Does Bankruptcy Identify a Type Of Real Estate Agent or a Stress-Induced Change in Performance?

Real estate brokers and agents play an essential role in housing markets and can have considerable influence on transaction outcomes, as a large empirical literature attests. While agent performance reflects inherent capabilities and the strategies they adopt, significant personal events may be distracting and thereby influence performance, too. This paper offers new evidence on how events associated with periods of personal stress affect real estate agent performance. It identifies the extent to which differences observed for agents filing for bankruptcy reflect stress that changes their choice of clients or how they serve them. We also consider the possibility that bankruptcy signals something more than a singular event since filing for bankruptcy may, at least in part, identify agents who typically conduct their business differently from agents who never file for bankruptcy.¹

The management and corporate finance literatures summarized in the next section provide evidence that some stressful events in individuals’ personal lives affect aspects of their job performance. And while the management and finance literatures examine high-profile positions like CEOs and board members, those studies consider settings in which it is impossible to identify the channels through which stress affects performance. Performance measures for those professionals are broad and only loosely reflect policy decisions as they cannot account for how subordinates and others affect observed outcomes. In contrast, real estate agents work in a setting that ties individual agent effort and ability more closely to observable outcomes.. Agents engage directly with clients and other agents, with the tangible outcomes of these interactions manifesting as observable results in individual transactions. The results of housing market transactions are not indicative of the efforts of numerous unobservable subordinates within a large organization. In this context, the housing market serves as a work environment where there is a more direct correlation between the efforts of individual listing and selling agents and the observable outcomes. The focus on agents in housing markets allows us to better identify the connection between individual effort and ability and observed outcomes.

This study examines real estate agents' performance in a single housing market, focusing on a single county in the Atlanta metropolitan area in order to reduce the effects of unobserved heterogeneity in the sample. The empirical approach controls for property, neighborhood, seller, buyer, and agent characteristics while modeling house price and liquidity as endogenous variables, explicitly taking into account that price and selling time are determined simultaneously in search markets. In addition, we recognize that financial distress may push agents to work with specific types of houses. We control for this possible selection effect by applying a repeat sales price-liquidity framework that allows us to directly compare the performance of bankrupt agents to other agents for a given house that sells several times in the sample period. We also recognize that listing agents and selling agents may have different skills or employ different strategies; we disentangle the effects of bankruptcy for agents on both sides of transactions rather than focusing solely on the listing side.²

Central to our question, significant financial distress, such as bankruptcy, may have separate short-run and long-run implications. To that end, we collect and aggregate the agent's publicly available bankruptcy filing records, not limited by current residence or specific years. We use this background data to identify agents whose business practices or personal lives lead to bankruptcy filings, contrasting with those who has not experienced any personal financial crises.³ Finally, we test for differences in the financial distress associated with bankruptcy and other types of personal stress associated with citations or arrest for legal infractions like hit and run, speeding, driving under impairment, trespassing, theft, battery, assault and other misdemeanors and felonies. Bankruptcy typically is not an unexpected event but rather the result of an extended period of financial distress. In contrast, legal infractions are more likely unexpected events, although the analysis here also allows for the likelihood that violations may reflect certain habitual behavior patterns.

Starting with the probability of property sale, we find that houses listed with agents who ever experience bankruptcy—whether in the past or future—are less likely to sell within any given time frame. A current bankruptcy event significantly reduces the probability of faster sales, but the marginal effects are weaker than those long run effects, and they fade to insignificance the longer the time frame considered. Turning to price and liquidity effects, properties listed with agents who ever file for bankruptcy sell at a lower price and take longer to sell. Similarly, properties sold by other agents bringing buyers to the table, who have also ever filed for bankruptcy, are sold at a discount and take longer to sell.

