Bankruptcy Shocks and Legal Labor Markets

Chad Brown Jeronimo Carballo Alessandro Peri University of Colorado University of Colorado University of Colorado

Abstract

We estimate the impact of Chapter 11 bankruptcy filings on local legal markets. Bankruptcy of U.S. based publicly traded firms act as unrelated demand shocks to legal sector’s local labor market. Constructing and employing a novel database on publicly traded firms bankruptcies for the period 1991-2001, we find that bankruptcies boost county legal employment. Exploiting a stipulation of the law referred to as Forum Shopping during the period 1991-1996, we confirm our interpretation of the results. We also document a counter-intuitive reduction in the average wage for the legal sector. We reconcile these joint implications in a model with skilled-unskilled workers, where bankruptcy shocks tilt the workforce distribution towards lower skilled legal workers (compositional channel).

JEL codes: K00, J00, G33, J40, J20, J82.

TO DO List - Finance Forum

Unbolded as completed.

•
Tables:
- –
  
  We need to find a proper label for LP in the Tables.
- –
  
  Change n_BR and n_FS to proper labels.
- –
  
  Drop the Lincom part of the tables.
- –
  
  Create one table that collects the Lincom and average effects for Table 4. We need to have standard errors for the coefficients and the corresponding stars, and compute the test between N_BR and N_FS_CS for the third column. Each set of test should be a new rows. The labels for these three additional test should be: Forum Shopped Bankruptcies, Forum Shopped Bankruptcies LP and Forum Shopped Bankruptcies no-LP. This table should go to the appendix. Also each test should be accompanied for an average effect test.
- –
  
  Make the sequence of the tables to follow the one in the text.
- –
  
  Combine Tables 1 and 2 to make it in one table. Drop Max and Total. Ideally, create the a Table that has two vertical panels one for all and the other for BR¿0.
- –
  
  Fixed Effects references need to be more change following the next example:
  
  Fixed Effects
  
  County Yes Yes Yes Yes Yes
  
  Year No Yes No No No
  
  Region-Year No No Yes No No
  
  State-Year No No No Yes No
  
  District-Year No No No No Yes
•
Descriptive Section
- –
  
  Map of the BR and FR. Use the either the US tile version in R or use a map by district courts.
- –
  
  Table showing the evolution by industry.
- –
  
  Compute the share of BR and FS that are in NY and DE. Make sure that the descriptive tables in the paper do not include NY and DE.
•
Additional Exercises
- –
  
  Do we have cases where all the BR are FS? If this is the case then perhaps we want to run another placebo. Estimate the impact of FS on the sample where BR==0 — BR==FS
- –
  Local conditions do not predict a BR in our dataset.
  - *
    
    What are the early indicators of recessions in the macro literature? Are any of this at the county level? Make a list of this early indicators such that we discuss how we are controlling for them and how they do not predict a BR in our sample. https://www.bea.gov/data/gdp/gdp-county-metro-and-other-areas change in gdp USE PERSONAL INCOME AS PROXY CITE ECONOMETRICA PAPER The Aggregate Implications of Regional Business Cycle
- –
  
  Estimate the impact using the number of establishments in the legal sector as our dependent variable. http://www.fpeckert.me/cbp/ USE ESTAB DATA
- –
  
  Compute the number of business that close/open in the county outside of the legal sector and include them as additional controls.
- –
  
  Which measure of wage are we using quarter or annual? Estimate the impact using the one that we are not including in our current tables.
- –
  
  Estimate the impact of BR and FR in the employment level of the counties where the district court are located. Count the number of BR and FS for the district courts and estimate the same regressions focusing on the counties that have a district court.
- –
  
  Replace Regions with Divisions in the table (or add divisions into the set of columns)
- –
  
  local effect tests.
•
Robustness Check:
- –
  
  Rewrite the bullets on the robustness section to make it paragraphs that connect each other.
- –
  
  Make sure that the empirical methodology details all our analysis that we include in the robustness section.
•
Data Work:
- –
  
  Scrape data so that CBP forum shopping captures in district shops like the LP database does.
•
Mechanism Section
- –
  
  Fix graphs (replace $\overline{L}_{BR}\text{ with }\overline{L}$ )
- –
  
  tie the writing to the data/results more than the model.

1 Introduction

Local economies and their labor markets depend on the health of their major employers. Much to the detriment of these local economies, when a large employer files for bankruptcy employment falls ([bernstein_bankruptcy_2019]). For instance, when Eastman Kodak Company, et al (henceforth Kodak) filed for Chapter 11 reorganization bankruptcy in 2012, unemployment spiked from 7.5% to 8.3% in Monroe County, New York.¹¹1 QUARTERLY https://fred.stlouisfed.org/series/NYMONR5URN) Similarly, when Trump Plaza Funding and their CEO Donald J. Trump filed for Chapter 11 reorganization bankruptcy in 1992, unemployment spiked from 8.8% to 10.2% in Atlantic County.²²2 (ANNUAL) BLS³³3Annual emp in 1991: 111509 in 1992: 110675 The detrimental effect of bankruptcy shocks on overall local employment hides complex adjustments in the local production network ([carvalho_micro_2014]). In particular, it fails to take into account the differential impact across sectors. We contribute to the growing body of the literature that studies the effect of bankruptcy shocks on local economies ([bernstein_bankruptcy_2019]), by restricting the attention to a particular sector: the legal labor market. In contrast to the rest of the economy, we show that the labor market for the legal industry acts as a potential channel for growth in the bankrupt firm’s local area. We estimate that during the early 90’s the Chapter 11 bankruptcy of a publicly traded firm is associated with a spike in county legal employment worth $4 million in 2020 dollars per year of bankruptcy proceedings.

In order to measure the effect of demand shocks to local legal services we use Chapter 11 bankruptcy reorganizations of publicly traded firms. Large firm’s Chapter 11 bankruptcy cases are complex phenomena, that arbitrate over the conflicting interests of several stakeholders: creditors, shareholders, workers, managers, and so forth. Accordingly, firms will often pay fees in the order of hundreds of millions of dollars⁴⁴4 The average professional fees amount to $125 million in 2020 dollars computed using the UCLA-Lopucki Bankruptcy Research Database, 1980-2014.⁵⁵5 For instance, the professional fees in Kodak bankruptcy case rounded up to $240M in 2013. Source: Reuters, https://www.reuters.com/article/us-kodak-bankruptcy/kodak-bankruptcy-advisers-likely-to-see-240-million-payday-idUSBRE9AD1DV20131114. for legal teams that range from a handful of highly trained consultants and attorneys, to teams of legal clerks and paralegals.⁶⁶6 For the legal industry specifically [freedman_how_2005] estimates that it takes upwards of 1.3 non-lawyer employees for each lawyer in a law firm. As well as indicating that this could increase as a firm grows in size. We contribute to the literature, by building a unique database that expands the universe of large bankruptcies in UCLA-Lopucki Bankruptcy Research Databasewith bankruptcy information of publicly traded firms in Standard and Poor’s Compustat Database, supplemented with court filing records in PACER.⁷⁷7 Web BRD (the “LoPucki”) database can be found at http://lopucki.law.ucla.edu. Professor Lopucki has been at the forefront of research into large chapter 11 bankruptcy cases over the last 30 years. As such, the Lopucki BRD has been the gold standard for researchers studying large corporate bankruptcies and has been used in over 200 publications since it’s creation. This database consists of firm level data with approximately 1000 observations of large public companies that filed for Chapter 11 bankruptcy in the United States between 1979 and 2014. Firms qualify as large public companies only if they file a 10-K form with the SEC that reports assets worth at least $100 million in 1980 dollars (about $300 million in current dollars). Our paper analyzes the impact of these demand shocks on both local legal employment and wages.

