We recover the value of curb appeal in residential housing by using photos obtained from Google Street View, a deep learning classification algorithm and a variety of hedonic controls. We show that own property curb appeal is worth about twice that of an across the street neighbor. Together, neighbor and own property curb appeal together may account for up to 7% of a house’s sale price. The curb appeal premium is more pronounced during times of housing market weakness and greater in neighborhoods with high average curb appeal. Results are robust to a variety of spatial controls and curb appeal specifications.
This is a preview of subscription content, log in to check access.
Buy single article
Instant access to the full article PDF.
Price includes VAT for USA
Subscribe to journal
Immediate online access to all issues from 2019. Subscription will auto renew annually.
This is the net price. Taxes to be calculated in checkout.
100 images for every curb appeal category. The classified images, model, labels and python code are publicly available at: https://github.com/erikbjohn/curb_appeal/tree/master/Replication. This will enable replication of training and model creation used in this paper.
Importantly, although the web interface allows for time-series selection of photos, the API only returns the most recent photo from the location. In the results section, we show that our results are robust to changes in the photo/sale date time windows. Thus, it appears that curb appeal scores are relatively stationary. This obviates concerns about the most recent photo restriction. This is consistent with the findings of Glaeser et al. (2018).
For example, if the assigned probabilities for curb appeal categories 1, 2, 3 and 4 are 0.01, 0.01, 0.03 and 0.95 respectively, then the weighted average curb appeal score is calculated as: 0.01(1) + 0.01(2) + 0.03 (3) + 0.95(4) = 3.92
We do not use the interaction of neighborhood and sale year dummies owing to the limited number of observations after applying the filters
It could very well be the case that most of the houses were built to be in the mid-curb appeal range with a smaller volume of houses that have the best curb appeal. The houses in the mid-range of curb appeal transform into lower curb appeal houses for a variety of reasons.
Due to the annualized data, in the event of a repeated sale within a given year, only the first observation is used and any remaining repeat observation during the year dropped.
Anselin, L., & Bera, A. K. (1998). Introduction to Spatial Econometrics. Handbook of applied economic statistics, 237.
Arraiz, I., Drukker, D. M., Kelejian, H. H., & Prucha, I. R. (2010). A spatial Cliff-Ord-type model with heteroskedastic innovations: Small and large sample results. Journal of Regional Science, 50(2), 592–614.
Chen, Jun, Jennifer S. Evans-Cowley, Ronald C. Rutherford and Brooke W. Stanley (2013). An Empirical Analysis of Effect of Housing Curb Appeal on Sales Price of Newer Houses, International Research Journal of Applied Finance, Vol. IV Issue – 11.
Conklin, James N., Moussa Diop and Herman Li (2017). Contact High: The External Effects of Retail Marijuana Establishments on House Prices, Real Estate Economics (Forthcoming).
Drukker, D. M., Egger, P., & Prucha, I. R. (2013). On two-step estimation of a spatial autoregressive model with autoregressive disturbances and endogenous regressors. Econometric Reviews, 32(5–6), 686–733.
Freybote, J., Simon, L., & Beitelspacher, L. (2016). Understanding the contribution of curb appeal to retail real estate values. Journal of Property Research, 33(2).
Glaeser, Edward L., Kincaid, Michael S. & Naik, Nikhil (2018) Computer Vision and Real Estate: Do Looks Matter and Do Incentives Determine Looks.
Kelejian, H. H., & Prucha, I. R. (1998). A generalized spatial two-stage least squares procedure for estimating a spatial autoregressive model with autoregressive disturbances. The Journal of Real Estate Finance and Economics, 17(1), 99–121.
Kelejian, H. H., & Prucha, I. R. (1999). A generalized moments estimator for the autoregressive parameter in a spatial model. International economic review, 40(2), 509–533.
Kelejian, H. H., & Prucha, I. R. (2010). Specification and estimation of spatial autoregressive models with autoregressive and heteroskedastic disturbances. Journal of Econometrics, 157(1), 53–67.
Liu, L., Silva, E. A., Wu, C., & Wang, H. (2017). A Machine Learning-based Method for the large-scale Evaluation of the Qualities of the Urban Environment. Computers, Environment and Urban Systems, 65, 113–125.
Samaha, S. A., & Kamakura, W. A. (2008). Assessing the market value of real estate property with a geographically weighted stochastic frontier model. Real Estate Economics, 36(4), 717–751.
Vandell, Kerry D and Jonathan S. Lane (1989). The Economics of Architecture and Urban Design: Some Preliminary Findings, AREUEA Journal; 17, 2.
Wachter, Susan (2005). The Determinants of Neighborhood Transformations in Philadelphia Identification and Analysis: The New Kensington Pilot Study.
Wolf, Kathleen L (2008). Community Context and Strip Mall Retail: Public Response to the Roadside Landscape, Transportation Research Record: Journal of the Transportation Research Board, No. 2060, Transportation Research Board of the National Academies, Washington, D.C., 2008, pp. 95–103.
We thank Ferdinand Wang and an anonymous referee for their comments. We also thank seminar participants at the DC Real Estate Valuation Conference, Federal Reserve Bank of Richmond’s Regional Economics workshop and the American Real Estate and Urban Economics Annual conference for their suggestions.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
In sample classification diagnostics
First, we start with the in-sample classifier accuracy by classifying photos used in the teaching sample. We compare the ‘ground-truth’ scores to their highest probability scores with a confusion matrix in Table 12. Overall statistics are shown in Table 13.
Class diagnostics are provided in Table 14. Discussion of the diagnostics may be found at https://topepo.github.io/caret/measuring-performance.html. Importantly, Precision is the same as size and Recall corresponds to Power.
Out-of sample classification diagnostics
We next examine out of sample analysis by manually scoring approximately 100 photos into each of the categories and compare the model predictions. The diagnostic statistics are similar to the in-sample training analysis, but with a bit less accuracy as is to be expected.
Overall statistics are shown in Table 15.
The confusion matrix for the out of sample classification is shown in Table 16. Class diagnostics are provided in Table 17 and in the manuscript. Discussion of the diagnostics may be found at https://topepo.github.io/ caret/measuring-performance.html. Importantly, Precision is the same as size and Recall corresponds to Power.
About this article
Cite this article
Johnson, E.B., Tidwell, A. & Villupuram, S.V. Valuing Curb Appeal. J Real Estate Finan Econ 60, 111–133 (2020). https://doi.org/10.1007/s11146-019-09713-z
- Machine learning
- Hedonic valuation