Sales Concessions in the US Housing Market

Transcription

1 J Real Estate Finan Econ DOI /s x Sales Concessions in the US Housing Market Darren K. Hayunga 1 # Springer Science+Business Media New York 2016 Abstract This article examines the use of concessions in the US housing market, specifically payments for closing costs, home warranties, and structural repairs. This is the first study to examine the motivations and characteristics of homeowners that utilize concessions. It also examines the impact concessions have on transaction prices and marketing durations. While the literature has attempted to determine if concessions can reduce marketing durations or increase transaction prices, the evidence is tainted by endogeneity and sample issues. Additionally, we find that relative bargaining power between buyers and sellers has a fundamental effect on how concessions alter prices and marketing durations. This aspect has been considered only narrowly in the extant literature. Our results demonstrate that when sellers have bargaining power, transactions including concessions exhibit higher prices and shorter marketing durations. Conversely, when buyers have greater negotiation leverage, transactions including concessions experience lower prices and longer marketing periods. Keywords Buyers incentives. Sellers concessions. Sellers motivations. Closing costs. Home warranties. Home repairs JEL Classification D12. R21. R31 Introduction In real property markets, owners use concessions to motivate potential tenants to rent and buyers to purchase their properties. This article examines the concessions used by homeowners to motivate potential purchasers, specifically payments for closing costs, I thank anonymous referees, editor C.F. Sirmans, Kelley Pace, and Anurag Mehrotra for helpful comments as well as the National Association of Realtors for the data. * Darren K. Hayunga hayunga@uga.edu 1 Department of Insurance, Legal Studies, and Real Estate, Terry College of Business, University of Georgia, Athens, GA 30602, USA

2 Hayunga home warranties, and structural repairs. We first investigate the motivations and characteristics of homeowners who include concessions to determine if any of these factors increase or decrease the propensity to use incentives. This analysis has not been considered previously by the housing literature. We examine motivations such as sellers urgency levels and reasons for selling as well as characteristics such as a seller s race, income, and age. We also investigate the impact concessions have on the transaction outcome of prices and time on the market (TOM). We question whether sellers are rewarded with shorter TOM as is commonly thought, and/or if they are able to capitalize concessions into transaction prices. The effect of concessions on transaction outcomes has been examined previously, but this study addresses a number of sample and econometric issues that cloud the extant literature and our understanding of the effect concessions have on prices and marketing durations. The first issue this study addresses is sample selection. Prior studies use MLS datasets, which primarily will capture concessions at listing and possibly during the marketing period. However, our discussions and experiences reveal that concessions are commonly introduced during negotiations between a specific prospective buyer and the seller. With the notable exception of Winkler and Gordon (2015), MLS datasets do not seem to record such late-stage concessions. For example, Soyeh et al. (2014) find only 3.3% of their sample include concessions. Also, Johnson et al. (2000) only analyze observations with transaction prices less than $100,000 because they do not find concessions used for higher-priced homes. We find that over 40% of the transactions in our sample include concessions and over 90% of those transactions have a transaction price over $100,000. The reason we are able to observe the greater use of concessions is that we use data from the National Association of Realtors (NAR). These data are ideally suited to our research objectives in four ways. The first is the higher capture rate, which is due to the NAR surveys being sent directly to sellers after the closing. The second is that the NAR survey is specifically tailored to collect the economic and demographic information that is pertinent to the sale but is generally unobserved in MLS and public datasets. These include the aforementioned sellers motivational factors and demographic information, as well as structural quality, nominal losses and gains, selling to acquaintances, and marketing methods beyond MLS. These measures thus allow us to document that the propensity to use concessions decreases when properties are sold to acquaintances, with greater sellers age levels, and when the sales are due to job relocations. We also find the propensity to include concessions increases when sellers indicate that their properties are too expensive to keep, when they move greater distances between the sale and purchase properties, and with longer holding periods The third way that the NAR dataset is well-suited to our research objectives is that its many unique variables provide strong instruments to control for simultaneity. This is critical because test statistics demonstrate that TOM, prices, and concessions are endogenous. The simultaneous solutions to the optimization of price and TOM while including incentives has not been considered by the literature. Concessions clearly are pecuniary benefits that directly impact net transaction prices. TOM can also be affected when comparing the net transaction prices to the property s service flow. For instance, adding an incentive without fully increasing the list price by the pecuniary benefit

