Brookings Institution Press

The last decade has not been kind to industries built on traditional models for providing agent-driven services to consumers. Between 1997 and 2006, employment in travel agencies fell by more than 35 percent. As late as the mid-1990s, commissions to trade a 100-share block of equity stock ranged from $75 to $150, but they have now dropped below $25, even at full-service brokerages, and they are lower still at online-only trade sites. The share of automobile insurance policies written directly to customers (thus bypassing insurance agents) rose by almost 70 percent between 1997 and 2005. These changes were not driven by secular declines in demand for the underlying services that the agents brokered. Domestic enplanements by paying passengers rose 24 percent over the past decade. Equities trading volume is at near-record levels. The total dollar value of auto insurance policies written rose by almost 50 percent between 1997 and 2005.1

The diffusion of the Internet and associated electronic-commerce platforms has been cited as a likely explanation for the contractions. Whether by making it easier for consumers to find lower-cost agents or allowing consumers to bypass agents altogether by creating cheaper distribution mechanisms (a process known as disintermediation), such technologies have created difficulties for agent-based [End Page 47] industries. The Internet provided airlines a low-cost means of distributing tickets via their own online sites or travel search engines, leading them to stop paying commissions on ticket sales by travel agents. Commission revenue fell from approximately $12 billion in 1997—60 percent of travel agency revenues—to essentially zero by the end of 2002.2 E-commerce facilitated the entry of online trading centers like E*Trade and the growth of discount brokerages with tech-heavy distribution platforms like Charles Schwab. The Internet lowered the consumer's cost of obtaining multiple quotes for an insurance policy and made it easier to use the direct policy-writing business model (employed by GEICO, for example). Whatever the specific mechanism, agency industries saw their consumers become less reliant on them to facilitate a match with the final service—airline tickets, stock trades, or insurance policies—presumably raising the utility of consumers in the process.

Residential real estate agents are a notable exception to the trend. Membership in the National Association of Realtors (NAR) nearly doubled between 1997 and 2006. Bureau of Labor Statistics data on employment of real estate agents (defined more broadly to include agents who are not NAR members) show a similar increase. Annual new and existing home sales increased almost 50 percent from 1997 to 2006, and the median sale price rose 42 percent. Although in some surveys commission rates appear to have declined slightly in the past five years, the median commission (the rate multiplied by the sale price) rose 26 percent over 1997–2005 due to rising sale prices.3

Why has the real estate agent industry been so much more successful in preserving its position than many other agent-based consumer industries? One explanation is that the same electronic commerce innovations that appeared in other sectors were not present in real estate. Yet that is inconsistent with even casual observation. Those involved in the market have turned to the Internet in multiple ways, greatly expanding consumers' access to residential real estate information. A large fraction of homes for sale are listed on the Internet, complete with detailed house specifications, virtual tours, and neighborhood profiles. Potential buyers can easily peruse dozens or even hundreds of listings, eliminating less appealing possibilities without ever taking the time to visit a house. Those interested in selling, buying, or simply holding residential real estate are now able to review public records of sales, ownership, and taxes, among others. Indeed, a 2005 survey indicated that 77 percent of recent buyers had used the Internet in the process of buying their [End Page 48] home (up from 2 percent in 1995), and 24 percent first found the home that they eventually bought on the Internet.4

A second possibility is that even though a substantial number of e-commerce innovations have taken place in the industry, housing has attributes that keeps those innovations from having as much of an impact as on travel services, stock brokerage, and insurance markets. Namely, housing is a big-ticket, highly heterogeneous item; perhaps that preserves an important role for a tightly knit agent-client relationship. Yet Emre, Hortaçsu, and Syverson (2007) has shown that e-commerce has impacted the structure of the new automobile sales industry, and autos are similarly big-ticket, heterogeneous goods (although, of course, not as big as housing). Therefore, while we cannot definitively rule out differences in product attributes as an explanation of the differences between real estate brokerage and other agency industries, it is not an obvious explanation for those differences.5

A third explanation for the relative strength of real estate agencies is that, compared with other agency-based industries, real estate was initially more competitive. If real estate agent commissions have always reflected marginal costs, if those costs have been stable or rising, and if agents continue to provide valuable services to customers, then there would be no reason for commissions to fall. Certainly entry into the industry is easy, given the low licensing barriers and the industry's ability to accommodate part-time agents. That satisfies one condition of a competitive market. However, as Hsieh and Moretti (2003) documents, the fact that commission rates exhibit so little variation across homes within or across markets implies that the marginal cost of selling a home must rise proportionately with the home's market value; that strikes many observers as implausible.6

A fourth possibility is collusion on the part of real estate agents. Such behavior was alleged in a recent suit brought by the Department of Justice (DoJ) [End Page 49] against the National Association of Realtors for their Virtual Office Website policy.7 It also is consistent with agents' efforts to institute and preserve state laws that either outlaw commission rebates (effectively making it illegal for agents to lower the price that they charge for their services) or establish minimum service requirements (forcing clients who would prefer low-cost, low-service real estate services to pay for more services than they wish). Finally, it is consistent with the alleged retaliatory conduct of traditional agents toward agents who operate a discount (for example, a flat-fee or rebate-based) agency.

If collusion does explain the difference between residential real estate and the harder-hit agency industries, the explanation begs the question of which factors facilitate collusion in real estate that are not present in the other agent-based business. An obvious difference is that most real estate transactions require the cooperative efforts of two agents, not one as in a travel agency, stock brokerage, or insurance sales business. The fact that a competitor's cooperation is necessary to conduct business creates an opportunity to set up a punishment regime to support collusion.

To see how, consider a travel agent who cuts her fees with the hope of gaining extra business. If a larger potential clientele comes through the door as a result, she does not need her competitor to sign off on the new business for their transactions to go through. That makes her unilateral incentive to break from the collusive regime (tacit or explicit) all the more tempting. Any drop in costs, then, is likely to be passed through to consumers, at the expense of the higher-cost operations in the industry.

The necessity of cooperation in the real estate industry makes things different. If an agent deviates from a collusive regime, other agents seeking to preserve collusion can punish that agent by withholding cooperation. That can make transacting business as a deviator prohibitively costly. If a buyer's agent offers clients low-cost access to online home listings, for example, other agents can refuse to make their own listings available through such channels. Or, if a seller's agent cuts her commission rate, other agents may be able to steer their potential buyers away from her listings. These examples are more than theoretical possibilities. The former scenario is essentially the key anticompetitive activity alleged in the DoJ lawsuit, and allegations of steering also have led to investigations, although evidence to this point has been insufficient to lead to a formal complaint.8 [End Page 50]

It is the ability of real estate agents to engage in immediate and targeted retaliation against discounting agents—and perhaps just as important, their ability to retaliate in turn against any nondiscount agents who fail to punish discounters when the opportunity arises—that can sustain a collusive equilibrium.

In this paper, we explore a range of issues related to allegations of antitrust violations in residential real estate, with a particular eye toward the performance of discount agents (whom we often refer to as flat-fee agents), the targets of the alleged illegal behavior. We begin by reviewing the specific antitrust concerns that have arisen in public policy debates and then consider what economic theory says about the potential for success of a collusive strategy in an industry in which cooperation of other industry operators is required to facilitate transactions (or at least greatly increases the probability of a transaction).

Next we empirically evaluate the question of whether discount agents provide a product to customers that matches, in terms of value, the product of traditional full-service agents. Using data from three real estate markets—Cook County, Illinois; a seven-county area including Sacramento, California; and Santa Cruz County, California—we find evidence that houses sold by flat-fee agents fetch the same prices as observably similar houses sold by full-commission agents. However, expected days on the market are substantially higher for flat-fee homes than for those listed by full-commission agents, because of both a lower probability of sale and a longer time on market (conditional on a sale taking place). Those results are consistent with allegations that traditional agents steer their own clients away from homes listed by flat-fee agents, although we consider other possible explanations. The results also are notably similar to those in Hendel, Nevo, and Ortalo-Magné (2007), which finds that homeowners who use a for-sale-by-owner (FSBO) website—not unlike many discount agent platforms—obtain prices that are similar to (in fact, slightly higher than) those obtained through full-commission agents but that FSBO homes also have a lower probability of selling and a longer time on market. Using the market outcome differences that we estimate, we calculate that homeowners in our database are able to sell their homes at a considerably lower cost using discount agents, even taking into account the longer expected time to sale and the selling costs that they must bear themselves (for marketing, hosting showings, and so forth) when using a flat-fee agent.

Antitrust Considerations in the Residential Real Estate Industry

The DoJ lawsuit brought antitrust concerns to the forefront of the economic debate about the real estate industry. Even so, the specific allegations in the [End Page 51] complaint comprise only part of the potentially collusive activities that have concerned policymakers. Here we provide a brief overview of those concerns.9

The central issue in the DoJ antitrust suit against the National Association of Realtors is the association's Virtual Office Website (VOW) policy. According to the NAR website,VOWs are "Internet sites operated by MLS participants through which they . . . work with clients and customers in cyberspace in ways similar to how real estate professionals interact with clients and customers in a 'brick and mortar' environment."10 Essentially, VOWs are designed to allow customers to access the multiple listing service (MLS) without coming to the agent's office. One discount brokerage model charges customers a flat fee for access to the VOW, allowing them to search MLS listings without the direct intercession of an agent. That saves agents time and expense, allowing them to provide their service at a relatively low price, and offers more flexibility in the search process to those consumers willing to forgo extra help from an agent. Some Internet-based agents also sell their VOW-searching clients as "leads" to sellers'agents for a fee contingent on the client's purchase of a selling agent's listing.

The NAR adopted a policy that would let agents choose to withhold their listings from particular agents'VOWs (a "selective opt-out") or all VOWs (a "blanket opt-out"). The DoJ's concern centers on the potential for traditional agents to withhold opportunities for trade from lower-cost VOW-based discount agencies. Agents could, especially if coordination was possible, withhold a large fraction of the houses listed for sale on the MLS from the VOWs of flat-fee agents, rendering their low-cost option essentially valueless to potential home-buyers. While the selective opt-out might seem to be the more powerful tool for achieving that goal, since it would allow traditional agents to target discount agents specifically, the blanket opt-out performs essentially the same role. Because agents themselves have direct access to the full MLS (including listings that other agents have withheld from VOWs), traditional agents can still transmit these listings to their clients along their traditional, largely in-person transmission lines. Discount agents, on the other hand, suffer because their model is built on reducing costs by using the VOW to limit intensive in-office interaction with clients. That is not possible with a VOW that is only a faint shadow of the local MLS.11 [End Page 52]

In addition to engaging in the actions that led to the federal case, real estate industry groups also have been encouraging state policies that have plausibly anticompetitive effects. (State legislation typically is exempt from federal antitrust enforcement.) Ten states have bans on agents offering rebates to their clients, which are essentially out-and-out bans on price cutting. Furthermore, laws outlining minimum service requirements are currently in force in eight states and are being considered in others. These statutes enumerate a set of services that all real estate agents must provide to clients. While their proponents argue that the provisions protect consumers, the Federal Trade Commission (FTC), the DoJ, and consumer groups like the Consumer Federation of America all have opposed their passage. Minimum service requirements make low-cost agency models more difficult or even impossible to operate. For example, the common discount agency model in which the agent simply lists the client's home on the MLS for a flat fee and offers no further services could easily be interpreted as illegal under the new requirements. Outlawing low-cost, low-service models reduces both the options available to consumers (the low end of the vertical product space distribution is essentially shut down) and, by raising the costs of lower-cost rivals, the competition that traditional brokers face.

