Media coverage and the Dutch housing market

(1)

Media Coverage and the Dutch Housing Market

Master of Science Thesis

June, 2018

Author

M.W. (Martijn) Schaaf

University of Amsterdam

Faculty of Economics and Business

MSc Finance

Finance and Real Estate Finance

Supervisor

(2)

Statement of Originality

This document is written by M.W. (Martijn) Schaaf who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Title

_{Media Coverage and the Dutch Housing Market}

Author

_{M.W. (Martijn) Schaaf}

Student number

_10360557

Supervisors

_{dhr. prof. dr. M.K. (Marc) Francke}

Amsterdam Business School, Faculty of Economics and Business

Ortec Finance

dhr. J.T. (Jesse) Groenewegen

Rabobank

Date

_{June, 2018}

Master

_Finance

Specialization

_{Finance and Real Estate Finance}

Faculty

_{Economics and Business}

(4)

Acknowledgements

Writing this thesis with the primary aim to graduate first, I soon realized there was more than meets the eye. Doing an internship at the research department of the Rabobank opened my eyes in this respect. Jesse, my supervisor at the bank, showed me the importance of performing research. He showed me the impact of research in terms of business and social aspects. Knowing a possibility existed that my research would not only result in graduation, but may also have a broader influence on society especially motivated me throughout the process. Because there is no research on this topic for the Dutch market yet as far as I am aware of, I sincerely hope my work paves the way for other students or researchers who are interested in this topic. This work would not have been possible without the help of people around me, who I would like to thank next.

First, I would like to thank my main supervisor Professor Marc. F. Francke of the Economics and Business Faculty at the University of Amsterdam. He is a true genius and gave nothing but clear support. Furthermore, I would like to thank Jesse, for the reason mentioned above, dr. Peter Boot, for ‘lending’ me the Dutch dictionary used in my linguistic analysis and Kristiaan van Drie from NU.nl, for providing me housing market articles from NU.nl. Also, I wish to thank all my friends and family for their support. Finally, I hope you will enjoy reading this study and I hope my work will bring about further research.

(5)

Abstract

This research explores the relationship between news media and the housing market of the Netherlands. It is based on international studies that show housing market variables are not only related to fundamental variables but also to media variables. Given the importance of the housing market for households and the overall economy, it is puzzling no research exists that includes such variables. Two media variables, sentiment entailed in housing market articles and the number of published articles, are incorporated in a vector autoregressive model with house prices and its fundamentals. Results confirm a significant role for fundamental variables such as the interest rate or income but also for the sentiment measure. This measure, obtained by running over 2.000 articles of traditional newspapers through content analysis software, contains predictive content for house prices. Additionally, results indicate its effect is more pronounced for online news, represented by articles posted on NU.nl. Furthermore, it appears the role of sentiment is larger for houses traded in lower price segments. This finding indicates especially lower-income households are prone to the influence of sentiment.

Keywords Real estate, media, sentiment, news, vector autoregression

(6)

List of Tables

Table 1: Literature on the role of media sentiment during financial distress ... 7

Table 2: Literature on the role of media sentiment during relative financial stability ... 8

Table 3: Literature on the political economy of media ... 14

Table 4: Dutch newspaper reach and LexisNexis coverage ... 24

Table 6: Descriptive statistics national data levels ... 33

Table 7: Descriptive statistics national data log differenced ... 34

Table 8: Descriptive statistics NU.nl coverage ... 37

Table 9: Relationship between media coverage and the housing market ... 44

Table 10: Sentiment (a)symmetry ... 49

Table 11: Relationship between NU.nl coverage and the housing market ... 50

Table 12: House price segments and media coverage ... 52

Table 18: Hypotheses and results ... 61

Appendix Tables Table A1: International studies that examine house price drivers ... 74

Table A2: Descriptive statistics of Dutch newspaper sample ... 76

Table A3: Unit root tests full sample (National newspapers) ... 77

Table A4: Unit root tests shorter sample (NU.nl) ... 78

Table A5: Cointegration tests full sample (National newspapers) ... 78

Table A6: Cointegration tests shorter sample (NU.nl) ... 79

Table A7: Granger-causal results national sample monthly lag specifications ... 80

