Creditor Rights, Technology Adoption, and Productivity: Plant-Level Evidence

Abstract

An influential body of research argues that legal institutions affect the economic organization of firms and the growth of an economy (King and Levine 1993; La Porta et al. 1997; Rajan and Zingales 1998). Although much has been learned, we still know little about the mechanisms by which specific legal rules and features govern firm financial and real outcomes. For example, the extent to which certain legal institutions lead to a relaxing of financial constraints and a consequent improvement in productivity via technological adoption and/or higher levels of investment is unclear. In this paper, I attempt to establish evidence along these lines by considering the role of secured creditor rights in bankruptcy on firm productivity and production processes, including the composition of both investment and workforce.

Stronger creditor rights, such as better legal enforcement, is a critical determinant of firm financing (Hart and Moore 1994; Rampini and Viswanathan 2013). More specifically, expansion of the menu of collateralizable assets or increased ability to recover collateral in bankruptcy may reduce the costs of lending for creditors, thereby leading to a lower cost of external financing for firms (Haselmann et al. 2010; Cerqueiro et al. 2016; Gopalan et al. 2016; Calomiris et al. 2017). With broader access to credit, firms can increase both their investment and employment (Campello and Larrain 2016; Gopalan et al. 2016; Aretz et al. forthcoming).

However, creditor rights can generate differences in not only the quantity but also the quality of investments and workforce and, consequently, productivity. Regarding investment, Eisfeldt and Rampini (2007) and Rampini (2019) argue that financial constraints constitute the main reason firms invest in used rather than new capital. Therefore, assuming that newer goods are more productive because of technological progress, relaxation in borrowing constraints might lead to higher technology adoption by firms, which could help them operate more efficiently (Midrigan and Xu 2014). Regarding workforce, Griliches (1969) and Krusell et al. (2000) argue that equipment and information technology (IT) capital are more complemetary to skilled labor than to unskilled labor. According to this capital-skill complementarity hypothesis, change in the composition of investment should lead to a change in the composition of workforce by increasing the demand for skilled labor.

Two obstacles hinder any attempt to study the effect of creditor rights on both firms’ productivity and composition of investment and workforce. First, standard corporate data sources generally report only aggregate items, such as capital expenditures or the number of employees, which does not allow a deeper look into the constituents. The disaggregated nature of the Census data reveals the type of capital and labor a firm invests in. The second obstacle relates to endogeneity: a variety of unobservable factors may be affecting both creditor rights in a country or state and firm behavior. To address this concern, I use enactment of antirecharacterization statutes as a source of exogenous variation in creditor rights (e.g., Li et al. 2016). These laws mainly affect the securitization industry and firms using a special purpose vehicle (SPV) to conduct secured borrowing. Collateral is transferred to a SPV for the purpose of protecting it from automatic stay in case of the bankruptcy of the debtor. However, before these laws, bankruptcy judges had the discretion to make the collateral in the SPV subject to automatic stay. These laws significantly increased the rights of secured creditors in bankruptcy by denying judges this discretion and allowing secured creditors to seize the collateral in the SPV.¹

I adapt the empirical methodology of Bertrand and Mullainathan (2003) to study the effect of antirecharacterization laws based on plant-level data from the U.S. Census Bureau. The granularity of the Census data coupled with the fact that antirecharacterization laws were adapted at the state of incorporation level helps me conduct two types of analysis. First, the Census data provide the exact geographic location and industry of each plant, which allows me to compare two plants in the same year, industry, and location. To illustrate, I compare the productivity change of two plants in Iowa, one of which belongs to a firm incorporated in Texas (a treated state) and the other that belongs to a firm incorporated in California (a control state). This plant-level analysis allows me to observe the productive efficiency of a plant far from the state of headquarters or state of incorporation, and separate out the effect of stronger creditor rights from the effect of local economic shocks contemporaneous with the laws. Second, the longitudinal nature of the plant-level data helps me observe the productivity of a plant for several consecutive years, which is crucial in the context of this study, given that effects on productivity may take time to materialize.

I find that following the adoption of antirecharacterization laws, total factor productivity (TFP) of plants belonging to treated firms increases significantly, by 2.6%. Further, with an increased access to financing, long-term and total debt significantly increase at treated firms.

Next, I explore the underlying mechanism through which strengthening of creditor rights and the resultant increased access to financing leads to greater productivity. To this end, I explore the unique features of the Census data and look at the composition of investments and workforce. First, I show that treated plants adopt more advanced technology by investing in newer capital and IT; new machinery and computer investments significantly increase at treated plants. Consistent with Eisfeldt and Rampini (2007), who show that investing in used capital rather than new capital is more common among credit-constrained firms, I document that the fraction of capital expenditures in new capital, new machinery, and IT significantly increases at treated plants.² Changing composition of investment is related to financial constraints: only plants belonging to financially constrained firms experience an increase in both the fraction of capital expenditures on new capital and productivity. This finding fits well with that of Midrigan and Xu (2014), who argue that financial constraints may reduce the productivity of individual producers by distorting their technology-adoption decisions.