In addition to these codified sources of antitrust concern, the industry has a history of allegations of reprisals against discount agents by other agents, ranging from petty harassment or slander to steering buyers away from the listings of discount seller agents. While such allegations have led to antitrust investigations—they were a major impetus for the 1983 FTC examination of the industry as well as more recent actions12—evidence of sufficient coordination has not yet been found (a single firm's refusal to work with another particular firm typically is not under the purview of competition regulation). Nonetheless, allegations of such activities are not uncommon, and the possibility of antitrust violations remains a concern for authorities.13

One issue that has puzzled observers is how a collusive equilibrium, if it exists, could be maintained in the industry.14 Legal barriers to competition can certainly facilitate collusion in areas where they are in force. Given the relative [End Page 53] uniformity across markets of features like commission rates, however, it seems that other collusion-supporting forces would also have to be at play. The puzzle arises because the industry has many firms, even within most metropolitan areas. Economic theory suggests that the difficulty of sustaining collusion grows with the number of actors.15 While Brock and Sheinkman (1985) shows that that prediction is not universal, it has been borne out in experimental studies, such as Isaac, Ramey, and Williams (1984) and List (2007).

Again, the answer here may lie in the fact that in the real estate industry, competitors' cooperation is needed in facilitating most transactions. In typical models of collusion, firms interact in the market in the standard way—their choices affect the payouts and therefore the decisions of their competitors. Given a competitor's decision, however, exchange takes place on the terms set without the imprimatur of other firms. They can take retaliatory action in the future, of course, but any such action affects only the environment in which the deviating competitor operates and does not directly impact its transactions. In real estate, however, the fact that collusive firms may have to cooperate with any deviating firm for it to even engage in a transaction can be a powerful tool for preserving collusive behavior. The traditional trade-off for a collusive firm considering deviating is that it prices low to enjoy a gain today at the expense of likely punishment in the future, but in a world were competitors must cooperate in transactions, a price cut may not lead to any extra sales.

Collusion and Cooperating Competitors: Theoretical Framework

Sustaining collusion is, by its nature, a dynamic process. In a static world, individual colluders typically have an incentive to deviate from collusive action, since generally they can raise short-term revenues and profits by charging a price lower than their competitors'. To make collusion possible, dynamic incentives must counteract the short-term gains; such incentives typically involve the threat of the loss of future rents from preserving collusion if firms deviate to earn short-term gains. Future losses are caused either because colluding firms passively revert to a more competitive equilibrium or because colluders actively retaliate against the deviator. Whether collusion can be sustained in equilibrium depends on the size of the short-term gain from deviating relative to the size of the long-term loss from the breakdown in collusion. [End Page 54]

An extra complication is involved when firms have different costs.16 In that case, firms have different preferences with respect to what the collusive price should be. Firms with lower costs want a lower price. If the cost asymmetry is large enough, low-cost firms may prefer to eschew collusion altogether, because in the competitive regime they may be able to drive higher-cost firms from the market or because the costs of administering collusive payouts among asymmetric firms outweigh any rents earned.

We assume in our analysis that the cost differences between flat-fee and traditional agents are large enough that the discount agents prefer the competitive market outcome. What we look at instead is the choice of a traditional agent to cooperate or not cooperate in transactions with a discount agent, since that decision strikes us as a key factor in preserving any collusion in the industry. Cooperation can take the form of allowing their listings to be posted on a discount agent's VOW, bringing clients to properties listed by discount agents as often as they would if the same properties were listed by traditional agents, or not engaging in one of the other forms of retaliation that have been alleged. If collusion in the industry is to succeed, the incentives for traditional agents not to cooperate with flat-fee agents must outweigh any benefits from cooperating.

Let's consider those incentives more specifically. Suppose a traditional agent cooperates in a pending transaction with a discount agent. He earns the commission from the sale, which we assume to be based on the standard market rate. By doing so, however, he realizes that by letting the discount agent operate successfully, his expected commissions will drop in the future. That will happen for two reasons. One is competitive: when transactions are done through discounters, other consumers become increasingly aware of discount agencies and any associated benefits (perhaps they witness the transactions themselves or hear about them through friends, neighbors, or relatives). That in turn increases the market share of flat-fee agents, raising the downward competitive pressure on traditional agents' commission rates. The second influence on the traditional agent's expected post-cooperation commission is the punishment response of other traditional agents. Because real estate transactions are publicly observable, traditional agents can see who among them cooperates with a discount agent. That makes a cooperating traditional agent susceptible to reprisals by other traditional agents using the tactics discussed above. Their retaliation lowers the probability that a cooperating agent can broker a successful transaction in the future, reducing expected future commissions.

To formalize this intuition slightly, suppose the prevailing commission per agent in the market is equal to F. We assume for simplicity that all homes in [End Page 55] the market are sold at a (normalized) value of 1, so the commission level per house also is F. To simplify our analysis, we do not endogenize the commission setting process, treating it instead as a reduced-form outcome of a more complex market-level equilibrium. We also assume that the traditional agents in the market take the commission individually as given. That seems to be a reasonable approximation of reality, especially given the evidence discussed above on the relative invariance of commission rates.

The traditional agent decides whether or not to cooperate with a discount agent in a transaction. He chooses the option that maximizes the present value of operating in the market given that the prevailing commission rate is F. That is, he maximizes the value function

inline graphic

where VC (F) is the present discounted value of cooperating in the transaction in the current period and VNC(F) is the value of not cooperating, given that the current size of the prevalent commission (the state variable) is F.

We can write the value function under cooperation in recursive form as follows:

inline graphic

If the agent cooperates with a flat-fee agent, the transaction occurs and he earns the commission F in the current period. However, the future payoff embodied in the discounted expected payoff found in the second term of the Bellman equation (β is the agent's discount factor) reflects the lower prevalent commission value expected in the next period. As described above, this reflects the lower expected payout due to competitive and punishment effects. We parameterize the degree to which cooperation reduces commissions with the value 0 < γ < 1.

If the agent refuses to cooperate, the transaction with the discount agent does not occur. However, we assume that there is some positive probability that a transaction occurs instead with another traditional agent. Perhaps the noncooperating agent finds a suitable house for his buying client that is listed by another traditional agent, or a buyer still finds one of the noncooperating agent's selling clients' listings even though the noncooperating agent withheld the listing from VOWs. Let this probability be 0 < δ < 1. Thus the value function under not cooperating in the current period is

inline graphic

Note that by refusing to deal with the flat-fee agent, the traditional agent has preserved his commission rate for the next period, either through stifling the competitive effect of flat-fee agents or by avoiding retaliation from other traditional agents. In not cooperating, the traditional agent trades off a lower payout today, in the form of a lower probability of having a transaction go through, for a higher expected payoff in the future (note that V'C (F) > 0, V'NC (F) > 0, and W'(F) > 0).

Three parameters embody the influences on the potential for success of a collusive outcome (which we define as traditional agents choosing to not cooperate): β, γ, and δ.17 The discount factor β can be taken to represent time preferences, as is the standard practice. The more the agent discounts the future (that is, the less patient he is), the more likely he is to cooperate. That is because cooperating with a discount agent involves trading off a lower future payment for a higher payment today, while not cooperating entails a lower payment today but higher expected payments in the future. The more the agent discounts the future, the less he cares about future payments relative to the current payout. That tilts his optimal decision in favor of cooperating.

The parameter β also can embody other influences. Suppose the agent will soon leave the market for one reason or another—perhaps he is near retirement or will be moving away. In that case, we would expect him to care much less about future outcomes. In the extreme, if the agent is working on what he knows to be his last transaction, there is a strong incentive to cooperate (although one might imagine ways that traditional agents could still impose costs on this agent even after he retires or moves out of the market). Therefore a low value for β could embody such "short-timer" influences on the agent's decision besides traditional time preference notions.

As we touched on above, γ embodies both the competitive and retaliatory effects of cooperating with a flat-fee agent. A smaller γ—that is, a greater competitive or retaliatory effect—lowers the payoffs a cooperating agent can expect in the future, leading to less cooperation with discount agents. Several market fundamentals would be expected to impact the magnitude of γ. For example, the competitive effect should be greater (that is, γ should be lower) in smaller markets. A successful transaction by a discount agent in a city with hundreds or thousands of traditional agents is unlikely to have a notable impact on commissions, but one in a smaller town may. Controlling for market size, the [End Page 57] competitive effect is likely to be inversely related to discount agents' market penetration. The notion is that if flat-fee agents already are a dominant force, much of the competitive effect will already have been realized. The greatest marginal competitive effect is therefore expected in markets in which discount agents still account for a modest fraction of firms (although perhaps the maximal marginal impact does not occur exactly when there are no discount agents—there may be some "increasing returns" in the influence of discount agent success that cause it to grow for some fractions above zero).

The retaliatory effect embodied in . will be larger when traditional agents can better coordinate their actions. To see why, note that the structure of the decision of other traditional agents to retaliate against traditional agents who cooperate with discount agents is very similar to the structure laid out above. As with the initial decision to cooperate with the discount agent, such retaliation involves trading off a reduction in short-term gain (retaliating agents lose the ability to have the cooperating agent as a partner in the current period's transaction) for a gain in future payouts (higher commissions are preserved).

If only a small number of other traditional agents choose or are able to retaliate, the future loss for the cooperating agent will be minimal. That implies that agents who are considering retaliating themselves are more likely to be better off by forgoing such actions and cooperating with the traditional agent who deviated in the first place. However, if the retaliation has a coordination mechanism, the threat can be very powerful, both in its direct effect on the initial cooperation decision (through making γ smaller) and in its indirect effect on "enforcing the enforcement"—ensuring that traditional agents remain willing to retaliate against deviators rather than hop on the deviation bandwagon themselves. That is also where the tie to the number of producers in the market discussed above comes into play. Presumably, with a large number of agents, deviations are less likely to be punished because the impact of any retaliation against those who fail to punish is small when agents are acting purely in their isolated self-interest.