Table A8: Granger-causal results NU.nl sample monthly lag specifications ... 81

Table A9: Granger-causal results segmented sample monthly lag specifications ... 82

Table A10: Granger-causal results national sample reach-weighted media variables ... 83

Table A11: Granger-causal results national sample quarterly lag specifications ... 84

Table A12: Granger-causal results national sample sentiment sub-categories ... 85

(9)

List of Figures

Figure 1: Media coverage of national newspaper sample ... 29

Figure 2: Media coverage NU.nl ... 36

Figure 3: Impulse response functions ... 60

Appendix Figures Figure A1: Dutch average house prices ... 87

Figure A2: Media coverage of De Volkskrant ... 87

Figure A3: Media coverage of NRC Handelsblad ... 88

Figure A4: Media coverage of Trouw ... 88

Figure A5: Media coverage of NRC.Next ... 89

Figure A6: Media coverage of Het Financieele Dagblad ... 89

Figure A7: Media coverage of Metro ... 90

Figure A8: Media coverage of Reformatorisch Dagblad ... 90

Figure A9: Media coverage of Algemeen Dagblad ... 91

Figure A10: Media coverage of De Telegraaf ... 91

Figure A11: Housing market fundamentals ... 92

Figure A12: House price development per sector ... 94

Figure A13: VAR stability ... 94

(10)

(11)

1 Introduction

After the global financial crisis hit the Dutch housing market around 2008, average house prices decreased by more than 16% until 2013. In the four years following this decline, prices increased by 24%, and average transaction prices rose by almost €50.000 (Statistics Netherlands, 2018). These figures represent large changes affecting both individual households and the Dutch economy as a whole. Because of the major costs associated with a home purchase, value reductions to this asset alter the occupants’ spending behavior and thereby affect the overall Dutch economy. On the contrary, increases to home equity reduce the relative mortgage payments of households and increase their purchasing power. Various researches exist that attempt to explain these Dutch house price developments on the basis of changes to fundamental macroeconomic variables like the interest rate, income, or unemployment (Galati, Teppa and Alessie, 2011; Kranendonk and Verbruggen, 2008; Verbruggen, Kranendonk, Van Leuvensteijn and Toet, 2005; De Wit, Englund and Francke, 2013). Such studies only examine the influence of traditional variables. Put differently, non-fundamental variables are usually not included in studies examining the housing market.

Niederhoffer (1971) is one of the first to study the influence of a non-fundamental factor in a financial market. Analyzing the effect of positive or negative word events on the volatility of United States (U.S.) stocks, he finds that these world events significantly influence the volatility of stock market returns. His study is one of the first to suggest non-fundamental factors contain explanatory power in financial markets. Subsequent literature on the influence of these factors in financial and non-financial markets stresses the importance of including non-fundamental factors (Case and Shiller, 2003; Baker and Wurgler, 2006; Tetlock, 2007).

Shiller and Akerlof (2009) provide an explanation for trading decisions not based on fundamental factors. According to them, these irrational trading decisions are driven by so-called ‘animal spirits’. They describe these ‘animal spirits’ as human instincts activated by investor sentiment. Investors then spontaneously feel the need to respond to observed changes in a market, not basing their responses on fundamental information (Shiller and Akerlof, 2009). Together with the herding behavior of investors, this results in irrational exuberance: general investors’ enthusiasm driving asset prices away from fundamentals (Shiller, 2005). Apart from the influence of fundamental variables, sentiment is then expected to drive the decision-making of households and thereby influence the Dutch housing market. Given the aforementioned findings, it is puzzling why only two studies include a non-fundamental element in their research on housing markets (Walker, 2014; Soo, 2015).

