Bankruptcy and steel plant shutdowns

The Quarterly Review of Economics and Finance 53 (2013) 165–174

Contents lists available at SciVerse ScienceDirect

The Quarterly Review of Economics and Finance journal homepage: www.elsevier.com/locate/qref

Bankruptcy and steel plant shutdowns Robert P. Rogers ∗ Dauch College of Business and Economics, Ashland University, Ashland, OH 44805, United States

a r t i c l e

i n f o

Article history: Received 15 January 2012 Received in revised form 11 January 2013 Accepted 30 January 2013 Available online 16 February 2013 Keywords: Bankruptcies The steel industry Firm governance Financial structure Empirical models of steel plant closure

a b s t r a c t The bankruptcies resulting from the American steel industry downturn in the period, 1999–2002, raise the question of whether the bankruptcy process itself led to permanent plant shutdowns and job losses. With information on 110 of the steel plants operating in the United States in 1994, this paper develops empirical models of steel plant closure and firm bankruptcy to see if the latter impacts on the former. Based on survival models, the results provide support for the hypothesis that the bankruptcy of steel companies could have led to viable steel plants closing, and thus, the bankruptcies in themselves may have caused permanent inefficient employment loss. © 2013 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved.

1. Introduction The recession of 2008 through 2010 led to financial distress and even bankruptcy for many large companies. These bankruptcies may have resulted in the large permanent layoffs. The prospect of these layoffs led to the demand for governments bailouts. In these situations, the governments would either pay off the firm’s creditors or force them to forego or delay the collection of their debt. Other than the economic inefficiency of the firm there are two major reasons that bankruptcy itself can cause lay-offs. The first is that for some products the threat of bankruptcy can lead buyers to cease patronizing a firm even when it is still in operation. This is called the “orphan effect.” A second reason for bankrupt firms shutting down viable plants arises from the financial structure of the firm. Economic and financial experts, however, are divided on this issue. Some posit that the interaction of the financial structure of the firm and the process of bankruptcy can lead to the shutdown of viable plants (Bolton & Scharfstein, 1996; Gertner & Scharfstein, 1991; Schleifer & Vishny, 1992). In certain circumstances, theoretical analysis confirms this assertion. Others, however, maintain that bankrupt firms can efficiently dispose of viable assets (Haugen & Senbet, 1978). They argue that firms often lose money on projects that have little to do with most of their operating plants. If a firm with a viable plant enters bankruptcy, either a new reorganized firm can take the plant over or it can be sold to another firm.

∗ Corresponding author. Tel.: +1 419 289 5739; fax: +1 419 289 5949. E-mail address: [email protected]

Empirical support can be found for either of these theories (Andrade & Kaplan, 1998; Maksimovic & Phillips, 1998; Opler & Titman, 1994; Pulvino, 1998). A case study can do much to illuminate this issue. The steel industry in the period between 1994 and 2006 provides such an example. Certain industry characteristics, however, may be relevant to the analysis. First, as will be shown, steel is a producer good, and steel customers are knowledgeable people who can readily ascertain the product quality of a given firm. Thus, if one firm leaves the market due to bankruptcy, another can readily replace it as far as the buyer is concerned. When a plant changes owners, customers will continue patronizing it if it produces the desired product. Consequently, if it is found that bankruptcy leads to efficient plants closing, then, this would provide strong evidence for the influence of the financial structure on the operation of efficient plants. Additionally, steel has always faced a cyclical demand situation. This puts firms in positions where they have to dispose of their assets to avoid or emerge from bankruptcy. In the mid and late 1990s, the pattern repeated itself. Domestic steel sales dropped, and problems in Southeast Asia increased imports into the United States. In the subsequent downturn, twenty-four American steel companies underwent bankruptcy. Among them were the large integrated companies and even minimill firms with the latest technology. During this steel slump, steel firms were not able to persuade the government to bail them out in the way the automobile and banking firms did in 2008 and 2009. The demise of these firms raises the question of whether the bankruptcy process led to plants shutting down and their consequent job losses. With information on 110 steel plants operating in the United States in 1994, this paper examines the influence of

1062-9769/$ – see front matter © 2013 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.qref.2013.01.006

166

R.P. Rogers / The Quarterly Review of Economics and Finance 53 (2013) 165–174

these bankruptcies on the survival of given steel plants. To accomplish this goal, we set up empirical models to test the connection between steel plant closure and firm bankruptcy. Before examining these models, the steel industry in the late 20th century is described. Then, we review the relevant literature on the question. Third, we present the models used to test our hypotheses. After that, we discuss the sample and the variables used in the analysis. Then, the empirical results are given. Last, some conclusions are drawn.

2. The steel industry To understand the relevance of steel to the bankruptcy issue, two aspects of the industry should be examined: technology and demand. In the late 20th century the industry underwent three major technological changes. The first change was the replacement of the open hearth, the old standard steel furnace for integrated mills, with the Basic Oxygen Furnace (BOF). This was a slow process beginning with the first BOF installation in 1954 and ending in 1991 with the last open hearth closing down (Lynn, 1982; Rogers, 2009). Second, the primary rolling mills which shaped the intermediate steel pieces, slabs, blooms, and billets, were replaced by the continuous caster. Third and most important, the electric furnace which usually employs scrap has become the dominant technology for making steel (Barnett & Crandall, 1986; Preston, 1992). By 2000, plants with this type of furnace (often called minimills) accounted for 55 percent of the steel produced. Costing millions of dollars, these changes resulted in plants being obsolete before the end of their apparently useful life. Thus, by the 1990s, different companies and plants were operating at quite different levels of efficiency. Additionally, the industry experienced a drastic fallback in demand in the 1970s and 1980s. Production dropped from 150 million tons in 1973 to 98 million in 1989. Many reasons have been given for this decline; among them backward technology, competition from imports and substitutes. The literature detailing this situation includes the following: Barnett and Crandall (1986), Barnett and Schorsch (1983), Crandall (1981), Duke et al. (1977), Hogan (1983, 1986, 1994), Preston (1992), Rogers (2009), and Tornell (1997). This fallback and the dislocations caused by the technological changes led to extremely difficult times for steel firms and steel workers. In the 1970s and 1980s, a number of steel firms underwent bankruptcy. Among them were Wheeling-Pittsburgh, and Sharon Steel. In the late 1970s, LTV acquired Youngstown Sheet & Tube which was on the verge of bankruptcy, and LTV acquired Republic Steel also on the verge of bankruptcy in 1984. In 1985, LTV underwent bankruptcy. To alleviate the steel industry problems, the federal government undertook a number of policies to attenuate imports. Among them were quotas on the import of steel from given countries and a pricing scheme called Trigger Price Mechanism (TPM) that was thought to keep the price of imports consistent with most American firm costs. Throughout the period, steel firms brought dumping cases against foreign companies that may have been charging prices below average costs. Sometimes, these actions were legally successful, but as a whole these cases and the quota and price policies did little to slow the decline of the steel industry (Mueller, 1984). In the 1990s, the steel industry stabilized, and its profitability revived. Between 1990 and 2000, production rose from 99 million tons to 112 million tons. As a whole, the industry experienced seven years of profitability between 1993 and 1999. In the late 1990s, however, the cyclical pattern of the industry reasserted itself. Domestic steel sales dropped, and problems in Southeast

