Chapter 7 or 13: Are Client or Lawyer Interests Paramount? * Lars Lefgren Brigham Young University. Frank McIntyre Brigham Young University

Transcription

1 Chapter 7 or 13: Are Client or Lawyer Interests Paramount? * Lars Lefgren Brigham Young University Frank McIntyre Brigham Young University Michelle M. Miller Boston University Draft July 2008 * We thank Tina Lindsay and Jocelyn Rick for helpful discussions regarding institutional factors affecting personal bankruptcy. We thank Paul Weitzel, Amanda Boren, Laura Summers, Justin Hansen, Brad Hunter, Mirinda Martin and Henry Tappen for excellent research assistance. We are grateful to David Sims, Joseph Price, Richard Butler, Mark Showalter, Brad Larsen, and seminar participants at Brigham Young University and the University of California at Berkeley for their thoughtful comments. Lars Lefgren can be contacted at [email protected], Frank McIntyre at [email protected], and Michelle Miller at [email protected].

2 Abstract Households often rely on professionals with specialized knowledge to make important financial decisions. In many cases, the professional s financial interests are at odds with those of the client. We explore this problem in the context of personal bankruptcy. OLS, IV, and fixed effects estimates all show that attorneys play a central role in determining whether households file under Chapter 7 or Chapter 13 of the bankruptcy code. We present evidence suggesting that some attorneys maximize profits by steering households into Chapter 13 bankruptcy even when the households objective financial benefits are low and the probability of case dismissal is high. An attorney-induced Chapter 13 filing increases household legal fees and reduces the probability of long-term debt relief.

3 Introduction Individuals rely on informed professionals for many of their important financial decisions. An emerging literature documents the extent to which professionals sacrifice their clients interests to increase profits. For example, Levitt and Syverson (2005) show that real estate agents price clients houses more cheaply than their own. Gruber and Owings (1996) provide evidence that doctors are more likely to perform expensive caesarian sections when there is relatively low demand for their services. Similarly, Harrington and Krynski (2002) discuss how funeral directors often induce households to bury their deceased loved ones in lieu of a less expensive cremation. Chevalier and Ellison (1997) demonstrate that mutual fund managers adjust the risk of their portfolio to maximize fund inflows instead of risk adjusted returns. In this paper, we examine the principal-agent problem in the context of personal bankruptcy. We document the extent to which bankruptcy attorneys take advantage of their clients lack of sophistication to steer them into a bankruptcy option that is lucrative for the attorney but poorly suited to the clients financial situation. Poor advice can cost insolvent households thousands of dollars and usually fails to provide debtors with longterm debt relief. Personal bankruptcy is one of the most important mechanisms through which Americans are insured against adverse financial, health, and personal shocks. Fifteen of every thousand households in the United States filed for personal bankruptcy in 2003, (Lefgren and McIntyre, 2007) on average discharging approximately $36,000 in debt (Culhane and White, 1999). Bankruptcy transfers more money than state unemployment insurance programs (UI) and Temporary Assistance for Needy Families (TANF) 1

4 combined. Despite the importance of personal bankruptcy as consumption insurance, limited information on the part of debtors induces a situation in which access to appropriate debt relief is based in large part upon which law office a debtor happens to enter. A debtor s bankruptcy experience crucially depends on whether he files under Chapter 7 or Chapter 13 of the bankruptcy code. 1 Under Chapter 7, often referred to as liquidation, households may only keep exempt property. All other assets of value are liquidated by a trustee and distributed to creditors. Most of the household s unsecured debts are then discharged (forgiven) and the debtor does not have to give up any of his future income. Alternatively, under Chapter 13, the debtor retains all of his financial assets but promises to follow a court approved repayment plan over a three to five-year period. Which chapter is optimal for a debtor depends on the particulars of his financial situation. Most households that file under Chapter 13, however, pay high legal fees and do not receive long-term debt relief. While it is to be expected that not all bankruptcies will work out as planned, evidence suggests that the chapter under which households file is not purely a function of their financial situation. 2 Using household level data from California, Texas, and Utah, we show that an attorney s fraction of other bankruptcies filed under Chapter 13 explains percent of the variation in chapter selection, even controlling for the financial situation of filing households. This relationship is by far the single most important 1 During the time period in question, Chapter 7 and Chapter 13 relief was available to all debtors. Relief under Chapter 11 was also available to individual debtors, but few debtors chose that option because of the significant cost of filing bankruptcy under Chapter For example, Lefgren and McIntyre (forthcoming) report that the fraction of personal bankruptcies filed under Chapter 13 ranges from 0.03 in North Carolina to 0.62 in Georgia, suggesting legal culture plays an important role in chapter choice. Sullivan, Warren, and Westbrook (1998) emphasize that debt loads and repayment ability appear similar for Chapter 7 and Chapter 13 filers. Braucher (1993) presents qualitative evidence on the attorney s role in the chapter decision. 2

5 observable predictor of which chapter gets filed and it holds after either (1) instrumenting the actual attorney s fraction of bankruptcies filed under Chapter 13 with a measure of average attorney filing behavior in the debtor s neighborhood or (2) using an identification strategy based on neighborhood fixed effects. Firms specializing in Chapter 13 collect more for these bankruptcies, although less payment is required upfront. Yet these firms do not offer benefits in terms of lower dismissal rates or more manageable payment plans. Indeed, our instrumental variables (IV) estimates, which focus on households that chose attorneys based on geographic convenience, suggest that those who file under Chapter 13 with Chapter 13 specialists are often objectively poorly suited for this type of bankruptcy and have high dismissal rates. While attorney specialization yields few benefits for clients, it appears consistent with firm profit maximization. Large firms and firms with a client mix better suited for Chapter 13 bankruptcies tend to specialize in Chapter 13. This is consistent with a model in which firms that expect to file a large number of Chapter 13 bankruptcies engage in fixed investments to reduce the marginal costs of filing such cases. Thus, firms encourage Chapter 13 when the cost is low enough to rationalize (from a profit maximization perspective) pursuing the higher court-regulated payment available under this chapter. While we cannot test the hypothesis that attorneys have different views regarding the advantages of each type of bankruptcy, attorney specialization is empirically consistent with debtor interests being subordinate to firm profits. Ultimately, a typical lawyer-instigated decision to file under Chapter 13, as opposed to Chapter 7, leads to a 3

6 substantial transfer of wealth from insolvent households to specialized attorneys with a reduced probability of long-term debt relief. Review of the Literature In addition to the empirical principal-agent literature cited in the introduction, our paper relates to a large existing literature on personal bankruptcy found in both the economics and legal disciplines. Most closely related to our analysis is a set of papers that explore factors driving a household s decision regarding bankruptcy chapters. Nelson (1999), Domowitz and Sartain (1999), Sullivan and Worden (1990), Li and Sarte (2002), and Sullivan et al. (1988) all find that financial incentives play a role in the choice of bankruptcy chapter. However, many other economists note that legal culture and the choice of attorney also play important roles in the chapter decision. Specifically, Lefgren and McIntyre (forthcoming) find that the propensity to file under Chapter 13 of the bankruptcy code varies greatly across localities. Because these differences are extremely persistent and exist across adjacent states with seemingly similar populations, they conclude that the differences in the proportion of Chapter 13 filings are likely due to legal culture. Sullivan et al. (1994) also argue that all of the variation across bankruptcy courts in chapter choice cannot be explained by state laws, by the behavior of particular individuals or by other non-legal factors. The authors develop a model in which the local legal culture is dominated by lawyers. According to this model, the lawyer, due to specialty, moral preference, or stereotype influences the chapter choice. Braucher (1993), Neustadter (1986) and Sullivan et al. (1988) all provide important qualitative 4

7 evidence that lawyers often steer households toward one particular bankruptcy alternative. Our paper is also closely related to the medical literature examining variation in treatment choice across providers and locations. Health economists have widely documented the variation in physician practice styles which cannot be explained by income, insurance, or patient preferences. Recent examples including Chandra and Staiger (2004), Epstein, Ketcham and Nicholson (2005), and Grytten and Sørensen (2003) show that choice of service provider plays an important role in the type of treatment patients receive. Chandra and Staiger (2004) and Allgood and Bachmann (2006) highlight the health benefits that patients receive from physician specialization. The current study makes three contributions to the existing literature on chapter choice. First, we quantify the magnitude of the attorney s role in chapter choice. Second, we address concerns that the apparent role of lawyers is driven by the endogenous sorting of clients to attorneys. Third, we show that attorney Chapter 13 specialization is on average detrimental to clients and plausibly driven by profit maximization on the part of the attorney. Our paper also adds fresh evidence regarding the problems of principal-agent relationships when the principal (the insolvent household in our example) is unsophisticated. Institution Background Personal Bankruptcy in the United States For households unable to service their debts, personal bankruptcy is a primary instrument of debt relief. Federal authority for bankruptcy is found in Article 1, Section 5

8 8 of the United States Constitution. For this reason, many aspects of bankruptcy are uniform across states. When households file for bankruptcy, creditors must stop all collection measures; this means that creditors must cease foreclosure proceedings, cannot send the debtor correspondence, and must stop all garnishments. Once they petition for bankruptcy, debtors must decide whether to file under Chapter 7 or Chapter 13 of the bankruptcy code. Under Chapter 7, households may have to liquidate their assets. Any proceeds are then distributed among their creditors. This process is fast and simple; Chapter 7 debtors are able to obtain a speedy discharge of most unsecured debts and remove some judicial liens. On the other hand, under Chapter 13, debtors keep their property and instead agree to repay their debts using their future income. Chapter 13 cases are comparatively complex and lengthy. From 1999 to 2001, Lefgren and McIntyre (forthcoming) report that 70.5 percent of all personal bankruptcies were filed under Chapter 7. A Chapter 7 case begins when the debtor files a petition with the bankruptcy court. In addition to the petition, the debtor must detail his assets, liabilities, monthly income, and average monthly expenses. After the petition is filed, a Chapter 7 bankruptcy trustee holds a meeting of creditors, during which the trustee and creditors may ask the debtor questions regarding these documents. After this meeting, the trustee can gather and sell debtor assets. The Bankruptcy Code, however, allows the debtor to keep certain exempt property. Specifically, debtors may keep assets with a value below the personal and homestead exemption levels. These exemption levels vary dramatically across states. For example, Florida has an unlimited homestead exemption while Delaware has none. After the debtor s assets are liquidated, the proceeds are used to 6