Singular bankruptcy events, however, have different effects on agent performance. For listing agents, concurrent bankruptcy filing has no effect on price but significantly affects the time it takes to sell the property. In contrast, selling agents bringing buyers to the transaction exhibit no systematic price or liquidity effects from current bankruptcy events. We also apply repeat sales analysis to control for any tendency by agents experiencing bankruptcy to focus on certain types of houses and find significantly longer selling times for listing agents who ever file for bankruptcy and lower prices for selling agents who ever file for bankruptcy. In the repeat sales framework, neither listing nor selling agents exhibit significant contemporaneous bankruptcy effects on either price or liquidity. These results indicate that the full sample estimates are to a large extent driven by the tendency of agents who experience bankruptcy to focus on certain types of properties during periods of stress. Taken together, the results imply that bankruptcy both signals a certain type of agent with particular business practices unrelated to stress as well as agents experiencing temporary periods of stress that negatively affect their performance in the short run.

As a point of comparison, we also examine agents' public records for motor vehicle violations (license or traffic violations, hit and run, speeding, or driving under impairment), trespassing, theft, assault, felony aggravated assault, and other crimes to consider whether the nature of the personal event matters. These events are stressful periods that, unlike bankruptcy, are not tied to significant financial issues. The question is whether different stress events in an agent's personal life have different effects on their work performance. We find that the effects do differ from the bankruptcy results, but the broad implications are similar. Like bankruptcy, the types of crime records considered here signal agents who generally conduct their business differently from other agents, both selecting different types of houses to sell and obtaining lower prices and liquidity for those properties. Comparing the full sample and repeat sales results also reveals that concurrent events elicit an additional adjustment in the types of clients or houses agents choose to work with, but not how they service their clients once selected.

Studies of real estate agent performance generally cover three aspects of agent productivity: what agent characteristics like experience, education and sex reveal about underlying inherent productivity (Bian et al., 2023; Pham et al., 2022); how agent strategies like concentrating in listings, sales, or geographic specialization affect performance (Anderson et al., 2014; Turnbull & Dombrow, 2006, 2007); and the role of incentives reinforcing or ameliorating principal-agent issues (Levitt & Syverson, 2008; Rutherford et al., 2005; Munneke et al., 2015; Turnbull, et al., 2022).

There is substantial literature outside the real estate context looking at how significant personal events other than financial shocks influence individuals' behavior and economic decisions in a variety of settings. An extensive corporate finance literature focuses on a CEO's personal life and firm outcomes: Yermack (2014) considers vacation trips, Bennedsen, Pérez-González, and Wolfenzon (2010, 2012) focus on family deaths and the CEO's or immediate relatives' hospitalizations, whereas Bernile et al. (2017) examine CEOs' early-life exposure to fatal disasters. In terms of mutual fund managers, Liu et al. (2022) study whether the bereavement of the manager affects performance and risk-taking behavior.

Among significant events, marriage enjoys particular attention. Building on Becker (1973), Korenman and Neumark (1991) and Bellas and Toutkoushian (1999) provide empirical evidence that married men are more productive than single men in terms of income and publishing papers, respectively. Chun and Lee (2001) probe more deeply into the reasons underlying the greater productivity and conclude that more productive men do not self-select into marriage; instead, their productivity increases by exploiting the increased specialization possible in family settings. Along similar lines, Wheatley et al. (1991) report survey evidence indicating that employee productivity declines after divorce. Neyland (2020) looks at divorce as an event rather than an enduring state and concludes that CEOs engage in less risk taking in the year of their divorce, which is reflected in lower risk for their firms. Lu et al. (2016) focus on the limited attention of hedge fund managers around their marriages and divorces. Viewing such events as distractions, they conclude that inattentive managers exhibit inferior performance relative to less distracted managers.

Another line of work examines the consequences of personal financial stress on worker performance. Maturana and Nickerson (2020) show that teacher bankruptcy is associated with lower student performance on standardized tests. They do not observe the productivity of individual workers, rather, they observe student performance at the campus grade level. Their main concern is to separate the effect of the financial distress of workers from the firm's underperformance. To do so, they argue that the public school setting has advantages over other environments for such a study because it is an environment with very stable employment, thus alleviating concerns that firm underperformance contributes to worker bankruptcy. Nonetheless, student performance reflects more than the single teacher's efforts, unlike the real estate agent case, where differences in transaction outcomes largely reflect the efforts of the individual agents directly engaged in the selling and closing process.