Walker (2014) and Soo (2015) employ media coverage in a statistical model to analyze its interrelation with the United Kingdom (U.K.) and local U.S. housing markets, respectively. Media

(12)

coverage is represented by the sentiment entailed in articles published by newspapers and the number of articles published. This sentiment measure then proxies for investor sentiment and is determined using linguistic analysis software (Shiller and Akerlof, 2009). Both studies find a significant relationship between their media variables and housing markets. Given the peculiar development of Dutch house prices, its effects on both individual households and the overall economy, the question arises why such research has not yet been performed for the Dutch housing market. This research fills the gap in this respect and answers the following research question: “What is the relationship between media coverage and the Dutch housing market?”. In specific, three relationships are examined: (1) between media coverage of traditional Dutch newspapers and the Dutch housing market, (2) between that of online media represented by NU.nl and the Dutch market and, (3) between media coverage of traditional newspapers and three price segments. These segments proxy for low, middle and high-income classes and allow the analysis of the role of sentiment throughout these segments. This analysis is based on literature that shows how lower-income households are more prone to the influence of sentiment (Lusardi and Mitchell, 2007).

To answer the main question, a comprehensive dataset including housing market variables, their fundamental determinants and media variables is used. To create this dataset, several sources are combined. Data on housing market variables and its determinants are obtained from Statistics Netherlands (CBS) and the National Institute Global Economic Model (NiGEM). In addition, a detailed dataset provided by the Dutch Association of Real Estate Brokers and Real Estate Experts (NVM) allows the creation of a hedonic price index used to construct the three price segments that proxy for income. For the media variables, LexisNexis serves as a source for traditional newspaper articles and NU.nl for those of online items. Based on the availability of newspaper articles, the full data set covers the period between July 2007 and December 2017.

The approach pursued to answer the main question is founded on four steps. First, studies on the role of media in financial and housing markets are discussed. Second, literature on international and Dutch housing markets is examined to determine the set of factors fundamentally related to the housing market. The approach then follows with a consideration of studies analyzing the role of media outside of financial markets because these researches provide additional explanations on how news media may affect households. Third, the approach to collect articles written about the housing market is examined, followed by the development of a measure for sentiment. This measure is based on a set of hand-collected articles describing the housing market. These articles are obtained from LexisNexis for the national and segment analysis and from NU.nl for the examination of the relationship between online news and the housing market. After

(13)

developing the sentiment measure and applying it to the sample of national newspapers’ articles and NU.nl’s posted items, the fourth and final step in the approach is to combine this data with the data on fundamental variables identified as explanatory for house prices. Then, by estimating a vector autoregressive (VAR) model, the interrelationship between the media variables and the Dutch housing market can be examined. Five VAR models are estimated, based on the three aforementioned relationships. One for the national sample, one for the NU.nl data and one for each price segment. These models allow causality to run either way: from the housing market variables to the media variables and the other way around. This is motivated by Walker (2014) and provides a clear image of the relationship between the variables, without imposing restrictions on the direction of these relationships.

The novelty of this research lies in the conversion of newspaper and online articles into a quantitative sentiment score. In order to determine this score, a custom Python program is employed combined with content analysis software. This content analysis software calculates the percentage of positive and negative words within each article, based on a set of predetermined psychosocial positive and negative words lists. Research shows words classified as negative in such psychological dictionaries may not be negative in a financial context (Loughran and McDonald, 2011). To mitigate the issue, these lists have been manually adjusted to accommodate for words that are truly positive or negative in a housing market context.

In light of the findings on sentiment as human instinct and its role in U.K. and U.S. housing markets, some implications of this research are expected (Shiller and Akerlof, 2009; Walker, 2014; Soo, 2015). Would the role of sentiment be confirmed for the Dutch market, the ramifications are major because it suggests irrational behavior is present and can explain prices beyond the effect of fundamental factors. Policies set out to change key variables may then not be as effective as desired, so this research is essential for the Dutch government. In addition, would the segment analysis suggest a larger sentimental effect for lower-income households, the Dutch government may want to increase the financial knowledge of these groups. Thus, this study is also important for private households who may be unknowingly influenced by sentiment. Furthermore, real estate investors could use a similar approach and include a sentiment measure to better forecast house prices. Finally, this is the first research performing such analysis for the Dutch market. Because it lays the groundwork for this topic, academics may use it to explore promising avenues of future research. The remainder of this research is structured as follows. First, Section 2 discusses the stance of the literature on the role of sentiment in financial markets, housing markets and non-financial markets. Subsequently, it provides an overview of international and Dutch housing market studies. Section 2 concludes with the formation of several hypotheses to investigate the role of the media

(14)

in the Dutch housing market. Subsequently, data collection and preparation methods of the housing market and media coverage variables are discussed in Section 3. Then, in Section 4, the methodology is explained and a model is set up for the national, NU.nl and segment analysis. The results of this methodology are considered in Section 5. Furthermore, several robustness checks are performed after which Section 5 ends with a summary of these findings with respect to the hypotheses defined in Section 2. Finally, Section 6 discusses conclusions and limitations and closes with suggestions for future research.

