Dr Yi Wu, Henley Business School, University of Reading, UK
Dr Steven Devaney, Henley Business School, University of Reading, UK
Tags: ESG, media sentiment, text mining, pricing, machine learning
Output type: Research briefing
Target stakeholders: Asset managers, property researchers, financial PR
"Newspaper colour" by NS Newsflash is licensed under CC BY 2.0.
What’s the issue?
While the role of sentiment in financial markets and the wider economy is well explored, there has been much less analysis of its impact on the real estate investment market. The pricing of real estate assets or investment vehicles should reflect rational expectations of future cash flows. However, as there is considerable illiquidity and information asymmetries in the underlying real estate market, sentiment is also likely to play a role in pricing and investment decision making. Arguably, measuring media influence on real estate market outcomes is not only of interest to academic researchers, but also to real estate professionals because it can help investors and their advisers understand anomalies in real estate pricing and the adjustments to pricing that might happen when sentiment changes.
Earlier thinking and research
Existing academic research on sentiment in real estate investment has either used published survey indicators or constructed bespoke measures based on economic data and/or capital flows. Survey indicators capture the perceptions and expectations of different industry players to gauge the sentiment of the real estate market using questionnaires. Examples in the UK include market surveys conducted by the RICS that include questions on the expectations and perceptions of surveyors. Such surveys can be time consuming to conduct and so results may only be available at relatively low frequencies such as quarterly or annual intervals. Similarly, the availability of composite indicators generated from economic data depends on the variables used to construct them and it can be challenging to isolate the contribution of sentiment from the other data contained within such measures.
Towards a new approach
The development of textual analysis techniques and the greater availability of textual data has created the possibility of going deeper into the analysis of soft data, including newspaper articles and other media-generated content. In our project, we started examining such data to gauge its potential for producing high frequency measures of real estate sentiment. We investigate whether the media has had an impact on real estate market pricing and performance, with an initial focus on media coverage of specific listed real estate firms. The project seeks to measure the sentiment that is embedded in the text of news articles related to such firms, and to compare this with the share price performance of those firms.
To detect sentiment in news articles, researchers in finance and economics have used textual analysis to interpret the words used by analysts and reporters. While most well-known sentiment scoring services are based on financial news providers, we examine articles published by real estate news providers, such as Property Week and EG Radius to capture data about the activities of listed real estate firms. This is because coverage of all but the very largest real estate firms in general financial media is often sparse. We tested different word lists to identify the presence of positive and negative sentiment in the articles. To our knowledge, there is little research of this kind in the real estate sector, but interest in this area is growing.
Our first trial is based on the textual analysis of a pool of UK real estate news articles. We downloaded a sample of articles related to 32 UK REITs from EG Radius and Property Week and produced a time series of sentiment scores using the TB, Vader and Finbert textual analysis packages used by previous research in social sciences and in finance. The graph below provides a comparison of the different approaches. Although the signals are noisy, the rises and falls mirror the ups and downs in the UK real estate market cycle in this period, especially since the mid-1990s.
Figure 1: Sentiment scores (average and range) for news articles about UK listed real estate companies – comparison of textual analysis packages
These measurements of sentiment towards real estate companies (both for individual companies and in aggregate) are novel in themselves. However, they also provide us with a platform for analysing whether and how the perception of company activities affects stock performance and company value. Hence, it differs from previous academic work in real estate by drawing on articles from multiple media providers and by examining the differences between individual real estate firms.
Future research and ESG
We also intend to analyse news articles that report on the ESG activities and initiatives of real estate companies. ESG metrics allow businesses across the economy to measure their environmental, social and governance performance so that they can be transparent with consumers and stakeholders. There has been marked growth in the development of ESG metrics by data providers such as MSCI, S&P and Morningstar, as well as rating schemes such as GRESB that monitor the ESG performance of real estate portfolios. ESG ratings from GRESB, Thomson Reuters and MSCI have been used in previous academic studies
While these schemes can assist investors in understanding the ESG attributes of a given fund or REIT, they do not capture how well received the initiatives are by other market participants. The perception of ESG initiatives is of interest owing to contemporary and regulatory concerns about whether all ESG activities have substance or whether ‘greenwashing’ is taking place. Understanding how different E, S and G activities are perceived could assist companies in framing their ESG policies in future. Therefore, we will identify the sentiment towards ESG initiatives publicised in the media and compare this with corporate share price movements. This provides an alternative approach for companies to assess the reception and success of their ESG activities.
Future applications for practitioners
Real estate has traditionally approached new technologies and approaches with curiosity and caution. Yet the increased volume and variety of data about the real estate market beyond traditional market indicators potentially has great value and deserves further examination. This project illustrates how textual data might be used. Such data could be further processed by machine learning systems, whose algorithms look for correlations or recurring patterns so that mathematical models can be built to represent those relationships. This might help investors to spot opportunities for arbitrage where prices have drifted away from the economic fundamentals that ultimately govern cash flows. Nonetheless, media coverage of real estate markets is not perfect, and the performance of models may depend on which news sources are used and how different parts of the market or different market events are reported in news media. This suggests that there is some way to go to achieve the delicate balance where machine learning can help human experts to spot new market opportunities.
Acknowledgements: We would like to acknowledge financial support from the Reading Real Estate Foundation, which has enabled us to conduct initial tests of our research ideas in this area. The assistance provided by our RA, Dr Zhenming Wu, was greatly appreciated.
 See Das, Freybote and Marcato, 2015; Marcato and Nanda, 2016; Freybote and Seagraves, 2017; Heinig, Nanda and Tsolacos, 2020.  See https://www.rics.org/uk/products/market-surveys/.  see Hausler, Ruscheinsky and Lang, 2018; Ploessl, Just and Wehrheim, 2021.  See Newell and Lee, 2012; Eichholtz et al., 2013; Fuerst, 2015; Westermann et al., 2018; 2019; Brounen and Marcato, 2018; Morri et al., 2020; Erol, et al, 2021