1# RECORD afc63bb9b7bf51c6364ea26f57de3de63b12d8bb2# TITLE: Irish Property Price Estimation Using A Flexible Geo-spatial Smoothing Approach: What is the Impact of an Address?3# YEAR: 2021.0 VENUE: Journal of real estate finance and economics DOI: 10.1007/s11146-022-09888-y4# AUTHORS: Aoife K. Hurley; James Sweeney5# PAGES: 39 CANDIDATE TABLE PAGES (1-based): [24]6# ABSTRACT: Accurate and efficient valuation of property is of utmost importance in a variety of settings, such as when securing mortgage finance to purchase a property, or where residential property taxes are set as a percentage of a property’s resale value. Internationally, resale based property taxes are most common due to ease of implementation and the difficulty of establishing site values. In an Irish context, property valuations are currently based on comparison to recently sold neighbouring properties, however, this approach is limited by low property turnover. National property taxes based on property value, as opposed to site value, also act as a disincentive to improvement works due to the ensuing increased tax burden. In this article we develop a spatial hedonic regression model to separate the spatial and non-spatial contributions of property features to resale value. We mitigate the issue of low property turnover through geographic correlation, borrowing information across multiple property types and finishes. We investigate the impact of address mislabelling on predictive performance, where vendors erroneously supply a more affluent postcode, and evaluate the contribution of improvement works to increased values. Our flexible geo-spatial model outperforms all competitors across a number of different evaluation metrics, including the accuracy of both price prediction and associated uncertainty intervals. While our models are applied in an Irish context, the ability to accurately value properties in markets with low property turnover and to quantify the value contributions of specific property features has widespread application. The ability to separate spatial and non-spatial contributions to a property’s value also provides an avenue to site-value based property taxes.789===== PAGE 1 (score 5, numeric-paren 0, table-words 0, var-words 5) =====1011 / Published online: 5 May 202212(2024) 68:355–39313https://doi.org/10.1007/s11146-022-09888-y14Irish Property Price Estimation Using A Flexible15Geo-spatial Smoothing Approach:16What is the Impact of an Address?17Aoife K. Hurley118·James Sweeney119Accepted: 15 January 202220© The Author(s) 202221Abstract22Accurate and efficient valuation of property is of utmost importance in a variety23of settings, such as when securing mortgage finance to purchase a property, or24where residential property taxes are set as a percentage of a property’s resale value.25Internationally, resale based property taxes are most common due to ease of imple-26mentation and the difficulty of establishing site values. In an Irish context, property27valuations are currently based on comparison to recently sold neighbouring proper-28ties, however, this approach is limited by low property turnover. National property29taxes based on property value, as opposed to site value, also act as a disincentive30to improvement works due to the ensuing increased tax burden. In this article we31develop a spatial hedonic regression model to separate the spatial and non-spatial32contributions of property features to resale value. We mitigate the issue of low prop-33erty turnover through geographic correlation, borrowing information across multiple34property types and finishes. We investigate the impact of address mislabelling on35predictive performance, where vendors erroneously supply a more affluent post-36code, and evaluate the contribution of improvement works to increased values. Our37flexible geo-spatial model outperforms all competitors across a number of different38evaluation metrics, including the accuracy of both price prediction and associated39uncertainty intervals. While our models are applied in an Irish context, the ability40to accurately value properties in markets with low property turnover and to quan-41tify the value contributions of specific property features has widespread application.42The ability to separate spatial and non-spatial contributions to a property’s value also43provides an avenue to site-value based property taxes.44Keywords House prices · Automatic valuation models · Spatial models ·45Generalised additive models · Hedonic regression · Non-linear smoothing46 Aoife K. Hurley47Aoife.Hurley@ul.ie48149Department of Mathematics and Statistics, University of Limerick, Limerick, V94 T9PX, Ireland50Journal of Real Estate Finance & Economics515253===== PAGE 24 (score 