Asia increased imports into the United States (Rogers, 2009, pp. 161–190). Production fell from 112 million tons in 2000 to 99 million tons in 2002. As a result of this downturn, a number of steel companies underwent bankruptcy. Among them were several traditional fully integrated BOF firms: for example Bethlehem Steel, LTV Steel, National Steel, Gulf States Steel, and WheelingPittsburgh. Some like LTV and Wheeling-Pittsburgh had been in bankruptcy before, but others like Bethlehem had never experienced such distress in the 20th century. Bankruptcy spread to even minimill firms with the latest technology such as Bayou Steel, Birmingham Steel, and Georgetown Steel. In reaction to these troubles, some companies brought more dumping cases, but the major response came from President George W. Bush. For a wide set of steel products, he brought a special type of tariff action called a Section 201 case. This action set up a quota system for given steel products, imposing substantial duties on imports above those quotas. During the whole of this slump, however, the steel firms were unable to persuade the government to bail them out of their bankruptcies. This contrasts with the experience of the automobile and banking firms in 2008 and 2009. Bush’s quota system was originally intended to last three years, from 2002 to 2005, but steel demand increased, and pressure from steel buyers led Bush to end the quotas in 2004. As of 2006, a large number of the mills that were closed temporarily were open. In the analysis below, we assume that if a plant was operating in that year, it was economically viable. Some statistics on the steel industry can help one understand the issues. Table 1 which defines the variables used in the analysis gives some idea of the structure of the industry. As of 1994, there were 110 steel plants in the sample (plant being defined as an establishment with a steel furnace which transforms raw iron or scrap metal into steel). Of these 110 plants, 22 were integrated Basic Oxygen plants where iron smeltered in on-premise blast furnaces is transformed into steel, and the others were plants with electric furnaces. The latter usually relied on scrap steel as an input. Most of the mills were concentrated in the traditional steel making region consisting of Pennsylvania and the eastern part of the Midwest. Just over one quarter of the mills (26.4 percent), however, were located in the southern part of the United States. With two exceptions, the southern mills were all electric furnace plants. The sample steel mills were owned by 73 companies. The bulk of them (50) were steel specialists. Some of these firms were focused mainly on other products: examples being Timken and Worthington. Some like Nucor and LTV started in other products but became mainly focused on steel. Fifteen of the firms are multiplant firm; Nucor being the one with the largest number of plants (eight). In 1994, foreign companies were operating as joint ventures with American companies such as U.S. Steel and Nucor. Furthermore, AK and National Steel were companies jointly owned by American and Japanese firms. The bankruptcy episode under discussion changed this situation; today large foreign companies such as ArcelorMittal, Severstal, and Gerdau are major factors in the American steel industry. Of the 73 companies having steel plants according to the source, 24 underwent bankruptcy. Of special importance to the hypotheses on governance was the ownership structures of these firms, 54 of the firms were listed on exchanges or the NASDAQ. Of the 24 firms that underwent bankruptcy, 20 were widely held companies traded on exchanges. Thus, for these firms, a potential exists for governance problems that can lead to bankruptcy shutting down efficient plants. Thus, all these situations raise the question of whether bankruptcy leads to efficient plants permanently closing down.

R.P. Rogers / The Quarterly Review of Economics and Finance 53 (2013) 165–174

167

Table 1 The basic statistics for the variables in the models. Variable

Mean

Median

Standard deviation

Definition

7 7

2.860 1.492

The number of years between 1994 and the year the plant shut down The number of years between 1994 and the year that the plant’s parent firm underwent bankruptcy

1,371,529

The rated capacity of the observation plant in tons of steel production per year Dummy equaling 1 if the plant had a Basic Oxygen Furnace, and 0 otherwise Dummy equaling 1 if the plant was located in Pennsylvania, and 0 otherwise Dummy equaling 1 if the plant was located in the Midwest,b and 0 otherwise Dummy equaling 1 if the plant was located in the SOUTH,c and 0 otherwise Dummy equaling 1 if the plant made stainless steel, and 0 otherwise Dummy variable equaling 0 if Barnett and Crandall either predicted that the plant would close or considered its future uncertain, and 1 otherwise Dummy equaling 1 if the plant was an electric furnace facility using the modern rolling process to make sheet steel, and 0 otherwise

a

The dependent variables TIMEPlant 5.91 7.20 TIMEBankr

The independent variables impacting only TIMEPlant (X1 ) 1,106,409 600,000 PlantCAP BOFF

0.200

0.402

PENN

0.173

0.380

MID

0.418

0.496

SOUTH

0.264

0.403

SS BCRAN

0.173 0.945

0.380 0.228

MSH

0.027

0.164

The independent variables impacting only TIMEBankr (X2 ) FirmCAP 2,992,345 1,120,000

4,073,009

CREDIT

0.327

0.471

LIST

0.800

0.402

The independent variables impacting both TIMEBankr and TIMEPlant (X3 ) MULT 0.473 0.502 CONGL

0.255

0.438

The total rated capacity of each plant’s parent firm in tons of steel production per year Dummy equaling 1 if the plant ( s parent company had a Dun & Bradstreet credit rating of 2 or over, and 0 otherwise Dummy equaling 1, if the parent firm was publicly traded and 0, otherwise Dummy equaling 1 if the parent company had more than one steel plant, and 0 otherwise Dummy equaling 1 if the parent company was a multi-product firm primarily concentrating on products other than raw steel, and 0 otherwise

a These statistics are pertain to only the 22 plants that shut down and the 32 plants that were with the 24 companies that underwent bankruptcy. Otherwise, the central tendency variables would be severely skewed. b This includes the states of West Virginia, Kentucky, Ohio, Indiana, Illinois, Michigan, Wisconsin, Minnesota, Iowa, and Missouri. c This includes the states of Virginia, North Carolina, South Carolina, Tennessee, Georgia, Florida, Alabama, Mississippi, Arkansas, Louisiana, Oklahoma, and Texas.