Other studies look at declines in housing wealth during the Great Recession as possible sources of personal stress. For example, Pool et al. (2019) investigate how shocks to fund managers' personal wealth translate into risk-taking in mutual fund portfolios. They collect data on personal real estate holdings during the housing market collapse for their sample of fund managers. They then compute the percentage change in wealth due to housing, defined as the dollar change in the value of a manager's home during 2007–2008 divided by their estimated wealth. Managers in the sample are assigned to the treatment group if their wealth change is below a given threshold. They show that a negative housing wealth shock prompts fund managers to reduce risk due to concerns about their careers. While the paper focuses on shocks to real estate assets, it also suggests that there are many other life events that can shock personal wealth, such as divorce or a personal lawsuit, which would have similar effects on delegated risk-taking.

Dimmock et al. (2021) find that financial advisors increase misconduct following declines in their home values. They use a panel of advisors' home addresses and examine within advisor variation relative to other advisors who work at the same firm and live in the same ZIP code. They observe a negative relation between housing returns and misconduct. The results are stronger for advisors with lower career risk from committing misconduct and for advisors with greater borrowing constraints.

Bernstein et al. (2021) use changes in housing wealth experienced by innovative workers during the financial crisis to compare individuals working at the same firm and living in the same metropolitan area who experience different housing wealth declines during the Great Recession. They find that workers who experience a negative shock to housing wealth are less likely to successfully pursue innovative projects, particularly high-impact, complex, or exploratory projects.

There is little empirical evidence concerning how stressful events affect the performance of key intermediaries in one of the major asset markets: real estate agents. It is this task to which we now turn.

The data are drawn from several sources. The real estate transaction data are from multiple listing service (MLS) records for Gwinnett County in Georgia over 2004–2019. The MLS dataset provides listed, sold, expired, and withdrawn houses with various property and location characteristics, listing and selling prices, and identifies the agents involved in each transaction. We consider only properties for resale and include houses that are at least two years old to avoid pricing effects that are unique to new construction (Munneke et al., 2015, 2019). Similarly, we exclude properties identified as agent-owned or agent-related to avoid the unique aspects of that market segment (Levitt & Syverson, 2008; Rutherford et al., 2005; Turnbull et al., 2022). Using several indicator variables and property descriptions, we carefully select only sales transactions of single-family detached and single-family townhouses and townhome condominiums. We also cull obvious errors. Liquidity, the standard indicator of difficulty of sale, is measured by time on the market (TOM), the difference between the reported off-market date and the listing date plus one. We also exclude outliers in the upper or lower 1% of the distributions of the observed sale price or time on the market.

The individual agent information is drawn from several sources. We retain only transactions with either the listing or selling agent operating as a full-time agent, which we define as handling at least 3.5 transactions per year on average. If the agent working with the other side of a transaction does not meet this criterion, we define that agent as a part-time agent. We exclude transactions by companies buying and selling real estate through technology (e.g., iBuyer) and those assisted by discount brokers⁴ because the different business models may have as yet undetermined effects on prices and liquidity (Buchak et al., 2020a, 2020b). We supplement the MLS data with active and non-active agent license records from the Georgia Real Estate Commission (GREC). GREC provides agents' full names and addresses, which we use to identify residence locations.⁵ If not reported in the above sources, we also find each full-time agent's age and full name or address using public background reports from the BeenVerified.com website.

For each full-time agent, we perform a manual background search via the BeenVerified.com website. BeenVerified provides a directory with access to an individual's date of birth, address, phone, employment histories, criminal and court records, property records, and other public records, which can be searched by name within a specific city, county, and state as well as nationwide. To test for differences in the financial distress associated with bankruptcy and other types of personal stress, we focus on two types of events – bankruptcy filings and criminal records.