9 repay creditors and the debtor is discharged (released from liability) of any remaining debts. In other words, all remaining debts are then forgiven. While the Chapter 7 bankruptcy code allows for contested filings and the liquidation of assets, the typical case involves neither of these. According to conversations with bankruptcy attorneys, often no creditors attend the meeting of creditors. Additionally, the large majority of Chapter 7 cases involve no non-exempt assets that are profitable for the trustee to sell. For this reason, the typical Chapter 7 bankruptcy is quick and simple both for the attorney and for the client. Furthermore, only a small fraction of such bankruptcies are dismissed. In our sample of Chapter 7 filings, only 3 percent were listed as being dismissed. Not all households can file for Chapter 7 bankruptcy. Households who already filed for Chapter 7 bankruptcy within the past seven years are ineligible. Additionally, the Bankruptcy Abuse Prevention and Consumer Protection Act of 2005 (BAPCPA) restricted eligibility for Chapter 7 bankruptcy on the basis of household income. 3 Finally, United States trustees, who represent creditor interests, may attempt to force individuals to file under Chapter 13 if it is clear that the debtors possess the ability to pay a substantial fraction of their debts. Rarely are the income restriction or trustee objections an obstacle for households considering bankruptcy. Under Chapter 13, households retain all of their assets and instead agree to repay some of their debts according to a court ordered payment plan lasting between three and 3 In a sample of 1,938 cases filed nationwide between November 1998 and August 1999, Flynn and Bermant (2000) only found two petitions which were filed under Chapter 7 but failed the means test. Similarly, Culhane and White (1999) estimated that only 3.6 percent of debtors would be barred from Chapter 7 when a means test was enacted. True to speculation, within the first year of its enactment, the Acting Director of the Executive Office of the U.S. Trustees testified that approximately 0.5 percent of Chapter 7 filers were being affected by the means test (Tabb and McClelland 2007). 7

10 five years. Debtors pay their projected monthly disposable income, calculated as the difference between their monthly income and monthly budgeted living expenses, into the Chapter 13 payment plan. An in a Chapter 7, a Chapter 13 case begins when the debtor files a petition with the bankruptcy court. Again, the debtor must detail his assets, liabilities, monthly income, and average monthly expenses. After the petition is filed, however, the legal proceedings around a Chapter 13 become more complex. Following the creditor s meeting, lawyers must complete a Chapter 13 repayment plan. This plan, described in further detail below, specifies the amount debtors will pay to the trustee every month. After a judge confirms the plan (decides that the plan is feasible and meets the standards set forth in the Bankruptcy code), the Chapter 13 trustee will begin to distribute funds to creditors. The funds are first disbursed to the debtor s lawyer, then to creditors with priority claims, and finally, any remaining funds are used to repay creditors with unsecured claims. Upon completion of the plan, the household s remaining debts are discharged. The Chapter 13 discharge is often referred to as the super-discharge; in addition to the debts discharged under Chapter 7, Chapter 13 debtors can discharge debts from property settlements following a divorce, willful and malicious injury, governmental fines and penalties, unpaid taxes, certain fraudulent tax filings, fraud, embezzlement, larceny, and damages from personal injury civil action. Significantly, Chapter 13 bankruptcy also allows debtors to retain possession of collateral even if clients are in arrears with their payments. Often, households will file under Chapter 13 to stay in their home or keep an automobile. Debtors can file under Chapter 13 as frequently as every two years. 4 The process of creating a household budget and securing 4 Prior to the Bankruptcy Abuse Prevention and Consumer Protection Act of 2005 (BAPCPA), households could file a Chapter 13 bankruptcy every six months. 8

11 the approval of the payment plan involves substantially more paperwork and time relative to a Chapter 7 bankruptcy. Relative to Chapter 7, Chapter 13 is advantageous for some households. But, it is a poor choice for many debtors. The benefit of a Chapter 13 is that debtors may keep all of their assets. However, in order to discharge its debts under Chapter 13, a household must complete its repayment plan. Thus, Chapter 13 is designed for households with a continuing ability to earn income and minimize future expenditures. If a debtor is unable to commit to a long-term plan, his case will be dismissed, and he will be liable for all his original debts as well as additional court and lawyer fees. According to Lefgren and McIntyre (forthcoming), while 29.5 percent of bankruptcies are filed under Chapter 13 of the bankruptcy code, the majority of these bankruptcies are dismissed (60 percent) largely due to nonpayment on the debtor s part. Another 12 percent of the Chapter 13 filings are ultimately converted to Chapter 7 for the same reason. Thus, Chapter 13 bankruptcies only represent 10 percent of bankruptcy discharges. For these reasons, the majority of households filing under Chapter 13 do not receive long-term benefits. Lawyer Specialization Lawyers vary significantly in the fraction of bankruptcies they file under Chapter 13. Figure 1 is a histogram showing the distribution of lawyers in California, Texas, and Utah by the fraction of bankruptcies they file under Chapter 13. This histogram demonstrates the large spread in lawyer behavior. 42 percent of lawyers file less than 10 percent of their cases under Chapter 13 of the bankruptcy code. On the other hand, 20 percent of lawyers file more than 40 percent of bankruptcies under Chapter 13. If 9

12 bankruptcy clients shop for lawyers who meet their specific financial situations, specialization may not indicate that lawyers play a significant role in determining chapter choice. On the other hand, if filing households are relatively unsophisticated when it comes to selecting a lawyer, the high degree of lawyer specialization suggests that the attorney is crucial in a household s choice of bankruptcy chapter. Braucher (1993) effectively documents the role that attorneys play in guiding typically unsophisticated clients towards one type of bankruptcy or another. 5 Attorneys can influence clients to file under Chapter 13 by emphasizing the benefits of retaining secured assets, the morality of repaying creditors, and access to future credit. 6 Other attorneys shift households towards Chapter 7 bankruptcy by highlighting the difficulty of maintaining the payment plan, the moral obligations to provide financially for their families, and the predatory nature of some creditors. Braucher finds that clients' guilt and loss of self-esteem makes them highly vulnerable to lawyers' influence, whether exercised unwittingly or deliberately. Thus, she concludes that attorney practices have more effect on chapter choices than features of the law conventionally thought to be important. Similarly, Sullivan et al. (1988) find that among other factors, attorneys greatly influence the choice of chapter. Using data collected during surveys, the authors find that nearly 32 percent of debtors consulted attorneys specializing in bankruptcy while the remaining debtors sought counsel from more general practitioners. The authors 5 As Braucher (193) noted, except for conversion, consumer bankruptcy clients are not typically repeat clients. First time bankruptcy clients typically have little information regarding the costs and benefits of each type of bankruptcy. Additionally, social stigma associated with personal bankruptcy likely reduces word of mouth transmission of peoples bankruptcy experiences. 6 A Chapter 13 bankruptcy stays on an individual s credit report for seven years as opposed to ten for Chapter 7 bankruptcies. Anecdotally, however, households filing under Chapter 7 bankruptcy have better access to credit immediately after filing for bankruptcy according to Braucher (1993). This is because they no longer have additional credit obligations and lose the option value of filing under Chapter 7 again for the next seven years. 10

13 discover a positive, moderate unconditional correlation (r = 0.27) between seeing a bankruptcy specialist and filing under the more complicated Chapter 13 of the bankruptcy code. This suggests that attorneys may exert an important influence over whether debtors file Chapter 7 or Chapter 13. While it seems plausible that many lawyers steer clients toward one type of bankruptcy, it is useful to consider what drives this behavior. Lawyer specialization may occur due to the complexity of Chapter 13 bankruptcies. Filing under Chapter 13 of the bankruptcy code requires additional paperwork relative to Chapter 7. First, lawyers must file a repayment plan for all Chapter 13 cases. This requires a detailed collection and investigation of a client s receipts and bank statements. Lawyers must then create a comprehensive itemized budget that their clients must follow over a three to five year period. The paperwork, planning, and organization required to complete this task is substantial. Additionally, the court must confirm Chapter 13 bankruptcies, a process which is often more difficult and time consuming than under Chapter 7. With the 2005 amendment of the bankruptcy code, filing for Chapter 13 bankruptcy became even more difficult. Lawyers must now complete an additional form, the Statement of Current Monthly Income and Calculation of Commitment Period and Disposable Income Form (Form 22C). With 60 lines of exceptionally detailed information, this form particularizes the debtor s monthly income and expenses. Each entry requires thorough documentation; without the proper clerical system, disorganized lawyers may find the task daunting. The form also requires lawyers to predict the debtor s future income and expenditure. While more difficult to file, the payment structure may provide a financial incentive for lawyers to file under Chapter 13, whether or not it is in the client s best 11

14 interests. Chapter 7 attorney fees are unregulated by the courts and households typically pay the entire fee upfront. With Chapter 13 on the other hand, households usually pay some fees upfront; the majority of the fees however are collected as part of the household s payment plan. Because Chapter 13 fees are rolled into a payment plan, households may be relatively insensitive to cost of such a bankruptcy. Perhaps for this reason, individual bankruptcy courts set norms for Chapter 13 compensation; fees above a customary limit are subject to special scrutiny. Thus, attorney fees are tightly clustered around this informal limit. 7 Typically, the customary Chapter 13 fee is much higher than the equilibrium Chapter 7 fee set by the market. From a 2007 sample of two Chapter 7 and two Chapter 13 filings from each of the 90 bankruptcy districts, attorney fees for Chapter 13 bankruptcies averaged $2,657 compared to $905 for a Chapter 7 bankruptcy. In the Chapter 13 repayment plan, attorney fees are given priority status relative to other creditors; their fees may displace payments to other creditors. Thus, even if the plan fails, attorneys collect the majority of their payment. 8 As an example of this, we drew a random sample of 54 dismissed (unsuccessful) Chapter 13 bankruptcies filed in Texas in For these 54 cases, the average lawyer fee was $2,055. The final trustee report showed that the average dismissed case lasted 13 months. Despite the rapid dismissal of the majority of these cases, lawyers still received $1,376 on average 66 percent of the nominal fee. In discharged cases (which we did not sample), we know the 7 See Braucher (1993). Note that these fee limits increased in the aftermath of the 2005 law change. The averages we report here are from February 1, When we look at fees as an outcome variable, we restrict our sample to pre Braucher (1993) documents that in some jurisdictions, the price differential between Chapter 13 and Chapter 7 bankruptcies is much smaller. In these jurisdictions, lawyers only rarely file bankruptcies under Chapter