Next, to shed further light on the underlying mechanism, I analyze the importance of asset redeployability. Eisfeldt and Rampini (2009) argue that leasing allows higher debt capacity for financially constrained firms because U.S. bankruptcy laws make repossession of assets easier for lessors than for secured lenders. In line with this reasoning, antirecharacterization laws that make it easier for secured creditors to repossess assets in bankruptcy can lead firms to switch from leased (Chu forthcoming) or used assets to new assets. However, not all assets have an active leasing or secondary market. Gavazza (2010, 2011 argues that asset redeployability is a crucial determinant of the thickness of the secondary markets. Therefore, changing composition of investments toward newer assets should be more pronounced for more redeployable assets for which there is an active secondary market. The findings support this hypothesis: using various redeployability measures, I find that plants operating in industries characterized by assets that are more redeployable experience a significant increase in the fraction invested in new capital and new machinery.

Finally, I look at the composition of workforce. The Census provides information about different types of employees, including production workers and nonproduction personnel consisting mainly of supervisors and white-collar employees in sales and marketing, financing, purchasing, and professional and technical roles. I show that total employment increases at treated plants. However, a closer look at different types of employees reveals that only the number of nonproduction workers significantly increases at treated plants. I see no significant change in the number of production workers. This shows that consistent with Griliches (1969) and Krusell et al. (2000), treated plants that change the composition of their investments toward newer and IT capital also change the composition of their workforce by hiring employees that are more skilled.

My findings contribute to the literature on creditor rights and optimal bankruptcy design. Strengthening of creditor rights can result in inefficient liquidation of firms by increasing the cost of continuing during financial distress (Aghion and Bolton 1992; Hart et al. 1999). This liquidation bias can affect ex ante financing and investment policies. Acharya and Subramanian (2009); Acharya et al. (2011); Berger et al. (2011); Seifert and Gonenc (2012); and Vig (2013) present evidence consistent with this view, showing that stronger creditor rights are associated with less innovation, as well as financing policies that are more conservative. A more recent set of studies shows effects contrary to the liquidation bias. Li et al. (2016) show that firms increase leverage significantly after the adoption of antirecharacterization laws. Favara et al. (2019) and Ponticelli and Alencar (2016) document increases in borrowing and investment when creditor rights are stronger. Finally, Mann (2018) and Cerqueiro et al. (2017) report a positive link between creditor rights and innovation output. I extend this prior work in two ways: First, there is little evidence on the productivity consequences of these creditor rights related financing and investment decisions.³|$^{,}$|⁴ Second, I provide new evidence that creditor rights including legal enforcement of financing affect not only the level or quantity but also the composition or quality of investments and workforce.⁵

In a closely related paper, Benmelech and Bergman (2011) show that airlines operating in countries with higher creditor protection use aircraft of a younger vintage and newer technology. I extend their work in three different ways. First, I identify a specific mechanism, rights of secured creditors in bankruptcy, and add to our understanding of which aspects of creditor rights matter more for financial and real outcomes (Calomiris et al. 2016). Second, exploiting the industry-level variation of Census data, I uncover the importance of asset redeployability in the relationship between creditor rights and investment. The fact that U.S. bankruptcy laws make repossession of assets easier for lessors than for secured lenders makes secondary markets, as well asset liquidity/redeployability, an important channel through which stronger creditor rights translate into changing composition of investments.⁶ Finally, I complement their work by showing that creditor rights affect both the quality of a firm’s capital and the quality of its workforce.

Finally, my study extends a set of studies looking at the effect of finance on different types of labor. Brown and Matsa (2016) and Ramin et al. (2018) show that financial distress prevents firms from hiring and retaining skilled labor. Caggese et al. (forthcoming) argue that financing constraints lead firms to fire short-tenured workers with high future expected productivity. Finally, using industry-level data, Larrain (2014) documents that capital-account opening and the resultant increased access to capital increase sectoral wage inequality by increasing the demand for skilled labor, rather than unskilled labor. I complement these studies by showing that stronger creditor rights can have differential ex ante effects on different types of labor through a relaxation of borrowing constraints and change in the composition of investments.

1. Antirecharacterization Laws

Antirecharacterization laws affect the securitization industry and firms using a SPV to conduct secured borrowing. The originating firm transfers the collateral to a SPV, which is a financial intermediary designed to be bankruptcy remote, and then sells it to investors as a security. The main reason for transferring assets to a different entity is to characterize them as sales, thereby protecting them from automatic stay in case of bankruptcy of the issuing firm. The bankruptcy remoteness of a SPV assures investors that obligations will be fulfilled even if the SPV-originating firm goes bankrupt. However, the bankruptcy remoteness of a SPV should not be taken for granted. Sometimes bankruptcy courts can recharacterize the asset transfer as a secured loan rather than as a true sale, making the SPV assets subject to automatic stay. Bankruptcy courts generally justify this by stating that the SPV will play an important role in the reorganization of the bankrupt firm. The enactment of antirecharacterization laws discards the possibility of recharacterization by bankruptcy courts.