(15)

2 Literature review

This section discusses relevant literature on the research topic and is structured as follows. First of all, Section 2.1 describes research about the influence of the media on financial markets. Section 2.2 reviews work on its influence on housing markets. Subsequently, Section 2.3 investigates the role of media outside of financial markets, after which price dynamics of the Dutch housing market are examined in Section 2.4. Finally, Section 2.5 ends with the formulation of hypotheses based on the literature.

2.1 Financial markets

Many articles indicate a prominent role of media in both financial- and non-financial markets. Niederhoffer (1971) is one of the first to quantify the relationship between world events and stock price volatility. Categorizing headlines of the New York Times on a seven-point good- to bad scale between 1950 and 1966, he finds that these world events significantly influence the volatility of stock market returns. His article shows world events may alter the public’s view of a market by the tone embedded in the headline of a newspaper. The housing market could thus be affected if newspapers decide to report on the market. Their journalists’ reporting consequently creates a ‘housing market event’, whereby the direction of the market depends on the amount of optimism – or pessimism – entailed in the headline. Subsequently, however, Cutler, Poterba and Summers (1989) note that news only affects a small fraction of the volatility of the returns in a market. In particular, they analyze the 50 largest one-day returns occurring on the S&P 500 using a vector autoregression (VAR) based on monthly data from 1946 to 1987. They conclude that these events cannot be linked to potential changes in future fundamentals. More recent research by Cornell (2013) also does not reject the findings of Cutler et al. (1989), when using the broader CRSP value-weighted index instead of the S&P 500 over a period of 25 years. Research of Cutler et al. (1989) and Cornell (2013) thus fails to explain large movements in share returns by quantitative macroeconomic news. Unfortunately, these articles only try to explain the influence of news for the largest return shocks after major macroeconomic events have occurred.

Studies examining firm-specific events covered by news media do find evidence returns of covered companies differ from those not covered. For covered firms, returns are affected by the media because it provides fundamental information to investors (Tetlock, Saar-Tsechansky and Macskassy, 2008; Antweiler and Frank, 2004; Fang and Peress, 2009). Cumming et al. (2016) add that the influence of media is larger for less sophisticated investors. As this research focuses on the Dutch housing market, mainly consisting of regular households (NVB, 2017), i.e. less sophisticated investors, this finding significantly demonstrates media may serve as a valuable source of

(16)

information for these market participants. Households only have a rough indication of the market value of their house, primarily through real estate listed for sale in their neighborhood. By providing general information on the current state and expectations of the market, the media allows households to devise a more precise market value and also enables them to better appraise the future value of their house. Apart from the role of news media to supply information on fundamentals, a rich body of research exists on the other component entailed in news articles likely to influence the market through non-fundamentals: sentiment. The influence of this component is scrutinized in the next section.

2.1.1 Sentiment in news coverage

The role of sentiment is best defined as an irrational action by Shiller and Akerlof (2009, p. 4), who describe it as “… a restless and inconsistent element in the economy. It refers to our peculiar relationship with ambiguity or uncertainty”. They explain how irrational trading decisions are driven by human instincts, also called ‘animal spirits’. Investors then spontaneously feel the need to respond to observed changes in a market, not basing their responses on fundamental information (Shiller and Akerlof, 2009). Together with the herding behavior of investors, this results in irrational exuberance: general investors’ enthusiasm driving asset prices away from fundamentals (Shiller, 2005).The degree to which media drives this enthusiasm depends on several factors. For instance, it depends on the type of the published story, the slant of reporting, the readers’ attraction to the story in terms of his or her preferences and, consequently, on the amount with which the readers’ image of reality has changed (Shiller, 2005; Kindleberger, 1978; Dellavigna and Kaplan, 2007). Case and Shiller (2003) illustrate why the influence of sentiment may be greater in the housing market, especially during times of financial distress. According to them, it is primarily explained by the presence of many amateur investors, who are for instance first-time buyers and have limited professional knowledge. Because these ‘amateurs’ are more exposed to forecasts of prices because of their lack of expertise, they might overreact based on an otherwise simple story, causing momentum in house prices (Case and Shiller, 2003).