However, the barriers to collusion raised by having a large number of producers can be overcome through coordination mechanisms, which can take many forms. A large literature has pointed out that publicly observable transaction information can serve as a coordination device to support collusion; early work in this regard includes Stigler (1964) and Green and Porter (1984). Certainly this feature is present in the real estate market. All the parties to the transactions are observable (agents cannot secretly cheat on quantity), and the commission splits between buyers' and sellers' agents are listed in the MLS entry for a house (although we cannot rule out unobserved side payments). The [End Page 58] owners of the brokerages for which the agents work also can serve as a coordination mechanism. They have leverage over the employment conditions of agents, typically including how commissions are split between the broker and agent, and thus they have leverage to enforce the orders that they give.

Furthermore, there are fewer (often many fewer) brokerages than agents in a market, which also tends to make coordination of retaliatory regimes easier. Finally, the local MLS systems themselves, which are supported by a consortium of local brokerages, can serve as a coordination mechanism. Indeed, individual MLSs have tried to ban listings of homes under certain types of listing agreements commonly used by flat-fee agents, apart from engaging in the anticompetitive behaviors previously described. Since there is usually only one MLS per market, any action that can be agreed on by its members will be highly enforceable even if there are a large number of agents in the market.18

The third parameter, δ, captures the diminution of the value of the current transaction when the traditional agent refuses to cooperate with the discount agent. The diminution reflects the fact that even if a traditional agent does not engage in transactions with the discount agent, she still may see a transaction through in the period by matching up with a traditional agent. Higher values of d imply that preserving collusion is easier because the loss from not cooperating with discounters is reduced.

The larger the fraction of traditional agents in the market, the higher the probability δ. When the market is thick with traditional agents, the loss from forgoing cooperation with discount agents is small. As discounters expand, however, it becomes more costly to ignore their listings (when representing buyers) and VOW clients (when representing sellers). Interestingly, that implies that there may be network effects in discounting. When there are few flat-fee agents, it will be hard to get traditional agents to cooperate because they lose little by not doing so. As the number of discounters grows, so does the loss from not cooperating, making it increasingly easy for new discounters to enter the market. That positive feedback mechanism, if strong enough, could in theory lead to a "tipping point" in which a market swiftly turns from being dominated by traditional agents to one with a large contingent of flat-fee agents.19

In sum, a number of agent- and market-specific features can be tied to predictions about the likelihood of success in sustaining a collusive equilibrium in a local real estate market. They range from measures of traditional agents' value of future transactions with their cohort, to market size, to the presence [End Page 59] of coordinating mechanisms, to the share of discount agents already in the market. In the end, all of those influences will be reflected in the relative performance of discount agents. The stronger a collusive regime, the poorer the outcomes for discount agents relative to those for traditional agents.

Data

We used data drawn from multiple listing service records covering Cook County, Illinois; a seven-county area including Sacramento, California; and Santa Cruz County, California.20 The data were provided to us by real estate agents in response to a March 2006 blog post soliciting the cooperation of agents for a research project to examine outcomes for home sellers using discount real estate brokers.21

In each of the three markets, we had all MLS listings for single-family homes for the period January 2004 to March 2006. The data include not only homes that sold, but also listings that expired or were cancelled before a sale took place. The data do not include housing transactions in which the seller did not enlist a real estate agent (for-sale-by-owner transactions).

MLS listings provide extensive detail about the homes that are for sale, the timing of events, listing and transaction prices, and the agents involved in a transaction. The data include basic facts about the homes, such as number of bedrooms and bathrooms, style (colonial, ranch, Cape Cod), presence and type of heating and air conditioning systems, and square footage (except in Cook County, where that field is frequently left blank). We had the ZIP code of the home.22 We also had access to the marketing description provided in MLS for each house (for example, "Well-maintained, spacious bungalow brimming with charm . . ."). Using text-extraction tools, we identified the presence of a wide array of keywords to control for less easily observable characteristics of home quality. [End Page 60]

Our analysis focuses on three primary outcomes: whether a sale was completed within the timeframe of our data, the number of days on the market for homes that sold, and the (logged) sale price of homes that sold. For the time-on-market and sale price variables we run specifications of the form

inline graphic

where h, t, and z index a particular home, the month and year in which the home was first listed in our data set, and the ZIP code in which the home was located, respectively. Outcome corresponds to either the number of days the house was on the market before a sale occurred or the logged sale price of the home. Flat-Fee is an indicator variable equal to 1 if the listing agency met our criteria for providing flat-fee real estate agent services.23 A vector of home characteristics is included in Xh: indicators for number of bedrooms and bathrooms, total rooms, size of garage, number of fireplaces, master bedroom bath, style of home, what material the exterior was made of, square footage, age of the home, keywords from the written description in the listing, and, where applicable, other characteristics such as whether the property was on the coast. Also included as controls are fixed effects for the ZIP code in which the property was located and the year and month in which the property listing first appeared in our data set. For the days-on-market and sale price outcome variables, we restrict the sample to homes in which a sale was observed. When we analyze whether a home ever sold, we include all homes listed, whether or not a sale took place. Because whether a home ever sold is dichotomous, we estimate this relationship using a probit model with the same explanatory variables as the specification above. For that specification, we report in the tables the marginal effects evaluated at the sample means.

Our data set is affected by two-sided truncation. We observe only listings beginning in January 2004. The same home, however, could be listed numerous times before it sold. Within our sample we are able to identify multiple listings on the same home and combine them into a single observation. However, with the data that we have we cannot reliably determine whether listings for that home existed before January 2004. Therefore, when we report time on the market, it is only time on the market beginning in January 2004. Our sample ends in March 2006, when the data were collected. Therefore, any home sale that occurred after that point is not recorded in our data. To minimize the impact of that truncation at the end of the data, we restrict our sample to homes listed in either 2004 or the first three months of 2005, leaving a minimum of one full year for us to observe the outcome for any listing in our sample. [End Page 61]

On the rare occasions in which a home sold twice within the study period, we include only the first sale. We exclude outliers in terms of sale price (any home that sold for more than five times the sample median in the area as well as any home that sold for less than $50,000). We drop listings from the sample that lack data on key housing attributes, such as the number of bedrooms. We also eliminate a small number of cases in which the data entered are inconsistent (for example, the home is shown as sold before the date that it is listed).

Table 1 presents selected summary statistics for the remaining observations in each of the three areas used in our analysis.24 In each case, we show data separately for full-service and discount agencies. Columns 1 and 2 correspond to Cook County, columns 3 and 4 to the Sacramento area, and columns 5 and 6 to Santa Cruz County. Standard deviations are shown in parentheses. We highlight in bold those pairs of entries for which we can reject at the 5 percent level equality of the means across full-service and flat-fee agents. Except where otherwise noted, the entries in the table include all home listings, regardless of whether a sale occurred within our sample.

The top row of table 1 reports the number of listings in the data, highlighting the low penetration rates of flat-fee agents in all of the sample markets. The Cook County data include 238 listings by discount agents, or 2.2 percent of all listings. The flat-fee shares in the other two markets are similarly low: 1.0 and 2.5 percent in Sacramento and Santa Cruz, respectively. One unfortunate implication of the uniformly low market share of flat-fee agents is that it greatly limits our ability to test some of the most interesting predictions of the theory above, such as the existence of network effects.

The next three rows of table 1 show our primary outcome variables: whether the home sold within our sample time frame, days on the market (conditional on selling), and sale price. Sellers using discount agents are roughly 10 percentage points less likely to ever sell their homes in Cook County and Sacramento. These differences in means are statistically significant at the 5 percent level. Recall that our data reflect only home sales made through the MLS. To the extent that home sellers who use flat-fee agents are more likely to eventually withdraw their homes from the MLS and instead pursue a for-sale-by-owner transaction, the differences between full-service and flat-fee agents may be exaggerated on this dimension. Even among homes that did sell through the MLS, time on the market is greater using flat-fee agents in Cook County and Sacramento (25 and 12 days longer, respectively). The raw data [End Page 62]

Table 1.
Summary Statisticsa

Characteristic Cook County Sacramento Santa Cruz



(1) (2) (3) (4) (5) (6)
Flat Full Flat Full Flat Full
fee commission fee commission fee commission

a. The data cover all MLS listings of single-family homes over the period January 2004 to March 2006. The odd-numbered columns present summary statistics for listings done by flat-fee agents; the even columns correspond to full-commission agents. Pairs of entries for which the difference in means across flat-fee and full-commission agents differ at the .05 level are shown in bold. Standard deviations are shown in parentheses for variables that are not dichotomous. See the appendix for further details about how the data set is constructed.
House ever sold 0.70 0.83 0.79 0.89 0.77 0.78
Days on market, if house ever sold 102.90 78.31 44.89 32.62 52.13 62.84
(90.77) (98.52) (45.06) (39.76) (50.44) (85.80)
Sale price, if house ever sold 439,744.95 421,649.92 315,751.25 311,698.89 654,477.00 728,931.12
(193,850.82) (275,883.88) (201,366.16) (165,562.61) (201,677.01) (353,084.97)
Square footage n.a. n.a 1793.37 1611.92 1656.41 1710.96
n.a. n.a. (629.95) (625.95) (673.98) (911.68)
Year built 1949.55 1952.76 1982.39 1977.26 1972.16 1960.18
(30.27) (25.72) (20.68) (20.23) (21.12) (30.25)
Bedrooms, number 3.45 3.42 3.39 3.19 2.91 2.98
(0.79) (0.86) (0.77) (0.79) (0.79) (0.88)
Bathrooms, number 2.21 2.14 2.23 2.05 2.01 2.05
(0.77) (0.81) (0.62) (0.63) (0.65) (0.77)
Fireplace 0.80 0.67 0.84 0.80 0.86 0.82
(0.73) (0.76) (0.37) (0.40) (0.35) (0.38)
Central air 0.87 0.86 0.93 0.92 0.07 0.04
Master bedroom bath 0.44 0.40 0.82 0.72 n.a n.a

Observations 238 10,746 270 26,100 83 3237

[End Page 63]

show that in Santa Cruz, flat-fee homes sold faster, but not significantly so. While these outcomes vary, there are no significant differences between the sale prices of homes sold by traditional and discount agents in any of the three markets.

The remaining rows in the table reflect home characteristics. In Cook County, homes sold through a flat-fee agent are similar on those dimensions to those sold through a full-service agent. In Sacramento, the flat-fee homes have more square footage, bedrooms, and bathrooms. In Santa Cruz, the only statistically significant difference is that homes sold through a flat-fee agent were built more recently.

Empirical Results

Our empirical analysis of discount agency outcomes focuses on two questions. First, do homes sold by discount agents take longer to sell—not just in terms of time to sale (conditional on a sale), but also after we account for any differences in the probability that a home will sell in the first place? Second, do customers who use discount agents receive lower prices for their homes? The answers to both can speak to antitrust concerns in the industry. Any differences between discount agents and traditional agents with respect to both the time on market and the sale price of homes can offer insight into the efficacy of discount agents as well as possible reasons for any observed differences. In addition, the prices obtained by home sellers using discount agents are relevant in addressing the wisdom of the state minimum service laws discussed earlier. If sellers using flat-fee agents appear to do worse on net after account is taken of sale price, commission paid, time on the market, and extra effort on the part of the home seller when using a flat-fee agent, that bolsters the case for minimum service standards. If they do better, then the minimum service argument is called into question.