3. Literature review There is strong support in the literature for either financial distress closing down viable plants or efficient plants surviving bankruptcy. On the first position, both demand-side and supplyside theories posit that financial distress can lead firms to shutdown viable plants. The demand side theory posits that buyers may shun the products of bankrupt firms. The types of product where this idea is most relevant are consumer durables. For example, consumers may stay away from the cars of companies on the verge of bankruptcy. There is a fear that people cannot obtain parts and/or that warranties would not be honored. This phenomenon is called the “orphan effect.” There is some evidence of this effect in this automobile industry (Hammond, 2010; Hortac¸su, Matvos, Syverson, & Venkataraman, 2010). This phenomenon is not relevant to the steel industry. Steel is a producer good, meaning that the customers are knowledgeable people. Most steel (just over 60 percent in 2000) is sold to distributors, construction firms, automobile companies, and oil producers. (See Rogers, 2009, pp. 164.) These firms and other users such as equipment, container, and appliance manufacturers are led by people knowledgeable about steel. Additionally, these users have such cyclical demand that they cannot make long term commitments on volume. Furthermore, once the steel has been installed there is little need for maintenance by the original firm. Thus, these customers can readily switch steel firms, and if one firm undergoes bankruptcy, they can readily find another supplier. Therefore, the orphan effect is absent in steel.

On the supply side, the literature is divided on issue of financial structure leading to viable plants shutting down. Classical economic theory and much finance literature maintain that absent the orphan effect financial distress will not prevent efficiently utilized resources from being redeployed. Essentially the operation of the firm is considered separate from its financial structure (Modigliani & Miller, 1958). Haugen and Senbet (1978) provide one of the most cogent expositions of this idea. They argue that in the absence of irrationality and legal barriers to mobility bankrupt firms will redeploy their assets in a value maximizing fashion no matter what the position of the asset owner is. Given this process, then, bankrupt steel firms would sell efficient plants to other firms who can operate them. Other scholars, however, posit that this may not be the case and even economically efficient plants may close due to bankruptcy. This strand of the literature asserts that firms in bankruptcy or other financial distress may not be able to optimally redeploy their assets. In other words, a viable plant might be closed in a bankruptcy, even when another operator could succeed with the plant. Essentially, this literature questions two of the key assumptions of neoclassical economics: first, that resources are mobile, and second, that the actors connected to the modern firm always move resources to their highest valued uses. Given actors may gain more wealth and utility when total social wealth is not maximized. These ideas arise from the theories on transactions costs and firm governance developed by Coase (1936) and Jensen and Meckling (1976). Two articles apply these ideas to the problems of firm capital structure and financial distress. Using agency theory and

168

R.P. Rogers / The Quarterly Review of Economics and Finance 53 (2013) 165–174

transactions cost economics, Williamson (1988) argues that debt and equity are not merely two methods of financing but also two different forms of firm governance. Thus, the financial structure can affect the firm’s operation. Jensen (1986) posits that debt can be used as a weapon by either the stockholders or the management to further their interests. Sometimes the promise of future debt enables management to undertake certain projects that would not be allowed otherwise. The significance of these papers to plant closures is that there may be situations where the people in control of a firm find it in their interest to shut down economically efficient plants. Other papers have further developed these ideas – examples being Bernanke and Gertler (1989), Gertner and Scharfstein (1991), and Bolton and Scharfstein (1996), but Schleifer and Vishny (1992) provide the most relevant theory on the connection between financial structure and plant shutdowns. They posit that firms will carry a large amount of debt in order to prevent management from undertaking projects that increase its utility but not the wealth of the firm. If a firm has a large debt load, then, management does not have a large cash flow to put into its own questionable projects. This large debt load, however, means that there is a thin cash flow, and if an industry downturn occurs, the firm is vulnerable to financial distress. Additionally, given this downturn, other companies within the industry termed by Schleifer and Vishny “inside” firms that could buy and operate a viable plant might have troubles of their own. With the same governance problem as the original distressed firm, they also carry a heavy debt load to constrain their managements. Thus, they may be in no position to take over the assets of a bankrupt firm. Another type of potential buyer is the outside firm wanting either to enter the industry or to use the plant assets for something else. Usually they cannot utilize the plant as effectively as inside firms. Consequently, these outsiders are left to take over the plant, but they usually do not have the technological knowledge of the experienced inside firms. Often, the outside firms may find it optimal to close plants that insider firms could have operated efficiently. These governance problems could arise in the steel industry. As stated above, 20 of the 24 bankrupt steel firms in the 1994 sample were traded on exchanges. Thus, the situation of conflict between outside owners and manager posited by Schleifer–Vishny and others are likely. To summarize, due to knowledgeable buyers and the lack of maintenance problems, the orphan problem does not exist in steel. On the supply side of the issue, however, Schleifer and Vishny posit situations where the governance role of financial structure could lead to steel and other firms abandoning economically viable plants. Nevertheless, Haugen and Senbet argue that financial distress will not lead to efficient plants being shut down. As stated above, empirical support can be found for either of these views (Andrade & Kaplan, 1998; Maksimovic & Phillips 1998; Opler & Titman, 1994; Pulvino, 1998). These papers cover a large number of samples, but none of them have focused on the problem from the perspective of the individual plant.

model of the interconnection of steel plant closure and steel firm bankruptcy is developed. This analysis consists of two models: one explaining the closing of the steel plants and the other explaining the bankruptcy of the plants’ parent companies. These models are best depicted by survival functions which explain the difference in time between a given starting point and the occurrence of given events (in this case the permanent shutdown of a steel plant and the bankruptcy of a steel firm). To model this situation, two survival models are developed (In a theoretical paper, He and Matvos, 2012 use a similar model to show the effect of the tax treatment of corporate debt on industry efficiency). The first depicts the ability of a steel plant to compete in the market during the period, 1994–2006. Using the 1994 sample of steel plants in an international steel plant directory (Serjeantson, Cordero, Cooke, Sexton, & Jordain, 1994) and other sources, we determine whether and when given steel plants were closed. The variable, TIMEPlant , is defined as the number of years between 1994 and the year the plant permanently shut down; it represents the likelihood of plant closure with the survival model. By no means, however, did all the plants permanently close. In fact, 82 out of the 104 plants for which sufficient data exist were still operating in 2006. Consequently, there is censoring problem with this sample; below we describe how it is taken into account. The second model depicts the ability of the plant’s parent firm to avoid bankruptcy. This ability is represented by TIMEBankr which equals the number of years between 1994 and the year when the plant’s company underwent bankruptcy. Here bankruptcy is defined as either chapter 7 or chapter 11 status. In the former case, the firm ceases operations, with a trustee dividing up its assets between the creditors (usually with nothing left over for the owners). In the case of chapter 11, the firm continues to operate under court supervision, while attempting to pay off its debts. Often firms come out from the chapter 11 process to return to solvency. In fact, two large steel firms, LTV and Wheeling-Pittsburgh, emerged from this process in the early 1990s. As with plant shutdowns, not all the firms underwent bankruptcy in the sample period. Of the 68 companies for which sufficient data exist, 42 remained solvent, leaving 24 firms that underwent bankruptcy. This censoring problem will be discussed below. The survival models assume that under given conditions any steel plant in the sample could be permanently closed and that any parent steel firm in the sample could undergo bankruptcy. The greater the viability of a plant the longer will be the time before the event of plant shut down, and the greater the strength of a firm the longer will be the time before bankruptcy. Therefore, this assumption does not distort reality because under given circumstances even the strongest plants and firms could close. Its advantage is that it allows one to posit continuous survival functions. From these functions, hazard functions can be derived. This function roughly represents the rate at which the subject entity ceases to exist (which is depicted by plant shut down or firm bankruptcy) at time, ti . (See Cox & Oakes, 1984; Greene, 2008, pp. 931–943; Kiefer, 1988.) For our variables, the hazard functions can be depicted as follows: H1 (t1 ) = the rate at which a given plant will permanently shut