15 lawyer is fully paid; thus a lawyer s unconditional expected payment from a chapter 13 filing was 80 percent of their fee. 9 The average unsecured debt in these dismissed cases was $19,803; households repaid $302 on average. Secured debt averaged $21,546 with $2,280 in principal repaid and $598 in interest. For this small Texan sample, even when a case was dismissed, the attorney received substantial compensation though a very small fraction of the other debts was repaid. In 22 out of 54 dismissals, the lawyer was paid more than all other creditors combined. Thus, while Chapter 13 bankruptcies are more difficult to file than Chapter 7 bankruptcies, they generate more revenue. Additionally, firms can engage in investments to lower the marginal cost of filing Chapter 13 bankruptcies. 10 These institutional factors suggest the economic model of attorney specialization contained in the appendix and summarized here. Suppose that firms can undertake investments in technology, 11 personnel, or human capital to lower the marginal costs of filing bankruptcies under Chapter 13. Firms that expect to file enough Chapter 13 bankruptcies would find it optimal to undertake such investments. Investments made to lower the cost of filing bankruptcies under Chapter 13 lead to law firm specialization. More specifically, investing firms find it optimal to steer a higher fraction of households into Chapter 13 than non-investing firms, even holding client mix constant. There are two testable implications of this model. First, larger firms will file a higher fraction of bankruptcies 9 60 percent of 2003 Chapter 13 cases in Texas were dismissed..4*1 +.6*.66 = Chapter 7 bankruptcies are relatively simple to file (at times they are even filed without the help of an attorney) which is why we do not consider the possibility that attorneys could engage in investments to reduce the cost of filing for Chapter 7 bankruptcy. 11 Specialized software programs automate much of the bankruptcy filing process. According to attorneys, this software is more helpful for the filing of Chapter 13 bankruptcies than the filing of Chapter 7 bankruptcies. 13

16 under Chapter 13 than smaller firms. The second implication is that firms with a client mix better suited for Chapter 13 will be more likely to file any given bankruptcy under Chapter 13. Similar implications can be obtained from a learning-by-doing model in which attorneys become better at filing Chapter 13 bankruptcies with practice. Our theoretical framework suggests that attorneys will steer households towards the bankruptcy option that maximizes profits, even at the expense of their clients best financial interests in some cases. Specialization may, however, be beneficial for bankrupt households. Households that file under Chapter 13 with firms specializing in this type of bankruptcy may enjoy better financial outcomes or lower costs than households that file Chapter 13 bankruptcy with unspecialized firms. We test this by comparing the filing costs and dismissal rates of households that have filed a Chapter 13 with different firms. Of course, specialization likely reflects different attorney beliefs regarding the relative benefits of Chapter 7 and Chapter 13 bankruptcy. Lawyers may also vary in how seriously they take their obligation to look after their clients best interests. These hypotheses cannot be tested, however, with available data. Description of Data We use several different data sources to examine the role of attorneys in the bankruptcy chapter decision. Bankruptcy data comes from Public Access to Court Electronic Records (PACER), the bankruptcy courts centralized registration and billing website. 12 While the names of bankrupt individuals are publically available, additional 12 Each court operates its own database with case information. Launched in 1997, courts slowly began to use PACER s electronic database system. The last court adopted the PACER system in

17 information is difficult to obtain. A filer s assets, debts, income, address, and lawyer, for example, are only available on his bankruptcy petition. Therefore, our dataset is limited to three bankruptcy courts which gave us access to their electronic records: Utah, Texas Northern, and California Northern from the period 2000 to The Utah sample is a census of bankruptcies from early 2000 to late 2004; our sample includes all 65,957 cases filed in Utah during that time. On the other hand, the Texas and California data are samples from 2003 to 2006; these samples include 30,575 of the 113,412 and 24,088 of the 78,263 cases filed in the Northern Districts of Texas and California respectively during these three years. 36 percent of the bankruptcies on our sample were filed under Chapter 13 relative to just under 30 percent in the U.S. prior to the 2005 bankruptcy reform and just over 20 percent in The filing rates for the states in our sample are also somewhat higher than the nation as a whole. 13 For each petition in our sample, we have information on bankruptcy chapter, filer address, lawyer identity, and lawyer location. In addition, we observe whether the case was dismissed or received a discharge of debts. Because we know the lawyer used in each case, we can calculate the fraction of bankruptcies (excluding the reference individual) filed under Chapter 13 for each attorney. 14 Filer address information was used to merge in 2000 Census block level data. This provides rough information on 13 The national average was 15 filings per 1,000 households in For approximately the same time period, California s filing rate was 16, while Utah s filing rate was 28, and Texas was 8. In terms of demographics, capita income in 2006 was $25,287 in the entire United States, $21,016 in Utah, $26,974 in California, and $22,501 in Texas. The foreign born percentage was 13 percent in the United States, 8 percent in Utah, 27 percent in California, and 16 percent in Texas. The black percentage was 12 percent in the United States, 1 percent in Utah, 6 percent in California, and 12 percent in Texas. Median age was 36 in the United States, 28 in Utah, 34 in California, and 33 in Texas. We control for observable demographic differences in the regressions. 14 In robustness checks, we redo our analysis calculating the fraction of bankruptcies filed under Chapter 13 by the law firm. In these specifications, the law firm is defined by the address on the bankruptcy records. The results are virtually identical. 15

18 demographics, income, and housing values of the bankrupt. Table 1 shows sample means of the filing information and block level demographics. The Census block data suggest that nearly two thirds of the population is married and another 11 percent is divorced. While the large majority at the block level has completed high school, about one fifth of the population has obtained an undergraduate degree. The average median income in the Census blocks is between $40,000 and $50,000, with the average unemployment rate at 3.5 percent. The block level data also indicates that approximately 60 percent of bankruptcy filers are homeowners. Finally, almost the entire sample, 92 percent, resides in an urban Census block. In addition to our primary data set, for 1,965 households that filed for bankruptcy in Utah during the first half of 2000, we collected more detailed information from the Statement of Financial Affairs. For these households, we manually collected income from the prior year, debt and asset levels, and household composition. Finally, we collected the same information for a sample of 16,735 Texas filings for which the court provided machine readable Statements of Financial Affairs. From this sample we exclude households with over $500,000 of unsecured debt. There were fewer than 20 of these outliers which were, in many cases, obviously erroneous. 15 Summary statistics for these two samples are also provided in Table 1. The means of the Census block group variables are generally similar to those of our baseline sample. We can also examine summary statistics of financial variables reported in the Statements of Financial Affairs. We see that in Utah and Texas, households earn about $2,500 and $3,100 per month, 15 This cleaning did not substantially affect our conclusions, although in a couple cases it had a large impact on reported means. For example, one filing reported debts in excess of seven billion dollars for an unsecured lease. 16

19 have secured debt levels of about $103,000 and $83,000 and unsecured debt levels around $46,000 and $47,000 respectively. We use two additional samples of data near the end of the paper. The first is a set of 3,878 Texas filers in 2004 for whom we have information on prior bankruptcy filings as well as their Statements of Financial Affairs. This allows us to perform our analysis on a set of first time filers. Doing so avoids the problem that prior filers of a Chapter 7 bankruptcy may be eligible to file only under Chapter 13 bankruptcy. Additionally, we address the concern that clients of lawyers who disproportionately file under Chapter 13 bankruptcy may be more likely to have filed under Chapter 13 bankruptcy already due to the high dismissal rate of such bankruptcies. Lastly, we wish to look at bankruptcy outcomes, and so, for 8,790 households in our Texas data, we collect information on dismissal rates, reported household budgets, and legal fees. Empirical Strategy One of our primary empirical objectives is to document the role attorneys play in the chapter decision; in this section we outline our methodology. We later use analogous methods for examining the consequences of filing with a particular attorney on other bankruptcy outcomes including the dismissal rate, appropriateness of Chapter 13, and legal fees. Note that we examine a sample of households that actually filed for bankruptcy. To the extent that a client s decision to file for bankruptcy, regardless of type, depends on lawyer behavior, our empirical models will be misspecified. It seems plausible, however, that most households are in sufficiently dire financial straits that the decision to file for bankruptcy is inelastic to attorney behavior. Braucher (1991) reports 17

20 By the time [debtors] consult a lawyer, they may have been living on credit card advances for six months or a year after losing a job or undergoing a divorce. We therefore assume that clients inelastically file for bankruptcy and simply choose between Chapter 7 and Chapter Our base empirical specification is a linear probability model of the following type: (1) chapter13 = X β + α frac13 + ε ij i ji ij where chapter13 ij is a binary variable equal to one if individual i filing with attorney j files under Chapter 13 of the bankruptcy code and zero otherwise. X i is a vector of household level characteristics correlated with the decision to file under Chapter 13, and frac 13 ji is the fraction of an attorney s clients who file under Chapter 13, excluding the reference household, i. Going forward, we will refer to this term as fraction 13. We use frac 13 ji as a proxy for the lawyer s underlying propensity to file households under Chapter If attorney assignment were random, or if the vector X i were sufficiently rich that lawyer assignment was conditionally orthogonal to the residual, α would represent the causal effect of being assigned to a lawyer with a higher propensity to file under Chapter The impact of lawyer behavior on the probability that a household files for bankruptcy cannot be measured using our data. It seems implausible that this would be an important concern, however, as nearly all lawyers file some bankruptcies of each type. To the extent that it is more profitable to file a client under the non-preferred chapter than to not file them at all, lawyers have a strong incentive to provide the type of services that keep the client from leaving their office. Additionally, households that are pushed in a direction contrary to their best financial interests retain the option to consult a different attorney. Finally, to the extent that pushing people towards Chapter 13 bankruptcy, which typically yields the fewest financial benefits, causes them not to file for bankruptcy at all, our estimates will be biased downwards. 17 We also experimented with adjusting frac 13 ji for the observable characteristics of the lawyer s client mix. The results, shown later in our robustness checks, are virtually identical. 18