3 Sales Concessions in the US Housing Market produces a housing service flow that is undervalued and therefore should result in a shorter TOM, all else held equal. The fourth way that the NAR observations fit our research objectives is that they form a national dataset. This allows us to understand the use of concessions beyond a local market. Further, we are able to examine the effect of concessions in hot (cold) markets when sellers (buyers) have more bargaining power. Recent literature has considered such situations but in narrower contexts. Winkler and Gordon (2015) examine foreclosed properties in Huntsville, Alabama and Soyeh et al. (2014) analyze the recent market downturn in Boca Raton, Florida. In each case, the authors are looking for unique market behavior and the impact of concessions based on buyers having more bargaining power. We take a more general approach by using the Carrillo (2013) measure of relative bargaining power. Consistent with concessions being introduced during negotiations, we find that relative bargaining power is a key factor in understanding the effect of concessions. When analyzing the sample prior to considering negotiation power, we find concessions have no significant impact on prices or TOM. However, this is an artifact of two countervailing effects. In markets where sellers have greater bargaining power, the results demonstrate that homes with payments for closing costs and credits for repairs experience higher transaction prices and shorter TOM. In markets where buyers have greater negotiating power, homes that include home warranties and credits for repairs exhibit lower transaction prices. Transactions that include home warranties also experience longer TOM in buyers markets. We detail these results along with the rest of our analysis in the balance of this article. We begin in the next section with further background on the various ways that concessions can impact prices and TOM. We then present the extant literature, the data sample, and the empirical results in subsequent sections. The final section presents our conclusion. Background and Motivation Since they have not been considered previously, one catalyst for this study is to examine sellers motivations and other characteristics that increase or decrease the inclusion of concessions in transactions. After addressing the econometric issues, we are also prompted to conduct this analysis due to the myriad impacts concessions can have on prices and TOM. The theories showing how concessions should affect transaction outcomes are standard search and bargaining. In contrast, the empirical findings can vary based on: the relative relations between the net transaction prices and total value of the housing service flows, the point in the marketing process that the concessions are introduced, the concession type, as well as the preferences and constraints of certain groups of buyers and sellers. Consider first concessions added at the time the home is listed on the market so that they are an additional amenity of the respective property. The effect of the concessions will be partly a function of the relative relations between net transaction prices and housing service flows. In the simplest case, if sellers increase their list prices by the full value of the concessions, expected net transaction prices will be comparable with other properties having the same service flows. Therefore, expected TOM should not change,

4 Hayunga the concessions will increase transaction prices given the higher list prices (conditional on the bargaining process), and so the econometrician should find a positive slope coefficient in price equations, ceteris paribus. If sellers do not increase their list prices or only partially relative to the concessions values, the properties are at least somewhat undervalued relative to other properties with the same service flows. Consequently, search theory shows that these properties should experience shorter TOM because the concessions increase the probability of sale (Wheaton 1990; Krainer 2001). Since the concessions are not fully priced, the slope coefficient on the incentive should be a reduced positive estimate or maybe even uncorrelated in the transaction prices model. Another situation that potentially undervalues certain properties can occur with the introduction of cash-constrained buyers. Assume again sellers increase their prices by the full amount of the concessions, but that the incentive in this case is payment of closing costs. All else held equal, a cash-constrained buyer will have a preference for the payment of closing costs over a substitute such as a list price reduction of the same amount. The net transaction prices will be equivalent, but a cash-constrained buyers will prefer the properties with the concession relative to others with the same service flow but without the added payment. The concession should be capitalized but, due to an increase in the probability of sale, search theory shows that TOM may also decrease conditional on the number of cash-constrained buyers in the market. At the other end of the price spectrum are possible situations when buyers do not fully value the concessions. Consider the case when a list price is too high due to necessary repairs, but a seller includes a credit that exactly offsets these costs. Conditional on the improvements meeting their tastes, buyers may prefer the repairs be completed by sellers prior to closing for two reasons. First, repairs completed by sellers are included in the total transaction prices and thus can be financed, which again helps cash-constrained buyers. Second, sellers may possess more information about costs relative to buyers. Sellers may have an information advantage in finding local contractors to perform the repairs, especially when purchasers move from other locations. The type of concession can also impact prices and TOM. For example, buyers may have a general preference for home warranties because sellers potentially know more about the quality of their properties and amenities. Buyers may thus prefer to reduce the costs of asymmetric information through the purchase of home warranties over list price reductions of the same amount. Payments for repairs offer an additional consideration from a seller s standpoint. Owners may prefer paying buyers in cash for repairs of, for example, outdated carpet or appliances rather than replacing these themselves. Sellers will not know a specific prospective buyer s penchant for quality and design so they prefer providing the payments at closing instead of guessing buyers tastes. A final factor to consider is the timing of concessions. When a concession is added at the time of listing or during the marketing period, the seller and prospective buyers can value it similarly to other transactional and structural features. However, what about the situations when buyers ask for concessions during the bargaining process? Our conversations with brokers and homeowners indicate that concessions are commonly included towards the end of the negotiations, and at two different points. The first is when a seller is attempting