We find that Chapter 11 bankruptcy reorganizations of publicly traded firms are associated with a 1% increase in legal employment. Our identification strategy relies on Chapter 11 bankruptcy filing of publicly traded firms to be an exogenous proxy for the local demand of legal services. The large/multi-national nature of the demand faced by publicly-traded firms ease endogeneity concerns, by decoupling the determinants of firms’ bankruptcy choices from the well being of the local economy. To illustrate this point anecdotally, consider the Chapter 11 bankruptcy of the well-known photography company from Rochester, New York: Kodak. Kodak, like the rest of the firms in our sample, was a large multi-national company who did business all over the world. With the worldwide decline in the popularity of film photography, Kodak began to struggle financially and had to file for Chapter 11 bankruptcy in 2012. While it is possible Kodak’s financial well-being may have been subject to high level macroeconomic trends, any economic fluctuations in Rochester itself would have been of little to no consequence for Kodak, whose demand comes from all over the world.

To bolster our identification strategy we exploit stipulations of the US corporate bankruptcy law which create variations in the location of the court where the bankruptcy is filed. This variation arises because large firms are able to exploit US corporate bankruptcy law’s flexible criteria for where a debtor must file. With this flexibility many firms choose to file in a forum far from where they are headquartered in hopes that the judges of that court will be more favorable to debtors. This phenomena of firms filing in a forum different from the local court is referred to as ‘forum shopping’. When a firm forum shops they are filing in a bankruptcy court different from the local court that typically services the area where the firm’s headquarters are located. For large publicly traded corporations who forum shop the average distance from their headquarters to the court they filed in is 749.7 miles. By comparison, this is a drastic increase from an average distance of only 8.7 miles for the firms who did not forum shop.⁸⁸8 Figures calculated using the Lopucki BRD. Lopucki (1992) show that this phenomenon was particularly relevant from 1991 to 1997, a period referred to as the court competition era (see Section 2). So, in our analysis we focus on the court competition era, which allows us to use forum shopping to better understand how the bankruptcy shock affects legal labor markets.

We document that observed changes in employment and wages for the local areas legal sector are contingent on firms filing for bankruptcy in their local court. When firms choose to file in a court that is far from their local area any impact on the legal sector is essentially exported away. Thus we show it is the actual filing of the bankruptcy that is important, and not just some confounding factor of the bankruptcy causing our observed results. By doing so, we contribute to the heated debate over the efficiency of forum shopping and court competition for bankruptcy outcomes ([Zywicki06], [ThomasRasmussen01],[EisenbergLopucki1999], [LopuckiKalin00], [Lopucki06])–by providing first evidence of the unintended consequences on the local economy. Our results suggest that when a firm chooses to forum shop the local area loses approximately $9.6 million worth of employment.⁹⁹9 Using $4 million each year stat, average length of bankruptcy in our sample was 2.4 years according to Lopucki BRD. Using $4 million each year stat, average length of bankruptcy in our sample was 3.2 years according to Lopucki BRD.

Counter-intuitively, we find evidence that Chapter 11 bankruptcy reorganizations of large publicly traded firms are associated with a drop in the county level average wage for the legal sector. We reconcile this puzzling finding with a compositional effect. To do this we specialize the canonical model from [Acemoglu_Autor_2011] to the legal labor markets. In the model economy, a representative firm produces legal services using two inputs of production that act as imperfect substitutes: skilled labor (attorneys) and unskilled labor (paralegals, law clerks, etc.). Skilled workers are more productive but their labor supply is more inelastic than unskilled workers. In this environment, bankruptcy shocks boost the demand of legal services, prompting an increase in the firm’s demand for both types of workers. Since skilled workers have a relatively inelastic labor supply, bankruptcy shocks shift the composition of legal workers towards unskilled workers, causing a drop in the average wage.

Finally, we bring the theoretical predictions of the model to the data in order to estimate the structural parameters. In doing so we contribute to the extensive body of the empirical literature (e.g. [bowlus_wages_2017]; [angrist_economic_1995]; [Autor18_lecture]) that aims to estimate the elasticity of substitution between high and low skilled labor for CES production functions. We differ from this existing literature by being the first to provide empirical evidence for the elasticity of substitution across skill levels in the context of the legal industry. With this novel setting we not only find evidence that supports previous estimates that skilled and unskilled labor are imperfect substitutes, but we are also able to find point estimates for this parameter.

2 Forum Shopping and The Court Competition Era

While U.S. Chapter 11 bankruptcy codes do contain provisions to restrict forum shopping, they still provide large firms with a wide range of options when deciding where to file. Currently debtors can file in any bankruptcy court that corresponds to the location of their (1) “principal place of business”; (2) “principal assets”; (3) “domicile”; (4) affiliates pending bankruptcy case.¹⁰¹⁰10 see 28 U.S.C. §1408 for Chapter 11 bankruptcy venue statutes. https://uscode.house.gov/view.xhtml?req=granuleid:USC-prelim-title28-section1408&num=0&edition=prelim As a result large firms can easily circumvent these restrictions with workarounds such as opening a small office near the court they wish to file in shortly before filing for bankruptcy.¹¹¹¹11 [lopucki_venue_1991] lists several examples of companies “venue hook” by quickly transferring headquarters to a small office near their preferred court shortly before filing.

Although the locations of the firms who choose to forum shop is distributed across the country, the courts these firms choose to file in is highly concentrated. The vast majority of firms who choose to forum shop do so in either the District of Delaware or the Southern District of New York. Prior to the 1990’s, New York was by far the favored district for forum shopping. This, however, took an abrupt turn in the early 1990’s as Delaware suddenly became the favorite district for shopping firms to file in.¹²¹²12 see [EisenbergLopucki1999] (p. 968) “Two jurisdictions have attracted most of the forum shoppers…”¹³¹³13[EisenbergLopucki1999] (p. 983-984) attribute this shift being the result of 2 key events in 1988. (1) A change in the New York court’s random assignment system which made it less likely for debtors to get their preferred judge; (2) a ruling from Judge Helen S. Balick that allowed corporations to use their place of incorporation as their “residence or domicile”. Delaware’s meteoric rise in popularity as a forum shopping destination marked the beginning of the ‘Court Competition Era’. This period saw intense competition between these two courts ([LopuckiKalin00], [Lopucki06]) driven by judges who wished to attract more of the prestigious and lucrative large Chapter 11 filings.¹⁴¹⁴14 [Cole03] (“[a]lmost all of the judges suggested that there is a level of prestige and satisfaction that attaches to hearing and deciding important cases….’Big Chapter 11 cases are interesting as well as prestigious.”) Consequently firms filing for bankruptcy often chose to ‘shop around’ for a venue where they believed the judges would be more likely to rule in the their (debtor’s) favor.

This trend of forum shopping and court competition continued until 1997 with Delaware emerging as a clear favorite for large Chapter 11 cases being shopped.¹⁵¹⁵15 see [EisenbergLopucki1999] Figure 2 In 1997, controversy over this trend caused the National Bankruptcy Review Commission to propose a reform that would have essentially eliminated debtors from filing in Delaware.¹⁶¹⁶16 see [skeel_1998] p. 1-2 To avoid this impending legislation the Chief Judge of the Delaware District Court made an order to reduce the number of filings from forum shopping debtors.¹⁷¹⁷17 see [skeel_1998] p. 33-35 (Also offers other explanations but stresses the importance of the proposal for the Judge’s order), [LopuckiKalin00] p. 234¹⁸¹⁸18Delaware Senator Joseph Biden controlled bankruptcy legislation in the Senate and is speculated to have not put the law to a vote in order to allow Delaware to deal with the problem on their own terms and allow forum shopping to continue in a reduced form. See [LopuckiKalin00] p.234 So while Delaware remains a favorite for forum shopping even today, the reduction in court competition among judges has made the motives and outcomes of forum shopping less clear.

3 Empirical Methodology

This section describes our empirical approach to estimate the impact of Chapter 11 bankruptcies on local legal markets (henceforth bankruptcy refers to only Chapter 11 cases). We start by describing the construction of our bankruptcy database. Then we present our empirical methodology and we discuss how we deal with our main identifications issues.