21 We start by estimating equation (1) using ordinary least squares. We include a specification in which we control for Census block group information; we also estimate the relationship using a subset of filers for whom we have income, debt, and asset information from their Statement of Financial Affairs. This richer set of covariates should control for most primary financial factors that drive chapter choice. The assumption that lawyer s fraction 13 is orthogonal to the residual (conditional upon covariates) may not hold. The high propensity some lawyers exhibit to file under Chapter 13 may reflect the unobserved characteristics of their clients. More specifically, it might be that all lawyers behave identically but that the variation in attorney filing behavior is driven by differences in client mix. Alternatively, some lawyers may specialize in a particular type of bankruptcy and match (through advertising or word of mouth) with individuals looking to file under that particular chapter. 18 In these cases, we would expect the residual to be correlated with the fraction of an attorney s other clients filing under Chapter 13. To address this concern, we also pursue an identification strategy in which we instrument fraction 13 with a weighted average of the fraction 13 of attorneys in the individual s area. This weighted average is defined by the following equation: (2) frac13 i = j frac13 * F / d j j ji j ij F / d ij 18 A brief perusal of the yellow pages in Utah found plenty of advertisements for bankruptcy attorneys, but little that suggested specialization in one chapter or the other. One attorney, though, did mention their flat fee for Chapter 7 a possible signal of Chapter 7 preference. Braucher (1993) provides more systematic evidence that attorneys do little to signal specialization. 19

22 where F is the annual number of filings for the lawyer and d is the distance (in miles) j 19 from debtor i to lawyer j. 20 Thus the instrument weights each lawyer by their closeness to the household and number of filings. The first stage equation in this specification is given by: (3) frac13 = X Γ+ γ frac13 + η. ji i i ij This strategy takes advantage of variation in filing behavior attributable to residential location. We will obtain consistent estimates of α if residential location is uncorrelated with a household s propensity to file for bankruptcy conditional upon observables. We ij test the plausibility of this assumption by examining the correlation of frac 13 i with households block level Census demographic information. We motivate the instrument as a reduced form approximation to a fairly complicated lawyer choice problem. Using a discrete choice framework, one could estimate a model with distance and other lawyer characteristics entering a household s optimal choice of attorney. Estimating such a set of equations requires fairly strong assumptions and adds additional complexity for relatively little additional insight. We view our first stage as the reduced form relationship between distance and attorney behavior that results from that consumer s optimization problem. Note that our IV strategy returns estimates based on the experience of those who choose their attorneys based on largely geographic convenience as opposed to quality or appropriateness of legal advice. 19 We use distance as a proxy for convenience. The number of filings captures factors such as firm visibility and the effectiveness of advertising. 20 We used Google to find the latitude and longitude of each household and law firm address. For 11.5 percent of the households, we could not obtain a valid longitude and latitude so we calculated distance from the centroid of their zip code. This reduces the power of our first stage but should not induce any bias. 20

23 Our instrumental variables specifications take advantage of variation in geographic location to identify the impact of lawyers on chapter choice. To the extent that unobserved neighborhood characteristics drive chapter decisions, our IV estimates may be misleading. An alternative approach is to completely control for geography and identify the attorney s role based on the decisions of households that live very close to each other but choose to file with different attorneys. This is almost the opposite identifying assumption as our earlier instrumental variables strategy. To see this, consider the fixed effects specification (4) chapter13 = δ + X β + α frac13 + ε ij k i ji ij where δ k is the neighborhood fixed effect. Fixed effects identification come from assuming that all the endogeneity is embedded in δ k, thus once it is purged the remaining variation is exogenous. Our instrument, on the other hand, is completely drawn from location and therefore uses only geographic variation, assuming that a particular kind of geographic variation is pure while individual variation is endogenous. We find the IV results more compelling, but present the alternative below to show the robustness of the conclusions. Findings OLS Results Table 2 reports the OLS estimates of the impact of lawyer fraction 13 on individual filing behavior. 21 The first column shows the OLS results for the full sample without controls. Unsurprisingly, the coefficient on lawyer fraction 13 is This 21 Recall that for our entire analysis we exclude the individual s own filing when calculating lawyer fraction

24 point estimate suggests that going from a lawyer who files 50 percent of bankruptcies under Chapter 13 to one who files 60 percent under Chapter 13 would increase an individual s own probability of filing under Chapter 13 by 9.8 percentage points. Furthermore, the r-squared is 0.26 suggesting that more than a quarter of the variation in filing rates is correlated with lawyer identity. This result is consistent either with lawyers having a large causal effect on chapter choice or strong sorting of individuals to those lawyers who specialize in their chosen type of bankruptcy. In the second column, we add Census block-group controls. The addition of these covariates has almost no effect on the lawyer fraction 13 coefficient and their partial r- squared is It may be that most of the variation in personal factors that affects filing decisions occurs within neighborhoods as opposed to across neighborhoods. In this case, controlling for block group averages will do little to control for the household heterogeneity that determines chapter choice. To investigate this possibility, we examine two samples for which we have collected household level debt, asset, and income data. In column 3, we examine our detailed Utah subsample, but do not include covariates. Consistent with our earlier results, we estimate a coefficient on lawyer fraction 13 that is again quite close to one. In column 4, in addition to Census blockgroup covariates, we control for individual covariates from the Statement of Financial Affairs, which drops the estimate to While the partial r-squared of these individual level covariates is 0.08, lawyer fraction 13 still explains more of the variation in filing behavior than even these covariates from the bankruptcy filing. 22

25 In the next column, we examine our detailed Texas subsample. When we do not include other covariates, the coefficient on lawyer fraction 13 is 0.90, the same as our Utah sample but a bit smaller than in our other sample. Adding covariates causes the coefficient to drop by about 0.11 to As with our prior findings, lawyer fraction 13 continues to be a very strong predictor of chapter choice. IV Results While the OLS coefficients suggest that lawyers play a central role in chapter choice, it may be that we have insufficiently controlled for the financial determinants of chapter selection. To further address this concern, we use our instrumental variables strategy. Table 3 shows the first stage regression results. The instrument, reported in the first row, shows that distance-weighted average of the lawyer fraction 13 is an excellent predictor of the filing rate for the lawyer a household actually goes to. The table also reports the F-statistics for the regressors, which suggest that the instrument is a statistically significant predictor of fraction 13 filed, though the instrument is substantially weaker in Utah than in Texas. 22 Table 4 reports the second stage estimates. The coefficient on lawyer fraction 13 exceeds 0.86 in all specifications. Furthermore, inclusion of either neighborhood or household covariates has little impact on the coefficient. This suggests our instrument has a low correlation with the observables that determine chapter selection. For the 22 To examine why instrument strength varies across states, we compared the standard deviation of lawyer fraction 13 in the two samples to the standard deviation of the instrument. The standard deviation of lawyer fraction 13 rates in the Texas and Utah subsamples are 0.25 and 0.23 respectively. The corresponding standard deviations of the instrument are 0.13 and Apparently in Utah, most households are nearly equidistant from several attorneys with different filing behavior and thus the variation averages out more than in our Texas sample. 23

26 Census and the Texas detailed samples, the standard errors are quite small. For the comprehensive sample, our confidence interval ranges between 0.90 and For the Texas detailed sample, the confidence interval ranges from 0.81 to Fixed Effects Results In Table 5, we show OLS estimates of the impact of lawyer fraction 13 when we include Census block fixed effects. Thus each neighborhood of approximately two to three thousand people has its own fixed effect. Identification comes from comparing results across individuals with the same unobserved neighborhood characteristics. The coefficients are almost identical to our OLS estimates and quite close to the IV results, providing convincing evidence that our results hold under a wide range of identifying assumptions. Comparing the adjusted r-squared values to those from the OLS in Table 2 suggests that, once one controls for observables, neighborhood fixed effects do not play a very important role with partial r-squared values around 1 percent. Since neighborhood observables were also found to be similarly unimportant, this result may not be too surprising, though it is reassuring. Further, the similarity with the OLS results suggests one of two things; either (1) there is no endogeneity problem at the neighborhood level, thus our IV results are unbiased, or (2) the neighborhood and individual-level unobserved effects together happen to be producing exactly the same bias as the individual level unobserved effects alone. For the remainder of the paper, we concentrate on the IV results, as we prefer the IV identifying assumption to that required for the fixed effects estimator. 24

27 OLS, IV, and neighborhood fixed effects estimates all suggest that lawyers play a central role in household decisions regarding bankruptcy chapter. To put the findings in perspective, the 25 th percentile lawyer in our comprehensive sample files no bankruptcies under Chapter 13 while the corresponding measure for the 75 th percentile lawyer is 41 percent. Even if we rely on a relatively conservative estimate, a fraction 13 coefficient of 0.9, happening to go to a 75 th percentile lawyer instead of a 25 th percentile lawyer would increase the probability of filing under Chapter 13 by 37 percentage points. No other covariate in our model has a similarly large impact on filing behavior. Parameter Heterogeneity The ability of a lawyer to affect a client s chapter choice varies widely by lawyer and client. Some clients already know what they need. Others cannot file under Chapter 7 due to legal restrictions. The parameter we estimate with OLS represents an average effect across the population. Our IV estimates capture a different local average treatment effect (LATE). 23 The IV strategy identifies the average influence of lawyers on households that choose lawyers based largely on geographic convenience. One might suspect that such households are particularly susceptible to filing under the bankruptcy chapter which is in the attorney s best interests. This may explain why, in some cases, the IV estimates are slightly larger than the OLS estimates. The OLS and IV estimates differ even more when we examine bankruptcy outcomes, likely for the same reason. In Table 6, we examine whether there are heterogeneous responses across observable groups. All regressions control for neighborhood characteristics. In rows 1 to 3, we use our comprehensive sample to examine the implied impacts of lawyers in the 23 See Imbens and Angrist (1994). 25