Table 2 presents estimates of the relationship between having a flat-fee agent and whether a sale was ever recorded for a home listed in our sample. The values reported in the table are the marginal effect of moving from a full-service to a flat-fee agent, evaluated at the sample means. Each entry in the table is from a different probit estimation. Although other covariates are included in the specifications, only the estimate on the discount variable is presented in the table. (Column 1 of appendix table 1 presents full results for the most saturated version of the model.) We report, in addition to the estimated marginal effect, robust standard errors and the pseudo R2 from the estimation. The number of covariates [End Page 64] included in the specification increase moving from left to right in the table. In column 1 the only controls are fixed effects for ZIP code and year and month of the listing. Column 2 adds basic features of the scale of the house: a quadratic in square footage; categorical indicators for the number of bedrooms, bathrooms, and other rooms; and the number of cars that the garage will hold. Column 3 adds additional controls capturing further aspects of the quality of the home—the presence of fireplaces, master bedroom baths, and central air conditioning—as well as indicators corresponding to the age, style, and exterior of the home. The final column adds in a wide range of keywords drawn from the written description of the home in the MLS listing. The top panel of the table shows results for Cook County, the middle panel for the Sacramento area, and the bottom panel for Santa Cruz County. In Cook County and Sacramento, homes represented by flat-fee agents are approximately 10 percentage points less likely to sell in our timeframe than those represented by full-commission agents. These estimates are highly statistically significant. In Santa Cruz County, flat-fee agents are actually slightly more likely to have homes that sell; these estimates are on the borderline in terms of statistical significance.

Table 2.
The Impact of Flat-Fee Agents on Whether a House Ever Solda

Market (1) (2) (3) (4)

a. The entries in the table are estimated marginal effects of having a flat-fee agent evaluated at the sample means using probit estimation. Robust standard errors are in parentheses. Psuedo R2 values also are reported. Each coefficient reported in the table is from a different specification. The number of covariates increases from left to right in the table. Only the coefficient on flat-fee agent is reported in the table. The full specifications corresponding to column 4 are reported in the appendix tables. See the appendix for a more complete description of the data set. **Significant at the .01 level.
Marginal effect of using a flat-fee agent
Cook County -0.128** -0.126** -0.127** -0.117**
(N = 10,999) (0.027) (0.027) (0.028) (0.027)
Pseudo R2 0.036 0.056 0.065 0.078
Sacramento -0.104** -0.100** -0.104** -0.097**
(N = 26,368) (0.024) (0.024) (0.025) (0.024)
Pseudo R2 0.018 0.033 0.040 0.048
Santa Cruz 0.060 0.051 0.052 0.044
(N = 3,298) (0.025) (0.027) (0.025) (0.024)
Pseudo R2 0.031 0.065 0.080 0.119
Covariates
   ZIP code fixed effects Yes Yes Yes Yes
   Month-year fixed effects Yes Yes Yes Yes
   Basic housing characteristics No Yes Yes Yes
   Measures of housing quality No No Yes Yes
   Keywords from descriptions No No No Yes

[End Page 65]
Table 3.
The Impact of Flat-Fee Agents on Total Days on the Market, Conditional on Salea

Market (1) (2) (3) (4)

a. The dependent variable is days on the market. The sample is restricted to homes that sold in our sample window, which ended March 2006. The entries in the table are the OLS coefficients on having a flat-fee agent. Robust standard errors are in parentheses. R2 values also are reported. Each coefficient reported in the table is from a different specification. The number of covariates increases from left to right in the table. Only the coefficient on flat-fee agent is reported in the table. The full specifications corresponding to column 4 are reported in the appendix tables. See the appendix for a more complete description of the data set. **Significant at the .01 level.
Coefficient on flat-fee agent
Cook County 35.27** 34.20** 33.82** 33.30**
(N = 9,103) (6.00) (5.83) (5.82) (5.86)
Pseudo R2 0.136 0.187 0.199 0.213
Sacramento 10.38** 9.12** 9.56** 8.75**
(N = 23,286) (2.38) (2.32) (2.32) (2.31)
Pseudo R2 0.060 0.102 0.108 0.118
Santa Cruz 6.408 7.726 7.874 8.309
(N = 2,564) (5.83) (5.86) (5.54) (5.76)
Pseudo R2 0.051 0.121 0.151 0.213
Covariates
   ZIP code fixed effects Yes Yes Yes Yes
   Month-year fixed effects Yes Yes Yes Yes
   Basic housing characteristics No Yes Yes Yes
   Measures of housing quality No No Yes Yes
   Keywords from descriptions No No No Yes

Table 3 reports results for time on the market for homes that sold in our time-frame. The structure of table 3 mirrors that of table 2. The only differences between the two tables are that the dependent variable changes; the sample in table 3 is restricted to homes that sell; and the model in this table is estimated with ordinary least squares rather than a probit. Flat-fee homes that sell in Cook County stay on the market more than a month longer than those represented by a full-service agent. The results are not sensitive to the set of controls included. In the other two areas, flat-fee homes take 6 to 10 days longer to sell, although the differences are statistically significant only in Sacramento.

Table 4 presents results on the relationship between flat-fee agents and the (logged) price at which a home sells. Table 4 is identical to table 3 except that the outcome variable differs. In contrast to the results for whether a home ever sells and time on the market, the price outcomes are much more mixed. In Cook County, discount agent homes sell for more only when a limited set of controls is included. As more covariates are included, however, any gap disappears. None of these differences are statistically significant. The same sort of pattern appears [End Page 66] in Sacramento, where large price differences (7 percent) appear when no controls for housing characteristics are included but the gap shrinks as more covariates are added. This pattern is consistent with the summary statistics in table 1, which shows that flat-fee agents sold homes that were larger on average. In the fullest specification, homes sold by flat-fee agents sell for a statistically insignificant 1.7 percent premium. In Santa Cruz, once controls are included, flat-fee homes also sell for a premium, although again the 2.3 percent difference is not statistically significant.

Table 4.
The Impact of Flat-Fee Agents on Logged Sale Pricea

Market (1) (2) (3) (4)

a. The dependent variable is ln(sale price). The sample is restricted to homes that sold in our sample window, which ended March 2006. The entries in the table are the OLS coefficients on having a flat-fee agent. Robust standard errors are in parentheses. R2 values also are reported. Each coefficient reported in the table is from a different specification. The number of covariates increases from left to right in the table. Only the coefficient on flat-fee agent is reported in the table. The full specifications corresponding to column 4 are reported in the appendix tables. See the appendix for a more complete description of the data set. *Significant at the .05 level; ** significant at the .01 level.
Coefficient on flat-fee agent
Cook County 0.023 0.019 0.004 0.000
(N = 9,103) (0.021) (0.013) (0.011) (0.010)
Pseudo R2 0.596 0.851 0.886 0.895
Sacramento 0.071** 0.023* 0.019* 0.017
(N = 23,286) (0.017) (0.009) (0.009) (0.009)
Pseudo R2 0.410 0.827 0.837 0.844
Santa Cruz -0.014 0.020 0.021 0.023
(N = 2,564) (0.032) (0.024) (0.022) (0.022)
Pseudo R2 0.329 0.679 0.709 0.729
Covariates
   ZIP code fixed effects Yes Yes Yes Yes
   Month-year fixed effects Yes Yes Yes Yes
   Basic housing characteristics No Yes Yes Yes
   Measures of housing quality No No Yes Yes
   Keywords from descriptions No No No Yes

Although theory provides a number of further predictions (for example, full-commission agents with a short time horizon will be more likely to cooperate with discount agents; full-commission agents who cooperate with discount agents will be punished; as the market share of discount agents rises, collusion will become more difficult to sustain; and so forth), the small share of discount agent transactions impedes our ability to test these auxiliary predictions at this time. As the share of discount agents rises, much richer tests of the theory will become possible. [End Page 67]

Interpreting the Results on Longer Market Times for Flat-Fee Homes

Our results suggest that homes listed with flat-fee agents take longer to sell (both conditional on a sale having occurred and in terms of the expectation of a sale, accounting for the chance of a delisting before a sale occurs) but eventually sell for prices similar to those for houses listed with full-service agents.

One possible explanation for these findings is that full-service agents purposely steer buyers away from properties listed by flat-fee agents. Such behavior would be aided by the fact that the MLS includes information fields that make it easy for other agents to determine whether a listing is done on a flat-fee basis. Discount agents cannot easily disguise themselves; the MLS that contains Cook County, for example, fines agents who incorrectly complete this field.25 Furthermore, it is not clear why, absent collusive aims, the terms of the relationship between the seller and the seller's agent is relevant to a prospective buyer.26

There are, however, other possible explanations for why homes represented by flat-fee agents might remain on the market for a longer period. One is that without the guidance of a full-service agent, sellers using flat-fee agents initially price their properties too high. To test that hypothesis, we estimate regressions with the (logged) original listing price as the dependent variable and the full set of controls (excluding the flat-fee indicator) on the right-hand side of the equation. In this estimation we include both homes that sold and those that did not. Using this regression, we compute a predicted original list price for each property. (The coefficients in the prediction equations were allowed to vary by market.) In Sacramento and Santa Cruz County, flat-fee properties list for less than observably equivalent full-commission properties by 1 percent and 6 percent respectively. In Cook County, flat-fee agent homes list for roughly 1 percentage point more than would be expected. Thus, only in Cook County is it possible that poor pricing could explain the observed differences in time on the market between houses represented by flat-fee and those represented by full-commission agents.

Further exploring Cook County, when we regress time on the market on our usual right-side variables but also add the deviation of the actual original listing price from the predicted one, each 1 percent higher original listing price predicts a (statistically significant) extra 0.7 days on the market. Since the listing price difference between flat-fee and full-commission listings in Cook [End Page 68] County was about 1 percent, this channel can explain a 0.7-day time-on-market difference. This is only a small fraction of the observed gap of 30-plus days. Further, the flat-fee coefficient remains essentially unchanged when the original list price residual is added to the regression. To put into perspective the size of the impact implied by the flat-fee variable in Cook County, note that the original list price would have to increase by roughly 50 percent in order to increase time on the market by as much as using a flat-fee agent increases it.27

A second reason why homes represented by discount agents might sell more slowly is that the homeowners are unwilling to accept offers that other sellers would accept.28 We have no direct way of testing for differences in discount rates across different groups of sellers, but one very crude piece of information available is the ratio of the sale price to the listing price at the time a deal is made. If homeowners using flat-fee agents are more patient than other sellers, one might expect that ratio to be closer to 1 for those using flat-fee agents. In none of the three markets do we find that to be the case: in two of the markets, the coefficient on this variable is small and of the incorrect sign, and in the other market it is extremely close to zero.

