Artificial intelligence, news sentiment, and property market liquidity
| Date | 29 November 2019 |
| Pages | 309-325 |
| Published date | 29 November 2019 |
| DOI | https://doi.org/10.1108/JPIF-08-2019-0100 |
| Author | Johannes Braun,Jochen Hausler,Wolfgang Schäfers |
| Subject Matter | Property management & built environment,Real estate & property,Property valuation & finance |
Artificial intelligence,
news sentiment, and property
market liquidity
Johannes Braun, Jochen Hausler and Wolfgang Schäfers
International Real Estate Business School,
University of Regensburg, Regensburg, Germany
Abstract
Purpose –The purpose of this paper is to use a text-based sentiment indicator to explain variations in direct
property market liquidity in the USA.
Design/methodology/approach –By means of an artificial neural network, market sentiment is extracted
from 66,070 US real estate market news articles from the S&P Global Market Intelligence database.
For training of the network, a distant supervision approach utilizing 17,822 labeled investment ideas from the
crowd-sourced investment advisory platform Seeking Alpha is applied.
Findings –According to the results of autoregressive distributed lag models including contemporary and
lagged sentiment as independent variables, the derived textual sentiment indicator is not only significantly
linked to the depth and resilience dimensions of market liquidity (proxied by Amihud’s (2002) price impact
measure), but also to the breadth dimension (proxied by transaction volume).
Practical implications –These results suggest an intertemporal effect of sentiment on liquidity for the
direct property market. Market participants should account for this effect in terms of their investment
decisions, and also when assessing and pricing liquidity risk.
Originality/value –This paper not only extends the literature on text-based sentiment indicators in real
estate, but is also the first to apply artificial intelligence for sentiment extraction from news articles in a
market liquidity setting.
Keywords Artificial intelligence, Sentiment, Deep learning, Commercial real estate, Market liquidity,
News analytics
Paper type Research paper
Introduction
Recent increases in media attention and public enthusiasm about the field of artificial
intelligence might lead one to draw the incorrect conclusion that artificial neural networks
(ANNs) are a new field of research. In fact, with Rosenblatt (1958) often being considered
the inventor of the first “real”ANN, the theoretical foundations of deep learning methods
are more than half a century old. Due to the vast computational requirements and lack of
mathematical algorithms to support the concept, research efforts dried up soon after the
initial suggestion of ANN approaches. Werbos’(1974) introduction of the seminal
backpropagation algorithm certainly pushed the borders of effi ciently trainin g complex
models. But only the possibility to accumulate massive amounts of exploitable data
through the internet, and an exponential increase in available computational power
during the last few decades, facilitated the recent renewal of interest in ANN research
and applications.
A moderate number of studies employing deep learning in a real estate context have
been published, although the majority of contributions addresses ANN-based property
valuation (see e.g. Kathmann, 1993; Worzala et al., 1995; Nguyen and Cripps, 2001; Din et al.,
2001; Lam et al., 2008; Peterson and Flanagan, 2009; Poursaeed et al., 2018). Apart from
valuation research, ANN studies in the field of real estate are sparse. Ellis and Wilson (2005)
suggest a portfolio-selection approach for the Australian property stock market, applying
ANNs, Zhang et al. (2015) use ANNs to identify real estate market cycles in China, and Chen,
Chang, Ho and Diaz (2014) develop an ANN-approach for REIT return forecasts.
Received 1 August 2019
Revised 3 November 2019
Accepted 6 November 2019
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/1463-578X.htm
News sentiment
and property
market
liquidity
JournalofProperty Investment&
Finance
Vol.38 No. 4, 2020
pp.309-325
©EmeraldPublishingLimited
1463-578X
DOI10.1108/JPIF-08-2019-0100
309
By introducing a deep learning-based approach to extract market sentiment from news
articles, this study not only extends research on sentiment in real estate markets, but also
the limited literature on investor sentiment as a factor explaining the variation in direct real
estate market liquidity. Scholars such as Fisher et al. (2003) and Clayton et al. (2009) have
pointed out the time-varying nature of direct real estate market liquidity compared to other
asset classes. Empirically demonstrated during the last market cycle, the “ease”of trading
increases during up-market periods, and decreases accordingly in down-markets. This
feature of the property market is caused partially by the characteristics of real estate assets
which are usually large-volume, heterogeneous and traded infrequently in segmented, local
markets. However, in accordance with Liu (2015), who shows a relationship between
sentiment and liquidity for the stock market, Freybote and Seagraves (2018) recently
demonstrated the influence of market participants’sentiment as an additional driver of
liquidity in the US office property market. The promising results of Freybote and Seagraves
(2018) are a good entry point for additional research. In particular, testing a model
incorporating shorter data-aggregation periods, the possibility of a lagged relationship and
the use of refined sentiment measures seems worthwhile.
Hence, in this study, an ANN is trained on a data set collected from the investment
advisory platform Seeking Alpha (SA). In a second step, the trained network is used to
evaluate news articles from the S&P Global Market Intelligence database regarding their
inherent sentiment. By averagingthe sentiment scores of the news articles within each month
of the study period,an aggregate index is calculated. The resultingmonthly market sentiment
indicator can then be analyzed for its influence on direct real estate market liquidity.
The chosen approach enables the extraction of a rich information structure from news
articles, as ANNs do not rely on a predefined set of rules to indicate the sentiment polarity
expressed by the respective article’s author. Unlike conventional deep learning sentiment
analysis, which requires the time-consuming and subjective practice of manually classifying
a sufficiently large training data set, this paper furthermore applies distant supervision
labeling. The available categorization of articles on the SA website into long and short
investment ideas is utilized as a natural indicator for the sentiment polarity prevailing in the
respective text. By automatizing the sentiment-extraction procedure and making use of a
vast online source for text data with a distinct sentiment polarity, the paper overcomes one
of the limitations of existing ANN sentiment analysis. In addition, it might be a starting
point for other studies exploiting alternative distant supervision labeled data sources.
For the observation period from January 2006 to December 2018, the findings provide
strong evidence of a dynamic link between sentiment and different dimensions of market
liquidity. While there is a significant contemporary link for both of the two tested liquidity
proxies, in the case of the market-depth proxy, sentiment leads market liquidity by up to
more than two quarters. Market participants in the direct commercial real estate market
seem to exhibit sentiment-induced behavior as a trigger of transaction decisions and by
doing so, stimulate future market liquidity.
The remainder of this paper is structured as follows. The second section provides an
overview of related literature, identifies existing research gaps and in this context outlines
the motivation for this study. The third and fourth sections describe the data sets, the
sentiment-extraction procedure and the econometric approach used to estimate the results,
which follow in the fifth section. The sixth section concludes.
Literature and motivation
The properties of market liquidity for the general stock market have been the subject of
extensive empirical research during the last few decades. Chordia et al. (2000) find a
market-wide co-movement, Amihud (2002) shows an effect of market liquidity on returns,
and Pastor and Stambaugh (2003) as well as Acharya and Pedersen (2005) provide empirical
JPIF
310
38,4
To continue reading
Request your trial