Information Value of Property Description: A Machine Learning Approach

Abstract

This paper employs a ML-Hedonic approach to quantify the value of uniqueness, a type of "soft" information embedded in real estate advertisements. We first propose an unsupervised learning algorithm to quantify levels of semantic deviation ("uniqueness") in descriptions, the textual portions of real estate advertisements. We then estimated the impact of description uniqueness on real estate transaction outcomes using linear hedonic pricing models. The results indicate textual data disseminate information that numerical data cannot capture, and property descriptions effectively narrow the information gap between structured real estate data and the houses by conveying "soft" information about unique house features. A one standard deviation (0.08) increase in description uniqueness compared to neighboring properties leads to a 5.6% increase in property sale prices and a 2.3-day delay in the closing time, controlling for house characteristics, transaction circumstances, and agent unobservables. This paper provides theoretical and empirical insights on how to utilize the emerging Machine Learning tools in economic research.

Documenting Loss Aversion using Evidence of Round Number Bias

“…At the same time, we provide evidence that the application of the standard empirical approach when applied to broad samples of housing transactions may overstate the extent of loss aversion due to a correlation between the types of houses and sellers on the market and the business or housing market cycle. Our findings are consistent with the evidence of compositional changes in the housing market over the business cycle, as documented by Nowak and Smith (2020) and Shen and Ross (2020). Our paper is also related to the focal point bias literature providing additional evidence that behavioral phenomenon like focal point bias and loss aversion are likely to be found together (e.g., Fraser et al, 2015) and that individuals who exhibit psychological biases are different on average from those who do not (Backus et al, 2019;Chava and Yao, 2017).…”

“…In this paper, we test for a relationship between the effect of expected losses and whether a seller exhibits focal point bias or a bias towards round numbers when selecting their purchase mortgage amount, i.e., amounts in multiples of $5,000. We focus on the initial mortgage amount because sellers (in the first sale) have far less stake in the mortgage amount than, for example, the sales price, and lenders are typically focused on loan to value and income ratios, rather than the 3 See Shen and Ross (2020) for evidence that the composition of owner-occupied housing on the market changes over both traditionally observable and unobservable housing attributes as the economy recovers from a downturn. Bayer et al (2016) show that the composition of borrowers changes over the housing cycle.…”

“…Bracke and Tenreyro (In Press) also find more modest estimates of 0.25 for residential housing in England and Wales, but their sample is restricted to cash-only sales, again possibly leading to a substantially more homogeneous sample of borrowers. Zhou et al (2020) also document a potential upward bias in estimates of the effect of loss aversion on sales prices based on bias in the composition of repeat sales transactions due to unobserved quality (Nowak and Smith, 2020;Shen and Ross, 2020).…”

Towards Accountability in Machine Learning Applications: A System-testing Approach

“…The triumph of ML applications has only started but has revolutionized commerce, personal interactions, entertainment, medicine, government services, state supervision -and research, already (Simester et al, 2020). In real estate and urban studies, a rapidly expanding literature explores the potential of ML algorithms, introducing novel measurements of the physical environments or using these estimates to improve the traditional real estate valuation and urban planning processes (Glaeser et al, 2018;Johnson et al, 2020;Karimi et al, 2019;Lindenthal & Johnson, 2021;Liu et al, 2017;Rossetti et al, 2019;Schmidt & Lindenthal, 2020;Shen & Ross, 2020). These studies, again and again, demonstrate the undisputed power of ML-systems as prediction machines.…”

Micro-Geographic Property Price and Rent Indices

“…Instead, we refer to European Commission (2013) for an overview. Recent notable developments in this literature are the use of matching approaches (Lopez and Hewings, 2018) to broaden samples beyond repeat sales (Bailey et al, 1963), adaptive weights smoothing to produce land value surfaces (Kolbe et al, 2015), or machine learning to capture otherwise unobservable housing characteristics (Shen and Ross, 2021). This strand of research is a manifestation of a broader trend to fit flexible functional forms to data in a way that supports out-of-sample predictions.…”