Hence, seminal literature stresses the important behavioral role of sentiment on investors. This implies, theoretically, that price movements and trading activity of the housing market may be affected through the tone of the information provided by news media, especially in times of bubbles or crisis. Before turning to the housing market, the influence of the media in financial markets is analyzed. First, the stance of the literature on the role of news media during times of financial distress is examined, followed by a section reviewing its role during more stable time periods. This discussion is divided into two sections because of the predicted differences in the

(17)

role of sentiment (Case and Shiller, 2003). The findings of these studies are summarized in Table 1 and Table 2, respectively.

2.1.2 Sentiment during financial distress

Among the most recent articles published on the role of news media during times of financial distress are those of Campbell et al. (2012) and Bhattacharya et al. (2009). These studies, summarized in Table 1, analyze the influence of news media during the British Railway Mania and the U.S. internet bubble. Media does not appear to influence markets during times of financial distress (Campbell et al., 2012; Bhattacharya et al., 2009). A potential conclusion may therefore be that media sentiment only marginally affects markets. This contradicts Case and Shiller (2003), who illustrate why the influence of sentiment is expected to be greater during times of distress. Another feasible explanation lies in the difficulty to develop a sound proxy for media sentiment and thereby correctly analyze the role of the media empirically.

For instance, the findings of Campbell et al. (2012) may partly mirror the obstacles when using software to analyze text. As explained later, using predetermined lists of words may result in misclassifications. On the contrary of machine-based textual analysis is the simple methodology employed in Bhattacharya et al. (2009) of manually categorizing articles as either good, bad or neutral. This approach might be too straightforward and this could be the reason why no relationship is found. Steps taken to minimize the influence of these issues are described later. Articles that examine the role of sentiment during more stable time periods do find evidence that sentiment affects markets, which implies the findings aforementioned may indeed be the result of methodological issues. These articles are discussed in the next section.

Table 1: Literature on the role of media sentiment during financial distress

Author(s) Study and scope Methodology Finding(s)

Campbell, Turner and Walker (2012) Media influence during U.K. British Railway mania, mid-1840s.

Using content analysis software to assign The

Times articles a pessimism score, running these

scores through a VAR.

Role of media only that of publishing information and did not contribute to magnification of bubble or price reversal. Bhattacharya,

Galpin, Ray and Yu (2009)

Media influence during U.S. Internet Bubble 2000s.

Classifying news items on 458 Internet initial public offerings (IPOs) and 458 similar non-internet IPOs as either good, neutral or bad, comparing returns.

Different stock returns between internet and non-internet IPOs cannot be explained by different news classifications.

(18)

2.1.3 Sentiment during relative stability

Covering a long time span of the 38 years between 1963 and 2001, Baker and Wurgler (2006) investigate if several sentiment proxies at the beginning of a given month can be used to predict patterns of U.S. firms’ returns at the end of the month. Proxies include the closed-end fund discount, the number of average first day-returns on IPOs and the dividend premium (Baker and Wurgler, 2006). Baker and Wurgler (2006) find that future cross-sectional stock returns are in fact conditional on these sentiment proxies. Obviously, similar sentiment proxies do not exist for the housing market. Their research provides evidence, however, for the psychological influence of sentiment and its ability to alter human behavior in financial markets. Several papers released afterwards confirm this behavioral element in financial markets, of which five are summarized in Table 2 and discussed below.

Table 2: Literature on the role of media sentiment during relative financial stability

Author(s) Study and scope Methodology Finding(s)

Edmans, García and Norli (2007)

Effect of investor sentiment after lost soccer match on asset prices for 39 countries (1973-2004).