Seven states enacted these laws: Texas and Louisiana in 1997, Alabama in 2001, Delaware in 2002, South Dakota in 2003, Virginia in 2004, and Nevada in 2005. Kettering (2010) states that these seven states fall into two groups in terms of the coverage of their antirecharacterization laws. Texas and Louisiana use Section 9.109 of the Uniform Commercial Code (UCC) to discard the possibility of recharacterization for all sales of receivables, whereas the remaining states use stand-alone statutes to discard this possibility only for sales under a securitization transaction.⁷

Although seven states enacted these laws, I consider only Texas, Louisiana, and Alabama as the treatment states because of the 2003 federal court ruling on Reaves Brokerage Company, Inc., v. Sunbelt Fruit & Vegetable Company, Inc. Reaves, a fresh food and vegetable seller, was making frequent sales to Sunbelt, a wholesaler. Sunbelt ceased operations in March 2000, owing Reaves |${\$}$|195,060.55. Reaves sought recovery from Sunbelt under the Perishable Agricultural Commodities Act (PACA). A couple of months later, Reaves sought recovery also from Fidelity Factors, LLC, a factoring company that had purchased several accounts receivable from Sunbelt. PACA was enacted in 1930 and later amended in 1984 to protect sellers of perishable agricultural products. PACA requires that perishable product sellers be paid promptly and their claims prime those of other secured and unsecured creditors in case the buyer gets out of business. Based on this, Reaves argued that the sale of accounts is not a true sale but just a disguised way of financing, which makes Fidelity a secured creditor whose claims must be subordinate to theirs. Fidelity defended itself based on the Texas antirecharacterization statute, arguing that they are buyers of accounts receivable and not lenders. The court ruled in favor of Reaves, stating that the relationship between Fidelity and Sunbelt was that of a secured lender and debtor, not that of a seller and buyer. This ruling created uncertainty around antirecharacterization laws because it created a precedent where federal courts could overrule state antirecharacterization laws.⁸

A deeper look at antirecharacterization laws shows that these laws affect a substantial number of transactions. Section 9.109 of the Texas and Louisiana UCC explicitly discards the possibility of recharacterization of the sale of the following four items: accounts, chattel paper, payment intangibles, and promissory notes.⁹ The most important item for the purposes of the study is accounts, defined in Section 9.102 of the UCC. Accounts means a right to payment of a monetary obligation, whether or not earned by performance, for a variety of items, including (a) property that has been or is to be sold, leased, licensed, assigned, or otherwise disposed of; (b) services rendered or to be rendered; (c) a policy of insurance; (d) licensing of intellectual property; and (e) credit cards. This definition extends beyond the traditional definition of accounts consisting mainly of trade receivables.

In Alabama, the third treatment state, and in the remaining four states, an act that precludes recharacterization of a sale of property of any kind included in a securitization transaction guarantees antirecharacterization. In Appendix A, Section 35-10A-2 of the Code of Alabama states that no property, assets, or rights transferred to a SPV can be subject to recharacterization. Assets typically transferred to a SPV include trade receivables, residential and commercial loans, equipment leases, licenses, and management contracts.

Taken together, these laws increase the pledgeability of any rights to future payment (i.e., receivables) by making them more accessible to creditors. Manufacturing firms, which constitute the sample, in particular generate a large amount of receivables. Korgaonkar and Nini (2010) state that firms involved in the manufacturing and production of consumer durables are heavy users of SPVs. For this reason, I expect these antirecharacterization laws to have a significant effect on manufacturing firms.

Manufacturing firms are not the only users of securitization. Feng et al. (2009) document that the percentage of firms using at least one SPV increased from 23% in 1997 to 59% in 2004, which confirms how common they became in the period the sample covers. Finally, these laws are fairly exogenous to the firms included in the sample. Kettering (2008) describes the enactment of antirecharacterization laws as a great success of the securitization industry. Janger (2003) constitutes another study regarding the big role that the financial industry played in the enactment of antirecharacterization laws. I will return to the endogeneity concern that is attributable to lobbying by industrial firms in Section 3.1.1.

