![Sample our Urban Studies journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/110826/TOC-Subject-Sample-2021-Urban-Studies.jpg"})
Abstract
Residential real estate transaction data from Savannah, Georgia, from 2008 to 2021 was used to explore the connection between the volume of competing homes for sale at the time of listing and the effort level of the listing agent. Five proxies for agent effort were identified and, controlling for price, characteristics of the house, timing of the sale, and agent, results suggest that a greater volume of competing homes led to increased agent effort. A 1 standard deviation increase in the number of similarly priced concurrently listed homes resulted in a 3.7 to 12% increase in agent effort, as measured by proxy variables at their means. This result was robust over all five proxies for agent effort and the effect was noticeably larger for competing homes in a similar price range as compared to those in different pricing segments. Sentiment analysis was also conducted on publicly displayed agent property descriptions and indicated that public descriptions became more positive in tone as the volume of competing listings increased.
Disclosure Statement
On behalf of all authors, the corresponding author states that there is no conflict of interest.
Notes
1 It could be argued that listing price, rather than sales price, could be used as a proxy for the expected sales price. Listing agents do, however, appear to sometimes employ heterogeneous listing price strategies (i.e., overprice with the expectation of subsequently reducing the price or underprice with the intent of triggering a bidding war). Given that professional real estate agents are likely capable of rationally estimating the final sales price based on their experience, final sales price was used. As a robustness check, the models were also estimated using list price as a substitute for sale price and the results were qualitatively unchanged.
2 The models were alternatively estimated with 30 5-digit zip code controls, with the average control representing roughly 1,260 observations. Results were qualitatively similar.
3 Brokerage firm fixed effects also could not be included due to the extremely high level of collinearity with agent fixed effects (>90%).
4 According to Katz and Walzer (Citation2022) via Redfin.com, approximately 1–2% of listed homes were removed from the MLS from November 2015 through November 2022 across 43 large real estate markets (https://www.redfin.com/news/home-delistings-november-2022/). Note that while the final analysis was conducted only with observations sold via agents with at least five total sales (see discussion above), competing homes on the market listed by less active agents (with fewer than five total sales) were included in the count of competing homes. The competition variables were generated early in the process, before any observation culling occurred and thus reflect the most accurate number of competing listings possible with available data.
5 The primary specification was estimated via OLS, but a Tobit specification was also estimated. Results were qualitatively similar.
6 A probit model was also estimated and the results were qualitatively similar. They are available on request.
7 This is inferred from a 100% completion rate for these variables.
8 This is inferred from a completion rate of >0%, but <100% in each year for these fields. It is necessary to examine these variables each year since some fields are dropped and new ones added over the 14 years of data. Thus, the 32 qualifying fields have some level of input, but are not 100% complete in each of the 14 years.
9 Inferences as to the maximum observed values of private_comments and public_comments were made from the highest values observed in each year and from information released by the MLS. Information from the MLS refers to the 2015+ caps as being 500 and 1,000 characters for private and public comments, respectively, though there was a small number of observations exceeding these amounts. There were 5 such occurrences for the private comments (with the maximum of 503) and 21 instances in the public comments (with the highest being 1,016).
10 The only difference in the five models, other than the dependent variable, was that the photo_count, private_comments and public_comments models could be estimated with only subsets of the data (as discussed in the Data and Model section).
11 Of the 37,883 observations, VADER was unable to produce a score for 4,100. Some comment cells were blank, contained nonsensical characters, or were otherwise unreadable by the algorithm. Thus, the evaluation of the sentiment analysis proceeded with the 33,783 observations for which a score could be generated.
12 Unsurprisingly, few properties were described in explicitly negative terms. Some examples of the lowest scoring homes, with a compound_score of –0.8 or lower read “FORECLOSURE NEEDS RENOVATION. FMFL OWNER OCCUPENT BIDS FIRST 15 DAYS THEN OPEN TO EVERYONE,” “Some fire damage. Sold AS-IS, no disclosure,” and “Use caution while showing. No sign on property.”
13 The highest scoring observation, with compound_score = 0.9982, read, in part: “Come home to this true Victorian BEAUTY! This home has been completely renovated but delicately restored with all the original character. … Unlike many, the Victorian dream offers secured, off-street parking!!! … Inside this glorious remodel, you will find gorgeous restored hardwood flooring throughout, stunning reconditioned windows. … Make no mistake, this Victorian home will be the most alluring home in the historic district!!! Don’t miss this amazing chance. Make this house your home!”