How the Evidence Separates Venue from Selection
For corporate bankruptcy forum shopping to Delaware, the central empirical problem is not finding a difference between Delaware cases and local cases. It is determining whether the venue caused the difference or whether firms with different prospects chose different courts. This chapter explains how the study builds that comparison from linked court and Census data, uses distance within Delaware’s neighboring states as a source of venue variation, and tests the assumptions behind the resulting counterfactual. The payoff is practical: the reader can interpret the later survival estimates as causal evidence with a defined scope, rather than as a simple before-and-after or Delaware-versus-everywhere story.
The data make survival visible beyond the docket
A bankruptcy docket can tell us where a firm filed, when the case ended, and whether it was converted from Chapter 11 to Chapter 7. Those are important facts, but they do not fully describe what happened to the business. A company can avoid formal liquidation while closing establishments or reducing employment. To observe that operating trajectory, the study links court records to the Census Bureau’s establishment-level business data.
The Federal Judicial Center’s Integrated Database supplies the basic case structure for bankruptcies closed after 2007: filing court and office, filing date, petition information about assets, liabilities, and estimated creditors, and several case outcomes. Those outcomes include conversion from Chapter 11 to Chapter 7, whether unsecured lenders received any recovery, and the date of disposition. The debtor’s primary ZIP code—usually the headquarters ZIP code, with rare exceptions—lets the researchers locate the firm relative to the Delaware bankruptcy court.
The study adds PACER information because the FJC data exclude cases closed before 2008 and do not provide all of the identifiers needed for the analysis. PACER case summaries can identify the debtor, employer identification number, law firm, judge, handling office, and case number. The authors obtained information directly for courts that granted access and used cleaned data providers for other courts. Linking these records makes venue choice and judge assignment observable in the same case-level dataset.
The second half of the linkage is the Census Longitudinal Business Database, or LBD. It follows establishments over time and records employment, payroll, industry, legal status, creation, closure, and the firm to which an establishment belongs. The Business Register supplies names, employer identification numbers, and physical and mailing addresses used to match bankrupt firms to Census firms. This changes the unit of analysis. Instead of asking only whether a bankruptcy case was converted, the researchers can ask whether a particular establishment remained active and how many people it employed before and after the filing.
The scale is important because earlier debates often relied on small samples of large public companies. The study’s full corporate Chapter 11 dataset contains 155,059 cases and captures 82% of corporate Chapter 11 cases filed since 2005. It is near comprehensive, not literally complete: earlier PACER coverage is limited, the data focus on cases with an observed employer identification number, and the public paper reports reviewed outputs rather than the confidential microdata themselves.
The main causal sample is narrower by design. It contains 16,094 bankruptcies involving firms headquartered in Maryland, New Jersey, or Pennsylvania, the three states neighboring Delaware. The authors use these states because physical distance is more likely to affect a venue decision when travel by car is feasible. They do not restrict the sample by filing court; a firm in one of the three states can file locally, in Delaware, or elsewhere. Of the cases in this sample, 66% can be matched to a Census firm. The resulting establishment-level data contain roughly 477,000 establishments active at some point within three years of filing.
That last time window has a technical reason with substantive consequences. If a firm files in year t, the researchers examine establishments around the filing and through t + 3. An establishment that closed shortly before filing can still be part of the firm’s recent operating footprint, while one that opened by the third post-filing year can reveal recovery or expansion. Because the current LBD ends in 2021, filings after 2018 are excluded from analyses that need three years of post-filing observations. The employment result therefore concerns a defined historical window, not an ongoing forecast.
The outcome definitions preserve the difference between a case and a business. For an establishment, employment change is employment in a later year minus employment in the year before filing. If the establishment is inactive in either year, the assigned employment for that year is zero. Closure is measured as inactivity three years after filing. At the bankruptcy level, conversion records a formal move from Chapter 11 to Chapter 7; positive unsecured recovery records whether unsecured lenders receive anything, not how much they receive. Duration is based on the days from filing to disposition. Each measure captures a different part of the survival question.
Distance turns venue choice into a testable counterfactual
The data alone do not solve selection. Firms choose whether to enter bankruptcy, where to file, and how to manage a case. A firm that files in Delaware may have better lawyers, a more complicated capital structure, or worse underlying prospects than a firm that stays home. The ordinary least-squares comparison between actual Delaware filers and actual local filers would mix the effect of venue with those pre-existing differences.
The study’s solution is to use distance from the firm’s headquarters ZIP code to the Delaware bankruptcy court as an instrument for filing in Delaware. The first-stage idea is straightforward. Among firms in Maryland, New Jersey, and Pennsylvania, proximity changes the practical cost of reaching Delaware. Firms closer to Wilmington may find an out-of-state filing easier for managers, employees, lawyers, and other participants; firms farther away face a greater travel burden. If distance predicts the Delaware filing decision, the researchers can use the distance-related portion of that decision in the second stage.