5 Sales Concessions in the US Housing Market to close negotiations with a specific buyer and the incentive is a final inducement to transact. The other is when a credit for repairs is prompted by the home inspection. To mitigate asymmetric information, buyers typically employ home inspectors and a new round of negotiation can occur based on the inspection report. Payments for discovered repairs may then become a new condition of sale after the initial price had been agreed upon. In either of these situations, it will be difficult for sellers to capitalize the concession values into prices at the end of negotiations, especially if a particular owner is anxious to sell and in the midst of bargaining with a specific interested buyer. Anxious sellers and concessions added during the negotiation phase introduce bargaining theory and participants relative leverage levels, which may also impact the effect concessions have on transaction outcomes. When sellers have bargaining power, concessions may be additional benefits that increase net transaction prices and/ or decrease TOM. Conversely, when they have bargaining power, buyers may seek economic rents in the form of concessions. Indeed, our findings support this reasoning regarding relative bargaining power. Literature Much of the concession literature in housing focuses on mortgage incentives. 1 The literature examining non-mortgage concessions consists of Asabere and Huffman (1997), Johnson et al. (2000), Soyeh et al. (2014), and Winkler and Gordon (2015). The two early papers examine transaction prices but not TOM. None of the papers consider the simultaneous determination of prices, TOM, and concessions. Based on statistical tests, we find endogeneity is a fundamental consideration in modeling incentives and transaction outcomes. With the exception of Winkler and Gordon (2015), the literature also does not include TOM in price equations. We find this exclusion causes an omitted-variable bias. Our results demonstrate that TOM (prices) subsume the effect of concessions in price (TOM) equations. As mentioned in the introduction, the extant literature uses MLS datasets, which provides some understanding of the use of concessions at listing and during the marketing period. However, MLS datasets probably do not capture concessions introduced during negotiations unless brokers are diligent in entering the information post-closing. Consequently, using keyword searches of the agent s comment section within MLS, Soyeh et al. (2014) find only 3.3% of their sample includes concessions. Similarly, Johnson et al. (2000) analyze only observations of low cost homes because they do not find concessions used for transactions priced above $100,000. In contrast, Fig. 1 shows that the use of concessions spans the distribution of selling prices in the NAR data, albeit with a lower use of closing costs at the highest price points. The maximum house price in the first quantile in Fig. 1 is $111,000, and thus almost 90% of the 1 Mortgage concessions are primarily assumable loans, buydowns, and discount points. Literature examining these incentives includes Zerbst and Brueggeman (1977), Guntermann (1979), Colwell et al. (1979), Brueckner (1984), Sirmans et al. (1983), Smith and Sirmans (1984), and Ferreira and Sirmans (1989).

6 * 60% 50% Hayunga 40% 30% 20% 10% All concessions Home warranty Closing costs Repairs 0% Fig. 1 Percentage use of concessions within quantiles, with intra-quantile median transaction prices ($) on the y-axis total concessions are found in transactions with selling prices greater than $111,000. The exception to the underreporting based on using MLS datasets is Winkler and Gordon (2015). They use the Huntsville Alabama MLS and find that approximately 43% of foreclosed properties use concessions and almost 70% of non-foreclosed properties include concessions. In describing the data sample in the next section, we report that 44% of the NAR transactions include at least one concession. Data Sample We construct a sample from the NAR surveys for the US market from 2010 to 2012, which includes 2009 sales. We use more recent surveys because older questionnaires do not include critical information. For example, purchase price is one of the questions asked in the most recent surveys and is important since it is required to compute variables such as expected nominal gains and losses from sale and structure quality. By design, the surveys obtain the information important to the sale, including factors that are often not generally observed in housing data but that we find do affect transaction prices, TOM, and concessions. Table 1 details the many demographic and economic measures. Table 10 provides additional definitions of the variables. The properties are representative of US housing. The median residence sold is 25 years old with 3 bedrooms, 2 baths, and 2000 square feet. The median purchase price is $177,000 with a selling price of $220,000. In comparison, the US Census reports the median home value for the nation as $174,600 during Regarding concessions, the survey responders indicate their use of closing costs, credit towards repairs, paying for a home warranty, and other non-realty concessions. 2 Responders mark their use of concession but they do not reveal the dollar amount. We 2 The closing costs can include payment of condominium association fees but we restrict the sample to singlefamily residences and townhouses.