28 three different bankruptcy districts in our sample. We find little difference in the point estimates. In rows 4 and 5 we report our findings for urban and rural Census block groups. In rows 6 and 7, we see whether the findings differ across low and high income block groups. Finally, in rows 8 and 9 we compare the results for block groups with high and low concentrations of minorities. There is very little evidence of heterogeneity in the parameter, which is above 0.77 in all the above specifications and typically between 0.9 and 1. Turning to the financial records, we use our Texas subsample to look at households with few secured debts (less than $20,000). These households have little collateral to protect from seizure and would therefore be expected to file under Chapter 7. Indeed, 24 percent of households with less than $20,000 in secured debt file under Chapter 13 bankruptcy compared to 48 percent of households with more than this amount. Comparing the OLS results from the first column of rows 10 and 11, households with little collateral are noticeably less responsive to lawyer filing behavior than those with more assets at risk. Since these low collateral households are well suited for Chapter 7, it makes sense that they are more difficult to sway (or, equivalently, lawyers are less willing to try and sway them). This makes the IV results all the more interesting. For low asset filers the IV estimate is 0.80 compared to 0.64 in the OLS specification. Though the difference is borderline in terms of statistical significance, the larger IV estimate suggests that households that match to their attorney on the basis of geographical convenience are more easily pushed into Chapter 13 bankruptcy even when they receive few financial benefits from doing so. 26

29 Robustness Checks While our OLS, IV, and fixed effects specifications in Tables 2, 4, and 5 yield remarkably similar estimates, it is worth considering specific threats to identification. We already considered the implications of the fixed effects strategy as a robustness check. Here we consider possible threats to the IV strategy. One concern is that our instrument, the distance-weighted average fraction 13, is correlated with the unobserved characteristics of the filing household. Our model of lawyer specialization (and evidence we present later) implies that firms tend to specialize in Chapter 13 bankruptcy when their client mix is better suited for this type of bankruptcy. This suggests that lawyer behavior will be correlated with observable and possibly unobservable neighborhood characteristics, potentially invalidating our instruments. In light of this, we regress our instrument on Census block and available household characteristics. The partial r-squared of neighborhood characteristics is 0.03 in the comprehensive and Texas samples (though 0.24 in the Utah sample). These regressions suggest that while our instrument is correlated with observable neighborhood characteristics, the relationship is quantitatively small in the Texas and comprehensive sample. Additionally, in our primary specifications, we control for a host of demographic characteristics this may absorb much of the variation in unobserved propensity to file under Chapter 13. In practice, these controls make little difference to the estimates. Even controlling for detailed financial characteristics has little impact on our estimates. To further show that our instrument does not reflect lawyers locating in neighborhoods with high unobserved demand for one type of bankruptcy or another, we 27

30 return to our examinations of households with fewer than $20,000 in secured debts. As Chapter 13 protects collateral, these households should find Chapter 13 relatively unattractive. If our instrument simply reflects the matching of local neighborhood needs and lawyer specialization, the lawyer choice of households with few secured debts should be insensitive to our instrument. The first stage coefficient for this subgroup is 1.08 compared to 1.01 for our full Texas sample. This suggests that even for households clearly suited for a specific type of bankruptcy, the geographic considerations captured in our instrument play a strong role in whether they select an attorney specializing in Chapter 13. A related problem is that our proxy for lawyer propensity to file under Chapter 13 may reflect the client mix as opposed to a lawyer s genuine preference for filing under a specific chapter. To examine this possibility, we estimate a linear probability model which includes attorney fixed effects in addition to other covariates. This can be represented by the following regression equation: (5) chapter13 ij = X i B + θ j + e ij. These attorney fixed effects represent the propensity to file under Chapter 13, controlling for client mix. To exclude each household s own contribution to the attorney fixed effect, we calculate a residual for each household equal to chapter X Bˆ. For each 13 ij i attorney, we average these residuals across all households, excluding the reference individual. Using our Census sample and census block group covariates, the correlation between our adjusted propensity to file under Chapter 13 and the raw fraction 13 is 0.953, suggesting that the adjusted measure is much the same as the old measure. As seen in 28

31 row 1 of Table 7, our findings are quite robust to the use of this adjusted propensity to file under Chapter 13. Another concern is that our primary specification is misspecified because our outcome is binary. In row 2 of Table 7, we present the marginal effects of lawyer fraction 13 on chapter choice using probit and IV probit specifications. The results again are qualitatively similar to our baseline estimates. In row 3, we perform our analysis on a subsample of our Texas data from 2004 from which we exclude households with prior filings. This addresses two concerns. First, households that filed a prior Chapter 7 bankruptcy are eligible only for Chapter 13 bankruptcy and thus cannot be swayed by their attorney. Second, because Chapter 13 filers are likely to have their cases dismissed, the composition of clients may differ between high and low fraction 13 attorneys or neighborhoods. More specifically, high fraction 13 attorneys may serve a higher fraction of households that are seeking to refile. Both the OLS and IV coefficients are smaller than our baseline estimates at.69 and.67 respectively. This difference is present in large measure, however, for the full 3,878 observations for which we have prior filing information even when we do not exclude prior filers. Additionally, the standard errors in the IV specification are sufficiently large that the baseline Texas estimates lie within the 95 percent confidence interval. In any event, the findings are qualitatively similar to our baseline estimates. Row 4 addresses the fact that new federal legislation went into effect in the fall of The Bankruptcy Abuse Prevention and Consumer Protection Act (BAPCPA) of 2005 made several changes to the bankruptcy code. A major amendment, known as the means test, aimed at making Chapter 7 more difficult. Additionally, the financial and 29

32 paperwork costs of filing for all types of bankruptcy increased. In row 4, we re-run our regressions only using data prior to October This has no effect on the estimates. Next, one may be concerned that the instruments weight each lawyer by number of filings. The number of filings may be endogenous, however. While not immediately clear how this would bias the results, it is easy to address. We reform the instrument without weighting by the number of lawyer filings. Row 5 uses this instrument and finds comparable results. In row 6, we define the law firm by the address of the filing attorney as opposed to the name. Thus multiple attorneys located at the same physical address are treated as a single entity. Both the IV and OLS results are similar to the baselines. The Causes and Consequences of Lawyer Specialization Our theoretical framework suggests that if lawyer specialization is due to fixed investments which lower the marginal costs of filing Chapter 13 bankruptcy, we should find that larger firms, as measured by number of bankruptcies, file a higher fraction of bankruptcies under Chapter 13. In Figure 2, we show how fraction 13 varies with firm size. This testable implication is borne out with startling clarity. There is a strong correlation between firm size and the fraction of bankruptcies filed under Chapter 13. Indeed, firms that file between 10 and 14 bankruptcies per year only file 15 percent of their cases under Chapter 13. On the other hand, firms filing more than 250 bankruptcies per year typically file over 40 percent of their cases under Chapter 13. A regression of fraction 13 on log firm size, controlling for neighborhood characteristics, finds that as firm size increases by 10 percent, the fraction of bankruptcies filed under Chapter 13 30

33 rises by 0.8 percentage points. This coefficient is highly significant with a t-statistic of A second implication of our theoretical framework is that firms with client mixes well suited for Chapter 13 bankruptcy will tend to specialize in Chapter 13. This will be mechanically true if we examine the correlation between fraction 13 and demographic characteristics (as measured by Census block-level covariates). To address this issue, for our base sample, we estimate lawyer fixed effects using the specification outlined in equation (5). A lawyer s empirical propensity to file under Chapter 13, holding constant client composition, is θ j. X j β represents the suitability of the attorney s client mix for Chapter 13 bankruptcy, where X j represents the average demographic characteristics of an attorney s clientele. To test whether lawyers with client mixes suited for Chapter 13 bankruptcy tend to specialize in Chapter 13, we regress θ j on X j β. The coefficient on X j β is 0.5 with a t-statistic of 6. This implies that if an attorney s case mix were 10 percentage points more likely to file under Chapter 13, the attorney will be 5 percentage points more likely to file under Chapter 13, holding constant client characteristics. This is also consistent with our model of attorney specialization. It appears that firms push individuals into Chapter 13 bankruptcy when the costs of doing so are low enough to justify the pursuit of the high payment for Chapter 13 cases. Aggressive lawyers may even induce households to file under Chapter 13 when it is a poor financial choice. Specialization may, however, benefit those households filing under Chapter 13 through lower costs and/or lower dismissal rates. We examine these 24 This finding is consistent with Sullivan, Warren, and Westbrook s (1988) result that attorneys specializing in bankruptcy law file a higher fraction of cases under Chapter 13 than other attorneys. 31