2. Data and Empirical Methodology

2.1 Data sources

The main data I use in this study are a combination of two data sets from the Census. Both data sets cover only U.S. manufacturing plants. The first manufacturing plant data set is the Census of Manufacturers (CMF). The CMF is a survey conducted every 5 years, which consists of all manufacturing plants in the United States with at least one paid employee. The second manufacturing data set is the Annual Survey of Manufacturers (ASM). The ASM is conducted in years when the CMF is not conducted, and it includes a subset of plants surveyed in the CMF. It includes all plants with more than 250 employees and some with fewer employees, included with a probability positively correlated with their number of employees. Both of these data sets include detailed information on industry, corporate affiliation, total shipments, employment, capital expenditures, and material inputs of each plant. Reporting for both of these surveys is mandatory, and misreporting is penalized, which alleviates misreporting and response rate concerns. The level of granularity of these manufacturing data sets plays a critical role in the analysis. First, it allows for the construction of various measures of productivity for each manufacturing plant, and thereby, for analysis of how the productivity of a plant and the creditor rights are related. Second, it makes it possible to study different types of capital and labor by providing granular and reliable machinery, computer, and employment numbers.

In addition to these two manufacturing data sets, I use another data set from the Census, the Longitudinal Business Database (LBD). The LBD keeps yearly track of all business establishments in the United States with at least one paid employee. It provides information on the number of employees, payroll, geographical location, industry, and corporate affiliation for each establishment. For the purpose of this study, I use the LBD to get information about the age of plants and the number of plants firms own.

Finally, I use firm-level data from Compustat. I confine the analysis to public firms because I can observe the state of incorporation only for public firms. To match firms in Compustat to plants, I use the Compustat bridge of the Census Bureau that ends in 2005. I extend the bridge to 2009 by making use of various employer characteristics, including name, address, and identification number (EIN). The sample covers the period from 1992 to 2009.

2.2 Variable construction and summary statistics

I consider firms headquartered or incorporated in Texas or Louisiana between 1997 and 2003, or Alabama between 2001 and 2003 as treated firms. The question of which state law will govern recharacterization is complicated.¹⁰ Article 9 of the UCC specifies the rules governing secured transactions, including securitization. A revised version of Article 9, effective as of July 2001, states that the law governing a secured party’s interest in receivables is the law of the location of the debtor, which is defined to be the state in which the entity is registered. For corporations, this definition of location of debtor means the state of incorporation.¹¹ However, the old version of Article 9, effective before June 2001, deemed the debtor’s location to be the location of its chief executive office, that is, the headquarters. The official comment to this older version of Article 9 explicitly states that the chief executive office is not necessarily the place of incorporation but is the place from which the debtor manages the main part of his business operations. Therefore, both the state in which the headquarters is located and the state of incorporation must be taken into account.¹²

The main dependent variable is the natural logarithm of total factor productivity, which I construct at the plant level following the methodology in Foster et al. (2016).

Using a perpetual inventory method, I construct capital stock that consists of structure and equipment. I take a plant’s first year in the CMF/ASM records as a starting point to construct the capital stock series. I then add capital expenditures, using investment price deflators from the Bureau of Labor Statistics (BLS) at the 2-digit SIC or 3-digit NAICS level, to the capital stock each year. I use an industry-level depreciation rate from the Bureau of Economic Analysis. I construct the capital stock series for structure and equipment separately. They sum to the total capital stock owned by the plant. I use “production worker equivalent hours” as the measure of labor input, which I construct by multiplying the number of hours worked by production workers with total wages divided by wages paid to production workers. Materials costs include both non energy and energy-related costs. Nonenergy materials costs include costs of materials and parts, as well as cost of resales and contract work. Energy costs include electricity and fuel costs. I deflate both types of costs by their corresponding industry deflators. Finally, I use industry-level factor cost shares for factor elasticities from the NBER Productivity Database.

In addition to TFP, I analyze the effects of stronger creditor rights on composition of investment and workforce. I use two new variables for investment: machinery investment and computer investment. I construct machinery investment by scaling machinery expenditures by lagged machinery and equipment stock. To represent investment in IT, I use the capital expenditures for computers variable from the CMF and ASM databases and scale it by lagged machinery and equipment stock. Specifically, capital expenditures for computers includes computer hardware, software, and network expenditures. I have this variable starting with 2000. As to composition of investment, the Census data allow me to see capital expenditures in new and used goods, separately. I define four new variables available until 2001: Fraction New Capital, new capital expenditures over total capital expenditures; Fraction New Machinery-1, new machinery expenditures over total machinery expenditures; Fraction New Machinery-2, new machinery expenditures over total capital expenditures; and Fraction computer, capital expenditures for computers expenditures over total capital expenditures.

Alongside composition of investment, I also analyze composition of workforce. Census microdata provide information about two types of employees: production workers and nonproduction personnel. Production workers consist of employees engaged in production operations at the plant up through the working foreman level. Nonproduction personnel consist of supervisors that are above the foreman level and office employees. To study the growth rate of these two types of labor, I calculate the symmetric growth rate of employment as two times the change in the number of employees scaled by the sum of the current and lagged number of employees. A big advantage of this measure is that it limits the effects of extreme values (Davis et al. 1998). To study the composition of workforce, I define Fraction nonproduction emp. as the natural logarithm of the number of nonproduction employees over the total number of employees. I winsorize all variables at 1% to ensure that the results are robust to outliers.

