Analyzing the Effect of Google Searches on the Czech Real Estate Market

• Název práce v češtině: Analýza vlivu Google Trends dat na trh nemovitostí v České republice
• Název v anglickém jazyce: Analyzing the Effect of Google Searches on the Czech Real Estate Market
• Klíčová slova: trh nemovitostí, Česká republika, Google Trends
• Klíčová slova anglicky: real estate market, Czech republic, Google Trends
• Akademický rok vypsání: 2020/2021
• Typ práce: bakalářská práce
• Jazyk práce: angličtina
• Ústav: Institut ekonomických studií (23-IES)
• Vedoucí / školitel: PhDr. Pavel Vacek, Ph.D.
• Řešitel: skrytý - zadáno vedoucím/školitelem
• Datum přihlášení: 20.05.2021
• Datum zadání: 20.05.2021
• Datum a čas obhajoby: 08.06.2022 09:00
• Místo konání obhajoby: Opletalova - Opletalova 26, O314, Opletalova - místn. č. 314
• Datum odevzdání elektronické podoby: 03.05.2022
• Datum proběhlé obhajoby: 08.06.2022
• Oponenti: Mgr. Roman Kalabiška

Seznam odborné literatury

Beracha, E. and Wintoki, M. B. (2013) ‘Forecasting Residential Real Estate Price Changes from Online Search Activity’, Journal of Real Estate Research, 35(3), pp. 283–312. doi: 10.1080/10835547.2013.12091364.
Dietzel, M. A., Braun, N. and Schaefers, W. (2014) ‘Sentiment-Based Commercial Real Estate Forecasting with Google Search Volume Data’, Journal of Property Investment & Finance, 32(6), pp. 540–569. doi: 10.1108/JPIF-01-2014-0004.
Hohenstatt, R. and Kaesbauer, M. (2014) ‘GECO’s Weather Forecast for the U.K. Housing Market: To What Extent Can We Rely on Google Econometrics?’, Journal of Real Estate Research, 36(2), pp. 253–282. doi: 10.1080/10835547.2014.12091387.
Lee, K., Kim, H. and Shin, D.H. (2019) ‘Forecasting Short-Term Housing Transaction Volumes using Time-Series and Internet Search Queries’, KSCE Journal of Civil Engineering, 23(6), pp. 2409–2416. doi:10.1007/s12205-019-1926-9.
Steegmans Joep (2019) ‘The Pearls and Perils of Google Trends : A Housing Market Application’, USE Working Paper series, 19(11).
Wu, L. and Brynjolfsson, E. (2015) ‘The Future of Prediction: How Google Searches Foreshadow Housing Prices and Sales’, in Economic Analysis of the Digital Economy. University of Chicago Press, pp. 89–118. Available at: http://www.nber.org/chapters/c12994.

Předběžná náplň práce v anglickém jazyce

Research question and motivation

The main research question I plan to study in my thesis is if it is possible to forecast the real estate market in the Czech Republic using Google search data. Specifically, I will try to forecast the price and the number of completed housing transactions on a statewide and regional level.

In modern society, where online activity has become part of everyday life, aggregated search data can be used as a database of intentions (Battelle, 2005), which can contain patterns that signal likely future outcomes. Readily available search intensity data allows us to examine consumer intentions of billions of people before any transactions even happen. This opens the door for researchers to include this information about consumer demand into their forecasting models and thus improve their predicting ability at virtually no cost. Wu and Brynjolffson (2015) found that predictions about the housing market made using only search indices outperformed predictions made by the National Association of Realtors. Beracha and Wintoki (2013) suggest that on average, cities associated with abnormally high real estate search intensity outperformed cities with low search intensity by 8.5% with relation to the overall U.S. housing market.

Contribution

Existing research papers on the topic of using Google search data in real estate forecasting were at first conducted mainly in the United States and the United Kingdom. Dietzel, Braun and Schäfers (2014) claim that models combining both macro and search data significantly outperform models only based on macro data, the mean squared forecasting error was reduced by 54% when predicting prices.

In recent years, researchers from other parts of the world have been utilizing Google search data more frequently, Lee, Kim and Shin (2019) found that in Korea incorporating search data decreased the mean average percent errors of forecasting models by about 50% when predicting transaction volume. More regionally importantly for this thesis, Cihlář (2020) analyzed whether it is possible to predict real estate prices in the Czech Republic using Google Trends data and found that including Google Trends data into macro-based models improved the models with statistical significance.

I would like to expand on the existing literature by analyzing whether the number of transactions on the Czech real estate market can be predicted using Google Trends data. Another important feature of my thesis is that in all my regressions I will be taking into consideration the sampling error, which is present in Google Trends data, as highlighted by Steegmans (2019). This issue of sampling error has been mostly overlooked by scholars, but if it is not accounted for, it can lead to incorrect results and conclusions.

Additionally, I will examine the forecasting power of the more refined Google search data (reduced sampling error) on prices on the Czech real estate market. Previous work on this topic by Cihlář (2020) did not account for the sampling error in Google Trends data which could have led to inaccurate conclusions.

Methodology

The data regarding online search activity will be obtained from the Google Trends webpage (www.google.com/trends), which reports the search intensity of personally chosen keywords or categories on a level from 0 to 100 (100 being the most) for a given time period. When looking at a long period of time (e.g., several years) the search data is reported in monthly frequencies which is the same as the data about the transaction volume I will be using, so no adjustment is needed. The list of keywords and categories I will be examining on the Google Trends webpage will be carefully chosen to representatively reflect the demand for real estate in the population.

The sampling error in Google Trends data appears, because Google only uses a sample of searches on any given day to compile the search intensity index. The usage of different samples everyday leads to varying values of intensity being reported daily. The way I plan to tackle this issue is to collect data on the same keyword or category on 30 consecutive days and use the average of the collected values to reduce the sampling error in my analysis.

The information on transaction volumes in the Czech Republic will be drawn from a dataset provided by the Real Estate and Construction branch of Deloitte Czech Republic, which reports monthly data on the number of transactions across Czech counties.

The data regarding real estate prices in the Czech Republic will be obtained from the Czech Statistical Office webpage (www.czso.cz), specifically the Prices of Observed Types of Real Estate section, which provides quarterly price indices for different types of real estate. Since this data is reported in a quarterly basis, the Google Trends data will be aggregated to match the frequency of the price indices.

The regression model used in my thesis will be a form of an autoregressive model where the past transaction volumes and prices will serve as an additional explanatory variable alongside the Google Trends data. To analyze the forecasting effect of the Google searches time series on the real estate market I will examine a variety of different lags on the search data in the regression to reflect the time it takes to complete a purchase on the housing market.


Outline
1. Abstract
2. Introduction
3. Literature review
4. Data
a) Google search data
b) Real estate transaction volume data
c) Real estate prices data
5. Methodology
a) Transaction volume regression model
b) Prices regression model
6. Results
7. Conclusions