34 possibilities using a subset of our Texas data for which we have information about outcomes. We have information on secured debt, whether the case was dismissed (and thus no debt relief granted), how much the lawyer reported for the filers household expenses, total legal fees, and upfront legal fees. We restrict data on dismissal and legal fees to filings before the 2005 law change. Due to this restriction and the fact that we focus exclusively on Chapter 13 filers, the sample size becomes noticeably smaller. Depending on the particular outcome, we have between 2,600 and 8,790 observations. Using this sample, we regress outcomes on the lawyer fraction 13 and covariates used previously as controls. 25 These results are shown in Table 8. We report both OLS and IV specifications for each outcome, providing progressively more controls in each regression. In the results that follow, it is important to keep in mind that the counterfactual is not the outcome for the household had they filed a Chapter 7 bankruptcy. Rather we compare the outcomes of household who filed a Chapter 13 bankruptcy with an attorney who files many Chapter 13 bankruptcies as opposed to an attorney who files primarily Chapter 7 bankruptcies. Because some attorneys induce households to file under Chapter 13, the types of households found filing for Chapter 13 bankruptcy may differ across attorneys. Thus the coefficients in this table capture both causal effects on a given household as well as a change in the composition of households. The first row examines the impact of going to a high fraction 13 lawyer on whether the household filing under Chapter 13 has less than $20,000 in secured debts. As mentioned earlier, a major advantage of Chapter 13 bankruptcy is the option to retain 25 When examining whether or not the households has secured debt, we control only for block-level covariates to take into account neighborhood composition. The household-level covariates include measures of a household s actual financial situation, which are mechanically linked to the outcome variable. 32

35 collateral. Chapter 13 bankruptcy would typically be an inappropriate choice for households with few secured debts. OLS estimates suggest that households that file Chapter 13 bankruptcy with a high fraction 13 attorney are as likely to possess secured debts as those who file with a low fraction 13 attorney. The IV estimates, however, imply that Chapter 13 specialists file households under Chapter 13, even when their financial situation is poorly suited for this type of bankruptcy. Indeed, increasing lawyer fraction 13 by 10 percentage points is associated with a 3.2 percentage point increase in the probability that households filing for Chapter 13 bankruptcy have few secured debts. Again, the IV results reflect the experience of those individuals who select their attorney on the basis of geographic convenience. Case dismissal means that the plan was not accepted and the household did not receive long-term debt relief. This sample is particularly small as we only consider filings from 2004 and earlier, so that cases had time to be dismissed. OLS estimates suggest households filing for Chapter 13 bankruptcy with high fraction 13 attorneys experience a slightly higher probability of dismissal than households that file with low fraction 13 attorneys. The IV point estimates, though, are noticeably larger and highly significant. Increasing attorney fraction 13 by 10 percentage points is associated with a 3.9 percentage point increase in the probability of dismissal. This suggests that Chapter 13 specialists file naïve households under Chapter 13 bankruptcy even when the probability of long-term debt relief is small. Each bankruptcy filing includes a budget of monthly income and expenditures. The difference between these is what the household must pay into the plan. Thus, controlling for income, households that report a higher set of monthly expenditures may 33

36 end up with smaller payments. While the proposal need not be accepted by the court, they generally are. There is no evidence that log expenditures vary much by lawyer. The point estimates are small and insignificant, though positive. Thus this metric shows no evidence of lawyer specialization aiding the client. The amount and timing of Chapter 13 attorney fees are reported in case filings. As noted above, even when payments are paid through the plan, lawyer fees are paid in the early months of the plan. In row 4 we regress total fees on the lawyer s rate of filing Chapter 13. Lawyers who file many cases under Chapter 13 tend to charge slightly more on average about $50. This is not a large premium since, in the Northern Texas Bankruptcy District, the average Chapter 13 bankruptcy costs about $2,000 before the BAPCPA and $3,000 after. Examining the data, it becomes clear that experienced lawyers are charging the court-accepted maximum rate, while lawyers who file few Chapter 13 bankruptcies sometimes charge less. The only evidence of a benefit being passed on to filers comes in the last row, where OLS results show that Chapter 13 specialists charge substantially less up front than other lawyers. As most bankrupt households are short on funds and credit, this could be a substantial short-term benefit, but has little long-term value given that our random sample of 54 dismissed filings suggests low rates of default on debt to the lawyer. Furthermore, the IV results are both much smaller and statistically inseparable from zero. All in all, there does not appear to be much evidence that lawyers who specialize in Chapter 13 bankruptcies secure better outcomes for their clients. When dealing with naïve households, however, they systematically push clients towards Chapter 13 bankruptcy even when the probability of dismissal is high and the household would 34

37 receive few financial benefits. We cannot rule out, however, that the negative selection effects may mask positive causal effects. Of course, all of this is conditional on the household choosing to file Chapter 13. The simple act of moving a household from filing Chapter 7 to filing Chapter 13 vastly increases their dismissal rate and their fees. Table 9 shows the impact of attorney fraction 13 on the dismissal rate and total attorney fees, independent of chapter choice. The instrumental variables estimates indicate that a 10 percentage point increase in attorney fraction 13 causes a 4.7 percentage point increase in the probability of case dismissal. Although up-front fees decline by over $1,100 dollars, total fees rise by $600 to $700. As mentioned above, our Texas sample suggests that lawyers recover the majority of their fees even when they do not collect up front. The OLS estimates give comparable but smaller results. Since a conventional endogenous matching story suggests the OLS estimate is biased upwards, one interpretation for the larger IV results is that our IV specification targets households that are particularly dependent on lawyer advice. Conclusion Choosing between Chapter 13 and Chapter 7 bankruptcy involves critical tradeoffs central to a household s future financial welfare. While Chapter 13 bankruptcy allows households to keep secured assets, discharge a broader set of debts, maintain better credit in the future, and preserve the option of a future Chapter 7 bankruptcy, most Chapter 13 filings are ultimately dismissed or converted to Chapter 7. Because the 35

38 failure rate of Chapter 13 bankruptcies is so high, it is important to understand what factors drive the household decision regarding bankruptcy chapter. Our analysis confirms earlier qualitative work that attorneys play a crucial role in helping their clients weigh the costs and benefits of each bankruptcy alternative. Ideally, a household would receive the same appropriate advice, regardless of which attorney they consult. Unfortunately, this does not appear to be the case. By far, the best observable predictor of a household s decision to file under Chapter 13 is the attorney they happen to consult. This is true even when we control for case level information regarding a household s assets and liabilities. The results are robust to instrumenting the fraction of cases the chosen attorney files under Chapter 13 with a distance-weighted average of fraction 13 s of attorneys in the area. They are also robust to using a neighborhood fixed effects strategy with a completely different identifying assumption. Moving from the 25 th percentile to the 75 th percentile in the distribution increases the probability that a household files under Chapter 13 by 37 percentage points. Attorney specialization can be explained in part by a model in which firms undertake fixed investments to lower the marginal costs of Chapter 13 cases. However, it seems that the benefits of specialization are enjoyed by the attorney as opposed to the clients. Households that file a Chapter 13 bankruptcy with an attorney specializing in that type of bankruptcy do not have higher success rates, more generous payment plans, or lower attorney fees. Indeed, IV estimates suggest that Chapter 13 specialists steer naïve households into Chapter 13 bankruptcy even when they are objectively poorly suited for this type of bankruptcy and have a high probability of case dismissal. A typical lawyer-instigated switch from Chapter 7 to Chapter 13 leads to a substantial transfer of 36

39 wealth from insolvent households to specialized attorneys with a reduced probability of long-term debt relief. With these findings, our paper joins a growing literature on the nature of principal-agent relationships when the agent possesses specialized knowledge. It would appear that many professionals are willing to sacrifice their clients best interests to maximize firm profits. Due to the high cost and low success rate of Chapter 13 bankruptcy, we have focused our analysis on the adverse consequences of being inappropriately pushed into Chapter 13 bankruptcy. Chapter mismatch can go the other way as well. A firm that has not undertaken investments to lower the cost of Chapter 13 bankruptcy may encourage a household to file under Chapter 7 bankruptcy even when Chapter 13 better suits the households specific needs. For instance, a household may be able to retain an automobile or a house under a successful Chapter 13 bankruptcy, which they would give up under Chapter The data to investigate such possibilities is not available in publicly available court records, however. Additionally, Chapter 7 s low cost and dismissal rate limit the adverse consequences of this type of chapter mismatch for most households. One potential solution to the mismatch between households and bankruptcy chapter is to better align attorney incentives with those of their clients. In particular, Braucher (1993) documents that in some bankruptcy districts with a high fraction of Chapter 13 cases, judges and trustees allow attorneys to charge far more for a Chapter 13 filing than for a Chapter 7 filing. While this encourages the filing of Chapter 13 bankruptcies, it may be justified as Chapter 13 cases are more difficult to file. However, 26 Of course, an unsuccessful Chapter 13 bankruptcy simply delays the date at which the assets must be surrendered. 37

40 because attorneys are often among the first to receive payment from the Chapter 13 plan, the financial benefit of filing under this chapter is largely invariant to the ultimate success of the bankruptcy case. By spacing out the attorney payments more evenly over the course of the payment plan, attorneys would have a smaller incentive to push households into Chapter 13 bankruptcy when the probability of dismissal is high. 38

41 Appendix This appendix provides a formal model that rationalizes chapter specialization. The central insight is that firms can engage in fixed investments to lower the marginal cost of filing under Chapter 13 bankruptcy. According to our model, firms specialize in Chapter 13 if they are relatively large or have a client mix better suited for Chapter 13. Consider a profit maximizing law firm which files n bankruptcies a year. We ll assume that the revenue generated from a Chapter 7 bankruptcy is fixed at r 7. The expected revenue associated with steering households to Chapter 13, however, varies across households based on how long the debtor is expected to service his payment plan. Thus we assume that a firm s expected revenue from filing a particular household under Chapter 13, r13, is characterized by r13 = μ13 + ε, where ε is normally distributed across 2 households with mean, 0, and variance, σ. 27 We emphasize that this equation captures the heterogeneity in expected revenue across households, not variation in realized revenue given a particular expectation, which is unknown at the time of filing. We assume the cost of filing a Chapter 7 bankruptcy, c 7, is fixed across firms and households. The cost of filing a Chapter 13 bankruptcy depends on whether or not a firm undertakes a fixed investment at a cost, F, to lower the marginal cost of filing. 28 Let N c 13 denote marginal costs when the firm does not invest and I c 13 denote marginal costs when the firm invests. Even if firms undertake the fixed investment, Chapter 13 bankruptcies 27 We could have allowed the cost of filing a Chapter 13 bankruptcy to vary across households as well. Doing so, however, adds complexity and yields equivalent predictions. 28 As discussed in the text of this paper, F incorporates investments in technology, personnel, and human capital. 39