Using local currency returns and international soccer results as mood variable in econometric model.

Significant decrease in market returns after losses of soccer match, stronger for smaller shares and for more important matches.

Baker, Wurgler and Yuan (2012)

Effects of sentiment components and whether sentiment spreads across six stock markets (1980-2005).

Decompose sentiment proxies into one global and six local indices.

Investor sentiment has significant effect on international volatility of markets and global sentiment can predict returns on the country-level.

Tetlock (2007) Relationship between U.S. stock market and media pessimism (1984-1999).

Collecting daily articles from a Wall Street Journal column, assigning pessimism score using software and implementing score in VAR.

Evidence that content of articles published can be used to forecast changes in broad indicators of stock market activity.

Dougal,

Engelberg, Garcia and Parsons (2012)

The relationship between the author of the U.S. Wall Street Journal’s column and stock market performance.

Similar to Tetlock (2007), but exploiting exogenous rotation across different journalists to analyze causal effect of author reporting on investor behavior.

Dow Jones Industrial Average (DJIA) future short-term returns can be predicted by journalist fixed effects. Baker and Wurgler

(2007)

Analyze the effect of U.S. investor sentiment on returns of individual firms and the stock market between 1966-2005.

Macroeconomic, ‘top-down’, approach: focus on attributing total sentiment effects to individual firms and stock market. Using similar proxies as in Baker and Wurgler (2006).

Sentiment measure feasible and patterns have detectable and regular effect on both individual firms and the stock market. Effect is bigger for stocks which are difficult to value or arbitrage.

(19)

First, Edmans et al. (2007) show how a national drop in investor mood, caused by the loss of a major international soccer game, drives a significant decline of that country’s stock market. This research reveals how mood, apart from the supply of fundamental information, can affect markets. Second, research by Baker et al. (2012) emphasizes sentiment spreads across markets and plays a major role in the volatility of international markets. Third, Tetlock (2007) investigates the relationship between news media pessimism and the U.S. stock market and finds evidence that the content of published columns can be used to forecast changes in broad indicators of stock market activity. Fourth, Dougal et al. (2012) complement the work of Tetlock and point out that columns written by more pessimistic journalists are associated with higher negative returns the day after. Fifth, Baker and Wurgler (2007) use the same proxies as in 2006 together with a top-down approach and infer sentiment has an important effect on the overall stock market, not only on individual firms. In addition, assets that are hard to either value or arbitrage are more influenced by sentiment (Baker and Wurgler, 2007). This may prove to be especially important for the housing market as houses are valued infrequently, a finding that is in line with the informational effect above-mentioned by Cumming et al. (2016): as news media caters to the information-demand of households through their supply of news articles, and households use this information to better estimate the value of their house, the sentiment entailed in those articles may significantly influence that households’ decision making.

The other factor that is found to be more influenced by sentiment, arbitrage, is likely to play only a minor role in housing markets. Only those that have enough capital available and are willing to accept the risk and opportunity costs associated with selling, finding a new house and moving could participate in housing market ‘arbitrage’ (Connock, 2002). In addition, wealthy market participants such as firms and investors face problems of market imperfections and costs. Because of the market’s heterogeneous nature, it would require very experienced people to construct a well-balanced portfolio of houses (Connock, 2002). In terms of derivative securities in housing markets, difficulties to arbitrage are present too. Research displays the difficulty in assessing whether the value of the security’s underlying assets are over- or undervalued (Barberis and Shleifer, 2003). Lewis (2010) states that even those who predicted U.S. mortgage-backed securities were overvalued before the crisis could have experienced significant losses if the boom had lasted a couple more years. Also, their success was made possible by financial innovations that allowed them to short-sell real estate related securities (Glaeser, 2013). In conclusion, housing markets are expected to be influenced primarily by the effect of media sentiment on value determination by households, because arbitrage is limited.