Table 1 presents summary statistics for the aggregate sample and the subsample of plant-year observations treated by antirecharacterization laws.¹⁵ The final sample contains 220,000 plant-year observations covering the period 1992–2009.¹⁶ The subsample of treatment plants accounts for a relatively small fraction of the total plant-year observations. However, the treatment subsample nearing 10,000 plant-year observations is sufficient, given that the treatment effect requires only that the treatment subsample be sufficiently large in absolute terms, as argued in Giroud (2013). Control plants are older and bigger, which constitutes one of the main reasons I include the age of plants and their total value of shipments to control for differences between treatment and control plants.

2.3 Identification and empirical model

I control for time-invariant characteristics at the plant level through plant fixed effects. I use state of incorporation fixed effects to control for time-invariant differences between treated and untreated plants. In addition, I use state of headquarters fixed effects, given that the definition of treatment includes both state of incorporation and state of headquarters. The state of location dummies interacted with industry and year dummies, |$\alpha_{lst}$|⁠, allow me to control for contemporaneous shocks at the state of location and industry levels. I use state and industry dummies for two main reasons. First, TFP shows considerable variation across industries (Foster et al. 2016). Second, the TFP measure is essentially a revenue measure. Demand shocks at different geographies may inflate or deflate the TFP measure in a way unrelated to the productive efficiency, which constitutes the main reason to control for state of location. For these two main reasons, I use state |$\times$| industry |$\times$| year fixed effects to control for contemporaneous shocks at the state of location level. I use fixed effects instead of demeaning the dependent variable because demeaning can lead to inconsistent estimates (Gormley and Matsa 2013).

The coefficient of interest is |$\delta$|⁠, which measures the effect of antirecharacterization laws on TFP, composition of investment, and composition of workforce. |$\delta$| measures the difference between the productivity change of two plants located in the same state (e.g., Iowa) and operating in the same industry (e.g., food manufacturing), one of which belongs to a firm incorporated in Texas (a treated state), and the other that belongs to a firm incorporated in California (a control state).

3. Empirical Results

3.1 Antirecharacterization laws and productivity

I first estimate Equation (2) with the natural logarithm of TFP as the dependent variable.¹⁹  Table 2 presents the results.

Column 1 presents results from estimation of Equation (2) with only plant and year fixed effects. The coefficient of interest on ARL is |$0.032$|⁠, and is significant at the 1% confidence level. This estimate corresponds to a 3.2% increase in TFP, which is in line with the prediction that stronger creditor rights will lead firms to operate more efficiently.

Column 2 adds state of incorporation and state of headquarters fixed effects to control for time-invariant differences between treated and untreated firms. The inclusion of these fixed effects decreases the estimated coefficient to |$0.027$|⁠. However, this coefficient is still significant at the 1% confidence level, and it translates into a 2.7% increase in TFP.

Column 3 adds state of location-industry-year fixed effects to the estimation in Column 2 to tighten the specification. As discussed in Section 2.3, these interacted dummies allow me to control for contemporaneous shocks at the state-of-location level. The inclusion of these controls makes a minor change to the estimated coefficient of interest, which is |$0.026$| and significant at the 1% confidence level. To put this estimate in economic terms, a 2.6% increase in TFP corresponds to a 2.6% increase in revenues with the same set of inputs (i.e., holding costs constant). Because profits equal revenues minus costs, the percentage increase in profits will depend on the profit margin. Given that the pretreatment average profit margin is 35%, a 2.6% increase in revenues will lead to a 2.6|$\times$|(100/35)=7.43% increase in profits, which is economically significant.²⁰|$^{,}$|²¹

Taken together, these estimates indicate that the passage of antirecharacterization laws has an economically large and statistically robust effect on TFP of treated plants. These estimates suggest that stronger creditor rights have a positive effect, leading firms to operate more efficiently.

3.1.1 Robustness checks

Next, I conduct several tests to ensure the robustness of the results in Table 2. First, in Appendix IA.IV, I estimate two alternative specifications. Specification 1, in Columns 1 to 3, includes a Post-ARL dummy equaling one for Texas and Louisiana from 1997 onward, and one for Alabama from 2001 onward, and a separate Reaves dummy equaling one for these three states from 2004 onward. Although estimated coefficients on Reaves are not statistically significant, estimated coefficients on Post-ARL are close to Appendix IA.I in terms of magnitude, and higher than benchmark estimates. This shows that the effect is mainly coming from the period before the federal court ruling. Specification 2, in Columns 4 to 6, further augments Specification 1 with a separate dummy, DSVN, which equals one for the states that pass antirecharacterization laws after the federal court ruling (i.e., DSVN equals one for Delaware, South Carolina, Virginia, and Nevada from 2002, 2003, 2004, and 2005 onward, respectively). Specification 2 reinforces the conclusion that the effect is stronger before the federal court ruling.