42 are more difficult than Chapter 7 bankruptcies they require additional paperwork and N I time. Thus c > c > c7 In essence, this is a two period model. In the first period, the firm decides whether to invest. In the second, a firm decides whether to file each case under Chapter 7 or Chapter 13 given its first period investment decision. A firm will file a case under the chapter which maximizes expected profits. Conditional upon having undertaken the I fixed investment, a firm will file a case under Chapter 13 if r 13 c 13 > r 7 c 7. Define expected net profit of filing a Chapter 13 relative to filing a Chapter 7 as ˆ I 13 = μ13 c I 13 r7 + c7 π ˆ π I 13 Chapter 13 bankruptcy as σ. Given this definition, we can rewrite the condition for filing a ε >. Accordingly, the proportion of cases an investing σ πˆ I 13 firm files under Chapter 13 equals Φ where Φ is the standard normal cdf. The σ πˆ N 13 corresponding proportion for non-investing firms is Φ σ with ˆ π N 13 defined as I expected net profits for non-investing firms. Since c 13 < c N 13, investing firms find it optimal to steer a higher fraction of households into Chapter 13 than non-investing firms. Now we will consider the first period in which a firm must decide whether or not to undertake the fixed investment. If the firm invests, expected profits are given by ˆ π ˆ π ˆ π Π = Φ + Φ + F, σ σ σ I I I I I (A1) n 1 ( r7 c7 ) n μ13 σλ c13 40

43 ˆ π I 13 where λ is the inverse Mills ratio and μ13 + σλ is the expected revenue of the σ Chapter 13 bankruptcy conditional upon the attorney finding it optimal to file under Chapter 13. On the other hand, if the attorney doesn t invest, his profit function is ˆ π ˆ π ˆ π σ σ σ N N N N N (A2) Π = n 1 Φ ( r7 c7 ) + nφ μ13 + σλ 13 c, Thus, a profit maximizing firm will invest if I N Π >Π. This occurs when: (A3) I I I ˆ π ˆ 13 π ˆ 13 π 13 I 1 Φ ( r7 c7 ) +Φ μ13 + σλ c13 σ σ σ N N N ˆ π ˆ ˆ 13 π 13 π 13 N 1 Φ ( r7 c7 ) +Φ μ13 + σλ c13 > σ σ σ F n The first term is the expected revenue minus cost per debtor with the fixed investment and the second is the expected profit per debtor in the absence of this investment. The left hand side of the equation is unambiguously positive because the cost of filing a Chapter 7 bankruptcy is the same while the cost of filing a Chapter 13 bankruptcy has been reduced. Note that as n increases, the fixed cost will be defrayed over a larger number of debtors. Furthermore, there exists an n sufficiently large that firms will find it optimal to undertake the fixed investment. Let s now consider what happens when we change μ 13. As this parameter increases, the attorney s client mix becomes better suited on average for Chapter 13 bankruptcy. By this we mean that filing a household under Chapter 13 becomes more profitable on average. To see how this changes the probability of investment, we can differentiate the left-hand side of equation (A3) with respect to μ 13. After some algebraic 41

44 ˆ ˆ Φ Φ > 0 I N manipulations, this derivative simplifies to π13 π13 σ σ. As an attorney s clients become better suited for Chapter 13 bankruptcy, per client revenue rises more for firms that have undertaken the fixed investment than for those that have not. This suggests that firms with a client mix well suited for Chapter 13 are more likely to undertake the fixed investment than firms with a client mix poorly suited for Chapter 13. According to our model, firms encourage Chapter 13 bankruptcy when the expected revenue minus cost of filing under this chapter exceeds that available under Chapter 7. Furthermore, firms will undertake fixed investments to reduce the marginal cost of filing Chapter 13 bankruptcy when they expect to file a large number of bankruptcies and when their client mix is well suited for this type of bankruptcy. Such firms will be more likely to file any given household under Chapter 13 bankruptcy, even controlling for household characteristics. 42

45 References Allgood, P.C. and M.O. Bachman (2006) Effects of Specialization on Treatment and Outcomes In Screen-Detected Breast Cancers in Wales: Cohort Study. British Journal of Cancer 94, pp Braucher, J. (1993) Lawyers and Consumer Bankruptcy: One Code, Many Cultures. American Bankruptcy Law Journal 67(4), pp Chandra, A. and D. Staiger (2004) Testing A Roy Model With Productivity Spillovers: Evidence From the Treatment of Heart Attacks. NBER Working Paper # Chevalier, J. and G. Ellison (1997) Risk Taking by Mutual Funds as a Response to Incentives. Journal of Political Economy 114(2), pp Culhane, M. B. and M. M. White (1999) Taking the New Consumer Bankruptcy Model for a Test Drive: Means-Testing Real Chapter 7 Debtors. American Bankruptcy Institute Law Review 7 (1). Domowitz, I. and R.L. Sartain (1999) Determinants of the Consumer Bankruptcy Decision. Journal of Finance 54 (1), pp Epstein, A., J. Ketcham and S. Nicholson (2005) The Welfare Effect of Physician Specialization: Do We Want Physicians to Practice Alike? Working Paper. Flynn, E. and G. Bermant (2000) Pre-Bankruptcy Planning Limits Means-Testing Impact. American Bankruptcy Institute Journal 19 (2), pp. 22. Gruber, J. and M. Owings (1994) Physician Financial Incentives and Cesarean Section Delivery. NBER Working Paper #4933. Grytten, J. and R. Sørensen (2003) Practice Variation and Physician-Specific Effects. Journal of Health Economics 22(30), pp

46 Harrington, D. and K. Krynski (2002) The Effect of State Regulations on Cremation Rates: Testing for Demand Inducement in Funeral Markets. Journal of Law and Economics 65(1), pp Imbens, G.W. and J.D. Angrist (1994) Identification and Estimation of Local Average Treatment Effects Econometrica 62(2), pp Lefgren, L. and F. McIntyre (forthcoming) Explaining the Puzzle of Cross-State Differences in Bankruptcy Rates. Journal of Law and Economics. Levitt, S. D. and C. Syverson (2005) Market Distortions when Agents Are Better Informed: The Value of Information in Real Estate Transactions. NBER Working Paper # Li, W. and P Sarte (2002) The Macroeconomics of U.S. Consumer Bankruptcy Choice: Chapter 7 or Chapter 13? Federal Reserve Bank Working Paper # McGuire, T.G. (2000) Physician Agency. in Culyer, A.G., and J.P. Newhouse (eds.) Handbook of Health Economics, North Holland pp Nelson, J.P. (1999) Consumer Bankruptcy and Chapter Choice: State Panel Evidence. Contemporary Economic Policy 17, pp Neustadter, G. (1986) When Lawyer and Client Meet: Observations of Interviewing and Counseling Behavior in the Consumer Bankruptcy Law Office. Buffalo Law Review 35(1), pp Sullivan, T.A., E. Warren and J.L. Westbrook (1988) Laws, Models, and Real People: Choice of Chapter in Personal Bankruptcy. Law & Social Inquiry 13(4), pp

47 Sullivan, T.A., E. Warren and J.L. Westbrook (1989) As We Forgive Our Debtors. Oxford University Press: New York, New York. Sullivan, T.A., E. Warren and J.L. Westbrook (1994) The Persistence of Local Legal Culture: Twenty Years of Experience From the Federal Bankruptcy Courts. Harvard Journal of Law and Public Policy 17(3), pp Sullivan, T.A., E. Warren and J.L. Westbrook (2000) The Fragile Middle Class. Yale University Press: New Haven, Connecticut. Sullivan, A.C. and D.B. Worden (1990) Rehabilitation or Liquidation: Consumers Choices in Bankruptcy. Journal of Consumer Affairs 24(1), pp Tabb, C.J. and J. McClelland (2007) Living with the Means Test. University of Illinois Law and Economics Research Paper # LE White, M.J. (2005) Economic Analysis of Corporate and Personal Bankruptcy Law. NBER Working Paper #

48 Figure 1: Histogram of Chapter 13 Filing Rate Fraction Fraction Chapter 13 Filings Figure 2: Chapter 13 Filing Rate by Annual Filing Rate Average Fraction Chapter 13 Filed Number of Annual Filings 46

49 Variable Table 1: Sample Means Census Block Sample Utah Detailed Sample Texas Detailed Sample Utah Texas California Filed Lawyer Fraction Urban Married Divorced Household of Household of Household of Household of Household of Household over Finished High School Finished college Black Hispanic Other Race Age Below Age 6 to Age 19 to Age 25 to Age 30 to Age 40 to Age 50 to Unemployed Self-Employed HH income Under $10, HH income $10-$20, HH income $20-$30, HH income $30-$40, HH income $40-$50, HH income $50-$60, HH income $60-$75, HH income $75-$100, Fraction Homeowners th Percentile of Log Housing Value th Percentile of Log Housing Value Monthly Income 2,483 3,141 Fraction with no Income Land Assets 94,855 76,156 Fraction with no Land Assets Personal Assets 16,845 31,115 Fraction with no Personal Assets Secured Debts 103,303 82,715 Fraction with no Secured Debts Unsecured Debts 45,936 47,124 Fraction with no Unsecured Debts Observations 122,350 1,965 16,735 47