4. The empirical model This paper uses a statistical analysis to determine which of the above theories best describes the situation in the early 21st century American steel industry. One theory posits that stopping a firm from going into bankruptcy might prevent an otherwise viable plant from closing, and the other states that plant survival is independent of the financial state of the firm. To test these theories, a

down during the time periodt1 , and H2 (t2 ) = the rate at which a given plant’s parent firm will undergo bankruptcy during period t2 . These hazard functions take on certain mathematical forms. Here, however, this form is not clear. Therefore, a model is

R.P. Rogers / The Quarterly Review of Economics and Finance 53 (2013) 165–174

estimated that makes no assumption about this form. This model is called the proportional hazards model (Cox & Oakes, 1984; Greene, 2008, pp. 939–943; Kiefer, 1988, pp. 664–669). To check the results of this method, the paper also estimates another model making explicit assumptions about the hazard function form. This particular model is called the Weibull form. The next step in the development of our model is connecting the hazard model to the variables that can change the time until plant shut down or firm bankruptcy. Essentially the hazards model can be modified to account for the likelihood that independent variables can impact on the time between the beginning period (1994 in our case) and the year of either plant shutdown or firm bankruptcy. This influence can be incorporated into the hazard function as follows: Hi (ti ) = H0 (ti )eXi ˇi ,

i = plant shutdown, parent

firm bankruptcy.

(1)

Here Xi equals a vector of independent variables impacting on the hazard function and ˇi equals the vector of impact coefficients for the variables. These specifications are called a “accelerated failure time models” (Greene, 2008, p. 937). The conditions reflected by Xi , can either accelerate or decelerate the failure of the observation entity. Thus, given information on ti and Xi , we estimate the impact of the Xi variables on the survival function for the two variables, TIMEPlant and TIMEBankr . As shown below, for the plant shutdown model, TIMEBankr can be in the Xi vector; and for the bankruptcy model, TIMEPlant can be included in its Xi vector. As stated above, the proportional hazards model which allows the data to determine the underlying hazard function is used, and Weibull function models are estimated to check the results. Having developed the basic model for the structural equations, we now examine the components of the model that impacts on the survival function, the Xi ·ˇi . The equations that impact on the survival function will be called ETIMEPlant for the plant shutdown model and ETIMEBankr for the bankruptcy model. In this case, these terms would act through their exponentials, eETIMEPlant and eETIMEBankr . The theories discussed above raise the possibilities that bankruptcy can lead to plants shutting down and, of course, plant shutdowns causing firm bankruptcies. This set of influences can be depicted by the following models: ETIMEPlant = b12 TIMEBankr + b11 X1 + b13 X3 + e1

(2a)

ETIMEBankr = b21 TIMEPlant + b22 X2 + b23 X3 + e2 .

(2b)

Here the bij ’s equal the components of the vectors, ˇ1 and ˇ2 . X1 represents exogenous variables affecting only plant viability thereby appearing only in Eq. (2a); X2 represents the exogenous variables affecting only firm strength appearing only in Eq. (2b), and X3 represents exogenous variables that may appear in both equations. For the proportional hazards version of this model, there is no constant, but for the Weibull versions, there is a constant. Before proceeding, we must deal with three econometric problems. The first is the possible simultaneity between TIMEPlant and TIMEBankr ; the second is the above mentioned problem of the censoring of the observations. The third is the sample selection problem in that it is not clear that the firms and plants for which data are not available are distributed in a random manner. The simultaneity problem can be best addressed by the use of instruments. The logical ones to use are forms of Two Stage Least Squares (TSLS) where we use as instruments, the predicted value of the reduced forms of the endogenous variable. For the TIMEPlant model, that variable would be TIMEBankr , and for TIMEBankr it would

169

be TIMEPlant . The reduced form equations for this model can be derived as follows: TIMEPlant = h11 X1 + h12 X2 + h13 X3 + v1 ,

(3a)

TIMEBankr = h21 X1 + h22 X2 + h23 X3 + v2 .

(3b)

Following Heckman (1978, p. 947), these instrumental variables are estimated by OLS, and the expected value will be fed back into the structural equation. Thus, the major problem would be to estimate the covariance matrix in order to get the necessary test statistics. A covariance matrix can be computed that takes into account the instrumental variables estimators. For the proportional hazards model, this estimate combines the Hessian matrix with the derivatives of the independent variables with respect to the instrumental variables. For the Weibull model, we use a slightly different method to compute the second derivative matrix which is, then, combined with instrumental variable derivatives. (See Greene, 2007, pp. E36–E37; Greene, 2008, pp. 333–334, 495, and 667.) The problem of censoring arises from the fact that as of the cutoff date, 2006, most American steel plants had not permanently closed, and most steel firms had not undergone bankruptcy. For both the proportional hazards and the Weibull survival functions, models can be estimated when a large part of the sample is censored (Greene, 2008, p. 936; Kiefer, 1988). Thus, we proceed accordingly with the censored models in a manner similar to the Tobit technique used in linear regression. The last problem is that for some plants in the 1994 sample certain information could not be found. For six plants, we do not have the year that they closed. This lowers the number of plants in the analysis from 110 to 104 and the number of companies from 73 to 68. A certain set of variables might affect the likelihood of finding data necessary for the model. They are included in a sample selection model which is estimated by probit. Two dummy variables are included to account for differences in information availability. One is for whether the plant was in a small town, and the other is for whether it was in an extremely large city. So we add two variables to the sample inclusion equation, SMALLT for a plant in a town of under 20,000, and BIGC for a plant in a city of over 100,000. The reason for these variables is that there are no newspapers or other media in very small towns, and for large cities a steel plant of the usual size would not be covered by large city news outlets. Therefore it is difficult to find the necessary information on the dates of the closing. The other variables in the reduced form model could affect the sample selection, and therefore they are included. From the result of the sample selection model, a Mill’s ratio is computed, and it is added to the estimating models. If the ratio passes the test for variable inclusion in the model, we assume that the sample selection affects the results, if not, we omit the ratio and assume that sample selection is not a problem. (Greene (2007, pp. E36–E44) suggests a maximization likelihood approach to the sample selection problem where a two equation system consisting of the hazard and the sample selection models is estimated. A major problem exists with this procedure in that misspecification in one equation can affect the results for the other. Therefore, the Mill’s ratio method is used.) We have developed here an estimation approach to our model. It is a survival model that takes into account the endogeneity of TIMEPlant and TIMEBankr . Furthermore, it deals with the problems of censoring and sample selection. The major focus is on the TIMEPlant equation (1) because it depicts the relationship of interest. The coefficient of interest is that of TIMEBankr on TIMEPlant or b12 . The primary hypothesis is that the incidence of firm non-bankruptcy has a positive impact on the incidence of plant survival through