Next, I further account for heterogeneity between treatment and control samples. First, to control for unobserved time-invariant and time-varying factors at the firm level, I augment Equation (2) with firm fixed effects and firm-level control variables, including sales, profitability, Tobin’s q, and tangibility. Appendix IA.V shows that benchmark results are robust to this new specification. Second, I adopt a nearest-neighbor propensity score matching scheme. The treatment group consists of firms headquartered or incorporated in Texas, Louisiana, or Alabama. One year before the enactment of antirecharacterization laws, I match each treatment plant in Texas, Louisiana, and Alabama to a control plant using a nearest-neighbor propensity score matching with replacement. I estimate propensity scores based on plant size, plant age, and firm size. Plant state of location and plant industry require an exact match. I also require that both sets of plants have at least one observation before and at least one observation after the enactment of antirecharacterization laws.

Appendix IA.VI displays the results. Panel A tabulates the means of the matched variables for the treatment and control groups 1 year before the enactment of antirecharacterization laws. The sample means of control variables for matched treated and control samples are not significantly different. Panel B presents estimates of the plant-level effect of antirecharacterization laws on TFP using these propensity-score-matched samples from the period 1992–2003. The estimated coefficient of interest is |$0.030$|⁠, which is very close to the benchmark estimate of |$0.026$|⁠.

One potential problem is related to the endogeneity of antirecharacterization laws. If firms lobbied for these laws, then their enactment might be correlated with unobserved factors affecting firms’ productivity. First, Kettering (2008) describes the enactment of antirecharacterization laws as a great success of the securitization industry, and argues that these statutes are the product of efforts by the financial industry to abolish the possibility of recharacterization.²² Second, I address this concern with a dynamic version of Equation (2) to investigate when the effects of these laws materialize. If these laws are the result of economic factors leading firms to lobby for them, then I should be able to detect their effect before their enactment.

Table 3 investigates the dynamic effects of antirecharacterization laws. First, the estimated coefficients on |$\text{Before}^{-2}$| and |$\text{Before}^{-1}$| are economically small and statistically insignificant. The estimated coefficient on |$\text{Before}^{0}$|, which represents the year antirecharacterization laws passed, is statistically insignificant. This shows that there is no statistically significant difference in the evolution of productivity at treated and control firms before the enactment of antirecharacterization laws, which is consistent with the parallel trends assumption. Second, the coefficients on |$\text{After}^{1}$|, |$\text{After}^{2}$|, and |$\text{After}^{3+}$| are economically and statistically significant, which shows that the effect of these laws on TFP starts materializing 1 year after their passage.

In Table 4, I conduct another test to check the validity of the results in Table 2 and Table 3. One might worry that regional shocks could affect firms headquartered or incorporated in treated and nearby states. Therefore, the estimates might simply be picking up these regional shocks rather than the effect of antirecharacterization laws. I address this issue in the following way: I falsely assume that states bordering Texas, Louisiana, and Alabama passed the antirecharacterization laws. Estimated coefficients in Columns 1 to 3 are statistically indistinguishable from zero, which shows that the results in Tables 2 and 3 are not artifacts of some regional or political shocks affecting states in the near geography of Texas, Louisiana, and Alabama.²³

A concern with the productivity measure is that it relies on structural assumptions, including the Cobb-Douglas production function. To address this, in Appendix IA.VIII, I use operating margin, labor productivity, and return on capital (ROC) as an alternative measures of productivity that do not involve any structural assumptions. I calculate operating margin by scaling total value of shipments minus labor and material costs by total value of shipments. I use the measure of labor productivity used in Brav et al. (2015) and Silva (2013): natural logarithm of value added per labor hour, which is the total value of shipments minus material and energy costs divided by the total labor hours. Finally, I define ROC as total value of shipments minus labor, material, and energy costs scaled by capital stock. Estimated coefficients of interest are statistically significant at the 1% confidence level in all three columns. Estimated coefficients for labor and capital productivity in Columns 2 and 3 are 0.068 and 0.095, and are economically significant given that they constitute about 7.3% (0.068/0.931) and 4.4% (0.095/2.136) of their standard deviation, respectively. The remarkable increase in labor productivity implies the importance of the machinery and IT investments I will analyze in in Section 3.2.3.

In ASM, plants with more than 250 employees are always included, whereas plants with fewer employees are chosen randomly, with a probability positively correlated with their number of employees. This sampling policy creates bias toward larger plants. To address this, I reestimate the baseline regression by weighting observations by their respective ASM sample weight. Results in Appendix IA.IX are very similar to those in Table 2.

Finally, I reestimate the baseline regression in Table 2 by allowing time shocks to affect plants of different characteristics differentially. To do this, I interact year dummies with plant size, plant age, and the number of plants belonging to the firm. Estimated coefficients in Appendix IA.X are very similar to those in Table 2, which shows that baseline results are robust to the inclusion of these fixed effects.