20, numeric-paren 0, table-words 8, var-words 2) =====5455A.K. Hurley, J. Sweeney56Table 8 Results from decision tree and random forest approaches57Model58R259RMSE60Median Absolute61Within 5%62Within 10%63Within 20%64% Error65of Price66of Price67of Price68Tree 1690.8070AC149,4447113.48%7220.22%7339.27%7467.49%75Tree 2760.8177AC148,9777813.32%7920.56%8039.13%8168.33%82Forest 1830.8784AC122,937859.54%8628.01%8751.71%8880.07%89Forest 2900.8891AC117,328928.82%9330.34%9455.38%9583.53%96We trained both decision tree and random forest models as an alternative. Results97from the implementation of two decision trees and two random forests are presented98in Table 8. ‘Tree 1’ is a decision tree that includes all variables apart from ‘Longi-99tude’ and ‘Latitude’, ‘Tree 2’ is a decision tree that includes all variables including100‘Longitude’ and ‘Latitude’ and ‘Forest 1’ and ‘Forest 2’ follow the same convention.101From Table 8, we see that the random forests are marginally more accurate at pre-102dicting the mean (R2 and RMSE) than our GAM models from Table 6, however are103less accurate in prediction interval coverage. Although we can compare the results of104the random forests and GAM models, in taking a GAM approach we reap the benefit105of being able to interpret our models and can assess the impacts of specific variables.106The Impact of Address Mislabelling?107The initial analysis used the given postcodes on the property listings. However, some108properties had given the neighbouring, typically more affluent, postcode rather than109their actual one. We assess the impact of postcode misspecification by correcting the110postcodes and re-running our analysis. 16% of properties in the data set (833 proper-111ties) gave the incorrect postcode in their advertisement. We investigate 15 postcode112changes (outlined in Table 9), based on a cross tabulation of the postcodes, shown113in Fig. 8, and prior knowledge of the Dublin property market. In Fig. 8, the white114lines for D20 and D22 indicate no changes in postcode labeling was required. In the115models ‘GAM 5’ and ‘GAM 6’ in Table 10, the postcode changes are modelled by116adding 15 dummy variables to previous models, one for each postcode change. GAM1175 and GAM 6 are the GAM 3 and GAM 4 models respectively, with the postcodes118corrected.119Figure 9 presents the estimates for the 15 postcode changes. The labels can be read120as ‘Given Postcode.Corrected Postcode’, e.g. “D11.D9” is a property listed as D11121but with correct postcode D9. The majority of the estimated coefficients are positive,122indicating a premium in sales price in each case where giving a neighbouring, more123affluent postcode. It is clear that “D11.D9” and “D16.D14” are different to the other124changes estimated, with scalings less than 1. These properties have listed postcodes125with a lower median price (less affluent) than their correct postcode. For properties126incorrectly listing a more affluent postcode, there is evidence of an additional pre-127mium being added to property values - for 13 out of the 16 postcode changes the mean128scaling estimates are greater than 1. The estimates have larger confidence intervals129than those seen previously in Fig. 6, as there are fewer transactions to estimate from.130378131 132133134===== PAGE 25 (score 4, numeric-paren 0, table-words 1, var-words 2) =====135136What is the Impact of an Address?137Table 9 15 Postcode Changes Investigated138Given Postcode139Corrected Postcode140Occurrences141Dublin 5142Dublin 131433144Dublin 6145Dublin 6W14612147Dublin 6W148Dublin 121493150Dublin 9151Dublin 31523153Dublin 9154Dublin 1715510156Dublin 11157Dublin 91587159Dublin 14160Dublin 1616119162Dublin 16163Dublin 141645165North County Dublin166Dublin 151679168South County Dublin169Dublin 417027171South County Dublin172Dublin 18173101174South County Dublin175Dublin 2417632177West County Dublin178Dublin 1517938180West County Dublin181Dublin 2218224183West County Dublin184Dublin 2418545186In Fig. 10 we present the estimated scalings associated with the corrected post-187codes. If the underlying spatial surface captured all spatial structure in the data, the188postcodes should be labels alone, with no obvious patterns. However we note there is189substantial evidence of a strong bias in postcode effects. The postcodes D7, D8 and190D9 have extremely significant impacts on price with these representing popular, up-and-191coming areas reflected in scaling estimates much greater than 1. D10, D22 and D24192all have scalings substantially less than 1 which would suggest these areas are under-193valued relative to their spatial location. This is perhaps due to negative connotations194Fig. 8 Heatmap of the changes that occur in postcodes195379196