170

R.P. Rogers / The Quarterly Review of Economics and Finance 53 (2013) 165–174

2006. An examination of the TIMEBankr model (2b) puts the situation in context.

5. The sample and the variables We now examine the data used to estimate the models. After discussing the sample, we first explain the dependent variables, and then the independent variables in X1 , X2 , and X3 . The sample consists of the 110 plants listed in the directory, Iron and Steel Works of the World (Serjeantson et al., 1994, 2006) for the years, 1994 and 2006. These data sets may not be as inclusive as the Census of Manufactures, but they give information about individual plants that cannot be obtained from Census such as location and capacity. Furthermore, these data sets are roughly consistent with the Census. For 1994, Serjeantson et al. lists 236 establishments in the American steel industry, while the 1992 Census includes 217; the 2006 comparison is similar. Most of these establishments do not make steel, rather they merely roll or shape steel bought from other plants. Census does not separate out plants that transform iron or scrap into usable steel, and since they do not list capacity, we have to depend on directories like Serjeantson et al. which are often more inclusive than other sources including Census. Consequently, the list includes most all the steel plants operating in the United States in 1994. We checked on whether the 1994 plants were still in the 2006 directory. If they were not, we initially assumed that they had been shut down. We then ascertained the year the plants were closed to construct the variable, TIMEPlant . Furthermore, we made checks to see if some of the plants not listed in 2006 were still in operation; we found a few discrepancies and corrected them. While the sample may not include all the plants in the United States, there are no systematic reasons certain plants would be excluded. The sample could be biased against plants in small towns and those outside the traditional steel centers (consisting basically of Pennsylvania, the Midwestern states of Ohio, Indiana, Michigan, and Illinois), but 41 percent of the plants listed are outside those states. Most steel plants are located in small towns or cities. Consequently, even if not complete, the sample is very likely to be a good and close-to-complete representation of the plants in the country. (For some plants in the sample, the closure year is not available. This is why the sample selection model is used.) Now, we discuss the dependent variables; they are displayed in Table 1. As stated above, the two dependent variables are TIMEPlant and TIMEBankr . TIMEPlant is the number of years between 1994 and the year the plant shut down, and TIMEBankr is the number of years between 1994 and the year that the parent firm underwent bankruptcy. Two firms, Lukens and Copperweld, were merged into other firms in the sample period; they are treated as survivors even though their new parents, Bethlehem and LTV, went into bankruptcy. In 2006, another company was operating Lukens’ main steel plant, and most of Copperweld’s plants were still operating and owned by other companies. TIMEBankr is the dependent variable in the second equation of our model. It also feeds into the first equation, the plant survival model, to test whether bankruptcy directly affects the survival of given steel plants. The following variables are hypothesized to impact on the TIMEPlant equation making them elements of X1 . PlantCAP is the total rated capacity in tons of steel production per year of each plant. This variable reflects the influence of economies of scale on firm efficiency. If economies of scale are present, PlantCAP will lead to plants lasting longer (Rogers, 1993; Stigler, 1958; Tarr, 1984). The next variable, BOFF, is a dummy variable equaling one if the main furnace of the plant is a Basic Oxygen Furnace, and zero, otherwise. The type of furnace used could affect the survival of the plant. Rising

energy costs and technological progress with the electric furnace have led to a difficult environment for the fully integrated Basic Oxygen Furnace plants. Given the weaknesses of the BOF plants, this variable should have a negative impact. PENN is a dummy variable equaling one if the plant was located in Pennsylvania, and zero otherwise. Historically, Pennsylvania was a major center of the steel industry, and there was a concentration of expertise in that state. Furthermore, the state was also a center for steel users, and consequently there might be a surfeit of customers. Therefore, we would predict this variable to have a positive sign in the TIMEPlant equation. MID is a dummy variable equaling one if the plant was located in the Midwest, and zero otherwise. The Midwest is defined here as the area consisting of the following states: West Virginia, Kentucky, Ohio, Indiana, Illinois, Michigan, Wisconsin, Minnesota, Iowa, and Missouri. The Midwest has been declining as an industrial center, and the demand for steel in that region has been dropping. Consequently, this variable would have a negative impact on TIMEPlant , other things equal. SOUTH is a dummy variable equaling one if the plant was located in the South, and zero otherwise. The South is defined here as the area consisting of the following states: Virginia, North Carolina, South Carolina, Tennessee, Georgia, Florida, Alabama, Mississippi, Arkansas, Louisiana, Oklahoma, and Texas. Over time, the South has grown as an industrial center, and its demand for steel has been rising. Therefore, this variable would have a positive impact on TIMEPlant , other things equal. BCRAN is a dummy variable based on the work of Barnett and Crandall (1986). In their analysis of the steel industry, Barnett and Crandall predicted which BOF plants would close. While they hedged on some plants in that they were not totally certain on them, their predictions can still be used. Even though their work was done in the 1980s, these scholars may well have discerned certain conditions that this author missed. This variable, then, is a proxy for these conditions. BCRAN equals zero for plants for which they predict closure or on which they are uncertain and one otherwise. SS is a dummy variable equaling one if the plant made stainless steel, and zero otherwise. Stainless steel had somewhat different demand conditions than most steel products, and therefore one might expect these different conditions to affect the plant’s chances of survival. MSH is a dummy variable equaling one if the plant was a minimill (electric furnace) facility making sheet steel using the new type of continuous caster developed in Germany in 1980s, and zero otherwise. These plants use a new efficient technology to make sheet steel, and presumably their chances of survival would be greater. The following variables are hypothesized to only impact firm strength, the ability of the firm to avoid bankruptcy, TIMEBankr , and thus, they would be in X2 . FirmCAP is the rated capacity in tons per years of the given parent firm; this variable reflects the influence of firm scale economies on the efficiency and thereby survivability of the firm. CREDIT is a dummy variable equaling one, if the plant’s parent company had a 1994 Dun & Bradstreet credit rating of 2 or over, and zero otherwise. Most of the firms in the sample were either in the low credit rating category or were unrated. Thus, firms with a 2 or over rating would be better able to avoid debt defaults than the others. LIST is a dummy variable equaling one, if the parent firm was publicly traded and zero, otherwise. Publicly traded firms are scrutinized by the Security and Exchange Commission and other government and private agencies for viability and honesty in reporting; consequently other things equal, they would be less likely to undergo bankruptcy. The below variables are hypothesized to impact on both TIMEPlant and TIMEBankr , making them components of X3 . MULT is a dummy variable equaling one if the parent company has more