A third explanation for the longer time to sale for homes represented by flat-fee agents is that sellers who select these agents live in homes with characteristics that generally lead their homes to sell more slowly. Arguing against that claim is the fact that our estimates of time on the market prove quite robust to increasing the number of controls from columns 1 to 4 of table 3. The model's R2 increases substantially as controls are added (although not nearly as much as when the sale price is the dependent variable—we are much worse at predicting time on the market than sale price). In order for this explanation to have validity, however, the unobserved differences between flat-fee and full-service homes would have to differ systematically from the observed differences, which appear to be essentially uncorrelated with time on the market.

A fourth possible reason for the divergent time-on-market results is that homeowners bear a greater responsibility for making the sale when they use a flat-fee agent; for example, the homeowner typically is responsible for staging and showing the home. The logic of specialization does indeed argue that real [End Page 69] estate agents will have a comparative (if not absolute) advantage at these tasks.29 On the other hand, since the homeowner has much more at stake in the deal than an agent, the homeowner is unlikely to shirk on those responsibilities.

Welfare Calculations for Customers Using Flat-Fee Agents

The Department of Justice, the Federal Trade Commission, and consumer groups like the Consumer Federation of America all have opposed minimum service requirements for real estate agents, which have the effect of making it difficult or impossible for discount real estate agents to function. The National Association of Realtors, on the other hand, has been a strong proponent of the laws, arguing that they protect consumers.

With some additional assumptions, the data analysis above allows us to make rough comparisons of outcomes for home sellers who use flat-fee agents and those for sellers who use full-service agents. The comparisons should offer guidance on whether consumers who hire discount agents save money at a level commensurate with the difference in service levels or instead suffer a disproportionately large quality loss without being fairly compensated for it through the commissions that they save. There are four dimensions to consider in this calculation: real estate commission paid, differences in the final price for which a home sells, costs associated with additional time on the market, and the extra costs borne by the home seller when using a flat-fee agent (showing the house, marketing expenditures, extra legal fees, and so forth).

With respect to real estate commissions, a homeowner using a full-service agent pays a share of the sale price of roughly 5 percent on average. (That number is probably slightly low, reducing our calculation of the possible savings from using a discount agent.) The average sale price across our three markets was roughly $450,000, which translates into an average real estate commission of $22,500. A seller hiring a flat-fee agent pays roughly $500 to her own agent, plus compensation to the buyer's agent equal to roughly 2.5 percent. Therefore the total amount of real estate fees paid by a home seller for the average house sold through a discount agent is $11,750—or $10,750 less than that paid for a full-service agent.

The second issue to consider is whether using a discount agent affects the final sale price. Our empirical evidence on this question suggests little impact of using a flat-fee agent. The point estimates are zero or positive in all three samples, but in two of the three cases the price differential falls as we control [End Page 70] better for observable characteristics. To the extent that unobservable and observable characteristics of these properties vary in a similar manner, our estimates of a positive impact of discount agents on sale price are likely to exaggerate any benefits that they provide. We conclude that a reasonable interpretation of the data is that flat-fee agents have no clear impact on the price at which a home ultimately sells.30

The third consideration is the time to sale. Homeowners using flat-fee agents can expect their property to stay on the market longer than a house sold using a full-service agent. Our estimates of this parameter range from 8 to 33 days across the three markets, conditional on a sale. Since flat-fee homes also are less likely to sell, the gap represents a lower bound on the time difference. Accounting for the homes that never sell increases the expected difference in time on the market to approximately 12 to 40 days. A crude way of approximating the costs of longer time on the market is simply to calculate the carrying costs of holding the property for the additional time. That approach is likely to greatly overstate the true costs, however, because in most cases the homeowner will be consuming the flow of housing services during that time (generally the house is not sitting empty). We are not including this benefit to the homeowner in our calculations.31 Assuming an interest rate of 8 percent a year and an average home value of $450,000, the carrying costs associated with holding the home for an additional 26 days (the midpoint of our estimate of 12 to 40 extra days) works out to be roughly $2,500.

The final consideration is the additional cost of selling a home borne by homeowners who use flat-fee agents. Marketing costs, which generally are paid by full-service agents, would be paid out of the homeowner's pocket with a flat-fee agent. Based on discussions with real estate agents, we estimate those costs to be roughly $100 per week on the market for a typical home. Incremental time spent selling the house (for example, conducting showings and open houses) is another cost borne by home sellers. While there are no precise statistics on this time cost, a reasonable guess might be 5 hours per week. With an average time on the market of 10 weeks, that translates into an incremental labor effort by the homeowner of 50 hours. If we value their labor at $30 an [End Page 71] hour, the opportunity cost of that time is $1,500, plus $1,000 for out-of-pocket marketing expenses.

We estimate the increased costs of using the flat-fee agent to total roughly $5,000. When that amount is subtracted from the commission savings of $10,750, the net benefit to the home seller is $5,750. Thus, at least for the set of homeowners currently using flat-fee agents, the trade-off appears to be a favorable one, and the justification for minimum service requirements is not well supported by the data.

An important caveat with regard to these calculations, however, is that the sellers who stand to benefit the most from using flat-fee agents (well informed, Internet savvy, and so forth) are in fact those that the data show to be most likely to use such agents. This selection effect influences the calculations in two ways. First, if such unobservable attributes of sellers influence outcomes like sale price and time on the market directly, the point estimates that we obtained above will be biased. Potentially just as important is the possibility that there is heterogeneity in the impact of using discount agents across individual consumers, whereby those electing to use discount agents—that is, those whose outcomes we observe in the current data—are the ones who benefit the most from the relationship. If that is the case, the average benefit of flat-fee agency across all consumers will be smaller than the benefit that we calculate here.

Conclusion

Residential real estate agencies have avoided the declines observed over the past decade in other agency-based consumer service industries like travel agencies, stock brokerages, and auto insurance companies, even though the growth of the Internet and associated e-commerce platforms, which made agents less relevant in those industries, also took place in the real estate industry. In this paper, we explore the possibility that collusion explains the real estate industry's relative good fortune and investigate the mechanisms through which collusion might be sustained. A critical issue is whether traditional agents cooperate in transactions with agents working under new discount agency business models.

We find that sellers who hire discount agents had a longer expected wait to sale. That reflects both a longer time on the market conditional on the house having sold (from one extra week to a month, depending on the market) as well as a lower probability of ever selling. However, we found no difference in the [End Page 72] sale prices of observably similar homes and no indication that accounting for unobservable differences would change that fact.

One possible explanation for the results is that traditional agents steer their clients away from homes listed by flat-fee agents, lowering the probability of a sale in any given period. Other explanations are possible, although they have elements that do not square as well with the data.

In our data, consumers who hired discount agents avoided having to make large commission payments to their own agent at the cost of waiting longer to sell their home. Interestingly, these outcomes are similar to those for FSBO sellers documented in Hendel, Nevo, and Ortalo-Magné (2007). That is perhaps not surprising, because the FSBO marketing mechanism in Madison, Wisconsin (where their sample was drawn), is similar to that for many discount agency models. Using our estimates of the relative market outcomes of the two agency models, we run some back-of-the-envelope calculations to see whether such consumers benefit, on net, from this arrangement. The results suggest that they do, by a fair amount. That finding throws into doubt the necessity of minimum service laws in force in several states and under consideration in others.

Our analysis comes with a caveat, however. We observe the impact of discount agency on those consumers who, specifically because the market share of flat-fee agents is still small, are plausibly the most likely to benefit from discount agency. These benefits may not be as large for the average consumer if flat-fee agents gain considerable market share in the future.

All in all, the results suggest that market outcomes for clients who use flat-rate agents are not systematically worse than outcomes for those who use full-service, full-commission agents. However, at the present, discount agency does entail a trade-off between a longer expected time to sale and savings on commissions. The longer time on market is consistent with traditional agents being less likely to bring their own clients to homes listed by flat-fee agents in an attempt to preserve an uncompetitive market outcome. Discount agents still constitute a very small share of the markets for which we have data, however. As their share grows, collusion theory makes the strong prediction that the ability of full-commission agents to maintain collusion will shrink, leading the observed time-on-market gap to narrow, providing an additional test of the theory. [End Page 73]

Steven D. Levitt
University of Chicago
National Bureau of Economic Research
Chad Syverson
University of Chicago
National Bureau of Economic Research

Fernando Ferreira: In this thought-provoking paper, Steven Levitt and Chad Syverson first point out that business has been good for real estate agents over the past ten years, although other agent-based businesses, such as travel agencies and financial brokerage firms, have had far less success. Why? Because real estate agents collude to maintain their market shares and profit margins. Based on that initial premise, Levitt and Syverson try to empirically detect collusion in the real estate industry.

It is important, first of all, to understand the authors'main underlying assumption, which is that real estate agents provide an easily replaceable service, so that the industry is successful only because of collusion. My main concern with respect to their premise is that real estate is very different from the other industries mentioned in the paper, such as the travel and insurance industries. First, housing is a very heterogeneous product, with potential "lemons" and high search costs. Housing is very expensive, and it is usually associated with a long-term commitment on the part of the buyer. Also, although there are thousands of buyers and sellers in this market, there is a lack of good information about them and about their marginal willingness to buy and sell a property. Finally, real estate involves a number of contractual and legal issues that are far more complex than those that arise when someone buys an airline ticket.

All of these factors indicate that it is not easy or even feasible to replace services provided by real estate agents solely with the use of the Internet. Agents can actually be very useful for buyers and sellers. Buyers, for example, need help searching for and evaluating properties, bidding the right price, getting answers to legal questions, and dealing with a number of other issues on which they may lack experience. On the other side, given their up-to-date knowledge of the local market, real estate agents can also help sellers with issues such as price setting, advertising, staging, and negotiating price. These services [End Page 89] can be very important, especially for risk-averse sellers and buyers who have poor knowledge of the local market. Of course, homebuyers and sellers are heterogeneous, so the relevance of the services offered will differ for different households.

Given the potentially high value of the service and the heterogeneity in customers, is the real estate agent's fee right? Fees usually are calculated at a fixed rate, 5 percent or 6 percent of the transaction price. It is hard to believe that a fixed-rate fee corresponds to the marginal cost of selling a house, however, especially with heterogeneous houses and customers. Collusion could explain why a $50,000 house and a $10,000,000 have a similar fixed-rate fee, and Levitt and Syverson are looking for evidence of the actions of a potential cartel.