For bankruptcy, we collect and aggregate each full-time agent's public bankruptcy filing records, not limited by their current state of residence or specific years. Each record contains the case number, filing and discharge date, bankruptcy chapter, and filing type. We focus on the filing date. Significant financial distress, such as bankruptcy, may have separate short-run and long-run implications. Therefore, we construct several indicators for a listing or selling agent experiencing bankruptcy anytime during their life (before or after the house transaction), LA_ever_bankrupt and SA_ever_bankrupt respectively, as well as experiencing bankruptcy within 6 months of the transaction, LA_bankruptcy_event and SA_bankruptcy_event.⁶

We also draw from all available public criminal records for each full-time agent to assemble alternative personal distress variables. These records typically contain the case number, criminal offense date and details, court date, and other relevant details. Most of the records for our sample of agents are associated with motor vehicle violations (license or traffic violations, hit and run, speeding, or driving under impairment), trespassing, theft, assault, resisting arrest and other misdemeanors while felony assault and other serious charges are rare. Following the same approach used for bankruptcy variables, we construct indicators for agents with criminal offenses occurring at any point of their life or within six months of the house transaction. The resultant data set contains information about agent characteristics, including the number of completed transactions, their residence location, and our financial and personal distress measures.

Buyer and seller data are drawn from Georgia Property Tax Assessor data, spatially merging these data with the MLS data using property spatial coordinates and transaction date. We select the name of the first individual reported as involved in each side of the transaction to determine whether a financial institution or a company is involved in the transaction as a direct buyer or seller.

Following standard practice, we include several socio-economic neighborhood controls in the models. We obtain census tract annual data related to age, education, and median household income from 2010–2020 American Community Surveys.⁷ We use the consumer price index (CPI) from the Bureau of Labor Statistics to express all monetary values in 2010 dollars. ZIP code fixed effects in all models control unobserved neighborhood characteristics. All models include year-quarter fixed effects as well. The resulting data cover 2004 through 2019. The ending date avoids having to deal with idiosyncratic pandemic effects in the housing market. The MLS and property data merge yields 125,201 listings, 82,817 completed transactions, and 12,648 repeat-sales transactions.

Table 1 defines key variables used in the analysis and Table 2 reports summary statistics. A complete list of property and neighborhood characteristics included in all estimating models but not reported here is available in the Appendix, as are complete sample summary statistics. As reported in the available appendix, the average house size is 2,457 square feet of the finished living area. The average listed property has 3.82 bedrooms, 2.5 bathrooms and is almost 18 years old. Recall that we exclude sales of all new construction. The average list price for the sample period is $211,974. It is important to note that our data is drawn exclusively from properties located in Gwinnett County, GA, an inner suburban county in the Atlanta MSA. As a result, our sample is likely to include newer and less densely situated housing units than those found in older central city neighborhoods in Atlanta.

Turning to the key variables reported in Table 2, almost 18% of listings are represented by agents who have or will at some point file for bankruptcy, LA_ever_bankrupt, but less than 1% are listings offered by agents who filed for bankruptcy within six months of the transaction date, LA_bankruptcy_event.⁸ Even smaller percentages of sold transactions are associated with agents who have experienced other personal stress related to being arrested for crimes or major driving violations. In terms of agent activity, about 34% of selling agents are identified as part-timers, defined as agents selling fewer than 3.5 houses on average during years they are active, who account for 18% of sold listings in the sample.⁹

Table 3 presents probit marginal probability estimates of how listing agent characteristics affect the likelihood of the house selling within the time frame indicated at the head of each column. Models (1)-(6) include a variable identifying properties listed with agents currently going through bankruptcy (LA_bankruptcy_event), and models (7)-(10) include the additional variable for agents who ever file for bankruptcy, whether in the past or future (LA_ever_bankrupt). Investor sellers and properties being sold after foreclosure have lower probabilities of quicker sales—results that are not surprising. Looking at the estimates reported across columns, we see that part time agents exhibit steadily increasing marginal probabilities as the time frame expands, indicating that they tend to take longer to sell. High-volume listing agents have negative marginal probabilities that decline with longer time frames, which is consistent with those sales being less likely to occur within six months. Properties in the listing agent's own neighborhood are more likely to sell earlier, and properties listed with multiple coagents are only modestly more likely to sell later.