3.1 Data

We use firm level data and county attribute data to construct a unique database consisting of county year panel data. We aggregate firm attributes to the county year level and then merge with local economic attribute data. The unique contribution comes from expanding existing firm-level bankruptcy databases.

3.1.1 Firm Level Data - Update

We make a unique contribution by building of Lynn Lopucki’s Bankruptcy Research Database (BRD).¹⁹¹⁹19 Web BRD (the “LoPucki”) database can be found at http://lopucki.law.ucla.edu. The Lopucki BRD has been the gold standard for researchers studying large corporate bankruptcies and has been used in over 200 publications since it’s creation. This database consists of firm level data with approximately 1000 observations of large public companies that filed for Chapter 11 bankruptcy in the United States between 1979 and 2014. Firms qualify as large public companies only if they file a 10-K form with the SEC that reports assets worth at least $100 million in 1980 dollars (about $300 million in current dollars).

By using firm level data from Standard and Poor’s Compustat Database (CS)²⁰²⁰20 Information on the S&P Compustat data can be found at http://sites.bu.edu/qm222projectcourse/files/2014/08/compustat_users_guide-2003.pdf we are able to expand the scope of the BRD to include smaller public companies. The CS database contains panel data on 27,726 publicly traded US firms between the years 1979 and 2014. Using accounting indicators listed in CS, we are able to deduce, with some degree of accuracy, which firms filed for bankruptcy and approximately when these firms filed. After finding the subset of firms that filed for bankruptcy in CS, we then turn to publicly available SEC²¹²¹21 SEC data comes from https://www.sec.gov/edgar/search-and-access filings and Court records.²²²²22 Court records can be found via https://www.bloomberglaw.com and https://pcl.uscourts.gov Using this information we are able to scrape the firm’s filing location for their bankruptcy case to deduce whether the firm forum shopped away from their local area.

3.1.2 Local Economic Attribute Data

To construct our dependent variable on employment and the controls for economic activity we use the County Business Patterns (CBP) data. The CBP data is published by the US Census Bureau for industry level employment for each county. We use the imputed CBP database provided by [CBP]. Our other dependent variable of interest, average wage, is constructed with BLS data.²³²³23 https://www.bls.gov/data/

The controls for economic activity used for our primary regression strategy are constructed using the following. County population data comes from United States Census Bureau’s intercensal data²⁴²⁴24 https://www.census.gov/en.html. State level GDP is obtained via the Bureau of Economic Analysis (BEA) website²⁵²⁵25 https://apps.bea.gov/itable/iTable.cfm?ReqID=70&step=1.

Table 1 shows descriptive statistics for the number of bankruptcies and the number of forum shopped bankruptcies for our main estimation sample (see Section 3.2 for more details on our main estimation sample).²⁶²⁶26See Table A in the appendix for an additional descriptive table that includes some additional observations dropped from the main estimation sample The total number of bankruptcies is approximately 770 and around third of cases are forum shopped. There are 451 county-year observations where a bankruptcy took place. On average for these observations with a bankruptcy, there were 1.70 bankruptcies fillings and 0.51 of the bankruptcies filings were forum shopped.

[]

Table 1: Descriptive Statistics

•

Full Sample: All counties with non-zero legal employment during the period 1991-1996, not including counties from New York and Delaware.
•

County-Year with BR: All counties where at least one bankruptcy took place during the period 1991-1996, not including counties from New York and Delaware.

3.2 Methodology

This section lays out our primary regression strategy. To explore the impact of Chapter 11 bankruptcy filings on the local legal sector, we use a linear regression empirical model as follows:

\ln y_{ct}=\gamma_{1}\times\text{\emph{bankruptcies}}_{ct}+\beta X_{ct}+\eta_{c}+\delta_{s(c)t}+\varepsilon_{ct}

(1)

$\ln y_{ct}$ captures our main outcome of interest: the level of employment in the legal sector in county $c$ in period $t$ . The estimation equation also applies to our second outcome, the average wage in the level sector. $\gamma_{1}$ is our parameter of interest and bankruptcies is our measure of the bankruptcy shock to county $c$ during period $t$ . We expect $\gamma_{1}>0$ based on our model. Our benchmark measure of bankruptcies is the number of firms headquartered in county $c$ that have an active Chapter 11 filing during period $t$ . Note that this measure includes all Chapter 11 bankruptcy filings, including the ones that are forum shopped.

The matrix $X_{ct}$ denotes all the variables that we include as additional controls such as county-year population, county-year employment level in non-legal sectors, and state-year level GDP among other variables. $\eta_{c}$ represents county level fixed effects and $\delta_{st}$ is state-year fixed effects. We use annual frequency data to define the period $t$ . Finally, $\varepsilon_{ct}$ is the usual error term.

The county level fixed effects, $\eta_{c}$ , captures all the county-specific constant factors that affect the level of legal employment. State-year fixed effects, $\delta_{s(c)t}$ , absorbs all the time-varying variables at the state-year level that have an effect on the our outcome of interest. These time-varying variables absorbed include measures of the business cycle at the state level.

We exploit corporate bankruptcies as plausibly exogenous shocks to the firm’s local market for labor in the legal sector. In this case, exogeneity comes from the fact that the probability of a large multinational corporation filing for bankruptcy is largely independent to fluctuations in their local economy at the county level. While the probability of a company filing for Chapter 11 bankruptcy may be dependent on constant county attributes, population, or large-scale economic trends, we are able to control for these via fixed effects and additional time-varying controls.

Next, we exploit the stipulation of the law to analyze the impact of forum shopping on local legal employment by estimating the following specification:

\ln y_{ct}=\gamma_{1}\times\text{\emph{bankrupcties}}_{ct}+\gamma_{2}\times\text{\emph{forum shopping}}_{ct}+\beta X_{ct}+\eta_{c}+\delta_{s(c)t}+\varepsilon_{ct}

(2)

where $\gamma_{2}$ captures the role of forum shopping on our main outcome. The term $\emph{forum shopping}_{ct}$ measures the number of bankruptcies that were filed in a different bankruptcy court from the one that typically handles cases for debtors from county $c$ . We measure forum shopping in the same manner as the Lopucki BRD defines it. In the U.S., federal courts have exclusive jurisdiction over bankruptcy cases, thus when a firm files for bankruptcy they must file in the courts for one of the 94 U.S. Judicial Districts. A firm is counted as forum shopping if the court they file in is either not located in the district they are headquartered in, or (for multi-court districts) is not the court their district has designated to serve the county they are headquartered in.²⁷²⁷27Note that by definition, $\text{\emph{forum shopping}}_{ct}\leq\text{\emph{bankruptcies}}_{ct}$ . In the Appendix section LABEL:App_Var_Def, we provide more details on our variable definitions.

This specification has a twofold objective. First, to better measure the overall impact of bankruptcy fillings on local legal employment by allowing a differential impact of bankruptcies in local legal employment for the regular and the forum shopped filings. Second, to support the interpretation that $\gamma_{1}$ measures the impact of bankruptcies on local legal employment. If the local companies going bankrupt is merely acting as a proxy for another determinant of demand for legal services (e.g. worker lawsuits, a spike in divorces, etc.) then the location of the filing should not affect our estimated results. However, if the estimated effect of forum shopping is negative and ‘cancels out’ the effect of the bankruptcy, then it seems plausible that our observed effect is in fact a result of the bankruptcy filing itself.

Following the forum shopping literature (most notably [LopuckiKalin00]), we define the era of court competition as the period from 1991 through 1996 inclusive (see more details in section 2). We set the start of our sample period in 1991, because that is when Delaware had fully emerged as a forum shopping destination, creating a duopoly competition between themselves and the Southern District of New York. The sample ends in 1996 because in early 1997 Delaware and the Southern District of New York toned down their rampant competition in order to avoid impending legislation. Most notably, Delaware took measures to reduce their standing as a popular destination for shopping debtors.