But do real estate agents really charge the same fee for high-priced houses? That is not very likely, since customers for such properties could request rebates, gifts, and so forth from agents. While evidence on this point is only anecdotal, if negotiation exists for customers who buy high-priced houses, why cannot all homeowners negotiate a lower fee? Again, collusion would explain this effect, since discounting by a real estate agent may be acceptable only among agents who sell high-priced houses. If that is true, then in the data we would potentially observe a small proportion of flat-fee agents involved in selling high-priced houses and a high proportion selling homes in the middle price range. Households in the low-price segment of the housing market would still use real estate agents since the fixed-rate commission translates into a small fee for such houses.

In addition to finding indirect evidence on collusion, such as by using the test proposed above, can we find any direct evidence in the literature? Very little is available, with the exception of the 1983 Federal Trade Commission report.1The commission's survey showed that newspapers had refused to print advertisements from more than one-third of discount brokers because of threats from real estate agents. Also, 90 percent of discount agents reported that real estate agents usually disparaged their services, labeling them as illegal, unethical, and unprofessional. Finally, real estate agents are in fact reluctant to show homes that are listed for sale by the owner. However, none of this evidence proves collusion, since steering away buyers can easily be justified by differences in housing quality or in the quality of the service provided by discount agents. This identification problem is compounded by the fact that discount agents account for only a tiny fraction of the market.

Despite all the difficulties involved in answering such questions, Levitt and Syverson do not shy away from trying to detect collusion in real estate markets [End Page 90] and using indirect evidence from housing market transactions to do so. Using MLS listings for three markets where they observed whether a housing transaction occurred through a flat-fee agent, the authors in practice estimate the correlation between the indicator for flat-fee agency and three important outcomes: final transaction price, whether a sale ever occurred, and the number of days that the property was on the market. They find that houses represented by a flat-fee agent sold for the same prices as similar houses sold by traditional agents. But flat-fee agent houses had a greater number of days on the market, which may indicate that real estate agents collude to steer buyers away from properties listed with flat-fee agents. However, as Levitt and Syverson point out, those results hold only under a number of strong assumptions. For example, discount agents (in practice, FSBO homeowners) would have to have the house-selling ability of traditional real estate agents (including the ability to handle advertising, staging, price setting, and so forth); in addition, it would have to be assumed that sellers do not self-select into discount agency services. Finally, the welfare implications of these estimates are subject to the same caveats.

1 Federal Trade Commission, "The Residential Real Estate Brokerage Industry" (1983).

In future work, several interesting lines of research could be explored:

  • —Is there a tipping point as the share of discount agents increases?

  • —Will discount agency business slow down during a recession? (Yes, if real estate agents provide a valuable service.)

  • —Why do discount agents still constitute only a small fraction of the market? (It is possible that traditional real estate agents' services are valuable for certain price ranges and types of buyers and sellers, as already argued in this comment.)

The key to answering these questions lies in collecting better data. Pre-transaction data sets would be especially useful, allowing researchers to understand how homebuyers and sellers choose real estate agents'services. Such data would also provide the chance to better estimate the value of certain types of services and the potential impact of collusion on their provision and pricing. The paper by Levitt and Syverson advances the literature in this direction, in addition to pointing out new and interesting testable implications of collusion in this vast market. More scholars should follow their lead. [End Page 91]

Appendix

The three data sets used in this analysis (for Cook County, Sacramento, and Santa Cruz County) were provided to us by real estate agents. In our data request we asked the agents to provide all listings of single-family homes posted between January 2004 and March 2006, regardless of whether a sale occurred. To minimize the number of sample homes whose listings were truncated, we restrict the actual sample to run from January 2004 to March 2005.

Exclusions

In order to link multiple listings for the same property, we exclude from the sample any listing that does not have a parcel identification number, as well as listings in which different properties appear to be tagged with the same identification number (if two properties in different ZIP codes carry the same identification number, for example). We also drop a small number of outlier properties that sold for less than $50,000, sold for more than five times the median in the ZIP code, or were missing identifiers for the listing agency.

Defining Variables

The variables included in the raw data differ across the three data sets depending on the information in each MLS and how its data fields were constructed. We use a core set of variables common to the three data sets, with supplementary control variables added when available in only one of the data sets (for example, an indicator for homes in a coastal region in Santa Cruz County).

The manner in which flat-fee agents are identified in the MLS data varies across region. For Cook County, we use two different indicators. The first is the Special Compensation Information (SCI) field included in the MLS, which asks whether a listing is classified either as "limited service" or "exception." The second indicator is whether the buyer's agent is instructed to contact the homeowner directly for arranging showings, instead of the usual arrangement in which the buyer's agent and the listing agent communicate directly. Both measures are imperfect, but they are positively correlated (ρ = 0.4). We define an agency as a flat-fee agency if more than 40 percent of its listings qualify under either measure of flat-fee agency, limiting the sample to agencies that [End Page 74] had at least ten listings over the 15-month period that we examine. The distribution, as would be hoped, is bimodal, with most agencies having very few listings that qualify and a small number of agencies having virtually all its listings qualify. In Sacramento, there is a field denoting whether a listing is "MLS only" or whether the listing is full service. In Santa Cruz County, the agent who provided the data also gave us a list of all flat-fee agencies operating in the area.

The Sacramento and Santa Cruz County data include the house's square footage, but in Cook County that field is nearly always empty. We exclude a small number of homes with fewer than 300 square feet or more than 8,500 square feet listed. We truncate the indicator variables for the number of bedrooms at six, the number of bathrooms at four and a half, and the garage size at three cars. We include indicator variables for house architectural style and exterior type; the set of included indicators varies according to the house types present in the different regions.

We include ZIP code fixed effects in the regressions, except in Santa Cruz County, where even more disaggregated MLS areas exist, and we include a fixed effect for each. In all cases, keywords from the marketing text that accompanies a listing were identified through a text search for these phrases. [End Page 75]

Table A1.
Full Regression Results, Cook County

(1) (2) (3)
Dependent House Days on
variable ever sold marketb ln(sale price)b

a. Column 1 reports the marginal effects from the probit estimation corresponding to row 1, column 4, of table 2. Columns 2 and 3 are OLS estimates corresponding to row 1, column 4, of tables 3 and 4 respectively. Included in all specifications, but not reported in the table, are ZIP code fixed effects, year-month interactions, and a large number of keyword descriptions. *Significant at the .05 level; ** significant at the .01 level.
b. Zeros and periods appearing alone in a row indicate an excluded dummy variable.
Flat fee -0.117 33.28** -0.0000280
(0.027) (5.68) (-0.00)
Age (years)
     1–5 -0.120 23.26* 0.0799**
(0.158) (9.122) (0.019)
     6–10 -0.031 0 0
(-0.133) . .
     11–25 -0.047 -1.720 -0.0536**
(0.136) (8.480) (0.018)
     26–50 -0.031 -3.212 -0.0811**
(0.120) (7.986) (0.017)
     51–100 -0.038 -2.240 -0.0609**
(0.120) (8.183) (0.018)
     100+ -0.057 -1.190 0.0114
(0.143) (10.601) (0.023)
     Age missing 30.38 -0.0874
(53.476) (0.086)
Bedrooms, number
     2 0.127 -7.529 0.230**
(0.044) (27.513) (0.049)
     3 0.179 -2.103 0.316**
(0.088) (27.473) (0.049)
     4 0.134 7.284 0.390**
(0.069) (27.555) (0.049)
     5 0.112 1.923 0.458**
(0.049) (27.760) (0.050)
     6 0.074 19.62 0.551**
(0.061) (29.433) (0.055) [End Page 76]
Bathrooms, number
     1 0.033 0 0
-0.025 . .
     1.5 0.041 4.384 0.0572**
(0.022) (2.596) (0.006)
     2 0.024 4.417 0.0721**
(0.023) (2.477) (0.005)
     2.5 0.030 8.934* 0.148**
(0.019) (3.843) (0.008)
     3 0.002 9.360 0.145**
(0.023) (5.165) (0.011)
     3.5 0.018 17.72** 0.261**
(0.019) (6.576) (0.013)
     4.5 24.85** 0.378**
(8.492) (0.019)
Other rooms, number
     3 -0.006 2.016 0.0236**
(0.019) (3.282) (0.007)
     4 -0.004 5.125 0.0571**
(0.019) (3.369) (0.007)
     5 -0.020 7.537* 0.0947**
(0.021) (3.725) (0.008)
     6 -0.031 10.01* 0.121**
(0.024) (4.464) (0.009)
     7 -0.045 21.93** 0.151**
(0.029) (6.319) (0.013)
     8 -0.013 25.28** 0.205**
(0.030) (8.851) (0.017)
Master bedroom bath 0.003 5.262* 0.0441**
(0.011) (2.589) (0.005) [End Page 77]
Fireplaces, number
     1 -0.002 -1.976 0.0762**
(0.009) (1.918) (0.004)
     2 -0.001 5.882 0.146**
(0.015) (3.872) (0.008)
     3 -0.002 11.88 0.304**
(0.024) (9.077) (0.021)
Air conditioning
     Central air 0.042 -2.235 0.0727**
(0.019) (3.925) (0.010)
     Other 0.057 4.139 0.0421**
(0.016) (4.566) (0.011)
Car spaces in garage, number
     1 0.015 -9.607 0.0351*
(0.027) (7.201) (0.016)
     2 0.013 -12.93 0.0717**
(0.028) (7.050) (0.015)
     3 -0.046 -2.508 0.130**
(0.035) (8.416) (0.018)
Style
     American four-square -0.010 5.180 0.0613**
(0.032) (0.62) (3.66)
     Bi-level 0.019 -6.250 -0.0134
(0.016) (-1.64) (-1.69)
     Bungalow 0.042 -10.47** -0.0399**
(0.015) (-2.58) (-4.66)
     Cape Cod 0.063 -9.201* -0.0495**
(0.014) (-2.24) (-5.62)
     Contemporary 0.024 19.83* -0.0146
(0.021) (2.15) (-0.84) [End Page 78]
     Colonial 0.029 -1.889 0.0644**
(0.014) (-0.46) (7.47)
     Cottage -0.023 -4.848 -0.0259
(0.042) (-0.48) (-1.25)
     English 0.037 -7.452 0.0750**
(0.024) (-0.96) (4.70)
     Farmhouse -0.006 -1.902 -0.0148
(0.032) (-0.28) (-0.78)
     French provincial 0.072 10.52 0.114**
(0.030) (0.48) (3.65)
     Georgian 0.052 4.295 0.00269
(0.018) (0.77) (0.22)
     Prairie 0.060 3.548 0.157**
(0.029) (0.26) (6.06)
     Quad-level 0.081 -11.44 -0.0400*
(0.024) (-1.27) (-2.47)
     Queen Anne 0.008 -7.877 0.0179
(0.042) (-0.74) (0.79)
     Ranch 0.050 -6.641 -0.0425**
(0.013) (-1.94) (-5.89)
     Step-up ranch -0.006 -3.254 -0.0421**
(0.032) (-0.49) (-2.83)
     Traditional 0.051 11.43 0.0687**
(0.016) (1.64) (5.31)
     Tri-level 0.036 -3.266 -0.0193*
(0.016) (-0.76) (-2.33)
     Tudor 0.088 -3.706 0.0879**
(0.022) (-0.44) (4.41) [End Page 79]
     Victorian 0.005 -8.897 0.0529**
(0.031) (-1.01) (2.63)
     Other 0.010 7.060 0.0429
(0.031) (0.74) (1.94)
Exterior
     Brick 0.016 -10.94 0.00144
(0.022) (-1.62) (0.10)
     ext_cd (cedar shingle) 0.018 -7.957 0.00710
(0.026) (-0.93) (0.40)
     ext_avs (aluminum, vinyl, or steel) 0.025 -9.920 -0.0535**
(0.022) (-1.47) (-3.79)
     ext_fr (woodframe) 0.018 -3.685 -0.0225
(0.026) (-0.49) (-1.35)
     Stucco 0.045 -9.355 0.0200
(0.024) (-1.17) (1.13)
Agent-owned house -0.042 -0.421 0.0270**
(0.017) (3.988) (0.008)
Constant 71.95* 11.98**
(30.495) (0.057)
N 10,999 9,103 9,103