This is not a claim that distance is random. Places closer to Delaware could differ from places farther away in their economies, industries, firms, or professional networks. The design therefore includes state-by-filing-year fixed effects, so comparisons are made among firms facing the same state laws and the same broad home-court option. It also uses county-year measures of local economic activity and, in more saturated specifications, controls for industry and petition-based financial characteristics such as binned assets, liabilities, and creditor counts. The intention is to make the relevant firms more comparable before using distance to isolate venue variation.
The first stage is economically meaningful and statistically strong. In the bankruptcy-level specification, a one-standard-deviation reduction in distance—62 miles in the study’s description—raises the Delaware filing rate by 5.1 percentage points. The reported Kleibergen–Paap instrument F statistic is 23.34. In an establishment-level specification, the corresponding estimate implies a 10.3-percentage-point increase, with an F statistic of 28.21. These are not two ways of stating one universal effect; they come from different units and specifications. Together they show that proximity is a relevant predictor of Delaware filing and make a weak-instrument explanation less likely.
The second stage asks a more demanding question: among the firms whose venue choice is moved by this distance incentive, what happens when the filing shifts toward Delaware? In plain terms, the method does not compare a handpicked Delaware firm with a handpicked home-court firm. It uses the predictable variation in venue choice associated with proximity, then estimates how that variation maps into case and establishment outcomes. The resulting quantity is a local average treatment effect.
“Local” describes the firms, not the importance of the result. The estimate is most directly about compliers—firms that file in Delaware when the practical cost is low but are more likely to stay elsewhere when distance rises. The first-stage patterns suggest that midsize firms may be especially represented among these compliers. The smallest firms may lack the resources or sophistication to forum shop even when nearby. The largest and most complex firms may gain enough from Delaware that they file there regardless of distance. A causal estimate for the distance-sensitive margin should not be casually relabeled the average effect for every distressed firm.
The design also requires an exclusion restriction: after the controls, distance must affect bankruptcy outcomes only through Delaware venue. The study tests this premise in several ways. First, it examines employment changes before the bankruptcy. Proximity to Delaware should not be able to cause an earlier employment trend through a later filing. The estimated effects on pre-bankruptcy employment changes are economically and statistically insignificant, which argues against distance simply identifying firms on different growth paths.
Second, the authors construct a placebo sample of establishments that have never been bankrupt. For each establishment in the main sample, they randomly select a never-bankrupt establishment in the same ZIP code, assign it the matched firm’s filing year and court characteristics, and then calculate employment changes around that assigned date. Across the pre- and post-assigned horizons, the estimates are near zero and statistically insignificant. This is a useful check because it holds local geography tightly while removing the bankruptcy event itself.
The placebos do not prove that the exclusion restriction is true. A null result is evidence against the particular alternative explanation that proximity marks different growth trends; it is not evidence that every unobserved channel is absent. The study therefore adds robustness checks for incorporation in Delaware, a prior Delaware establishment, law-firm experience and common law firms, major cities, distance to New York City, exact financial estimates when available, and tighter geographic comparisons. It also uses alternative distance measures, including commute time and distance relative to the nearest bankruptcy court. These checks narrow the range of plausible objections but cannot turn an observational instrument into a randomized assignment of venue.
A further assumption is monotonicity: moving farther from Delaware should not make some firms more likely to file there while making others less likely, in a way that cancels out the overall first-stage relationship. The study’s nonparametric evidence shows a residualized probability of Delaware filing that declines monotonically with residualized distance. This supports the interpretation that the instrument is moving firms in one general direction. It still does not mean every firm responds identically.
The logic can be summarized as a sequence, but each arrow has a different evidentiary status:
Proximity is the instrument; Delaware filing is the treatment; employment, closure, liquidation, recovery, and duration are outcomes. The placebos are not extra outcomes of the treatment. They are checks on whether the instrument is carrying a hidden growth difference. Keeping those roles separate prevents a common mistake in reading causal studies: treating a strong predictor as if it were automatically a valid cause.
What would change the conclusion? Evidence that distance predicts pre-bankruptcy growth or outcomes for never-bankrupt establishments would weaken the exclusion argument. Evidence that firms can systematically time filings to select particular judges would complicate the assignment logic. A materially different result under credible alternative distance measures or tighter geographic controls would narrow confidence further. Conversely, the judge-assignment analysis and the court-congestion shock provide independent ways to test whether the venue effect behaves as the judicial-capability mechanism predicts.
The study therefore offers more than a large sample and a technical instrument. It creates a disciplined bridge from venue choice to a counterfactual outcome, while making the assumptions visible. The next question is what that bridge shows about the firm itself: whether filing in Delaware changes employment and operational continuity, or merely changes the paperwork surrounding an inevitable liquidation.