The sample we use for our main set of results is panel data for all U.S. counties over the years 1991 through 1996 inclusive. We include all U.S. counties with available data, but we omit New York (NY) and Delaware (DE) from our main set of results. We omit New York and Delaware counties because of the relationship between bankruptcies, forum shopping and local legal employment is different for these states since they were the most popular forum shopping destinations (see Section 2 for more details).²⁸²⁸28In the robustness section, we explore the sensitivity of our results to include the counties from New York and Delaware.

Additionally, we perform several analyses to support our main estimation results. More specifically, we run placebo tests, we use different measure for the bankruptcies and forum shopping, we modify our sample to include different set of counties and different years, we replicate our results with alternative data sources, and we also include different sets of controls. We discuss all these exercises in our result section.

4 Results

This section provides our main results of primary interest to our work. First, we present our results on the impact of bankruptcies on the level of legal employment. Then we discuss the findings of exploiting the forum shopping era and the robustness checks that we performed to support our main results. Finally, we show the impact for the average wage in the legal sector.

4.1 Legal Employment

Table 2 reports the results of our benchmark estimations following specification (1) where the level of legal employment at the county is our outcome variable and the number of bankruptcies is our main independent variable. Our main estimation sample correspond to all U.S. counties with non-zero legal employment but omitting Delaware and New York for the period 1991-1996 as we discuss on our methodology section (see Section 3.2 for more details). We clustered our standard errors at the county level for all the results that follows.

The results in Table 2 indicate that a bankruptcy filing from a firm headquartered in the county is associated with a significant increase in the level of legal employment. This impact is robust to different fixed effects specifications where we impose more demanding fixed effects along the geographical and time dimension.²⁹²⁹29We follow the Census Bureau definition of region and use 4 categories: Northeast, Midwest, South and West. In the case of the column (5) we introduce judicial district-year fixed effects. There are 94 judicial districts in the US. Hence this is akin to controlling for time-varying within state factors. Also, the additional controls included in the specifications, county population and employment level outside the legal sector, have the expected sign and significance.

Our preferred specification controls for county and state-year fixed effects (see Column (4) in Table 4) and absorb all determinants of the legal employment level that are constant at the county level and absorb all the time-varying determinants at the state level, including state GDP, state unemployment rate, as well as any other state level measure of the business condition. In this specification, the effect is around an increase of 0.72% per active bankruptcy filing in the county (see column (4) from Table 4). However, note that the impact is very stable across the different specifications varying between 0.072% and 0.085%. Hence, for the counties that have at least one bankruptcy filing, bankruptcies represent an increase of 1.69% of the local legal employment in the county. This translates to an increase in the wage bill of around 3.5 millions per year based on the average wage and the average number of workers in the legal sector for these counties.

Table 2: Employment Level Regressions

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies; (ii) Fixed Effects: county; county and year; county and region-year, where region is one of 4 Census Bureau-designated regions; county and division-year, where division is one of 9 Census Bureau-designated divisions; county and state-year; county and district-year, where district is one of 94 judicial districts; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

4.1.1 Forum Shopping

In this subsection, we exploit the stipulation of the law known as Forum Shopping to support our interpretation of the results and we estimate the specification (2). Filing in a different court than the one were the firms has it headquarter was particularly easy to achieve during the period 1991-1996 (see Section 2 for more details). This additional flexibility implied that the demand shock to the legal sector in the county where the firm was headquartered did not took place for certain bankruptcies filings. To get a sense of this, a standard bankruptcy filing for a large publicly traded firm takes place 8.7 miles from the headquarter of the firm on average but a forum shopped bankruptcy takes place 749.7 miles away.

We exploit this stipulation of the law in two way: first, as a type of placebo exercises and, second, to better measure the role of bankruptcies in the local legal employment level. Starting on the later, once we allow for heterogeneous effect for bankruptcy filings in the county and bankruptcy filings that are forum shopped, we find that the positive and significant effect on legal sector slightly increase to almost reach a 1% impact per bankruptcy filing (see column (2) from Table 4). We confirm this finding also when we further disaggregate our forum shopping measurement by distinguishing by number of cases by the size of the company that is forum shopping.

Interestingly, when we exploit the stipulation of the law and focus on the bankruptcies that are forum shopped we find that the effect of these forum shopped bankruptcies is null for legal employment. More specifically, the estimated impact on local legal employment for a forum shopped bankruptcy is -0.08% and it is not significant (see column (2) for the second panel in Table 4). Hence, forum shopping exported away the gains from bankruptcies filing in the county. The fact that bankruptcies only have a null effect on legal employment when they are filed in the county where the firm is headquartered support our interpretation of the relationship between local legal employment and the bankruptcies filings.

Exploring the heterogeneity of the impact of forum shopping depending of the size of the filing, Column (3) in Table 4 highlights that the null effect associated with forum shopping is mostly associated with the big bankruptcies cases. If the firm that forum shopped its bankruptcy has assets worth less than $340 million in 2021 dollars ($100 million in 1980 dollars), the bankruptcy case has a positive impact of 0.75% on legal employment. An effect very close to our initial estimate as reported by column (1) in Table 4. Finally, Table LABEL:tab:avg_values_BR shows that on average 0.50 of the 1.70 bankruptcies cases are forum shopped during our sample. This implies that the average impact of bankruptcies after including the role of forum shopping is around 0.1% according to our results on Column (2) in Table 4.³⁰³⁰30Our results show that this impact is significant with a p-value of 0.0512 This implies that forum shopping reduces the impact by 0.07% on legal employment.³¹³¹31We also estimate alternative version of Table 4 where we use different set of fixed effects following the approach of Table 2. The tables are reported in Appendix D.10 and in all cases they confirm our results.

Table 3: Employment Level Regressions

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Table 4: Employment Level Regressions

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

4.1.2 Placebo Tests

Table 5: Employment Level Regressions with lags for bankruptcy shock.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies lagged one, two, or three years; (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

We perform several placebo test to evaluate whether there were differences in the local legal employment level in the counties affected by bankruptcies before the filings took place. In order to do this, we lagged the variable that captures the number of bankruptcies. More specifically, we use three alternative lag structures: one, two and three years before the bankruptcies actually took place. Additionally, we combined all the three lagged measures in the same specification.

Results in Table 5 show that none of the bankruptcies measures have a significant effect and the estimated coefficients shrink by an order of magnitude. Hence, there are no significant differences in the level of local legal employment for the counties affected by bankruptcies before the bankruptcies actually take place. These results point in the same direction that our findings exploiting the forum shopping period and reassures our finding that bankruptcies filings are associated with an positive impact on the local legal employment level.³²³²32We also run some additional placebo tests where we use binary indicators as measures for bankruptcies. The results in Table 11 in the appendix also confirms our findings.

4.1.3 Robustness Checks

In this subsection, we perform several robustness exercises in order to ease the main concern regarding our results.