[End Page 80]
Table A2.
Full Regression Results, Sacramento

(1) (2) (3)
Dependent House Days on
variable ever sold marketb ln(sale price)b

a. Column 1 reports the marginal effects from the probit estimation corresponding to row 1, column 4, of table 2. Columns 2 and 3 are OLS estimates corresponding to row 1, column 4, of tables 3 and 4 respectively. Included in all specifications, but not reported in the table, are ZIP code fixed effects, year-month interactions, and a large number of keyword descriptions. *Significant at the .05 level; **significant at the .01 level.
b. Zeros and periods appearing alone in a row indicate an excluded dummy variable.
Flat fee -0.097 8.739** 0.0166
(0.024) (2.305) (0.009)
Age (years)
     1–5 -0.119 8.704 0.0361
(0.236) (9.618) (0.050)
     6–10 -0.093 9.499 0.0414
(0.236) (9.630) (0.050)
     11–25 -0.098 12.41 -0.00115
(0.197) (9.617) (0.050)
     26–50 -0.108 14.64 -0.00644
(0.202) (9.624) (0.050)
     51–100 -0.137 15.88 0.0807
(0.251) (9.601) (0.050)
     100+ -0.249 17.89** -0.00194
(0.321) (3.645) (0.021)
     Age missing -0.265 27.52 -0.0725
(0.396) (20.574) (0.058)
Square footage 0.000 0.0120** 0.000453**
(0.000) (0.003) (0.000)
Square of square footage 0.000 0.00000100 -2.55e-08**
(0.000) (0.000) (0.000)
Bedrooms, number
     2 0.030 -12.51** 0.160**
(0.017) (3.269) (0.015)
     3 0.042 -15.01** 0.232**
(0.021) (3.354) (0.016)
     4 0.026 -16.32** 0.215**
(0.020) (3.444) (0.016)
     5 0.020 -17.02** 0.168**
(0.020) (3.768) (0.017) [End Page 81]
     6 -0.006 -19.46** 0.151**
(0.033) (5.825) (0.026)
Bathrooms, number
     1 -0.980 -5.997* -0.0416**
(0.001) (2.784) (0.012)
     1.5 -0.931 -4.723 -0.0731**
(0.002) (2.979) (0.013)
     2 -0.880 -4.438 -0.0123
(0.010) (2.855) (0.012)
     2.5 -0.976 -3.529 -0.0334**
(0.001) (2.943) (0.013)
     3 -0.971 -4.517 -0.0254
(0.001) (3.013) (0.013)
     3.5 -0.910 -0.687 0.0107
(0.002) (4.235) (0.018)
     4.5 -0.910 -2.577 0.0134
(0.002) (5.427) (0.022)
Master bedroom bath 0.017 -1.886** 0.00157
(0.006) (0.631) (0.003)
Fireplace 0.012 -2.794** 0.0377**
(0.006) (0.647) (0.003)
Central air -0.003 -2.505* 0.0215**
(0.007) (0.993) (0.005)
Car spaces in garage, number
     2 0.008 -1.476 -0.0232
(0.013) (2.514) (0.015)
     3 0.028 -3.828 -0.0254
(0.011) (2.196) (0.014) [End Page 82]
     4 0.050 -6.181** 0.0815**
(0.014) (2.253) (0.014)
     5 0.050 -10.76** 0.0995**
(0.010) (2.398) (0.015)
     6 0.004 -5.693 0.178**
(0.078) (5.064) (0.048)
Style
     A frame -0.022 3.129** -0.00108
(0.009) (1.086) (0.004)
     Colonial -0.013 1.571 0.00330
(0.026) (4.462) (0.018)
     Contemporary 0.017 -0.653 0.00515*
(0.005) (0.542) (0.002)
     Cottage 0.025 -0.950 0.0501**
(0.009) (1.340) (0.007)
     Mediterranean 0.012 0.263 0.00808
(0.010) (1.342) (0.005)
     Ranch 0.018 -1.117 0.0294**
(0.005) (0.614) (0.003)
     Spanish 0.010 -4.169 0.0111
(0.023) (2.748) (0.015)
     Tudor 0.023 -7.283* 0.108**
(0.019) (2.899) (0.014)
     Victorian -0.004 -7.557 0.0124
(0.037) (6.457) (0.034)
     Other -0.006 12.10* 0.0343
(0.046) (5.955) (0.036)
Exterior
     Brick 0.015 1.636 0.0303**
(0.015) (1.729) (0.008) [End Page 83]
     Shingle 0 0
. .
     Siding 0.009 2.521 0.0189*
(0.015) (1.690) (0.008)
     Stone -0.010 0.861 0.0264**
(0.020) (2.022) (0.009)
     Stucco 0.006 1.633 0.0122
(0.015) (1.619) (0.008)
     Wood -0.001 2.986 0.0122
(0.015) (1.671) (0.008)
     Other -0.030 4.377* 0.0224*
(0.021) (2.142) (0.010)
Agent-owned house -0.003 -1.014
0.00803
(0.009) (1.018) (0.005)
Constant 63.48**
11.60**
(11.532) (0.059)
N 26,368 23,286 23,286 [End Page 84]

Table A3.
Full Regression Results, Santa Cruz County

(1) (2) (3)
Dependent House Days on
variable ever sold marketb ln(sale price)b

a. Column 1 reports the marginal effects from the probit estimation corresponding to row 1, column 4, of table 2. Columns 2 and 3 are OLS estimates corresponding to row 1, column 4, of tables 3 and 4 respectively. Included in all specifications, but not reported in the table, are ZIP code fixed effects, year-month interactions, and a large number of keyword descriptions. *Significant at the .05 level; **significant at the .01 level.
b. Zeros and periods appearing alone in a row indicate an excluded dummy variable.
Flat fee 0.043 8.535 0.0222
(0.024) (5.738) (0.022)
Age (years)
     1–5 0.00874 -35.27** -0.0662
(0.032) (10.132) (0.038)
     6–10 0.007 -31.26** -0.106**
-0.035 (10.311) (0.040)
     11–25 0.001 -31.99** -0.0814*
(0.030) (8.978) (0.035)
     26–50 0.022 -31.79** -0.0719*
(0.030) (9.103) (0.035)
     51–100 -0.006 -29.65** -0.0376
(0.033) (9.286) (0.037)
     100+ -0.010 -27.54** -0.0678
(0.035) (9.371) (0.037)
     Age missing 0.000 0 0
0.000 . .
Square footage 0.000 0.0255** 0.000354**
(0.000) (0.007) (0.000)
Square footage missing -0.031 19.51* 0.450**
(0.040) (8.364) (0.044)
Square of square footage 0.000 -2.40e-6 -2.25e-08**
(0.000) (0.000) (0.000)
Bedrooms, number
     2 0.001 -9.552 0.0445
(0.028) (5.277) (0.033)
     3 -0.006 -9.619 0.0467
(0.031) (5.670) (0.035)
     4 -0.005 -15.57* 0.0469 [End Page 85]
(0.035) (6.439) (0.040)
     5 -0.041 -20.24 -0.0344
(0.056) (10.909) (0.054)
     6 -0.022 18.21 -0.0570
(0.068) (43.302) (0.101)
Bathrooms, number
     1 0.018 -9.640 -0.188**
(0.021) (13.237) (0.063)
     1.5 -15.70 -0.181**
(12.909) (0.064)
     2 0.039 -13.81 -0.167**
(0.019) (12.564) (0.062)
     2.5 0.012 -16.92 -0.206**
(0.023) (12.523) (0.061)
     3 0.000 -13.25 -0.150*
(0.026) (12.624) (0.061)
     3.5 -0.006 -6.131 -0.100
(0.036) (14.582) (0.060)
     4.5 -0.004 0 0
(0.041) .
Fireplace 0.000 3.559 0.0381**
(0.014) (2.447) (0.012)
Central air -0.012 -6.310 0.0249
(0.025) (3.980) (0.019)
Car spaces in garage, number
     2 0.044 2.589 -0.0774**
(0.024) (8.072) (0.029)
     3 0.062 -3.526 -0.0348
(0.021) (7.986) (0.029) [End Page 86]
     4 0.082 -8.265 -0.0179
(0.034) (7.903) (0.029)
     5 -0.059 -1.185 0.110**
(0.017) (10.288) (0.041)
Style
     Cabin -0.023 10.59 -0.0505
(0.021) (5.658) (0.032)
     Cape Cod -0.030 25.65* 0.116*
(0.026) (10.207) (0.049)
     Country -0.004 -0.440 0.0938**
(0.031) (6.036) (0.036)
     Colonial 0.097 -7.684 -0.0110
(0.133) (18.925) (0.068)
     Cottage -0.036 -1.887 0.0179
(0.014) (2.824) (0.016)
     Mediterranean 0.012 8.119 0.0631*
(0.029) (6.544) (0.031)
     Ranch -0.024 -2.371 0.000346
(0.015) (2.540) (0.013)
     Spanish -0.037 -1.252 -0.00462
(0.026) (4.945) (0.038)
     Traditional -0.018 0.340 -0.00404
(0.014) (2.606) (0.013)
     Victorian 0.034 1.561 0.0691*
(0.038) (7.850) (0.035)
     Other -0.042 10.54 0.0476
(0.027) (14.295) (0.052)
     Missing -0.001 3.540 -0.0193
(0.015) (2.881) (0.013) [End Page 87]
Exterior
     Brick -0.028 -5.069 0.0590
(0.058) (9.137) (0.040)
     Shingle -0.016 15.36 0.142**
(0.057) (10.915) (0.046)
     Stucco -0.044 0.268 0.0428
(0.050) (5.556) (0.028)
     Wood -0.043 2.379 0.0483
(0.064) (5.477) (0.028)
     Other -0.032 -33.78** 0.0267
(0.084) (11.698) (0.077)
     Missing -0.018 2.853 0.0512
(0.056) (5.655) (0.030)
Agent-owned house -0.008 4.321 0.00910
(0.017) (3.393) (0.013)
Coast 0.027 9.642 0.238**
(0.028) (5.396) (0.040)
Fixer-upper -0.013 0.656 -0.123**
(0.018) (4.794) (0.025)
exc1031 (special tax status) 0.024 5.599 0.0373
(0.025) (4.679) (0.026)
Constant -9.517 12.10**
(21.663) (0.112)
N 3,298 2,564 2,564 [End Page 88]

Acknowledgment

This paper was prepared for the 2007 Brookings-Wharton Conference on Urban Affairs. We thank Gary Burtless, Fernando Ferreira,Aviv Nevo, Janet Rothenberg Pack, and Canice Prendergast for helpful discussions. Marina Niessner and Margaret Triyana provided excellent research assistance. Correspondence can be addressed to either of the authors at Department of Economics, University of Chicago, 1126 E. 59th Street, Chicago, IL 60637.