3.2 The channel behind creditor rights and increasing productivity

3.2.1 Receivables

From this point on, I investigate the channel through which stronger creditor rights translate into increasing productivity. Antirecharacterization laws apply to all sales of receivables. Therefore, effects of these laws should be more pronounced for firms whose receivables decrease.

I test this hypothesis in Table 5. I classify each firm to be undergoing a change in receivables depending on its lagged change in receivables (recch) scaled by lagged assets (at). Columns 1 and 2 classify firms below the median of the change in receivables distribution as undergoing a decrease in receivables, and firms above the median as undergoing an increase in receivables.

Estimated coefficients show that effect of these laws is concentrated among plants belonging to firms whose receivables undergo a decrease. This result is in line with the mechanism facilitated by antirecharacterization laws.

3.2.2 Debt

Next, I analyze whether stronger creditor rights as a result of antirecharacterization laws lead firms to increase their borrowing. On the theoretical front, the effect of stronger creditor rights on the borrowing behavior of firms is not obvious. On the one hand, relaxation of collateral constraints would induce firms to borrow more, as predicted in Hart and Moore (1994). On the other hand, Vig (2013) argues that increasing access to collateral by creditors might lead firms to decrease their use of secured debt. Further, leverage also depends on whether or not the originator firm consolidates the SPV onto its balance sheet. In case of consolidation, the balance sheet include both the underlying asset and the borrowing by the SPV. In case of unconsolidation, the originator firm removes the underlying asset from its balance sheet, and does not include the borrowing by the SPV as additional debt in its balance sheet.²⁴

Table 6 shows that treated firms increase their borrowing; this result is consistent with that of Hart and Moore (1994) and Li et al. (2016). In Columns 1 and 2, long-term debt scaled by total assets increases significantly. The estimated coefficient of interest is statistically significant at the 1% confidence level. In Columns 3 and 4, leverage, defined as the sum of long-term and short-term debt divided by total assets, increases significantly. In terms of economic magnitude, the estimated coefficient corresponds to approximately 5% increase relative to the sample mean of 0.267. Finally, in Columns 5 and 6, I define the dependent variable as the natural logarithm of the sum of long-term and short-term debt to ensure that the results in Columns 1 to 4 are not driven by the asset values in the denominator. Estimated coefficients are in line with Columns 1 to 4. These results confirm that antirecharacterization laws relax the borrowing constraints of treated firms by allowing creditors more access to collateral.

3.2.3 Composition of investment

To shed further light on the underlying mechanism, in this section, I establish the link between increased borrowing and increased productivity by looking at the composition of plant-level investment and employment.

An important body of literature argues that relaxation of collateral-based lending constraints will result in higher investment by firms.²⁵ However, not all investment is the same in terms of productivity consequences. Firms might be investing in pet projects, or invest in land or buildings that may have little effect on the efficiency of the production process.²⁶ Therefore, the composition of investment plays a critical role for productivity.

The main reason for relatively limited research on the composition of investment or the composition of a firm’s workforce is data limitations. For investment, aggregate items including capital investment expenditures or plant, property, and equipment reported by standard corporate data sources cannot inform us about the vintage of the productive assets in a manufacturing plant. As to employment, difficulty of finding reliable firm- or plant-level data constitutes an additional obstacle against any attempt to study the composition of a firm’s workforce.

I focus on two aspects of composition of investment relevant for productivity. The first aspect is vintage of capital. Eisfeldt and Rampini (2007) show that investing in used capital rather than new capital is very common among credit-constrained firms. If new capital is more productive because of technological progress, then firms can become productive by investing in new capital rather than used capital. The second aspect is IT. Research confirms the relationship between IT, computers, and output.²⁷ Two recent studies in the finance literature examine the effect of IT-related investments. First, Brav et al. (2015) show that IT-related investments by hedge funds contribute to the productivity of target firms. Second, Ashwini and Tambe (2016) argue that employees in target firms benefit from IT-related private equity investments by acquiring new skills. They show that workers treated with a private equity investment earn higher wages on average. Therefore, IT-related investment can be a channel through which stronger creditor rights translate into increasing productive efficiency.

First, in panel A of Table 7, I analyze the effect of antirecharacterization laws on the level of investment. Columns 1 and 2 show that Total investment, defined as total capital expenditures scaled by lagged capital stock, significantly increases at treated plants. In terms of economic magnitudes, the estimate corresponds to a 7% increase relative to the sample mean, which is significant.²⁸ In Columns 3 to 6, a breakdown of total investment indicates that both machinery and computer investments experience a similar significant increase.²⁹|$^{,}$|³⁰

Next, I examine the composition of investment in panel B. The disaggregated nature of the Census data allows us to see capital expenditures in new and used goods, separately. In Columns 1 and 2, the dependent variable is Fraction New Capital, defined as new capital expenditures over total capital expenditures. I find that the fraction of investment composing new capital significantly increases at treated plants. Estimated coefficients of interest in Columns 1 and 2 are statistically significant at the 1% confidence level, and they indicate an increase of |$2.8$| and |$2.4$| percentage points, respectively. This constitutes a significant increase, given that the pretreatment fraction of investment in new capital is 94%.