R.P. Rogers / The Quarterly Review of Economics and Finance 53 (2013) 165–174

171

Table 2 The results for the plant survival model using the proportional hazards assumption. (The dependent variable is TIMEPlant , the number of years between 1994 and the year the plant shut down. For TIMEFirm , PlantCAP, SOUTH, and BCRAN one tail tests are appropriate for the relevant hypothesis, while for the Mill’s ratio a two tail test is appropriate.). Variable name

Coefficients

Cox model using a strictly OLS style estimation procedure −0.1540 TIMEFirm PlantCAP −0.0004 SOUTH −1.1717 1.3118 BCRAN −1.4862 Mill’s ratio 18.805 Chi-squared value Cox model using an instrumental variables estimation procedurea TIMEFirm −0.7136 PlantCAP −0.0006 SOUTH −0.9605 BCRAN 1.2135 Mill’s ratio −2.4148 Chi-squared value

z values

Estimation probability values

−1.900** −1.277 −1.509 2.055** −0.754

0.0287 (for a one tail test) 0.1009 (for a one tail test) 0.0657 (for a one tail test) 0.0199 (for a one tail test) 0.4508 (for a two tail test) 0.0021

−1.766** −1.866** −1.247 1.136 −1.175

17.708

0.0387 (for a one tail test) 0.0310 (for a one tail test) 0.1062 (for a one tail test) 0.1280 (for a one tail test) 0.2400 (for a two tail test) 0.0033

a

The z values in this model are computed from the covariance matrix adjusted for the instrumental variables described in the paper. *Significant at the 5 percent level on a two tail test. ** Significant at the 5 percent level on a one tail test.

than one steel plant, and zero otherwise. It is included to depict the influence of multiplant operations on a plant’s and firm’s efficiency (Scherer et al., 1975). This variable could have either a positive or negative sign. There is some debate over whether multiplant operations improve firm and plant performance. Even so, we cannot rule out the possibility that this variable would have an impact. CONGL is a dummy variable equaling one if the parent company is a multi-product firm primarily concentrating on products other than raw steel, and zero otherwise. It is included to depict the influence of product diversification on both plant and firm success. Its sign could be either positive or negative. Again debate exists over the influence of conglomeration. Theoretically we could include all the hypothesized independent variables in the system. The theory, however, is ambivalent on most of these variables. Therefore, if the theory is unclear and the variables do not pass a standard inclusion test, they are omitted. 6. The results In this section, we present the results. As stated above, two variants of the survival model are estimated, but the greatest reliance is placed on the proportional hazards model. Therefore, the results

of this model are displayed in Tables 2 and 3. The former shows the model for the incidence of plant survival, and the latter shows the model of the incidence of firm bankruptcy. The Appendix discusses the results of the estimation of the models with the Weibull assumptions. The conclusions from these models are consistent with those of the proportional hazards models. Consequently, this section focuses on the result for the proportional hazards model. One unavoidable issue is the possible simultaneity between the variables, TIMEPlant and TIMEBankr in both models. To deal with this problem, we estimate a plant closure model with an instrumental variable for the TimeFirm and a firm bankruptcy model with an instrument for TimePlant . The instruments used are the predicted values for TimeFirm and TIMEPlant from the reduced form model of a system consisting of the TimeFirm and TIMEPlant (Eqs. (3a) and (3b)). Following Heckman (1978), we estimate the predicted reduced models for TIMEPlant and TIMEBankr using the Ordinary Least Squares method. With this method, we eliminate the problem of correlation between the variables and the residuals. With these estimators, we construct predicted values for the endogenous time variables. To estimate the covariance matrix, then, we use the procedures described above (on page 10) which combine the Hessian matrix with estimates of the instrumental variable variances.

Table 3 The results for the parent firm bankruptcy model using the proportional hazards assumption. (The dependent variable is TIMEBankr , the number of years between 1994 and the year that the parent firm underwent bankruptcy. For TIMEPlant a one tail test is appropriate for the relevant hypothesis, while for FirmCAP, MULT, CONGL, and the Mill’s ratio two tail tests are appropriate.). Variable name

Coefficients

Cox model using a strictly OLS style estimation procedure −0.0538 TIMEPlant 0.00009 FirmCAP −1.1347 MULT −0.8517 CONGL 3.8310 Mill’s ratio 15.127 Chi-squared value Cox model using an instrumental variables estimation procedurea TIMEPlant −04118 FirmCAP 0.00008 −0.8995 MULT −0.6350 CONGL 4.5242 Mill’s ratio Chi-squared value a

19.570

z values

Estimation probability values

−1.014 1.744* −2.306* −1.332 1.725*

0.1554 (for a one tail test) 0.0812 (for a two tail test) 0.0211 (for a two tail test) 0.1828 (for a two tail test) 0.0845 (for a two tail test) 0.0098

−1.322 1.649* −1.212 −0.411 1.263

0.0931 (for a one tail test) 0.0496 (for a two tail test) 0.2255 (for a two tail test) 0.6811 (for a two tail test) 0.2066 (for a two tail test) 0.0015

The z values in this model are computed from the covariance matrix adjusted for the instrumental variables described in the paper. * Significant at the 5 percent level on a two tail test. **Significant at the 5 percent level on a one tail test.