Footnotes

1. Travel agency employment is from the U.S. Bureau of Labor Statistics (2007); enplanements are from the U.S. Bureau of Transportation Statistics (2007). Stock trade commission data are from Bakos and others (2005). Direct insurance sales figures are from Flannagan and Yates (2001) and Flannagan and Aartrijk (2007).

2. Global Aviation Associates (2002).

3. Home sales are from the NAR (2007) and the U.S. Census Bureau (2007). Sale price data are from the U.S. Department of Housing and Urban Development (2007), and the median commission fee is from Federal Trade Commission and U.S. Department of Justice (2007).

4. National Association of Realtors (2006).

5. One piece of evidence against this explanation is the fact that in two of the three markets that we examine in this paper (Cook County and Sacramento), the share of home listed by flat-fee agents is higher in zip codes in which the housing stock is more heterogeneous. If heterogeneity increases the value of the agent, then the opposite pattern would be expected. In the third market (Santa Cruz), no clear relationship emerges.

6. See, for example, Federal Trade Commission and U.S. Department of Justice (2007) and the citations therein. The Hsieh-Moretti point is actually more subtle. When commission rates are fixed (or nearly so) and there is free entry of agents into the market, the cost to a given agent of selling a home necessarily rises with overall house prices in the market, because more agents vie for each listing. However, the cost increases because agents have to work harder to obtain each listing, not because it becomes harder to sell a home once a listing contract is signed. It is the latter cost (what one might term the "technological cost" of selling a home) that seems unlikely to be proportional to the sale price, even though that is what is implied by the competitive market model.

7. United States v. National Association of Realtors, Civil Action No. 05 C 5140. In a proposed settlement filed in May 2008, the NAR agreed to end several challenged policies and practices, including the Virtual Office Website policy described here.

8. Federal Trade Commission and U.S. Department of Justice (2007).

9. Much of what we cover here is discussed in greater detail in Federal Trade Commission and U.S. Department of Justice (2007).

10. National Association of Realtors, "Questions and Answers on Virtual Office Websites (VOWs)" (www.realtor.org/mempolweb.nsf/pages/VOWQandA).

11. On the day that the DoJ filed the initial complaint, the NAR modified its VOW policy to remove the selective opt-out option, claiming that it had removed the provision of the policy that was at issue in the antitrust action. Some months later, the NAR filed a motion to dismiss the DoJ's complaint. The DoJ, which had in the meantime filed an amended complaint responding to the NAR's modified policy, argued in both the amended complaint and the response to the motion that the blanket opt-out and the selective opt-out were similarly anticompetitive. The NAR motion was denied in November 2006.

12. FTC (1983).

13. See, for example, Birger and Caplin (2004).

14. For example, Levitt and Syverson (forthcoming).

15. For example, Tirole (1987).

16. Vives (1999).

17. We use "parameter" very loosely with regard to . and d. They do not reflect fundamentals of the market's supply or demand side; they instead are reduced-form embodiments of (possibly very) complicated equilibrium outcomes that we do not attempt to model here. We simply find them a useful device for summarizing some of the economic influences on traditional agents' cooperation choices.

18. Hahn, Litan, and Gurman (2006).

19. We do not have a sense of the size of the feedback mechanism. We are not aware of any market that has yet experienced such a shift; that suggests that it is fairly modest.

20. Our data on Cook County do not include the city of Chicago (which has a small share of single-family homes) and is limited to the thirty-three suburbs with the greatest number of sales of existing single-family homes.

21. See Steven D. Levitt, "Flat-Fee Real Estate Agents . . . I Need Your Help!" (http://freakonomics.blogs.nytimes.com/2006/03/04/flat-fee-real-estate-agentsi-need-your-help/). We received roughly fifty inquiries from curious real estate agents. When the agents realized the magnitude of the task and the technical sophistication required to fulfill our request for data, most elected not to participate.

22. In Santa Cruz County, the MLS-defined areas are generally smaller than ZIP codes, so we use those areas in place of ZIP codes.

23. For precise definitions of how we define each of the variables used, see the appendix.

24. Summary statistics for variables that take on many possible values, like housing style, housing exterior, and ZIP code are not shown in the table, but full summary statistics are available from the authors.

25. The MLSNI rules committee voted in a tiered structure of fines for agents who engage in two types of deception: artificially manipulating time on the market and not properly completing the field that denotes flat-fee or discount seller relationships.

26. Except, perhaps, to facilitate showing of the house—information already conveyed in other MLS fields.

27. Another possible indicator of any poor pricing of discount agent houses would be if there was a systematically larger divergence between the actual and predicted original list prices for flat-fee homes. (That would imply that the market value of a home as reflected in the sale price was further from the list price.) In fact, however, the mean squared deviations from the fitted original listing price are smaller for flat-fee homes in all three markets.

28. It is important to note that such a difference could arise endogenously when a full-service agent manipulated the seller to accept a suboptimal offer because the incentives of the agent and the seller diverged. See Rutherford, Springer, and Yavas (2005) and Levitt and Syverson (forthcoming).

29. Many discount agents provide these services on a fee-per-hour basis (typically at a cost of $50 to $100 an hour). However, our discussions with discount agents suggest that take-up rates for the services are low.

30. It is perhaps surprising that homes represented by flat-fee agents take longer to sell but eventually sell for the same price. One might expect that both the time margin and the price margin would be affected. Certainly if a homeowner wanted to trade off between those dimensions, there is the possibility. One partial answer to this puzzle is that the implied cost of the extra time in the market (discussed below) is so small that it would not be worth it to a homeowner to accept substantially lower offers to partially offset the longer waiting time.

31. On the other hand, the variance in time on the market also increases with the use of a flat-fee agent. Increased uncertainty about the timing of a sale carries costs separate from the length of time on the market.

References

Bakos, Yannis, and others. 2005. "The Impact of E-Commerce on Competition in the Retail Brokerage Industry." Information Systems Research 16 (4): 352-71.
Birger, Jon, and Joan Caplin. 2004. "The 4 ½ Percent Solution." Money 33 (10): 104-9.
Brock, William A., and Jose A. Scheinkman. 1985. "Price Setting Supergames with Capacity Constraints." Review of Economic Studies 52 (3): 371-82.
Emre, Onsel, Ali Hortaçsu, and Chad Syverson. 2007. "E-Commerce and the Market Structure of Retail Industries." Unpublished manuscript.
Federal Trade Commission. 1983. "The Residential Real Estate Brokerage Industry."
Federal Trade Commission and U.S. Department of Justice. 2007. "Competition in the Real Estate Brokerage Industry."
Flannagan, Madelyn, and Peter van Aartrijk Jr. 2007. "The 2005 Property-Casualty Insurance Market: Softening Prices, Competitive Threats, Big Opportunities for Independent Agents and Brokers." Alexandria, Va.: Independent Insurance Agents and Brokers of America.
Flannagan, Madelyn, and Jeffrey M. Yates. 2001. "The Independent Agency System: A Time of Great Opportunity." Alexandria, Va.: Independent Insurance Agents of America.
Global Aviation Associates. 2002. "The Economics of Travel Distribution in an Internet Driven Environment." Washington (http://govinfo.library.unt.edu/ncecic/).
Green, Edward J., and Robert H. Porter. 1984. "Noncooperative Collusion under Imperfect Price Information." Econometrica 52 (1): 87-100.
Hahn, Robert W., Robert E. Litan, and Jesse Gurman. 2006. "Bringing More Competition to Real Estate Brokerage." Real Estate Law Journal 35: 86-118.
Hendel, Igal, Aviv Nevo, and François Ortalo-Magné. 2007. "The Relative Performance of Real Estate Marketing Platforms: MLS versus FSBOMadison.com." Working Paper 13360. Cambridge, Mass.: National Bureau of Economic Research.
Hsieh, Chang-Tai, and Enrico Moretti. 2003. "Can Free Entry Be Inefficient? Fixed Commissions and Social Waste in the Real Estate Industry." Journal of Political Economy 111 (5): 1076-122.
Isaac, R. Mark, Valerie Ramey, and Arlington Williams. 1984. "The Effects of Market Organization on Conspiracies in Restraint of Trade." Journal of Economic Behavior and Organization 5 (2): 191-222.
List, John A. 2007. "The Economics of Open Air Markets." Unpublished manuscript.
Levitt, Steven D., and Chad Syverson. Forthcoming. "Market Distortions When Agents are Better Informed: The Value of Information in Real Estate Transactions." Review of Economics and Statistics.
National Association of Realtors. 2007. Existing Home Sales. (www.realtor.org/research/research/ecoindicator).
Rutherford, R. C., T. M. Springer, and A. Yavas. 2005. "Conflicts between Principals and Agents: Evidence from Residential Brokerage." Journal of Financial Economics 76 (3): 627-65. [End Page 92]
Stigler, George J. 1964. "A Theory of Oligopoly." Journal of Political Economy 72 (1): 44-61.
Tirole, Jean. 1987. The Theory of Industrial Organization. MIT Press.
U.S. Bureau of Labor Statistics. 2007. "Employment, Hours, and Earnings from the Current Employment Statistics Survey" (data series CEU6056151001) (www.bls.gov/ces/home.htm).
U.S. Bureau of Transportation Statistics. 2007. "Air Carrier Summary: T1: U.S. Air Carrier Traffic And Capacity Summary by Service Class" (data series RevPaxEnplaned) (www.transtats.bts.gov/Fields.asp?Table_ID=264).
U.S. Census Bureau. 2007. New Residential Sales (www.census.gov/const/www/newressalesindex.html).
U.S. Department of Housing and Urban Development. 2007. "U.S. Housing Market Conditions" (www.huduser.org/periodicals/ushmc.html).
Vives, Xavier. 1999. Oligopoly Pricing. MIT Press. [End Page 93]

Previous Article

Comments

Next Article

Comment

Share