7 Sales Concessions in the US Housing Market Table 1 Descriptive statistics of data panel (N = 3302) Mean Median Std. Dev. Minimum Maximum Purchase price ($) 222, , , , ,600, List price ($) 300, , , ,250, Sale price ($) 277, , , ,590, Time on market (weeks) Home age (years) Square feet Number of bedrooms Number of bathrooms Holding period (years) Purchase home warranty Payment of closing costs Credit for repairs High urgency Some urgency No urgency Reason for selling Avoid foreclosure Relocation Family change Too expensive First time seller Short sale Sold to a friend Number of children African American Asian Caucasian Hispanic Income to 25k Income 25 35k Income 35 45k Income 45 55k Income 55 65k Income 65 75k Income 75 85k Income k Income k Income k Income k Income k Income k Income k Income k

8 Hayunga Table 1 (continued) Mean Median Std. Dev. Minimum Maximum Income 1000k Ages Ages Ages Ages Ages Ages Ages Ages Ages Ages Ages Ages Sellers ages (continuous variable) MLS listing Open house Internet marketing Magazine marketing Print marketing Sign in yard Social media Distance between sale and purchase properties 1 to 5 miles to 10 miles to 15 miles to 20 miles to 50 miles to 100 miles to 500 miles to 1000 miles More than Year Year Year Year Detached SFR Townhouse Suburb City Small town Resort

9 Sales Concessions in the US Housing Market thus code the concession as a binary variable equaling one when used in the transaction and zero otherwise. We analyze the primary concession types of payments for home warranties, closing costs, and repairs. Of the 3328 total properties sold during the sample period, 819 sellers (24.6%) provide home warranties. The next most common concession is assistance with closing costs, which is included in 707 transactions (21.2%). Credit toward repairs is the other main concession occurring in 230 transactions (6.9%). The total number of transactions that use at least one incentive is 1476 (44.4%). Figure 2 details the annual use of concessions, which is generally stable across the years. The survey provides a number of categories that record various reasons owners are selling their properties. Nineteen percent indicate a job relocation, 8% note a change in the family status such as divorce or the birth of a child, and 3% indicate the property is too expensive to keep. Three percent of the sample indicate a desire to avoid foreclosure. Note that since the surveys are sent to individuals, these properties are not realestate-owned properties. Regarding demographics, sellers incomes are generally distributed unimodally across the categories with the largest percentage having incomes between $100,000 and $125,000. Our analysis examines each of these categories coded as a binary variable and also a continuous measure of log income using the midpoint of each category with the top-coded upper category set to $1.5 M. When examining the sellers motivations to utilize concession, we find the continuous variable is more informative. In the analysis of the transaction outcomes, we use the binary variables. Sellers provide their ages and the proportion of responders at the various levels are consistent with the home-ownership lifecycle. The percentage of responders increases from 4% in ages to 11% across most of the working years. The proportion begins to decrease at ages and falls monotonically across the traditional retirement ages. Similar to sellers income levels, we compute a continuous variable for age 50% 40% 30% 20% All concessions Home warranty Closing costs Repairs 10% 0% Fig. 2 Annual use of concessions

10 Hayunga and find that it is more informative than the individual categories in the analysis of sellers motivations; otherwise, we use the binary variables. Other demographics indicate that first-time sellers are 38% of the sample, 18% of the sellers self-identify as highly urgent, and 43% are somewhat urgent. We note a majority of responders are Caucasian, which we use as the control group. While 93% of the responders identify themselves as Caucasian, this is less than the 95.8% in Harding et al. (2003a). Since the dataset provides the ZIP codes of the sold homes, we can control for property market conditions, labor markets, and other local considerations using fixed effects. We use up to 455 fixed effects at the 3-digit ZIP code level. The dataset also provides the ZIP code of the home purchased by the survey responder. We thus control for possible additional search and transaction costs using the separation distance between the sale and the purchase. Table 1 shows that 24% of the responders move 5 miles or less while 13% indicate a move of greater than 1000 miles. Using a continuous variable computed similar to the income and age measures, we find the median separation distance is 18 miles. There are a number of other transactional aspects that may correlate with prices, TOM, or concessions. These include the home age, location types such as city or small town, whether the home is a resort property, and numerous marketing methods in addition to MLS. We also create other possible determinants of prices, TOM, and the use of concessions. Since the dataset includes purchase prices, the first set are expected nominal losses and gains, structural quality, and holding period. Genesove and Mayer (2001) find that sellers set their list prices higher and experience longer TOM when they expect to suffer nominal losses upon sale. Bokhari and Geltner (2011) extend the model to incorporate expected gains. Losses and gains are the percent differences sellers will realize between the log purchase price and the expected log transaction price at the time of listing. The expected log transaction prices come from a hedonic price model using the sold properties in the sample. A positive value for expected losses indicates the percentage loss at the current average market price and is truncated from below at zero. Expected gain is the same measure but having a negative value and is and truncated above at zero. 3 Structure quality is the residuals from a hedonic model at the time of purchase. The residuals are the portion of the previous sale prices that the regression did not predict. To the extent these qualities do not change significantly over time, the residuals are a noisy but reasonable measure of their impact on future transaction prices. We also compute and include measures for structural atypicality in all the models. If atypicality results in thin markets for instance, sellers may be motivated to decrease TOM through the use of concessions. Following Harding et al. (2003b), we compute binary variables representing the upper and lower 3 It may seem more logical to code a loss as negative and a gain as positive. However, we follow the literature in our coding, which provides the benefit of a more straightforward interpretation in the price and TOM models. That is, the literature demonstrates that sellers who expect a loss will set higher list prices. With loss as positive values, the slope coefficient in the price models is positive, which coincides with the increase in prices. If the loss is coded as a negative, the positive relation between loss and prices will produce a negative coefficient in a price model and may cause more confusion in explaining how the negative parameter estimate equatestoanincreaseinprices.