172

R.P. Rogers / The Quarterly Review of Economics and Finance 53 (2013) 165–174

Table 2 shows the result for the two TIMEPlant models, one using the actual value for TIMEBankr and the other using the predicted value from the reduced form estimation. The Chi squared tests for both models indicate that the probability of them occurring by chance is very low (0.0021 and 0.0033). By the use of Akaike’s Information Criterion (AIC), all the proposed variables for the TIMEPlant equation except PlantCAP, SOUTH, BCRAN, and the Mill’s ratio were eliminated. (See Maddala, 1992, p. 500 for a description of this specification criterion. It is often used for non-linear models. Actually given the size of our model, the critical value is 0.908 – not a lot different from 1.00 which is often used in linear models.) Of the proposed exogenous variables, three seem to impact on the survival model. PlantCAP shows the plant size decreases the likelihood of shutdown; this is consistent with economies of scale. The results for South weakly support the hypothesis that the expansion in the southern part of the United States leads to plant surviving. In contrast, BCRAN increases the likelihood of plant closure. The Mill’s ratio indicates that the sample selection was not random for this model. Most important are the results for TIMEBankr . The z values are significant at the 5 percent level on a one tail test. The z values for the second model come from the instrumental variable adjusted covariance matrix. These results indicate that the length of time between 1994 and when the parent firm underwent bankruptcy decreases the likelihood of plant shutdown. The longer the firm survival before bankruptcy the less likely it was that the plant would close. Thus, the firm financial situation does impact on probability that a plant will survive. The results, then, support the Schleifer–Vishny hypothesis. Apparently in the steel industry, the financial situation of the parent firm can lead to efficient plants shutting down. Table 3 displays the results for the TIMEBankr model. At this juncture, it is helpful to point out what is being estimated in this model. Here the observation is not merely the company undergoing bankruptcy during the sample period. Rather the observation is the steel plant, and therefore, TIMEBankr indicates the length of time between 1994 and when the plant’s parent company went through bankruptcy. As with the plant closing model, two models are estimated: one with the actual value of TIMEPlant , and the other with the predicted value of TIMEPlant from the Instrumental Variable estimation. The former is comparable to an OLS estimate. For these models, the Akaike criterion indicates that the variables, FirmCAP, MULT, CONGL, and the Mill’s ratio should be included in the equations. For both models, the coefficient for FirmCAP indicates that the firm size increases the likelihood of bankruptcy. Thus, other things equal, companies with larger capacity are more prone to bankruptcy. The appearance of the Mill’s ratio indicates that the sample selection mechanism may not be random for this equation. The most important result is that on the time variable, TIMEPlant . Apparently the connection between plant shutdown and firm bankruptcy is not strong. The coefficients are not significant at the 5 percent level. While this result might indicate that the OLS type equation for the plant shutdown model is the most appropriate, this lack of significance does not preclude a problem with two-way causation, and thus, we should rely on both TIMEPlant models since the probability of simultaneity is not zero. Essentially, then, the estimation results support the Schleifer–Vishny prediction that firm bankruptcy leads to plant shutdown, but there may be problems with the bankruptcy process itself. 7. Conclusion The major finding of this paper is that survival of steel plants can be impacted by parent firm bankruptcy. In contrast,

permanent plant shutdowns apparently do not affect the incidence of bankruptcy. Two questions arise from these results. First, what do they imply for policy? Second, what do they show about the controversy between Haugen and Senbet (1978) and Schleifer and Vishny (1992)? The former sees bankruptcy as an efficient way to sort out and deploy the assets of failing firms, while the latter posits that rational ways of setting up firm structure can lead to bankruptcies closing perfectly efficient plants. There are two policy implications of these results. First, even though the orphan effect is absent, given the knowledgeable steel buyers, it may be efficient for the government to bail out a failing steel firm. The problems posited by Schleifer and Vishny may lead to firms shutting down efficient plants in the face of bankruptcy. Consequently, a government bailout of such a firm could save efficient facilities. The second policy implication is that there may be good reasons for a reform of the bankruptcy laws. Perhaps, attempts at reform can take into account the governance problems implied by these results. As for the controversy between Haugen and Schleifer–Vishny, this paper’s results show that there is a statistically significant relationship between firm bankruptcy and the shutdown of the steel plants. This lends support to the latter’s hypothesis that the governance problem may be real, and the occurrence of firm bankruptcy may result in efficient plants closing. Making this connection especially important are two conditions in the steel industry. First, as mentioned above, the orphan effect is not present. Second, the majority of plants owned by the bankrupt firms were bought by other companies or reopened by new reorganized firms; this shows that there was a fairly open market in the United States for steel plants. In spite of these conditions, our results indicate that bankruptcy led to the shutdown of some efficient steel plants. Thus, even in industries with no orphan effect and a quite active market in plant assets, the Schleifer–Vishny effect may be present. Consequently, since governance with respect to financial distress is a problem that appears in many situations, it should be further addressed. Perhaps, parts of the bankruptcy process preventing the efficient redeployment of the resources of bankrupt firms could be changed.

Appendix. To check the proportional hazards model results, we estimate these models assuming a Weibull hazard function. For both the plant shutdown and parent firm bankruptcy models, statistical tests indicate that the maintained hypothesis of the exponential model, another form of the hazard function can be rejected. The exponential form is encompassed by the Weibull in that it has an assigned parameter value. The assumption of this value does not pass the Akaike Information Criterion. Therefore, the results for the Weibull models are discussed in this Appendix. Tables A1 and A2 display these results. The results are similar to the proportional hazards model, but for all models, the variables are never significant at the 5 percent level. For the instrumental variable models, the z values are less than one for all variables except TimeFirm in the plant closing model. The Mill’s ratio is significant for the parent firm model but not for the plant shutdown model. The direction implied by the TimeFirm is consistent with the results for the proportional hazards model. For the Weibull model, positive coefficients mean the same thing as negative coefficients for the proportional hazards model. (See Kiefer, 1988, pp. 664–666.) For the Weibull model, increasing the length of time between 1994 and plant shutdown increases the length of time between 1994 and plant closing. In contrast, increasing TimeFirm lowers the

R.P. Rogers / The Quarterly Review of Economics and Finance 53 (2013) 165–174

173

Table A1 The results for the plant survival model using the Weibull assumption. (The dependent variable is TIMEPlant , the number of years between 1994 and the year the plant shut down. For TIMEFirm , PlantCAP, and SOUTH one tail tests are appropriate for the relevant hypothesis, while for MULT a two tail test is appropriate.). Variable name

Coefficients

Weibull model using a strictly OLS estimation procedure 1.5941 Constantb 0.1294 TIMEFirm 0.0002 PlantCAP SOUTH 1.1076 MULT 0.7648 Weibull model using an instrumental variables estimation procedurea b Constant −0.9511 0.3668 TIMEFirm 0.0004 PlantCAP SOUTH 1.1120 MULT 0.5806

z values

Estimation probability values

1.642 1.432 0.948 1.441 1.501

0.0504 0.0762 (for a one tail test) 0.1715 (for a one tail test) 0.0749 (for a one tail test) 0.1333 (for a two tail test)

−0.269 1.021 0.237 0.079 0.172

0.3940 0.1537 (for a one tail test) 0.4064 (for a one tail test) 0.4685 (for a one tail test) 0.8634 (for a two tail test)

a

The z values in this model are computed from the covariance matrix adjusted for the instrumental variables described in the paper. The intercept term is needed in the Weibull models, while it is not necessary in the proportional hazards model. *Significant at the 5 percent level on a two tail test. **Significant at the 5 percent level on a one tail test. b

Table A2 The results for the parent firm bankruptcy model using the Weibull assumption. (The dependent variable is Timebankr , the number of years between 1994 and the year that the parent firm underwent bankruptcy. For TIMEPlant a one tail test is appropriate for the relevant hypothesis, while for FirmCAP, MULT, CONGL, and the Mill’s ratio two tail tests are appropriate.). Variable name