Looking at the main variables of interest, houses listed with agents who are currently going through bankruptcy are somewhat less likely to sell earlier, although the results reported in the separate columns indicate that the negative marginal probability of sale declines with longer time windows. This pattern is stronger in models (6)-(10) that introduce controls for agents who ever file bankruptcy, illustrating the importance of controlling for this aspect of bankruptcy experience as well. Finally, houses listed with agents who ever file bankruptcy are less likely to sell early, although the negative marginal probability declines for longer windows. The results suggest what we suspected at the outset, that filing for bankruptcy both signals a certain type of agent and indicates a period of elevated stress that changes agent behavior.

Considering completed transactions, note that the housing market is a search market in which selling price and time on the market or liquidity are simultaneously determined for each given property, reflecting the interplay of the distributions of heterogenous buyers, sellers and properties listed for sale during the same time frame (Krainer, 2001). One consequence of empirically modeling search market equilibria is that price and liquidity are determined by the same factors. To consider this, we use the empirical price-liquidity model based on the generalized seller search framework presented by Turnbull and Zahirovic-Herbert (2012).¹⁰ This discussion only summarizes the approach developed there.

Consider a house with characteristics vector X, which includes physical property characteristics as well as its use as a rental property, i.e., seller is an investor (Turnbull & van der Vlist, 2022), whether currently vacant, and location. The seller of a given house with characteristics vector X sets a reservation price with the usual stopping rule: sell if an arriving buyer's offer exceeds the reservation price otherwise, do not sell and wait for the next buyer who will make an offer. A higher reservation price eliminates possible lower offers by buyers, thereby yielding a higher expected selling price E[SP] for a house that eventually sells. Eliminating potential buyers, however, also reduces the probability of an acceptable buyer arriving during any given time interval, thereby lowering liquidity or increasing expected time on the market E[TOM]. The seller, therefore, confronts the sales-opportunity constraint taking these relationships into account, summarized here by the implicit function the shape of which reflects the distribution of types of buyers in the market (in terms of willingness-to-pay) as well as the number and characteristics of competing houses in the market at the same time as the subject property, included in the vector M. Constraint (1) spells out how the seller's choice of reservation price translates into possible selling price and liquidity outcomes determined by market conditions. The constraint also includes the vector of agent characteristics A to allow for possible real estate agent influence on sellers and buyers in the search and matching process affecting potential price and liquidity outcomes.

To complete the model, seller utility is an increasing quasiconcave function of expected selling price and a decreasing function of expected holding cost or the cost of waiting for the sale to occur, \(U\left(E\left[SP\right],hE\left[TOM\right]\right),\) where h seller holding cost per unit of time. The functional form or shape of the utility function reflects the seller's rate of time preference, degree of risk aversion, and other attributes influencing the seller’s willingness to trade of liquidity and selling price. The seller's problem is to choose the strategy (i.e., reservation price) yielding the expected price and holding cost that maximize utility subject to the market determined price-selling time constraint, which is equivalent to choosing the expected selling price and expected liquidity to maximize utility,

This format makes it easy to draw the connections between standard consumer demand theory and the generalized search theory to motivate the empirical approach taken here. Intuitively, the seller's optimal strategy yields the resultant expected selling price and liquidity that satisfy the two necessary conditions for the maximization problem (2), the tangency between the indifference curve defined by U(.) and the market constraint summarized in the market-determined price-liquidity trade-off \(\Phi = 0\):

Applying the implicit function theorem to solve the above optimization conditions simultaneously for the expected selling price and expected liquidity yields

These are reduced form equations that describe the equilibrium expected price and liquidity transaction outcome.¹¹

Adding error terms to (5) and (6), the observed selling price and liquidity for house i sold at time t are E[SP_it] + u_it and E[TOM_it] + v_it, respectively. Assuming semi-log functional forms for (5) and (6), the observed selling price and liquidity outcomes for house i sold at time t are

The generalized search model therefore yields an empirical framework describing the two dimensions of the equilibrium outcome for the transaction of property i sold at time t as functions of property characteristics, market conditions, and agent characteristics. As in standard consumer theory, these reduced form equations are functions of the same variables. In addition, since the reduced form equations are the simultaneous solution to the same set of equations, the errors may be correlated across equations.