11 Sales Concessions in the US Housing Market one percent of the distribution for multiple structural features. We define a new home as 2 years old or less while an old home is equal to or greater than 120 years old. A large home is greater than 5000 square feet and a small home is less than 900 square feet. A home has many bathrooms if there are 5 or more and many bedrooms if 6 or more. As another possible measure of atypicality, Harding et al. (2003b) include the inverse Mills ratio (IMR) using the Heckman two-stage correction model. The NAR data allow us to compute the IMR and potentially control for unique price-tom preference for those sellers who may be fishing for high transaction prices and do not mind staying on the market longer than expected. Table 11 details the first stage model used to compute the IMR. Motivations and Determinants of Concessions Use Our first empirical analysis investigates for factors that prompt as well as dissuade homeowners to include concessions in their transactions. We regress the binary concession variables in probit specifications against the many economic, demographic, structural and transactional variables. Table 2 reports the findings. Column 1 reports the determinants when a transaction includes any one of the concessions and the other three specifications model the specific incentive named in the column heading. Somewhat unexpectedly, we note first that urgency does not increase the propensity of sellers to include concessions. It appears that an overall urgency effect is being subsumed by the reasons for sale and other transaction characteristics. For example, consistent with a preference for lower TOM, owners who indicate they are selling because the property is too expensive to keep exhibit an increase in the use of concessions, specifically home warranties. Another factor is the various marketing methods, which generally exhibit increases in the probability of using concessions. Positive parameter estimates suggest sellers using additional marketing methods such as open houses and Internet marketing do so to increase buyers interest levels, which may be a proxy for urgency. Similarly, concessions provide a feature not found in other properties with the same service flow and can increase buyers interest levels. The increase in the use of concessions with properties that are MLS listings is also consistent with brokers advising their clients to include concessions to increase the probability of sale; the sale being necessary for commissioned brokers to receive compensation. In addition, brokers may earn compensation from the companies that sell the home warranties. Owners selling due to job relocation significantly decreases the propensity to use concessions, specifically home warranties. This may be a function of a third party in the form of a corporate relocation firminvolvedinthesetransactions. Another transaction that involves a third party is short sales, which consistently decrease the propensity to include concessions. This result is consistent with lending institutions being more apt to short properties in BAs is^ condition. In Table 2 and consistent throughout our analysis, we find that selling to a friend, acquaintance, or relative decreases the use of concessions. This is

12 Hayunga Table 2 Propensity to use concessions All Closing Home concessions costs warranty Repairs High urgency (0.026) (0.021) (0.023) (0.014) Some urgency (0.019) (0.016) (0.017) (0.010) Too expensive to keep *** *** (0.053) (0.052) (0.048) (0.030) Job relocation ** * (0.028) (0.023) (0.024) (0.015) Family change (0.031) (0.028) (0.030) (0.016) Avoid foreclosure * ** ** (0.046) (0.042) (0.043) (0.031) Sold to a friend *** *** *** * (0.038) (0.036) (0.036) (0.024) Log sellers incomes ** (0.015) (0.013) (0.013) (0.008) Log sellers ages *** *** *** (0.050) (0.043) (0.045) (0.030) African American * ** (0.067) (0.048) (0.053) (0.031) Asian (0.059) (0.051) (0.053) (0.035) Hispanic (0.057) (0.049) (0.047) (0.032) First time seller (0.021) (0.018) (0.019) (0.012) Expected loss *** *** *** *** (0.078) (0.068) (0.069) (0.038) Expected gain *** *** *** *** (0.062) (0.056) (0.055) (0.034) Short sale *** *** *** (0.052) (0.047) (0.052) (0.031) Log number of earners ** * * (0.045) (0.038) (0.040) (0.024) Log number of children (0.018) (0.016) (0.017) (0.010) Log separation distance *** * ** (0.005) (0.004) (0.004) (0.003) Quality *** *** *** *** (0.064) (0.056) (0.056) (0.033) Holding period *** *** *** *** (0.003) (0.003) (0.003) (0.001) Log square feet *** * **