•

Include New York and Delaware counties: We drop counties from New York and Delaware due to the fact that they benefit from the forum shopped cases in our main estimation sample. In order to evaluate if this is affecting our results, we reestimate our main results using an augmented version of our sample where New York and Delaware counties are included. Results are reported in Table 12. Results show that a bankruptcy filing in the county increases the local legal employment level by almost a 0.1% and that forum shopped bankruptcies has null impact on the employment level. We also consider whether bankruptcies and forum shopped filings have a different effect for New York and Delaware. We find that bankruptcies filings and forum shopped bankruptcies have no significant impact on local legal employment for counties in New York and Delaware.
•

Counties with Headquarters: One potential issue with our estimating sample is that by including all counties where legal employment is non-zero we are including several counties that could never experience a bankruptcy filing of big corporation because there are none of this type of firm located in the county. To ease this concern, we replicate our baseline results where the estimation sample is restricted to all counties that have a publicly traded company located in the county at any point between 1980 and 1997. Imposing this restriction significantly reduces our estimation sample to a slightly more than one third of our main one. Notwithstanding this, the results in Table 15 confirm all our main findings.
•

Extending the Period: In order to explore whether our results hold when we consider a longer period, we rerun our main benchmark Table for the period 1991-2001. The results in Table 11 are similar to our main finding that bankruptcy filings are associated with an increase in the local legal employment level.
•

Controls: in our benchmark estimations we include the log of population at the county and the log of the number of employee outside of the legal sector in the county. We also run regressions were these controls are lagged one period in order to avoid any impact of bankruptcies in the control variables. The results of this estimation are presented in 16 and they confirm our findings. Note that this approach also avoids potential indirect effect of bankruptcies on the level of employment outside of the legal sector. Additionally, we incorporate an additional time-varying control, the unemployment rate at the county-year level. Again, our results remain the same when we include this additional control (see Table 17).
•

Business Conditions: A possible concern with our main results is that they might be very sensitive to the overall macroeconomics conditions. First, note that our state-year fixed effects play an important role easing this concern. Second, the years in our sample, 1991 through 1996, can be considered largely free of recessions on an aggregate level. To further check this issues, we estimate our main results dropping the year 1991 due to the fact that the early 1990’s recession ended in 1991.³³³³33See [hamilton_2020] for more information. Results are reported in Table 20 and show that our main results are robust to dropping this year. Since the unemployment rate at the county could also be considered correlated with the business condition, our results presented in Table 17 that include the unemployment rate in the county show that our main results are robust to using different measures of the business conditions in the county.
•

Alternative Bankruptcy Measures: In the Table 22 we show the results for our main estimation benchmark of measuring bankruptcies and forum shopped bankruptcies using binary variables instead of count one. We find that all our results hold, albeit with a lower significant level. Note that this could be expected due to the fact that our measurement for bankruptcies and forum shopping is less precise compared to our benchmark estimation strategy.
•

Alternative Data Source: We evaluate whether our results are robust to constructing our employment variables an alternative data. More specifically, we use Bureau of Labor Statistics (BLS) data rather than CBP data. We expect the CBP database to be more complete and less prone to errors. This is evidenced by the fact that the number of observations drops by nearly 6,000 when using BLS data to construct measures of employment due to missing data. Even with this less complete database we see that bankruptcies still have a significant positive impact on employment and the combined effect of bankruptcies and forum shopping is not significantly different from zero. These results are reported in Table 23.
•

Adjacent County Bankruptcies: We include these regressions in order to quell potential concerns that spillover effects from nearby counties could be affecting our results. To do this we create a measure of the number of bankruptcies filings in directly adjacent counties to county $c$ in year $t$ . Table 24 show that bankruptcies in adjacent counties have no impact on the local legal employment in the county.

4.2 Wage in the Legal Sector

Table 6 reports our results for average wage in the legal sector. Columns (1), (2), and (3) fail to show significance for the estimated effect of a bankruptcy on average wage. This result is consistent with the idea that wages tend to be ”stickier” than employment levels in the short run and require larger shock to record a meaningful change. Column (4) corroborates this hypothesis, by showing that large firms bankruptcy shocks n_BR_LP indeed have a statistically and economically significant effect.

Yet, the direction of the effect is puzzling. Shouldn’t positive demand shocks have a positive effect on wages? The next section specializes a model with skilled and unskilled workers and reconciles this puzzling observation with a compositional effect. Bankruptcy shocks increase the demand in the county of both high paid attorneys and low paid paralegal and legal clerks. However, the increase in demand is relatively larger for the lower skilled workers, who tend to have a more elastic labor supply. As a result, the average wage goes down.

Finally, at the bottom of Table 6 the test of the linear combination of the estimates for bankruptcy and forum shopping shows that their combined effect is not significantly different from zero. Thus, as we would expect, it appears that forum shopping is canceling out any shock to the local labor market for legal employment that a firm going into bankruptcy otherwise would have had.

Table 6: Average Annual Pay for Legal Sector.

Note: OLS regression estimates of log Average Wage for Legal Sector on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR, BLS.

5 A Model of the Legal Labor Market

This section specializes the canonical model with skilled and unskilled workers (e.g. [acemoglu_technical_2002]; [Acemoglu_Autor_2011]; [bowlus_wages_2017]) to study the effect of bankruptcy shocks on legal labor markets and reconcile the puzzling effect on wages.

We consider a static partial equilibrium model of the legal sector. The economy is populated by a mass of heterogeneous households and a representative firm that produces legal services. Households differ in their labor productivity: skilled and unskilled. Skilled workers can be thought of as lawyers, who require an extremely large investment in training before they can work. Whereas unskilled workers can be thought of as para-legals, law clerks, and legal sector employees who require considerably less upfront costs in training relative to skilled workers. The representative firm hires skilled and unskilled workers to produce legal services (the numeraire). The demand of legal services is exogenous. Labor markets are competitive.

5.1 Households

The economy is populated by heterogeneous households that differ in the type ( $j$ ) of labor supplied: skilled ( $j=s$ ) and unskilled ( $j=u$ ). Households of type $j$ choose consumption $c\in\mathbb{R}_{+}$ and labor $j\in\{s,u\}\in\mathbb{R}_{+}$ in order to maximize Greenwood–Hercowitz–Huffman preferences.³⁴³⁴34 As such, the marginal rate of substitution between labor and consumption will only depend on labor. See [GHH_Preferences_1988] (p. 404).

\underset{c_{j},j}{\max}\frac{1}{1-\sigma}\left(c_{j}-\gamma\frac{j^{1+\rho_{j}}}{1+\rho_{j}}\right)^{1-\sigma}

(3)

In this setting, $\rho_{j}>0$ determines type $j$ ’s Frisch elasticity of labor supply, and the terms $\sigma>0$ and $\gamma>0$ are shaping parameters.

By taking FOC, we obtain the labor supply functions:

s(w_{s})=\left(\frac{w_{s}}{\gamma}\right)^{\frac{1}{\rho_{s}}}

u(w_{u})=\left(\frac{w_{u}}{\gamma}\right)^{\frac{1}{\rho_{u}}}

It follows that the Frisch Elasticity of Labor Supply ( $\epsilon_{j}$ ) for $j=\{s,u\}$ can be expressed as:

\epsilon_{j}=\frac{1}{\rho_{j}}

(4)

Grounded on the vast empirical evidence on the relative elasticity of labor supply of skilled versus unskilled workers () we make the following assumption.

Assumption 1.

The elasticity of labor supply of skilled workers is lower than unskilled workers, i.e. $\rho_{u}<\rho_{s}$

5.2 Firms

A representative firm hires skilled $(s)$ and unskilled $(u)$ workers to produce legal services

L(s,u)=\left[(A_{u}u)^{\frac{\delta-1}{\delta}}+(A_{s}s)^{\frac{\delta-1}{\delta}}\right]^{\frac{\delta}{\delta-1}}

(5)

where $\delta\in[0,\infty)$ is the elasticity of substitution between the input factors. In accordance with the broad consensus of estimates for $\delta$ in the literature, we consider the case where $\delta>1$ .³⁵³⁵35 [bowlus_wages_2017] provides a summary of the literature on this topic, where they refer to even the low end estimates being in the range of 1.4 to 1.8^,³⁶³⁶36 [Autor18_lecture] describes the consensus estimate to be approximately 2 in the U.S. while the most commonly used estimate is 1.4.^,³⁷³⁷37 [angrist_economic_1995] from the AER estimates implied elasticity of substitution between 16+ years of schooling and less than 12 to be approximately 2. This corresponds to skilled and unskilled labor being imperfect substitutes. $A_{u}$ and $A_{s}$ are the factor-augmenting technology terms for unskilled and skilled labor, respectively.

Assumption 2.

