A robust house price index using sparse and frugal data

Abstract

In this article, we describe a house price index algorithm which requires only sparse and frugal data, namely house location, date of sale and sale price, as input data. We aim to show that our algorithm is as effective for predicting price changes as more complex models which require detailed or extensive data. Although various methods are employed for determining house price indexes, such as hedonic regression, mix-adjusted median or repeat sales, there is no consensus on how to determine the robustness of an index, and hence no agreement on which method is the best to use. We formalise an objective criterion for what a house price index should achieve, namely consistency between time periods. Using this criterion, we investigate whether it is possible to achieve strong robustness using frugal data covering only 66 months of transactions on the Irish property market. We develop a simple multi-stage algorithm and show that it is more robust than the complex hedonic regression model currently employed by the Irish Central Statistics Office.

The Better-Than-Average Effect Is Observed Because “Average” Is Often Construed as Below-Median Ability

“…Many researchers have struggled with the term “average.” Although the statistical mean or median might seem to be the most representative average, the statistical mean or median no longer stands as a representative standard when there is only limited information and/or the data is skewed. According to information theory, “representativeness” converges to the exemplar that holds the most meaningful information ( Maguire et al, 2016 ). For example, researchers who construct house price indices can sometimes find that their “average” house price is, in fact, less than both the statistical mean price and the median house price.…”

“…In this scenario, people may see the comparison target ( the average person ) not as the statistical mean or median but as someone with below-median ability or in other words, mediocre ability. We believe that when trying to conjure up an average, people choose a target they believe is the most representative of the group, and this comparison target is more often than not someone with below-median ability ( Maguire et al, 2016 ), particularly in the traditionally measured ability domains in the BAE literature. In other words, the BAE may not be an accurate reflection of self-enhancement bias, if people perceive “average” not as a neutral statistical term but as a slightly negative term connoting mediocrity, found somewhere below median.…”

A Re-Assessment of House Price Indices: Evidence From the Spanish Market

“…Encapsulated within this is the key question concerning how to control by quality given the different nature of the commodity (housing) and the quantity consumed. However, beyond this initial point of agreement there is an apparent lack of consensus on how an index should be constructed though three main methods to build house prices indices are frequently reported in the literature (Rapport, 2008;Coulson, 2012;Goh et al, 2012;P. Maguire, Miller, Moser, & R. Maguire, 2016).…”

“…Third, that houses which resale frequently tend to appreciate at higher rates. Fourth, that short holding periods may capture significant improvement to properties thereby violating assumptions on constant quality (Maguire et al, 2016;Bollerslev et al, 2016;Goh et al, 2012). Indeed, Goh et al (2012) identified that of five different models/variations they assessed, the repeat sales model was the least preferred.…”

New adventures with Schrödinger's cat: The role of statistical uncertainty in stabilizing measurement

“…Always, we are led back to the idea that the ultimate measurement is one that reflects the aggregation of a large number of independent samples (see Ref. 7). Independence is the cornerstone of measurement, not truth, nor the eradication of uncertainty.…”