In the remaining columns of panel B, I analyze how the fraction invested in new machinery and computers is affected. The dependent variable is defined as new machinery expenditures over total machinery expenditures in Columns 3 and 4, and as new machinery expenditures over total capital expenditures in Columns 5 and 6. Estimated coefficients indicate a significant increase in the fraction invested in new machinery. Further, Figure 1 plots the evolution of new machinery expenditures over total capital expenditures over time. The figure is consistent with the “parallel trends” assumption. Finally, Columns 7 and 8 indicate that IT-related expenditures over total capital expenditures significantly increase at treated plants.

Figure 1

The effect of the adoption of antirecharacterization laws on the “fraction” of new machinery investments

This figure shows the effect of the adoption of antirecharacterization laws on the “fraction” of new machinery investments, defined as new machinery expenditures scaled by total capital expenditures. I estimate Equation (2), except that I replace ARL with dummy variables indicating the year relative to the adoption of antirecharacterization laws, where year t is the year antirecharacterization law is adopted in a given state. The solid line represents the point estimates associated with each of these dummy variables. The dashed lines represent the 95% confidence interval at which robust standard errors are clustered at the state-of-location level.

Open in new tabDownload slide

In summary, I find strong evidence that not only the level of investment but also the composition of investment change at treated plants. The fraction invested in new capital rather than used capital as well as the fraction invested in IT increase significantly following the enactment of antirecharacterization laws.

3.2.4 Composition of investment and financial constraints

Next, I analyze how financial constraints affect the composition of investment and productivity. Eisfeldt and Rampini (2007) argue that financial constraints constitute the main reason firms invest in used capital rather than new capital. To reiterate, if new capital is more productive because of technological progress, then firms can become more productive by investing in new capital rather than used capital. This reasoning is in line with Midrigan and Xu (2014), who argue that distortion of technology adoption decisions is an important channel through which financial frictions decrease individual firms’ productivity. According to this argument, following the relaxation in borrowing constraints, firms can become more productive by increasing their fraction of investment in new capital.

I use two different measures of financial constraints. I take the median value 1 year before the treatment across all companies to classify firms as financially constrained or financially unconstrained. The first measure of financing constraints is the size-age (SA) index first used in Hadlock and Pierce (2010). I classify firms as financially constrained (unconstrained) if their SA value is above (below) the median. The second measure is the Whited-Wu index first used in Whited and Wu (2006). I classify firms as financially constrained (unconstrained) if their Whited-Wu value is above (below) the median.

Table 8 presents the results. Columns 1 to 3 of panels A and B show that the adoption of antirecharacterization laws has little to no effect on the investment composition of relatively unconstrained firms. Plants belonging to relatively constrained firms, however, experience a significant increase in the fraction invested in new capital and new machinery. Further, changing the composition of investment is associated with higher productivity. Column 4 in both panels shows that plants belonging to relatively constrained firms experience a significant increase in productivity, in line with the predictions of Midrigan and Xu (2014).³¹

3.2.5 Composition of investment and asset redeployability

Next, I analyze how asset redeployability affects the relationship between creditor rights and composition of investment. Eisfeldt and Rampini (2009) argue that because U.S. bankruptcy laws make the repossession of assets easier for lessors than for secured lenders, leasing allows for increased debt capacity. This higher debt capacity is valuable, especially for financially constrained borrowers, despite the agency costs associated with leasing. Consistent with this, I show in Section 3.2.4 that after the enactment of antirecharacterization laws allowing secured creditors to seize their assets more quickly during bankruptcy, financially constrained firms switch to new assets from used or leased assets. This changing composition should be more visible for more redeployable assets with an active secondary-asset markets making the relationship between primary and secondary markets more substitutable (Gavazza 2010; Chu forthcoming).

I use two industry-level redeployability measures, kindly provided by Kim and Kung (2009). To build these industry-level measures, first, a redeployability score for each asset is constructed. For both measures, this score represents the proportion of firms using the asset with the only difference that the second measure also incorporates correlations of outputs among industry peers in the construction of asset redeployability score. The industry-level redeployability measures are then constructed by value-weighting the asset redeployability score of each asset in the industry. I define an industry redeployable if its corresponding redeployability measure ranks above the median in a given year.

Table 9 presents the results. First, estimated coefficients on ARL are statistically insignificant, which shows that there is no significant change in industries with assets that are less redeployable. Second, plants operating in industries characterized by assets that are more reployable experience a significant increase in the fraction invested in new capital and new machinery. Estimated coefficients on ARL |$\times$|  Redeployable are statistically significant at the 1% confidence level. This shows that asset redeployability, which makes the relationship between primary and secondary markets more substitutable, allows firms to switch to new assets following a strenghtening of creditor rights in bankruptcy.