50 Table 2: OLS Estimates of Lawyer Impact on Chapter 13 Choice (1) (2) (3) (4) (5) (6) Census Census Utah Utah Texas Texas Lawyer Fraction [0.006]** [0.006]** [0.038]** [0.038]** [0.014]** [0.016]** Urban [0.007] [0.053] [0.016] Married [0.025]** [0.188] [0.066] Divorced [0.042] [0.300] [0.099] Household of [0.030] [0.214]* [0.075] Household of [0.035]* [0.255] [0.083] Household of [0.038] [0.254]* [0.087]** Household of [0.044] [0.314] [0.111] Household of [0.047] [0.286] [0.150] Household over [0.059] [0.401] [0.160] Finished High School [0.032] [0.262] [0.066]** Finished college [0.021]** [0.158] [0.061] Black [0.018]** [0.655] [0.024]** Hispanic [0.019] [0.175] [0.040] Other Race [0.017] [0.258] [0.072] Age Below [0.059] [0.407] [0.161] Age 6 to [0.047] [0.365] [0.123] Age 19 to [0.051] [0.345] [0.110] Age 25 to [0.052] [0.349] [0.138] Age 30 to [0.039] [0.312] [0.090] Age 40 to [0.044] [0.336] [0.107] Age 50 to [0.049] [0.475] [0.130] Unemployed [0.068] [0.482]* [0.159] Self-Employed [0.026] [0.188] [0.058] HH income Under $10, [0.042] [0.366] [0.105] HH income $10-$20, [0.037] 48 [0.304] [0.109]**

51 HH income $20-$30, [0.037] [0.243]* [0.104]* HH income $30-$40, [0.034] [0.254] [0.090]** HH income $40-$50, [0.033] [0.247] [0.092] HH income $50-$60, [0.034] [0.255] [0.103] HH income $60-$75, [0.035]* [0.273] [0.093] HH income $75-$100, [0.034]* [0.259] [0.097] Fraction Homeowners [0.011]** [0.078] [0.023] 25th Percentile of Log Housing Value [0.008] [0.100] [0.016]* 75th Percentile of Log Housing Value [0.008] [0.084] [0.019] Log Monthly Income [0.019] [0.009]** Dummy- No income [0.153] [0.072]** Log Land Assets [0.016]** [0.005] Dummy- no Land Assets [0.190]* [0.049] Log Personal Assets [0.009]** [0.004]** Dummy- no Personal Assets [0.076] [0.080]* Log Secured Debts [0.013]** [0.005]** Dummy- no Secured Debts [0.118]** [0.043]** Log Unsecured Debts [0.011]** [0.004]** Dummy- no Unsecured Debts [0.144]* [0.036]** State and Year Dummy Variables No Yes No Yes No Yes Observations 122, ,350 1,965 1,965 16,735 16,735 R-squared Partial R-squared of Chapter 13 Filing Rate Robust standard errors in brackets, clustered at the zip code level. * significant at 5%; ** significant at 1% 49

52 Table 3: First Stage Relationship Between Lawyer Fraction 13 and Distance Weighted Average (1) (2) (3) (4) (5) (6) Census Census Utah Utah Texas Texas Distance Weighted Fraction 13 (Instr.) [0.049]** [0.066]** [0.236]** [0.208]* [0.024]** [0.027]** State and year dummy variables No Yes No Yes No Yes Neighborhood Covariates No Yes No Yes No Yes Personal Filing Covariates No No No Yes No Yes Observations 122, ,350 1,965 1,965 16,735 16,735 R-squared F-Test for Excluded Instrument * significant at 5%; ** significant at 1%. Neighborhood Covariates are those in the first column of Table 2 -- "Neighborhood controls" are those used in the first column of Table 2 -- Urban, Marital Status, Household Size, Education, Race, Age, Employment status, Household Income, Homeownership, and Housing Value. Personal Filing Covariates are the additional covariates in Table 2, Columns 2 and 3 -- Monthly Income, Land and Personal Assets, Secured and Unsecured Debt. Robust standard errors in brackets, clustered at the zip code level. 50

53 Table 4: IV Estimates of Lawyer Impact on Chapter 13 Choice (1) (2) (3) (4) (5) (6) Census Census Utah Utah Texas Texas Lawyer Fraction [0.018]** [0.025]** [0.265]** [0.431]* [0.030]** [0.029]** State and year dummy variables No Yes No Yes No Yes Neighborhood Covariates No Yes No Yes No Yes Personal Filing Covariates No No No Yes No Yes R-squared Partial R-squared of Chapter 13 Filing Rate Observations 122, ,350 1,965 1,965 16,735 16,735 * significant at 5%; ** significant at 1%. Neighborhood Covariates are those in the first column of Table 2 -- "Neighborhood controls" are those used in the first column of Table 2 -- Urban, Marital Status, Household Size, Education, Race, Age, Employment status, Household Income, Homeownership, and Housing Value. Personal Filing Covariates are the additional covariates in Table 2, Columns 2 and 3 -- Monthly Income, Land and Personal Assets, Secured and Unsecured Debt. Robust standard errors in brackets, clustered at the zip code level. 51

54 Table 5: Neighborhood Fixed-Effects Estimates of Lawyer Impact on Chapter 13 Choice (1) (2) (3) (4) (5) (6) Census Census Utah Utah Texas Texas Lawyer Fraction [0.006]** [0.006]** [0.059]** [0.058]** [0.017]** [0.017]** Neighborhood Fixed Effects Yes Yes Yes Yes Yes Yes Year dummy variables No Yes No Yes No Yes Personal Filing Covariates No No No Yes No Yes Adjusted R-squared Partial R-squared of Chapter 13 Filing Rate Observations 122, ,350 1,965 1,965 16,735 16,735 * significant at 5%; ** significant at 1%. Personal Filing Covariates are the additional covariates in Table 2, Columns 2 and 3 -- Monthly Income, Land and Personal Assets, Secured and Unsecured Debt. Robust standard errors in brackets, clustered at the zip code level. 52

55 Table 6: Parameter Heterogeneity Specification OLS IV (1) Utah [0.009]** [0.164]** (2) Texas [0.011]** [0.042]** (3) California [0.011]** [0.027]** (4) Urban [0.007]** [0.027]** (5) Rural [0.020]** [0.084]** (6) Low Income [0.015]** [0.059]** (7) High Income [0.016]** [0.096]** (8) High Minority [0.019]** [0.057]** (9) Low Minority [0.009]** [0.090]** (10) Low Assets (Texas Sample) [0.029]** [0.093]** (11) High Assets (Texas Sample) [0.018]** [0.094]** * significant at 5%; ** significant at 1% All regressions include time and state dummies (where applicable) and neighborhood census block covariates; Urban, Marital Status, Household Size, Education, Race, Age, Employment status, Household Income, Homeownership and Housing Value, but no personal filnig covariates. Robust standard errors in brackets, clustered at the zip code level. 53

56 Table 7: Robustness Checks Exogenous Lawyer Choice Specification Endogenous Lawyer Choice Full Sample Utah Texas Full Sample Utah Texas (1) Lawyer Propensity Regressor [0.006]** [0.040]** [0.017]** [0.029]** [0.629]* [0.076]** (2) Probit Marginal Effects [0.013]** [0.068]** [0.044]** [0.038]** [0.519]** [0.107]** (3) Census Block Fixed Effects [0.010]** [0.058]** [0.016]** (3) No Prior Filings n=2852 [0.032]** [0.136]** (4) Pre-BAPCA law change [0.007]** [0.018]** [0.031]** [0.085]** (5) Instrument not weighted by lawyer's # of filings [0.032]** [0.315]* [0.092]** (6) Firm is defined by address, not lawyer [0.017]** [0.075]** * significant at 5%; ** significant at 1% All regressions include the complete set of covariates used in Table 2, Column 3, (Urban, Marital Status, Household Size, Education, Race, Age, Employment status, Household Income, Homeownership, Housing Value, Stae and Year Dummies, Monthly Income, Land and Personal Assets, Secured and Unsecured Debt) except the "No Prior Filings" row includes only 2004 Texas data. Robust standard errors in brackets, clustered at the zip code level. 54

57 Table 8: Does Lawyer Specialization Improve Bankruptcy Outcomes (Conditional on Choosing Chapter 13) Specification Dependent Variable Bivariate OLS Neighborhood Controls Full Controls Bivariate IV Neighborhood Controls Full Controls (1) Few Secured Debts n= 6959 [.03]+ [.026] [.122]* [.096]** (2) Case Dismissed n= 2600 [.053]** [.051]** [.051]* [.177]** [.168]* [.158]* (3) Log Monthly Expenditure n= 6959 [.034] [.032] [.027] [.069] [.046] [.038] (4) Lawyer Fees n= 4136 [17]** [17]** [17]** [33] [33] [34]* (5) Up Front Fees n= 6959 [30]** [29]** [28]** [99] [86] [84] significant at 10%; * significant at 5%; ** significant at 1%. All data are from Texas-- dismissal regression is only for 2003 and 2004 data; fee regressions are pre-bapca law change. Bivariate regressions include year dummies but no other controls. "Neighborhood controls" are those used in the first column of Table 2 -- Urban, Marital Status, Household Size, Education, Race, Age, Employment status, Household Income, Homeownership, Housing Value, and Year Dummies "Full Controls" Include the Neighborhood Controls plus the personal filing information included in the third column of Table 2 -- Monthly Income, Land and Personal Assets, Secured and Unsecured Debt. Robust standard errors in brackets, clustered at the zip code level. 55

58 Table 9: Does Lawyer Specialization Improve Bankruptcy Outcomes (Unconditionally)? Specification OLS IV Dependent Variable Bivariate Neighborhood Controls Full Controls Bivariate Neighborhood Controls Full Controls (1) Case Dismissed n= 5131 [.025]** [.025]** [.022]** [.081]** [.086]** [.081]** (2) Lawyer Fees n= 8790 [29]** [29]** [30]** [99]** [107]** [101]** (3) Up Front Fees n= 8790 [33]** [32]** [34]** [109]** [112]** [112]** All results are significant at the 1% level **. All data are from Texas-- dismissal regression is only for 2003 and 2004 data; fee regressions are pre-bapca law change. Bivariate regressions include year dummies but no other controls. "Neighborhood controls" are those used in the first column of Table 2 -- Urban, Marital Status, Household Size, Education, Race, Age, Employment Status, Household Income, Homeownership, Housing Value, and Year Dummies. "Full Controls" include the Neighborhood Controls plus the personal filing information included in the third column of Table 2 -- Monthly Income, Land and Personal Assets, Secured and Unsecured Debt. Robust standard errors in brackets, clustered at the zip code level. 56