Market conditions M_it include time fixed effects to capture market-wide conditions and neighborhood market conditions measured by competition from surrounding houses for sale at the same time as the subject property. This last variable measures the number of competing houses on the market each day the subject property is for sale, defined as the listing density LD. Following Turnbull and Dombrow (2006) and others,¹² the listing density is calculated as the distance-weighted number of competing houses for sale within 1 mile and within 20% of living area of the subject property per day the subject property is on the market. Substituting these terms into M yields the empirical framework describing the two dimensions of the equilibrium outcome for the transaction of property i sold at time twhere the T_it are time period (quarter) fixed effects. We estimate this reduced form system using seemingly unrelated regression (SUR) to obtain consistent and asymptotically efficient error estimates that take into account possible cross equation correlation in the error terms.

Table 4 reports key parameter estimates for the SUR model (9)-(10); the complete set of parameter estimates are reported in the appendix. We begin by noting that agent characteristics coefficient estimates unrelated to bankruptcy are robust across the 3 models reported. Transactions involving agents who are 65 and older have significantly higher prices but less liquidity. Selling price and liquidity are significantly affected by part-time listing agents when compared with full-time agents. Part-time listing agents yield lower selling prices and take longer to sell. On the selling agent side (agents who bring buyers to the transaction), part-time agents are also working with houses that are on the market longer. The variables indicating transactions completed by agents in their home neighborhoods exhibit significantly positive coefficients in the price equations for both listing and selling agents, indicating that they tend to obtain higher prices for houses near their own residences. Listing agents in this case, however, do not have a significant effect on liquidity, while selling agents are associated with significantly less (at the 10% level) time on the market for houses in their own neighborhood. The lnLA_Vol and lnSA_Vol variables measure the number of completed transactions on both listing and selling sides during the previous 12 months. These variables capture one aspect of experience (Gilbukh & Goldsmith-Pinkham, 2023) but for listing agents in particular, indicate the agent's inventory of listing clients over the preceding 12 months. The estimates indicate that higher volume listing agents tend to take longer to sell and sell at lower prices, which is in line with existing evidence that agents with larger inventories of listings essentially face greater marginal effort cost of servicing the larger number of listings and so end up spreading their effort more thinly across clients (Bian et al., 2015).

The Dual_agent coefficient estimates show that individual listing agents who also bring the buyer to the transaction sell at lower prices and less liquidity. On the other hand, houses that have more than one listing coagent sell more quickly and at higher prices than those represented by single listing agents. This may be attributable to coagents' abilities to exploit their comparative advantages by specializing in different phases of the listing and selling process identified by Turnbull and Dombrow (2007). Listings of agents who adopt farming, or geographic concentration, as a strategy exhibit faster selling times, and but no consistent price effects across the models. Listing agents with a larger share of local market listings obtain higher prices and greater liquidity in all of the models.

Turning to the variables of central interest, the LA_ever_bankrupt and SA_ever_bankrupt coefficients in model (1) in the table show similar effects of agents who ever file for bankruptcy, whether listing or selling agents. These agents obtain lower prices and take longer to sell the property. In model (2), we see a difference in listing and selling agent effects when considering bankruptcy events around the time of sale; the LA_bankruptcy_event coefficients show listing agents obtain lower prices and take longer to sell, and these effects are significantly greater than found for agents who ever file for bankruptcy, but the SA_bankruptcy_event coefficients show selling agents who have just filed for bankruptcy exhibit no significant price or liquidity effects. Model (3) incorporates both types of bankruptcy variables, ever bankrupt and bankruptcy around the time of sale. The ever bankrupt effects on price and liquidity remain virtually unchanged when compared with model (1); introducing the bankruptcy event control does not matter in this case. On the other hand, the bankruptcy event variable itself for the listing agent is now insignificant in the price equation, although the significantly slower sale or lower liquidity effect still pertains. As before, the bankruptcy event for selling agents bringing buyers to transactions has no significant effect on either price or liquidity.