Skilled labor is more productive than unskilled labor, $A_{u}<A_{s}$ .³⁸³⁸38 While it is reasonable to assume a lawyer could do the work of a paralegal/law clerk we recognize that there must be some minimum threshold of attorneys in order for the firm to practice law. For simplicity, this requirement has been omitted, however it is without loss of generality to assume firms are operating in the range where $s>0$ .

$L(s,u)$ is the production of all legal services for a local area. When a firm files for Chapter 11 bankruptcy the demand for legal services increases, which results in an increase in the equilibrium level of $L$ in the market for legal services. These bankruptcy shocks are used as instruments for variation in $L$ , and as such appear in the model as an increase in $L$ .

Since markets are competitive, factors receive their marginal product as wages

w_{s}=A_{s}^{\frac{\delta-1}{\delta}}\left[A_{u}^{\frac{\delta-1}{\delta}}\left(\frac{s}{u}\right)^{-\frac{\delta-1}{\delta}}+A_{s}^{\frac{\delta-1}{\delta}}\right]^{\frac{1}{\delta-1}}=A_{s}^{\frac{\delta-1}{\delta}}\left(\frac{L}{s}\right)^{1/\delta}

(6)

w_{u}=A_{u}^{\frac{\delta-1}{\delta}}\left[A_{u}^{\frac{\delta-1}{\delta}}+A_{s}^{\frac{\delta-1}{\delta}}\left(\frac{s}{u}\right)^{\frac{\delta-1}{\delta}}\right]^{\frac{1}{\delta-1}}=A_{u}^{\frac{\delta-1}{\delta}}\left(\frac{L}{u}\right)^{1/\delta}

(7)

From (6) and (7) we obtain the following relationship for average wage in terms of relative factor intensity ( $s/u$ ).

\overline{w}\equiv\frac{uw_{u}+sw_{s}}{u+s}=\frac{\left[A_{u}^{\frac{\delta-1}{\delta}}+A_{s}^{\frac{\delta-1}{\delta}}\left(s/u\right)^{\frac{\delta-1}{\delta}}\right]^{\frac{\delta}{\delta-1}}}{1+s/u}

(8)

Lemma 1.

If the skill premium is positive ( $\frac{w_{s}}{w_{u}}>1$ ), $\implies\overline{w}$ is increasing in $\frac{s}{u}$
Where the skill premium is defined as:

\pi\equiv\frac{w_{s}}{w_{u}}=\left(\frac{A_{s}}{A_{u}}\right)^{\frac{\delta-1}{\delta}}\left(\frac{s}{u}\right)^{-1/\delta}

(SP)

5.3 Equilibrium

Lemma 2 uses market clearing in the input markets of skilled and unskilled workers to derive equilibrium wages and employment for each type of worker

Lemma 2.

Labor Market Equilibrium for the Legal Sector:

s^{*}(L)=[\gamma^{-\delta}A_{s}^{\delta-1}L]^{\frac{1}{1+\mathbf{\delta\rho_{s}}}}

(9)

u^{*}(L)=[\gamma^{-\delta}A_{u}^{\delta-1}L]^{\frac{1}{1+\mathbf{\delta\rho_{u}}}}

(10)

w^{*}_{s}(L)=\gamma\left(A_{s}^{\delta-1}L^{\rho_{s}}\right)^{\frac{1}{1+\delta\rho_{s}}}

(11)

w^{*}_{u}(L)=\gamma\left(A_{u}^{\delta-1}L^{\rho_{u}}\right)^{\frac{1}{1+\delta\rho_{u}}}

(12)

Equations (9) and (10) show that in equilibrium both $s$ and $u$ are increasing in $L$ .

The next lemma shows that the frisch elasticity of labor supply $\rho_{j}$ plays a crucial role in determining the responsiveness of the equilibrium levels of $s^{*}$ and $u^{*}$ to a change in $L$ due to a bankruptcy shock. In particular, by taking the ratio of 9 and 10, we formalize the compositional effect of exogenous increases in the demand of legal services.

Proposition 1.

Compositional Channel. Given Assumptions 1 and 2, an increase in the demand of legal services $L$ tilts the composition of workers towards unskilled workers

\frac{\partial\left[\frac{s*}{u*}\right]}{\partial L}<0

As expected, bankruptcy shocks ( $L>0$ ) increase wages for both skilled and unskilled labor (Eqs. (11) and (12)). Yet, the effect on average wage is less clear. Proposition 2 uses Lemmas 1, 2, and 1 to derive the relationship between $L$ and average wage.

Proposition 2.

Given $\rho_{u}<\rho_{s}$ and $\frac{w_{s}}{w_{u}}>1$ , an increase in $L$ will result in the following:

i) Equilibrium levels of $s^{*},u^{*},w^{*}_{s},w^{*}_{u}$ increase.

ii) $\frac{s*}{u*}$ decreases $\implies\overline{w}=\frac{\left[A_{u}^{\frac{\delta-1}{\delta}}+A_{s}^{\frac{\delta-1}{\delta}}\left(s/u\right)^{\frac{\delta-1}{\delta}}\right]^{\frac{\delta}{\delta-1}}}{\left(1+s/u\right)}$ decreases.

5.4 Forum Shopping

This section incorporates forum shopping to the model through the demand for legal services $L$ . We define $BR$ to be a measure of the bankruptcies, such as the number of local firms currently undergoing bankruptcy proceedings. Similarly let $FS$ be a measure of how much of the bankruptcy shock is being forum shopped away. When there is no shock, $BR=FS=0$ . If there is a shock, then $BR\in\mathbb{R}^{++}$ and $FS\in[0,BR]$ . The proportion of the bankruptcies that forum shops is the ratio $f\equiv\frac{FS}{BR}\in[0,1]$ .

The magnitude of the bankruptcy shock is $h(BR,f)\in\mathbb{R}^{+}$ . Which takes the following functional form using $\lambda>0$ as a scaling parameter.

\begin{split}h(BR,f)&=\lambda(1-f)BR\\ &=\lambda(BR-FS)\end{split}

(13)

Next, we split $L$ into two parts, $L_{BR}$ the quantity of legal services for bankruptcy, and $L_{other}$ the quantity of all other legal services. The quantity of bankruptcy legal services $L_{BR}=\overline{L}_{BR}+h(BR,f)$ . Where $\overline{L}_{BR}$ is the steady state level of legal services in the bankruptcy sector.

\begin{split}L&=L_{BR}+L_{other}\\ &=\overline{L}_{BR}+h(BR,f)+L_{other}\end{split}

(14)

Thus a bankruptcy has the following effect on the quantity demanded of legal services:

\begin{split}\frac{\partial L}{\partial BR}=\left(\frac{\partial L}{\partial h(BR,f)}\right)\left(\frac{\partial h(BR,f)}{\partial BR}\right)=\lambda(1-f)\geq 0\end{split}

(15)

Similarly, we obtain the effect of a bankruptcy on employment in the legal sector.

\begin{split}\frac{\partial s}{\partial BR}&=\left(\frac{\partial s}{\partial L}\right)\left(\frac{\partial L}{\partial BR}\right)=\left(\frac{\partial s}{\partial L}\right)\lambda(1-f)\end{split}

(16)

\begin{split}\frac{\partial u}{\partial BR}&=\left(\frac{\partial u}{\partial L}\right)\left(\frac{\partial L}{\partial BR}\right)=\left(\frac{\partial s}{\partial L}\right)\lambda(1-f)\end{split}

(17)

Clearly both of these are decreasing in $f$ , which gives us the following relationship between forum shopping and the shock:

Proposition 3.

Skilled and unskilled labor both increase with bankruptcies;

\frac{\partial s}{\partial BR}\geq 0\qquad\frac{\partial u}{\partial BR}\geq 0

(18)

However, this effect is decreasing with respect to forum shopping;

\frac{\left[\frac{\partial s}{\partial BR}\right]}{\partial FS}<0\qquad\frac{\left[\frac{\partial u}{\partial BR}\right]}{\partial FS}<0

(19)

Proposition 3 is illustrated by figures 1(c) and 1(d) in the following section.

