, but not by integrated firms. However, without the inclusion of company-specific effects, we are hesitant to assert this interpretation.