The estimates reported in Table 4 reveal performance differences between agents who ever file for bankruptcy and those who do not, indicating that they pursue their professional duties differently. One question is whether this difference in their approaches means they choose to work with different types of houses or simply that they adopt different selling behaviors when selling the same mix of houses. We use a repeat sales approach to remove possible house selection effects for bankrupt agents, allowing us to examine differences across the different agents selling the same house repeatedly over time.

The repeat sales approach in the price-liquidity framework is new. The empirical model follows directly from differencing the price-liquidity Eqs. (9)-(10) for properties sold at time t + s and t in the sample period, which yields

The first right hand side terms in Eqs. (11)-(12) capture the two periods the repeated sales occur, and in that way resemble the standard single-equation repeat sales price model. The additional listing density variables capture differences in localized competition between the two periods, accounting for changes in neighborhood-specific market conditions. The final vector differences ΔA = A_it+s—A_it capture the focus of our interest in this study, agent characteristics impact on price and liquidity. We estimate the system using SUR to control for cross equation correlation of error terms.

Table 5 reports the key parameter estimates for the bankruptcy-related variables alone; full estimates are available by request. These results differ from the standard price-liquidity model results reported in Table 4. In particular, in models (1) and (3) in the table, listing agents who are ever bankrupt no longer obtain significantly lower prices, although they do continue to take longer to sell their listed properties. In the repeat sales models (2) and (4), bankruptcy events have no effect on listing agent performance in terms of either price or liquidity. In contrast, the selling agent side shows no significant effect from selling agents who ever declare bankruptcy but significantly lower selling prices by selling agents currently or recently going through bankruptcy. Drawing the full sample and repeat sales sample results together, it appears that filing for bankruptcy identifies listing agents who conduct their business activities differently. The price effect observed in the full sample comes from these agents choosing to work with lower priced houses in the market, but the full sample liquidity effects carry over into the repeat sales sample, which indicates that their way of conducting business leads to slower sales when compared with agents who do not file for bankruptcy. On the selling agent side, those agents who ever file for bankruptcy also appear to work with lower priced houses but do not appear to be doing business differently in other respects. Selling agents currently going through bankruptcy, however, tend to obtain lower selling prices, reflecting that such periods of temporary stress work in their buyer's favor.

A question remains: does the nature of the stressful event matter? Bankruptcy is a large event signaling a period of substantial financial distress. Therefore, we consider other personal events that, though stressful, are not tied to the possibility of long running financial issues. To that end, we consider criminal records for agents' traffic violations (hit and run, speeding, driving under impairment), or arrests for trespassing, theft, battery, assault and other misdemeanors and felonies. The question is whether these events in an agent's personal life affect their work performance differently than observed for bankruptcy.

The analysis follows the bankruptcy analysis. We define two types of event variables for listing and selling agents, if they ever have crime records (LA_ever_crime and SA_ever_crime) or if they have criminal records within six months of the house transaction (LA_crime_event and SA_crime_event).

Table 6 reports coefficient estimates for key variables for the price-liquidity models using the full sample; Table 7 reports coefficients for the crime record variables using the repeat sales models. Aside from the crime record variables, the agent-related variables reported in Table 6 resemble those for the full sample with the bankruptcy variables in Table 4. The crime record variables, however, do exhibit some differences when compared with the bankruptcy results. The listing agent and selling agent ever crime record coefficients in (1) and (3) in Table 6 indicate that houses represented by these agents sell at lower prices and liquidity. The crime event variables in (2) and (3) show no listing agent effect but a persistent negative selling agent price effect. Interestingly, in both cases, there are no significant crime event liquidity effects.

Referring to the repeat sales models that control for systematic differences in agents' choice of houses to list or sell, Table 7 reports significant negative price and liquidity effects for listing agents who ever have a criminal record, indicating that the criminal record is a signal of agents who conduct their business differently. The difference between the observed full sample and repeat sample effects indicates that some of this difference in behavior comes from those agents choosing to work with lower priced houses but expending less effort on selling houses so that they sell at a discount and take longer to sell. Selling agents show a different pattern. The price and liquidity effects of these agents ever having a crime record observed in the full sample largely disappear (except for a marginally significant liquidity effect) in the repeat sales sample, indicating that the effects observed in the full sample are tied to the differences in the houses these agents choose to present to their buyer clients. The significant criminal report event negative price effect found in the pooled sample disappears entirely in the repeat sales sample, suggesting that the event price effect reflects a change in the type of house sold by these agents rather than a stress induced short term modification in effort.

