A "Perceived" Hedonics Approach for Dealing "Un-obervables" in Real Estate & Moderated Effect of Urban Greenary

（The following is my personal research diary, where I record my ongoing research thoughts. So, it’s not written in formal academic language.）

This is actually our new work for the upcoming AsRES International Conference 2025, it's a joint work with two of my postgrad colleagues Tai YI and Yanfu BAI from Georgia Tech, Centre for Urban Resillience and Analytics.

Key words: Machine Learning, Hedonics, Real Estate Pricing, Human feelings, Urban Green Space (UGS)

My reflections about real estate valuation studies:

In Real Estate valuation field, actually 2 models are always "competing with each other": the Hedonic pricing model and the Repeat Sales model. Although the Repeat Sales model may offer better accuracy when estimating pure price appreciation over time, especially by controlling for property characteristics through repeat transactions, the Hedonic model has historically been more flexible in capturing the marginal value of specific housing attributes.

In recent years, the mainstream academic field of real estate (e.g., top journals such as REE, JREFE, and JRER) has barely featured any studies on traditional valuation modelling, shifting instead toward more "empirical" researches that emphasizes "how to tell an interesting story". Also, the recent good story-telling papers on REE often aligned with urban studies or behavioral economics, where DiD has seemingly become a standard feature in these papers.

In Hedonics field, KW Chau is the most well-known professor in Hong Kong (HKU). This may be partly because the HKU real estate department is bonded under the Faculty of Architecture, which tends to foster stronger interest in theoretical and computational approaches. Yet, Since the 2010s, however, research on hedonic pricing models has become less common in real estate studies. Some professors argued, because most major points of interest (POIs) and amenities have already been thoroughly examined, leaving only non-decisive POIs that contribute limited explanatory power.

Another major limitation of the hedonic pricing model is "un-observables"—not all attributes/amenities will be fully perceived or measurable by individuals. To address this challenge, scholars in real estate research have developed various approaches. Liu et al (2020) used textual analysis tool (such as on Offering Memorandums) to filled the gap of "un-observable data” in the data sciences. Thies Lindenthal (2020) applied Machine Learning (ML) for quantifying the aesthetic quality of building beauty. Erkan Yonder (2022) applied a big data approach to transactional records to better capture the "dynamic" evolution of housing prices. They all showcase how Machine Learning may be a new tool for empowering hedonics valuation.

Why "Urban Green Space"?

The inspiration behind my interest in “Urban Greenery” traces back to an exchange semester I spent at the University of Manchester, where I studied in the Department of Urban Studies. During that time, I took a class taught by Professor Ian Mell—one of the UK’s leading scholars on Green Infrastructure, known for his deeply humanistic approach to landscape and planning. His teaching not only introduced me to the conceptual foundations of green infrastructure, but also profoundly shaped my understanding of how urban greenery reflects social values, memory, and ecological justice.

At the beginning of the class, my data-driven mindset and economics background often put me at odds with Professor Mell's more humanistic perspective. Both in class discussions and private conversations, he would often remind me: “Numbers are just numbers, they can’t capture how people feel.” As the head of the Urban Department, perhaps his frequent interactions with quantitatively-oriented real estate scholars made this tension especially salient to him. Ironically, it was through these disagreements that I became increasingly interested in the question: "Can numbers, in fact, be pushed to capture emotion and perception?" This became the seed for my current research interest, using quantitative and formalized methods to measure and represent people’s feelings and lived experiences within the real estate and urban context.

MIT team was the most arlist that applied urban coding technics into Real Estate Valuaiton

Globally, there are a few major academic approaches to real estate research.

• MIT tends to focus on computational methods and urban analytics;

• Dutch scholars (mainly Uni of Masstrict) emphasize fundamentals, such as multi-century price trends or REIT fundamentals, as well as their response to the natural disasters;

• Researchers in Singapore and Hong Kong typically favor empirical studies, which have very strong "bussiness school" / "finance dicipline" approach.

OUR APPROACH for quantifying people's feelings:

Yet, the value of Urban Greenary in terms of "Real Estate" still mostly remains at euclian distance-based valuation. After inspired one of the professors that Ian Mell (2025); Rachel Lauwerijssen (2025) pointed out People’s perceptions of green gardens heavily depend on their childhood experiences. (Qualitative research in the Netherlands).

1, 我们全面替换了“Euclidian

MIT lab

The 量化of 人体感受似乎并不是传统房地产界会去做的东西，更多像是近几年来Urban Econ或者一些GIS方面的Geo-coding或者Urban analytics 会做的东西。

2, 人们感官角度出发

Wall-Street Index

Figure 3: Visualization of people on two streets with high density of buildings Figure 4: Visualization of people on two streets with low density of buildings

3，ML Vision Calculation for Google Street View

Peer Sugguestions:

Prof. Desmond Tsang (CUHK) for pointing out, that people rarely decide how much they are willing to pay based on how it feels to walk through a nearby park. Moreover, addressing endogeneity is particularly difficult in this context, as there may be reverse causality at play.

Prof. William Cheung (Uni of Ackland) for suggesting that "Mobility" data could be a strong supplementary, for showcasing whether people actually made the move to the nearest park, if their "theoratical" preceived greenary accessibiliy is higher. It could further tackle the question "why people would walk to the park". Yet, William only got the human mobility data in New Zealand based on their mobile device records, currently it seems quite impossible for us to obtain such individual-level mobility data.

Julian Chen (CUHK)

Hangming YU (NUS) for the suggestion that for the robustness test, key POI could be replaced into "transit station" / "shopping mall" (those kind of other neighborhood amendities, let's say) to see whether high correlation would appears on other POI as well, rather than treating other POIs as solid Fixed Effect / Control Variable.

Data avaliablity:

• Residential Real Estate sales prices are from Co-Star, thanks for access granting from University of Manchester, Center for Real Estate, during my exchange life in Manchester.

•