Certain Words in Listing Comments Positively Affect Closing Price

Setting up a reasonable list price for a property is a challenge to both the seller and the listing agent. Each property is unique due to its location, condition and internal/external design. In a competitive, active market, all physical attributes of a property should be fairly and accurately priced. However, CoreLogic research finds that properties with certain words in their listing comments, on average, sell higher than those without such comments.

In previous blogs, Turning Words into Data, Part I and Part II, my colleague Matt Cannon discussed how public comments can affect property values. Following that, Bin He discussed in his blog which two-gram combinations affect days-on-market for homes being sold. The purpose of this analysis is to evaluate whether certain wording in Multiple Listing Service (MLS) public remarks are associated with overpricing or underpricing. The analysis is based on 81,025 MLS-listed property sales between January 2, 2015 and January 29, 2016 in Los Angeles County. The target variable is defined as the percentage price change of the closing price relative to the original list price[1]. Each of these properties has a public remarks comment, and from that we can generate more than 10,000 binary variables based on unique key words in the comments. We can then narrow the list down to 251 binary variables (key words) based on the success rate of these binary variables. In other words, infrequent variables are excluded from this stage of the analysis. Finally, we applied a shrinkage regression method called “elastic net” to further narrow down the list to 14 variables based on their correlations with the target variable. Shrinkage regression method is a state-of-the-art data mining approach that balances the biases and variances of the estimated coefficients and sets the coefficients of insignificant variables to zero. Figure 1 shows the list of the final 14 variables together with their coefficients from the shrinkage regression model.

The finalized predictor variables are grouped into three categories: location, condition and design. Attributes with positive coefficients contribute to the price change in a positive way. For example, properties with key words related to great locations such as overlook, step (e.g., steps to the beach), hill and park, on average, sell higher, relative to the final list price, than those without these positive location attributes. Finish and best might be good words indicative of good property conditions; slide, central (central AC), dual and spanish (Spanish style) are attractive design styles for most property buyers. Homes with a fireplac (fireplace) are generally found in more unique, upscale units in Los Angeles. As my colleague Bin He pointed out in his previous blog, luxury features are often associated with higher property values. These properties tend to have less market appeal and hence may take a longer time to find a buyer and sell at a higher discount from the list price (or less of a premium to list price).

In general, properties in good locations, under good conditions and with desirable designs are more attractive than properties without these attributes. These attributes are true measurements of the properties and, in many ways, contribute more to the prices than expected by the sellers/realtors, if marketed appropriately.

© 2016 CoreLogic, Inc. All rights reserved.

CoreLogic.com