Drawing the results together, criminal record effects do differ from bankruptcy effects, but the broad implications are similar. Like bankruptcy, records of legal infractions appear to identify agents who generally conduct their business differently from other agents, both in the choice of houses they sell and the prices and liquidity they obtain for them. Comparing the full sample and repeat sales results also reveals that concurrent events elicit an additional adjustment in the types of clients or houses agents choose, but apparently not in how they service the clients with whom they work.

Finally, we re-estimate the full sample (Table 8) and repeat sample models (Table 9) including both bankruptcy and legal infraction variables. Reviewing the estimates, while sorting through the expanded number of coefficients becomes a bit more daunting, it is clear that the conclusions from the analysis treating these variables separately do not fundamentally change when both are included together in the same models. This is not surprising given the weak covariation between bankruptcy and legal infraction variables.

The growing literature dealing with real estate agent performance in the resale housing market focuses on agent characteristics like experience, education and sex, strategies like concentrating in listings, sales, or geographic specialization, and the structure of incentives. This paper begins to fill in an overlooked factor, how aspects of agents' private lives affect their job performance. We specifically consider events indicating periods of stress indicated by bankruptcy filing and crime records, adopting an empirical approach that allows us to identify the extent to which these events identify agents who normally conduct their business differently and to what extent these events identify temporary periods of personal stress affecting job performance.

Properties listed with agents who ever experience bankruptcy in the past or future are less likely to sell within a given time frame. Current bankruptcy events also reduce the probability of faster sales, but the marginal effects are weaker than the long run effects and fade with longer time horizons. The price and liquidity effects are more instructive. Houses listed with agents who ever file for bankruptcy sell at a lower price and take longer to sell. Similarly, properties sold by other agents bringing buyers to the table, who have also ever file for bankruptcy, sell at a discount and take longer to sell. In contrast, listing agents going through bankruptcy exhibit no significant effect on price but take significantly longer to sell their listed property. On the other hand, selling agents going through bankruptcy exhibit no significant price or liquidity effects beyond the long term effects identified earlier.

We introduce a new price-liquidity repeat sales approach in order to control for any tendency for agents with bankruptcy experience to focus on certain types of houses. The selection effect turns out to be important. The price discount associated with the type of listing agent who ever experiences bankruptcy reflects their tendency to focus on lower priced houses. In contrast, selling agents with bankruptcy experience do not tend to concentrate on lower priced houses; the performance differences observed for selling agents largely reflect temporary behavior changes when experiencing bankruptcy. Overall, our results indicate that bankruptcy both signals a certain type of agent with particular business practices as well as agents changing their behavior during temporary periods of stress.

We also consider whether the type of stress in agents' personal lives matters. To address this question, we search agents' public records for charges related to hit and run, speeding, driving under impairment, trespassing, theft, battery, assault and other misdemeanors and felonies to consider whether these types of stressful personal life events also affect productivity. Although stressful events, unlike bankruptcy these are not necessarily associated with significant longer running financial issues. Nonetheless, it turns out that records of legal infractions and arrests and bankruptcy have similar general implications for agent productivity. Like bankruptcy, the pooled sample and repeat sales estimates indicate that listing and selling agents with criminal records in the past or future generally choose lower value houses to sell and obtain even lower prices and liquidity. Listing agents do not exhibit concurrent crime record effects beyond property selection; selling agents, however, do obtain lower prices for selected properties.

In general, both bankruptcy and legal infractions affect how real estate agents do their jobs. This conclusion adds to the broader management and finance literature looking at how a variety of personal stress events affect job performance. In addition, though, we are able to begin sorting out the role of these events as signals indicating certain types of agents who normally conduct business differently versus changes in behavior during temporary periods of increased personal stress. The differences observed in these agents' productivities reflect differences in the agents and the event itself. Nonetheless, the signaling role of these events appears to be the major component at work.