5.5 Mechanism

Refer to caption — (a) Isoquant $L(s,u)=\overline{L}_{BR}$

This section discusses the economic intuition behind the propositions derived in the model section. Consider the isoquant for the production of legal services $L(s,u)=\overline{L}_{BR}$ in a local area, shown in Figure 1(a). Here, $\overline{L}_{BR}$ refers to the steady state level of demand for bankruptcy legal services. Given the level of demand $\overline{L}_{BR}$ the firm then chooses their optimal input bundle, $A=(s_{1},u_{1})$ , satisfying the tangency condition for cost minimization $MRTS=MP_{s}/MP_{u}=w_{s}/w_{u}$ . In terms of our model, this tangency condition characterizes the skill premium $(\pi)$ from Lemma 1.

\pi\equiv\frac{w_{s}}{w_{u}}=\left(\frac{A_{s}}{A_{u}}\right)^{\frac{\delta-1}{\delta}}\left(\frac{s}{u}\right)^{-1/\delta}

(SP)

This relationship is shown in Figure 1(b). Using (SP) the slope of the dashed line corresponds to the left side of the equality and the slope of the isoquant (MRTS) is given by the right side. Thus these two expressions are only equal in optimum, i.e. when the tangency condition is satisfied.

When one or more firms file for bankruptcy locally we see an increase in demand for legal services. Given that all the firms are filing locally the quantity of legal services produced increases by the full amount of the shock, $h(BR,0)=\lambda BR$ . This is shown in Figure 1(c) by the outward shift in the isoquant, changing the firms optimal choice of inputs from point $A=(s_{1},u_{1})$ to point $B=(s_{2},u_{2})$ . Proposition 2 characterizes the relationship between these two bundles. Proposition 2 tells us that while the amount of both types of labor increases, the relative factor intensity of skilled labor decreases, $\mathbf{\frac{s_{2}}{u_{2}}<\frac{s_{1}}{u_{1}}}$ .

However, the magnitude of this shock is dampened if some, or all, of the bankrupt firms forum shop to a venue outside their local area. Accordingly, the increase in production due to the shock, $h(BR,f)=\lambda(1-f)BR$ , depends on the proportion of the firms who chose to forum shop, $f$ . This can be seen graphically in Figure 1(d), where $C=(s_{3},u_{3})$ is the optimal bundle when forum shopping occurs. In this setting, using Proposition 2 and Proposition 3, it is clear that the relationship between the optimal bundles satisfies $s_{2}>s_{3}\geq s_{1}$ , and $u_{2}>u_{3}\geq u_{1}$ . However, due to the inelastic nature of skilled labor $\mathbf{\frac{s_{2}}{u_{2}}<\frac{s_{3}}{u_{3}}\leq\frac{s_{1}}{u_{1}}}$ . Where the weak inequalities approach equality as $f$ approaches one.

Figure 1(d) reconciles the joint dynamics on the legal sector employment and wages observed in Table 2 and Table 6, respectively.

6 Structurally Estimating Model Parameters (Tentative)

The following section looks into estimating the elasticity of substitution between skilled and unskilled labor ( $\delta$ ) proposed by the CES production function from our model. Using the model’s characterization of equilibrium in the market for unskilled labor, (Lemma 2) we develop a strategy for estimating $\delta$ .³⁹³⁹39This can equivalently be done using the equilibrium for skilled labor as well. In particular, by taking logs of Equations 10 and 12,

u^{*}(L)=[\gamma^{-\delta}A_{u}^{\delta-1}L]^{\frac{1}{1+\mathbf{\delta\rho_{u}}}}\qquad w^{*}_{u}(L)=\gamma\left(A_{u}^{\delta-1}L^{\rho_{u}}\right)^{\frac{1}{1+\delta\rho_{u}}}

we obtain the following,

\ln(u)=\alpha_{u}+\beta_{u}\ln(L)

(20)

\ln(w_{u})=\alpha^{w}_{u}+\beta^{w}_{u}\ln(L)

(21)

where $\beta_{u}=\frac{1}{1+\mathbf{\delta\rho_{u}}}$ and $\beta^{w}_{u}=\frac{\rho_{u}}{1+\mathbf{\delta\rho_{u}}}$ . To obtain an expression for $\delta$ Equations 21 and 20 can then be rearranged algebraically and simplified as follows:

\delta=\frac{1/\beta_{u}-1}{\frac{\beta^{w}_{u}}{\beta_{u}}}

(22)

To estimate Equations 21 and 20 we use BLS data for unskilled workers in the legal industry over the period from 1991 to 2001 for the dependent variables $u$ and $w_{u}$ .⁴⁰⁴⁰40 We use NAICS code 54119 for a measure of unskilled labor in the legal sector.⁴¹⁴¹41Equivalently we are able to follow the above steps using skilled labor, which ultimately gives an estimate in line with the one presented here using unskilled labor. However, the Poisson Maximum Likelihood regression failed to converge so we have elected not include those results until we can verify their validity. Then, using bankruptcy shocks to instrument out the demand of legal services ( $L$ ), we run a Poisson regression with county and state-year fixed effects to obtain estimates for $\beta_{u}$ , and $\beta^{w}_{u}$ . Which then allows us to construct an estimate of $\delta$ as in Equation 22. In doing so we are able to provide first empirical estimate for the elasticity of substitution between skilled and unskilled labor for the production of legal services. We find this estimate to be in line with the body of literature that estimates $\delta>1$ for skilled and unskilled labor across broader sections of the economy (e.g. [bowlus_wages_2017], [angrist_economic_1995]). Table 7 reports the estimates for $\beta_{u}$ , $\beta^{w}_{u}$ , and $\delta$ .

Table 7: Structural Estimation of CES parameter

\delta

	(1)	(2)	(3)
	$\beta_{u}$	$\beta^{w}_{u}$	$\delta$
	0.0431^∗∗∗	0.0877^∗∗∗	21.185^∗∗
	(0.000709)	(0.00731)	(9.229)
Standard errors in parentheses
^∗ $p<0.1$ , ^∗∗ $p<0.05$ , ^∗∗∗ $p<0.01$

Note: Columns (1) and (2) are Poisson regression estimates of log of employment level for unskilled labor in the legal industry and the log of average wage for unskilled workers in the legal industry on: (i) Treatment: number of bankruptcies; (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All county-year observations that had atleast 1 bankruptcy, omitting Delaware and New York. Period: 1991-2001. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR, BLS.

7 Conclusion

This paper provides first evidence of the effect of Chapter 11 bankruptcy reorganizations of publicly traded firms on local legal labor markets. To do so, we build a unique database that expands the universe of large bankruptcies in UCLA-Lopucki Bankruptcy Research Database with bankruptcy information of publicly traded firms in Standard and Poor’s Compustat Database, supplemented with court filing records in PACER.

We find that Chapter 11 bankruptcy reorganizations of publicly traded firms are associated with a 1% increase in legal employment. To corroborate our identification, we use stipulations of the US corporate bankruptcy law which allow firms to file in a court located far away from their headquarter. To exploit this stipulation of the law–often referred to as forum shopping–we restrict our sample to a period where forum shopping was prevalent: the Court-Competition Era Period (1991-1997). In doing so, we contribute to the heated debate on the economic efficiency of court competition for bankruptcy outcomes by shifting the attention from the bankruptcy per se to the unintended consequences of forum shopping for local legal labor markets. Our result suggests that when a distress firm forum shops the local area loses approximately $9.6 million worth of employment.

In conclusion, we find that Chapter 11 bankruptcy reorganizations of publicly traded firms are associated with a 1.5% drop in the average county wage in the legal sector. We reconcile this counter-intuitive finding with a compositional effect, by specializing a model with skilled and unskilled workers to the legal labor markets.

\addappheadtotoc\appendixpage

Appendix A Summary Stats

[H] Descriptive Statistics on Bankruptcies and Forum Shopping

•

Sample: All counties with non-zero legal employment during the period 1991-1996, including counties from New York and Delaware.

