Which Sentiment Indicators Matter? Evidence from the European Commercial Real Estate Market

Abstract

Measuring sentiment has been at the center of research for many years. While direct sentiment measures are assumed to be more suitable, we examine the construction and U.S. ability of indirect measures in the absence of direct measures in Europe. Driven by the unavailability of direct measures for most of the European countries in our study, we follow the general belief that each imperfect sentiment proxy—at least to a certain extent—contains some pure market sentiment. We present a set of four sentiment indicators relevant to real estate. In order to examine their relative importance, we apply them in a cap rate modeling framework with a sample of both developed and less developed European real estate markets. Forecast evaluations and goodness-of-fit measures reveal that models incorporating our sentiment measures improve upon the standard model specification of cap rates. We further find that sentiment has a stronger impact on economically weaker regions with less developed real estate markets in Europe, suggesting that investors may be relying on sentiment for price discovery in these markets. In addition, the use of a constructed search volume index suggests that markets with less publicly available information may experience a higher degree in searches. Our study highlights an application of online-search-based sentiment measure in those informationally inefficient and less developed markets.

Keywords:

• Sentiment
• Google Trends
• yield modeling
• forecast evaluation

JEL Classification:

• C53
• C82
• E37
• R31

Notes

1 Google Trends Data 2016, https://www.google.co.U.K./trends/, accessed on 5 May 2016

2 In unreported tests, we construct the sentiment indicators following the Kaiser criterion. Here, all principal components with an eigenvalue greater than 1 are used. We then construct a weighted average of the number of recommended components based on their explanation power. However, the results do not suggest that this method is superior in any way.

3 Formerly known as Thomson Reuters Datastream.

4 The source code of the GT webpage uses these codes for each of the categories.

5 Unfortunately, the authors do not explain where they get this information. Up to this point, we have not been able to get in contact with Google about this and other questions. Google does not offer any service line for GT and emails remain unanswered.

6 We have chosen the London West End market, since the market provides both the office and the retail market. Other London regions lack one or other of the markets.

7 Both Theil’s U1 and U2 measures, have been proposed by Theil (1958, 1966). Both measure the accuracy of a forecast. Theil’s U1 is an inequality measure, which draws back on the RMSE metrics. It calculates the difference between actual and the predicted values in terms of change. The value ranges between 0 and 1. A smaller value indicates a better forecast prediction. Theil’s U2, evaluates the forecast against a naïve rule or an alternative model. Values above 1 are assumed to be good indicators, that the model predictions are better than the naïve forecast.

8 These results can be obtained upon request from the authors.

9 These results can be obtained upon request from the authors.