13 Sales Concessions in the US Housing Market Table 2 (continued) All Closing Home concessions costs warranty Repairs (0.029) (0.024) (0.026) (0.016) Log home age *** *** *** (0.013) (0.011) (0.012) (0.007) MLS listing *** *** *** (0.025) (0.021) (0.023) (0.014) Open house *** ** *** (0.018) (0.016) (0.016) (0.011) Internet marketing *** ** *** (0.033) (0.029) (0.031) (0.020) Print marketing (0.020) (0.017) (0.018) (0.011) Sign in yard *** ** *** (0.023) (0.020) (0.021) (0.013) Detached SFR (0.028) (0.024) (0.025) (0.015) Suburban ** (0.032) (0.027) (0.029) (0.016) City (0.030) (0.025) (0.027) (0.016) Small town (0.019) (0.016) (0.017) (0.011) Resort property *** (0.086) (0.101) (0.081) (0.036) New home (0.127) (0.108) (0.132) Old home * * (0.090) (0.083) (0.107) (0.050) Small home * ** *** (0.071) (0.059) (0.077) (0.038) Large home (0.070) (0.078) (0.057) (0.032) Many bathrooms (0.092) (0.107) (0.080) (0.048) Many bedrooms (0.101) (0.081) (0.080) (0.041) Observations Pseudo-R Log-likelihood Using probit models, the table presents the partial derivatives and heteroscedasticity-robust standard errors in parentheses for each incentive type. The first column equals one if any one of the concessions is included in the transaction and zero otherwise **, ** and * denote p <0.01,p <0.05,andp < 0.10 respectively

14 Hayunga notable because the a priori relation is not entirely clear. Owners may offer better values to those they know by providing concessions but the result indicates the opposite. We find in analysis of the transaction outcomes in the next sections that selling to an acquaintance significantly and consistently decreases TOM but does not impact prices. The overall findings indicate that selling to acquaintances is done close to expected transaction prices so there is less motivation to include concessions as indicated in Table 2, and the transaction is conducted early in the marketing period. The results demonstrate that owner demographics also impact the propensity to include concessions. Greater sellers income and age levels decrease the probability of using concessions. These findings are consistent with lower preference for a quicker sale with higher incomes and being older. Race also impacts the use of concessions. Specifically, African American sellers provide payments for closing costs more often than the control group of non-hispanic Caucasian sellers. Splitting the sample on whether owners are expecting a gain or a loss upon sale is a highly significant determinant of including concessions. The positive slope coefficients on expected losses indicates that, while sellers may set higher list prices to mitigate a loss, they have a higher likelihood to offer concessions. Thus, observed transaction prices may be greater when sellers are expecting a loss, as in Bokhari and Geltner (2011), but net transaction prices will be closer to the expected market value when concessions are considered. Expected gains is a negative loss so the positive parameter estimate in Table 2 denotes a lower probability to use concessions. Also highly predictive of the propensity to use concessions is structural quality, which exhibits an inverse relation. The finding is consistent with higher quality homes requiring fewer repairs; vice versa for lower quality properties. The negative marginal slopes across all incentive types may also indicate buyers preference for higher quality homes. The amount of time sellers have owned their properties affects the use of concessions. Across concessions types, longer holding periods increase the propensity to use concessions. The positive coefficients suggests greater deferred maintenance. Owners tend to improve their properties near the time of sale; thus, longer holding periods indicate greater deferred maintenance that increases the propensity to include concessions such as repairs and home warranties. This is similar to structural quality since deferred maintenance lowers quality and thus increases the probability of incentives in the transaction. Two structural attributes are significant predictors of concession use. An increase in the size of a home decreases the use of closing costs, but increases the use of home warranties and repairs; vice versa for smaller homes. The negative slope in the closing costs model is consistent with the decrease in Fig. 1 across home price segments. Additionally, the positive parameter estimates on home age are highly intuitive. All else held equal, older homes should require more repairs and have older appliances and equipment. Home age does not impact closing costs, but older homes exhibit an increase in the probability of including home warranties and repairs.