Coefficients

Weibull model using a strictly OLS estimation procedure Constantb 2.3053 0.0247 TIMEPlant −0.00004 FirmCAP 0.6483 MULT Mill’s ratio −2.3066 Weibull model using an instrumental variables estimation procedurea b Constant 0.2583 0.2208 TIMEPlant FirmCAP −0.00004 MULT 0.4883 Mill’s ratio −2.6589

z values

Estimation probability value

9.032 1.020 −1.571 2.191* −2.095*

0.0000 0.1538 (for a one tail test) 0.0581 (for a two tail test) 0.0284 (for a two tail test) 0.0362 (for a two tail test)

0.006 0.625 −0.016 0.033 0.061

0.4976 0.2660 (for a one tail test) 0.9872 (for a two tail test) 0.9736 (for a two tail test) 0.9514 (for a two tail test)

a

The z values in this model are computed from the covariance matrix adjusted for the instrumental variables described in the paper. The intercept term is needed in the Weibull models, while it is not necessary in the proportional hazards model. Significant at the 5 percent level on a two tail test. **Significant at the 5 percent level on a one tail test. b

*

probability of plant shut down in any given year in the Cox model. Both of these results have the same implication a longer time to bankruptcy lowers the likelihood of plant closure. It is reasonable to suppose that the restriction imposed by the imposition of a given functional form distribution explain the weakness of the Weibull results. References Andrade, G., & Kaplan, S. (1998). How costly is financial (not economic) distress? Evidence from highly leveraged transactions that became distressed. Journal of Finance, 53, 1443–1493. Barnett, D. F., & Crandall, R. (1986). Up from the ashes: The rise of the steel minimill in the US. Washington, DC: Brookings Institute. Barnett, D. F., & Schorsch, L. (1983). Steel: Upheaval in a basic industry. Cambridge, MA: Ballinger Publishing Company. Bernanke, B., & Gertler, M. (1989). Agency costs, net worth, and business fluctuations. The American Economic Review, 79(1), 14–31. Bolton, P., & Scharfstein, D. (1996). Optimal debt structure and the number of creditors. Journal of Political Economy, 104, 1–25. Coase, R. H. (1936). The nature of the firm. Economica, 4(16), 386–405. Cox, D. R., & Oakes, D. (1984). Analysis of survival data. London/New York: Chapman and Hall. Crandall, R. (1981). The U.S. steel industry in recurrent crisis. Washington, DC: Brookings Institution. Duke, R., Johnson, R., Mueller, H., Qualls, P., Roush, C., & Tarr, D. (1977). The United States steel industry and its international rivals: Trends and factors determining international competitiveness: Staff report. Washington, DC: Federal Trade Commission, Bureau of Economics, U.S. Government Printing Office. Gertner, R., & Scharfstein, D. (1991). A theory of workouts and the effects of reorganization law. Journal of Finance, 46, 1189–1222.

Greene, W. H. (2008). Econometric analysis. Upper Saddle River, NJ: Pearson/Prentice Hall. Greene, W. H. (2007). LIMDEP version: Econometric modeling guide. Plainview, NY: Econometric Software, Inc. Hammond, R. G. (2010). Quantifying consumer perception of a financially distressed company. Manuscript presented at Western Economic Association International, Portland, OR. Haugen, R. A., & Senbet, L. W. (1978). The insignificance of bankruptcy costs to the theory of optimal capital structure. The Journal of Finance, 33(2), 383– 393. He, Z., & Matvos, G. (2012). Debt and Creative Destruction: Why could Subsidizing Corporate Debt be Optimal? Cambridge, MA: Working Paper 17920, National Bureau of Economic Research. Heckman, J. J. (1978). Dummy endogenous variables in a simultaneous equation system. Econometrica, 46(4), 931–959. Hogan, W. T. (1983). World steel in the 1980s: A case of survival. Lexington, MA: Lexington Books. Hogan, W. T. (1986). Steel industry future. Duluth, MN: D.N. Skillings. Hogan, W. T. (1994). Steel in the 21st century. New York: Lexington Books. Hortac¸su, A., Matvos, G., Syverson, C., & Venkataraman, S. (2010). Are consumers affected by durable goods makers’ financial distress? The Case of auto manufacturers. Manuscript http://home.uchicago.edu/∼syverson Jensen, M., & Meckling, W. (1976). Theory of the firm: Managerial behavior, agency costs, and capital structure. Journal of Financial Economics, 3, 305–360. Jensen, M. C. (1986). Agency costs of free cash flow, corporate finance, and takeovers. Papers and Proceedings of the Ninety-Eighth Annual Meeting of the American Economic Association. The American Economic Review, 76(2), 323–329. Kiefer, N. M. (1988). Economic duration data and hazard functions. Journal of Economic Literature, 26(2), 646–679. Lynn, L. H. (1982). How Japan innovates: A comparison with the U.S. in the case of oxygen steelmaking. Boulder, CO: Westview Press. Maddala, G. S. (1992). Introduction to econometrics. New York: MacMillan Publishing Company.

174

R.P. Rogers / The Quarterly Review of Economics and Finance 53 (2013) 165–174

Maksimovic, V., & Phillips, G. (1998). Asset efficiency and reallocation decisions of bankrupt firms. The Journal of Finance, 53(5), 1495–1532. Modigliani, F., & Miller, M. (1958). The cost of capital corporation finance, and the theory of investment. American Economic Review, 48(3), 261– 297. Mueller, H. (1984). Protection and competition in the U.S. steel market: A study of managerial decision making in transition. Murfreesboro, TN: Business and Economics Research Center Middle Tennessee State University. Opler, T., & Titman, S. (1994). Financial distress and corporate performance. Journal of Finance, 49, 1015–1040. Preston, R. (1992). American steel. New York: Avon Books. Pulvino, T. (1998). Do asset fire-sales exist? An empirical investigation of commercial aircraft transactions. Journal of Finance, 53, 939–978. Rogers, R. P. (1993). The minimum optimal steel plant and the survivor technique of cost estimation. Atlantic Economic Journal, 21(3), 30–37.

Rogers, R. P. (2009). An economic history of the american steel industry. London/New York: Routledge. Scherer, F. M., et al. (1975). The economics of multi-plant operation: An international comparisons study. Cambridge, MA: Harvard University Press. Schleifer, A., & Vishny, R. W. (1992). Liquidation values and debt capacity: A market equilibrium approach. Journal of Finance, 47, 1343–1366. Serjeantson, R., Cordero, R. R., Cooke, H., Sexton, S., & Jordain, S. (2006, 1994). Iron and steel works of the world. England: Metal Bulletin Books, Ltd. Stigler, G. J. (1958). The economies of scale. Journal of Law and Economics, 1, 54–71. Tarr, D. (1984). The minimum optimal scale steel plant. Atlantic Economic Journal, 12(2), 122. Tornell, A. (1997). Rational atrophy: The U.S. steel industry. Cambridge, MA: National Bureau of Economic Research. Working Paper 6084. Williamson, O. E. (1988). Corporate finance and corporate governance. The Journal of Finance, 43(3), 567–659.