Table 8: Summary stats for county-year observations with non-zero legal employment from 1991-1996, omitting NY and DE.

Table 9: Summary stats for county-year observations with at least 1 BR in sample 1991-1996, omitting NY and DE.

Appendix B Tests and Average Effects

Table 10: Employment Level Regressions

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies; (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-2001. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Appendix C Graphs

Appendix D Robustness Checks

D.1 Legal Employment and BR from 1991 to 2001

Table 11: Employment Level Regressions

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies; (ii) Fixed Effects: county; county and year; county and region-year, where region is one of 4 Census Bureau-designated regions; county and division-year, where division is one of 9 Census Bureau-designated divisions; county and state-year; county and district-year, where district is one of 94 judicial districts; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-2001. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

D.2 Include New York and Delaware Counties:

Table 12: Legal sector employment level with NY and DE included.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS), as well as interaction terms for whether the observation occurred in NY or DE; (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

D.3 Counties with Firm Headquarters:

Table 13: Employment Level Regressions over sample of only HQ counties

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: subset of counties that had a publicly traded companies HQ at any point between 1980 and 1997, omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

D.4 Counties with Bankruptcy Courts:

Table 14: Employment Level Regressions over sample of only counties that have a bankruptcy court.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies; (ii) Fixed Effects: county; county and year; county and region-year, where region is one of 4 Census Bureau-designated regions; county and division-year, where division is one of 9 Census Bureau-designated divisions; county and state-year; county and district-year, where district is one of 94 judicial districts; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: subset of counties that had a publicly traded companies HQ at any point between 1980 and 1997, omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Table 15: Employment Level Regressions over sample of only counties that have a bankruptcy court.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: subset of counties that had a publicly traded companies HQ at any point between 1980 and 1997, omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

D.5 Controls:

Table 16: Employment Level Regressions with controls lagged by 1 year.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of previous years population for each county-year observation, log of previous years employment in all non-legal sectors for each county-year observation. Standard errors clustered at county level. Sample: subset of counties that where a bankruptcy court is located, omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Table 17: Employment Level Regressions with control for unemployment rate.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors, and county-year unemployment rate. Standard errors clustered at county level. Sample: subset of counties that where a bankruptcy court is located, omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR, BLS.

Table 18: Employment Level Regressions with control for change in number of Establishments outside of legal sector.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors, and county-year change in the log of the number of establishments in non-legal sectors from previous year. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Table 19: Employment Level Regressions with control for change in the log of number of Establishments outside of legal sector.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors, and county-year change in the log of the number of establishments in non-legal sectors from previous year. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

D.6 Business Conditions:

Table 20: Employment level regressions dropping 1991 from sample.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1992-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Table 21: Average wage regressions dropping 1991 from sample.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1992-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR, BLS.

D.7 Alternative Bankruptcy Measures:

Table 22: Employment level regressions replacing bankruptcy and forum shop regressors with their dummy variable counterparts.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: dummy for bankruptcies, dummy indicating whether any forum shopping occurred, dummy for forum shopping split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

D.8 Alternative Data Source:

Table 23: Employment Level Regressions using BLS data

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, U.S. Census Bureau, BEA, SEC EDGAR, BLS.

D.9 Adjacent County Bankruptcies:

Table 24: Employment Level Regressions controlling for bankruptcies in adjacent counties.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS), number of bankruptcies in adjacent counties; (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

D.10 Fixed Effects

Table 25: Legal sector employment level with county fixed effects.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors, log of state year GDP. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Table 26: Legal sector employment level with county and year fixed effects.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors, log of state year GDP. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Table 27: Legal sector employment level with county and region-year fixed effects.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and region-year, where region is one of 4 Census Bureau-designated regions; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors, log of state year GDP. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Table 28: Legal sector employment level with county and division-year fixed effects.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and division-year, where division is one of 9 Census Bureau-designated divisions; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors, log of state year GDP. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Table 29: Legal sector employment level with county and district-year fixed effects.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and district-year, where district is one of 94 judicial districts; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

D.11 Legal Establishment Regressions

Table 30: Regressions for Number of Establishments in the Legal Sector.

Note: OLS regression estimates of log of the annual average number of establishments in the legal sector. on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR, BLS.

Table 31: Regressions for Number of Establishments in the Legal Sector.

Note: OLS regression estimates of log of the annual average number of establishments in the legal sector. on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR, BLS.

Table 32: Regressions for Number of Establishments in the Legal Sector.

Note: OLS regression estimates of log of the annual average number of establishments in the legal sector. on: (i) Treatment: number of bankruptcies, number of forum shopped bankruptcies, number of forum shopped bankruptcies split by database (LP or CS); (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR, BLS.

D.12 Placebo Test

Table 33: Legal Wage Regressions with lags for bankruptcy shock.

Note: OLS regression estimates of log Average Wage for Legal Sector on: (i) Treatment: number of bankruptcies lagged one, two, or three years; (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR, BLS.

Table 34: Employment Level Regressions with lags for dummy indicator of bankruptcy shock.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: bankruptcy dummy lagged one, two, or three years; (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Table 35: Legal Wage Regressions with lags for dummy indicator of bankruptcy shock.

Note: OLS regression estimates of log Average Wage for Legal Sector on: (i) Treatment: bankruptcy dummy lagged one, two, or three years; (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: All U.S. counties omitting Delaware and New York. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR, BLS.

Table 36: Legal Wage Regressions for county-year observations where there were no bankruptcies or all bankruptcies were forum shopped away.

Note: OLS regression estimates of log Legal Employment on: (i) Treatment: ???; (ii) Fixed Effects: county and state-year; (iii) Controls: log of county-year population, log of county-year employment in all non-legal sectors. Standard errors clustered at county level. Sample: ???. Period: 1991-1996. Sources: Lopucki BRD, Compustat, CBP, U.S. Census Bureau, BEA, SEC EDGAR.

Appendix E Database Creation

1.
Clean Compustat American Annual Database.
- •
  
  Remove duplicate entries, remove non USA firms, remove firms in legal sector, and remove years outside of LP range.
- •
  
  Create bankruptcy indicators. Ch11 if ”fresh start accounting” and no deletion, Ch11 if ”bankruptcy” and no deletion, Ch11 if ”bankruptcy” and deletion with merger delete reason in CS.
2.
Collect data for potential bankruptcies from Compustat.
- •
  
  Use bankruptcy indicators from Compustat to create list of companies that appear to have undergone Chapter 11 bankruptcy.
- •
  
  Use SEC EDGAR to find each companies EIN code.
- •
  
  Search PACER Case Locator using companies EIN codes to find court records for bankruptcies. (Includes filing location)
- •
  
  Supplement missing cases with SEC filings found using Bloomberg Lawwebsite when possible.
3.
Create firm level database.
- •
  
  Merge Lopucki BRD, Compustat, and data scraped from PACER, to create complete panel data of publicly traded companies with information Chapter 11 bankruptcies.
- •
  
  Drop firms in the legal industry from database. (SIC = 8111)
- •
  
  Determine bankruptcy occurred when indicated by either Lopucki BRD or one of the bankruptcy indicators from Compustat.
- •
  
  Determine if forum shopping occurred if either indicated by Lopucki BRD or if the PACER filing location does not match the district where the firm was headquartered (Note this does NOT capture within district forum shopping!). If filing location data missing, assume firm did not forum shop.
4.
Create County-Year panel data used for regressions.
- •
  
  Aggregate firm level database to county-year level and merge with CBP employment data.
- •
  
  Include U.S. Census Bureau and BEA data for controls.

Fixed Effects
County	Yes	Yes	Yes	Yes	Yes
Year	No	Yes	No	No	No
Region-Year	No	No	Yes	No	No
State-Year	No	No	No	Yes	No
District-Year	No	No	No	No	Yes