15 Sales Concessions in the US Housing Market Transactions Prices We now consider the impact concessions have on transaction outcomes, beginning in this section with transaction prices. To do this, we must control for the simultaneity between prices, TOM, and concessions since the incentives we are examining are monetary benefits that influence net transaction prices. We use two stage least squares (2SLS) and instrumental variables (IV) in a system of equations. While finding highquality instruments is often a challenge using MLS datasets, the NAR data offer strong instruments that correlate with the various incentive types but with neither TOM in price equations nor prices in TOM specifications. 4 Our empirical framework is accordingly a system of simultaneous equations, with the following specifications detailing the price system: lnðp i Þj¼ α þ TOM i þ C i þ X i þ Z þ ω þ ϵ i ; lnðtom i Þ j¼ μ þ X i þ Z þ ω þ ξ i ; C i ¼ jθ þ X i þ Z þ ω þ η i ; P i is the price of house i and TOM i is the marketing duration for the same property. C i is equal to one if a concession is included in the transaction and zero otherwise. X i is a vector of demographic and economic covariates. Z and ω are spatial and annual fixed effects that capture market conditions. ϵ i, ξ i and η i are time-variant error terms that are assumed to be randomly distributed. When we include only the concession and initially withhold TOM (prices) as an independent variable in the price (TOM) models, we are able to compute the test statistics of endogeneity and report them at the bottom of each column. Because our models estimate a heteroscedasticity-robust variance-covariance matrix, we first report Wooldridge s (1995) score test of exogeneity that indicates whether endogeneity exists between the dependent and independent variables. A second test we include detects weak instruments. We report both Shea s partial R 2 and the F statistic. The partial R 2 measures the correlation between the IV and the instruments after partialling out the effect of the exogenous variables. Stock et al. (2002) note that the F statistic is often statistically significant even with weak instruments. They argue for the F statistic being greater than a threshold, generally set at approximately 10. To check the correlation between the instruments and the structural error term, a third test statistic we report is the Wooldridge (1995) robust score of overidentifying restrictions. Again, the Wooldridge score considers the robust variance-covariance matrix. All Concessions To confirm the endogeneity between marketing duration and prices, the specificationincolumn1intable3 includes the predicted value of TOM but initially 4 The concession IVs use linear probability models to avoid the forbidden regression and the incidental parameters problems. Appendix 3 further explains these issues and specifies the models.

16 Hayunga Table 3 Log prices using 2SLS (1) includes TOM IV Standard error (2) includes incentive IV Standard error Log TOM IV *** (0.015) All concessions IV *** (0.049) High urgency *** (0.017) *** (0.017) Some urgency *** (0.013) *** (0.013) Too expensive to keep ** (0.037) (0.038) Job relocation *** (0.019) *** (0.019) Family change (0.021) * (0.021) Avoid foreclosure (0.032) (0.032) Income 35 44k (0.048) (0.048) Income 45 54k (0.043) (0.043) Income 55 64k *** (0.041) ** (0.041) Income 65 74k *** (0.041) *** (0.041) Income 75 84k *** (0.038) *** (0.038) Income 85 99k *** (0.041) *** (0.042) Income k *** (0.039) *** (0.039) Income k *** (0.041) *** (0.041) Income k *** (0.043) *** (0.043) Income k *** (0.044) *** (0.043) Income k *** (0.044) *** (0.045) Income k *** (0.043) *** (0.043) Income k *** (0.057) *** (0.056) Income 1,000k *** (0.085) *** (0.085) Ages (0.027) (0.028) Ages (0.029) (0.029) Ages *** (0.030) *** (0.029) Ages *** (0.032) *** (0.030) Ages *** (0.035) *** (0.032) Ages * (0.037) *** (0.033) Ages *** (0.037) *** (0.033) Ages *** (0.039) *** (0.036) Ages ** (0.049) *** (0.047) Ages ** (0.058) *** (0.055) Ages (0.069) *** (0.069) African American *** (0.054) *** (0.056) Asian *** (0.036) *** (0.037) Hispanic (0.041) (0.040) Expected loss ** (0.080) (0.077) Expected gain ** (0.070) *** (0.067) Short sale *** (0.038) *** (0.037) Log number of earners (0.034) (0.034) Log number of children (0.013) (0.013) 1 5 miles S.D *** (0.023) * (0.023) 6 10 miles S.D *** (0.025) *** (0.025) miles S.D *** (0.027) *** (0.026) miles S.D *** (0.027) *** (0.028) miles S.D *** (0.026) *** (0.026)

17 Sales Concessions in the US Housing Market Table 3 (continued) (1) includes TOM IV Standard error (2) includes incentive IV Standard error miles S.D ** (0.032) ** (0.032) miles S.D ** (0.023) ** (0.023) miles S.D (0.025) (0.025) Quality *** (0.071) *** (0.067) Holding period ** (0.003) *** (0.003) Log square feet *** (0.019) *** (0.019) Suburban ** (0.021) * (0.021) City (0.020) (0.020) Small town *** (0.012) *** (0.012) Resort property *** (0.068) *** (0.066) New home ** (0.064) ** (0.060) Old home *** (0.050) *** (0.048) Small home (0.066) (0.065) Large home (0.060) (0.060) Many bathrooms *** (0.060) *** (0.060) Many bedrooms (0.066) (0.068) Log home age (0.009) First time seller (0.015) Detached SFR (0.018) Inverse Mills ratio (0.275) Constant *** (0.188) *** (0.184) Observations Adjusted R Robust score χ 2 (p-value) Partial R Robust F Over-identification (p-value) The models present determinants of housing prices. Model (1) includes the predicted value of TOM while Model (2) includes the predicted use of concessions. The models include fixed effects at the 3-digit ZIP code level as well as the year of listing. Heteroscedasticity-consistent and IV-robust standard errors in parentheses ***, ** and * denote p <0.01,p < 0.05, and p < 0.10 respectively withholds the use of concessions. The first stage equation is provided in Tables 13 and 14. The endogeneity statistics at the bottom of Model 1 demonstrate that TOM is quite endogenous with prices and the IV is neither weak nor overidentified. The TOM coefficient is positive and significant, which is consistent with search theory and the understanding that higher (lower) priced homes will tend to be on the market longer (shorter). Model 2 replaces TOM with the concessions IV. The reduced form equation is again detailed in Tables 13 and 14. The test statistics at the bottom of the Model 2 indicate that concessions are endogenous with prices, and the IV is neither weak nor over-identified. The parameter estimate on the predicted use of concessions is positive indicating an increase in prices when any incentive type is included in the transaction.

18 Hayunga Since TOM and concessions demonstrate endogeneity with transaction prices, we next model the full system of equations and report the three specifications in Table 4. The instruments are the same as those in Table 3. Column 1 details the price model. We observe that the inclusion of TOM subsumes the impact of concessions. TOM maintains a relation similar to Model 1 in Table 3 but the concessions measure is insignificant. The other independent variables in Model 1 meet with many of our expectations and offer a number of noteworthy fundamental findings that have not been reported previously in the literature. The first is that higher sellers urgency levels decrease prices. Using Kennedy (1981) for proper interpretation of a binary variable in a logarithmic equation, the coefficient on high urgency equates to a discount of 7.57% compared to non-urgent owners, while somewhat urgent sellers realize price reductions of 6.34%. Sellers expressing that the home is too expensive to keep experience an increase in transaction prices. At first glance this may seem counterintuitive if these sellers are cash constrained but concessions help to explain this finding. These owners have a greater probability of including concessions and to let their properties stay on the market longer. Thus, the positive coefficient in the price model is consistent with search theory: being on the market longer and introducing incentives results in higher transaction prices. Many seller demographics correlate with prices. The slope coefficients increase monotonically across the income levels, which is consistent with sellers who earn higher incomes also own more expensive homes. Age also increases prices, which can be a function of older individuals owning more expensive homes as well as older sellers having a lower preference for liquidity. Additionally, race impacts prices. African Americans experience a decrease in values while Asian sellers realize higher prices than non-hispanic Caucasian sellers. Sellers who expect to experience losses upon sale realize higher transaction prices. This is consistent with the setting of higher list prices found by Genesove and Mayer (2001) and Hayunga and Pace (2016). We note that these sellers also have a significantly greater propensity to use concessions in column 2 and stay on the market longer as indicated in column 3. Again, these results are consistent with search theory. There are additional structural and transaction characteristics that correlate with prices. Similar to the findings by Aroul and Hansz (2014), short sales realize a loss of approximately 20%. Throughout our analysis and in Table 4, short sales also exhibit longer TOM and a negative propensity to use concessions. As can be expected, home quality levels are a strong determinant of prices. Higher (lower) quality homes obtain higher (lower) prices. Longer holding periods indicate a slight decrease in transaction prices. This is consistent with the previous discussion indicating that longer holding periods may indicate greater deferred maintenance. Individual Concessions Each concession type can be used by sellers differently and thus may have unique economic characteristics that are not captured using an all-inclusive

19 Sales Concessions in the US Housing Market Table 4 Log prices using a system of simultaneous equations Log price All concessions Log TOM All concessions IV (0.065) Log TOM IV ** (0.022) High urgency *** (0.018) (0.027) (0.060) Some urgency *** (0.013) (0.020) (0.044) Too expensive to keep ** *** ** (0.040) (0.058) (0.127) Job relocation *** *** (0.021) (0.031) (0.068) Family change (0.022) (0.033) (0.073) Avoid foreclosure (0.030) (0.046) (0.100) Income 35 44k * (0.040) (0.062) (0.135) Income 45 54k (0.038) (0.059) (0.129) Income 55 64k *** ** (0.038) (0.057) (0.124) Income 65 74k *** (0.035) (0.055) (0.119) Income 75 84k *** (0.035) (0.054) (0.118) Income 85 99k *** (0.034) (0.052) (0.115) Income k *** (0.033) (0.050) (0.110) Income k *** ** (0.035) (0.053) (0.116) Income k *** ** (0.037) (0.057) (0.125) Income k *** ** (0.040) (0.061) (0.133) Income k *** *** (0.040) (0.059) (0.130) Income k *** ** (0.039) (0.059) (0.130) Income k ***