(20)

Although some articles only find a minor role for the media (Bhattacharya et al., 2009; Campbell et al., 2012), literature exists that indicates a more prominent role for media sentiment (Baker and Wurgler, 2006;2007; Edmans et al., 2007; Baker et al., 2012). These papers reject neoclassical theories that assume households make rational decisions and are instead more in line with behavioral models of financial markets. In these models, two assumptions are commonly made that are based on two types of investors: rational- and irrational investors (De Long, Shleifer, Summers and Waldmann, 1990). De Long et al. (1990) argue that irrational investors are exposed to – and trade on the basis of – the influence of sentiment, defined as non-factual beliefs about prospects. Additionally, Shleifer and Vishny (1997) state as second assumption that arbitrage possibilities are limited. Rational investors can bet against the irrational, sentimental, investors but only at high risk and costs (Shleifer and Vishny, 1997). Applied to the housing market, the risk and costs would naturally be even higher, as research by Connock (2002) shows. Behavioral models confirm the potential role of sentiment on households. Given the expected role of media sentiment in the housing market, it is remarkable that only two studies look at the effect of media coverage on the housing market: the work of Walker (2014) and Soo (2015). Their researchis most closely related to this study and is reviewed in the next section.

2.2 Housing Markets

Walker (2014) investigates news media coverage in the housing market of the U.K. and applies a methodology that resembles the methodology used by Tetlock (2007). He collects over 30.000 articles on the housing market of several U.K. newspapers between 1993 and 2008 by using the predefined ‘Housing Market’ tag of database LexisNexis. This tag unbiasedly returns all articles related to financial conditions in the residential property industry. These articles are then run through Diction 5.0. Similar to the General Inquirer used by Tetlock, Diction analyses text, but instead of assigning a pessimism score, it assigns an optimism score per article. The aggregated monthly score is then used in a VAR to determine the interrelation between this optimism score, changes in real house prices, frequency of published articles and the volume of housing sales. In this model, real average income, cost of borrowing and the unemployment rate are added as control variables (Walker, 2014). After estimating the model, Walker (2014) concludes media sentiment, not the number of articles published or the volume of housing sales, is related to real house price changes. This finding could prove the role of the media to influence households but may also be driven by misclassifications of content analysis software.

In general, this type of software uses categorized lists of predetermined words from psychosocial dictionaries. For instance, ‘good’ and ‘nice’ would be listed under a category of words

(21)

that describe positivism. In order to determine an article’s positivism score, the number of words that fall under this category are added up and divided by the total number of words in the article. Loughran and McDonald (2011) investigate the use of psychosocial word lists in financial markets and find that words are often misclassified. To give an example, ‘tax’ is defined as negative in psychosocial dictionaries but of course is not necessarily negative in a financial context, which could have driven the results of Campbell et al. (2012). Walker (2014) reasons that because news articles are read by the general public and not professional traders, these issues are largely avoided. The validity of this reasoning is doubtful, because the word ‘tax’ itself is not negative and, moreover, may even be placed in a positive context. For instance, if a tax reduction is discussed, the optimism score will have a downward bias due to this misclassification. Later research by Loughran and McDonald (2015) confirms the above: of the specific words used by Diction to gauge optimism and pessimism, 83% and 70% are misclassified for the use of financial text analysis, respectively. Unfortunately, Diction 5.0 does not allow modifications of its inbuilt dictionaries so these issues cannot be resolved.

Soo (2015) limits the influence of these issues when studying the predictive power of sentiment in housing news on future prices throughout several U.S. cities. Soo (2015) customizes the standard list of Harvard’s General Inquirer to accommodate for articles related to the housing market. To collect these articles for several U.S. cities, Factiva.com is used. Similar to the ‘Housing Market’ tag of LexisNexis, Factiva.com provides an objective algorithm to classify articles that discuss local real estate markets. The result is a dataset of nearly 40.000 articles covering the major newspapers of 34 U.S. cities, between January 2000 and December 2013 (Soo, 2015). Defining sentiment as the share of positive words minus the share of negative words in the sample of housing-related articles, she finds sentiment precedes house price changes on average by two years. This result was obtained after estimating a linear framework consisting of a set of economic variables that have been found to predict house price development, previous price changes and the sentiment measure.

On the city-level, some interesting results arise: house price variation in cities experiencing small price appreciation are more related to traditional factors, while variation in cities that experience unstable price appreciation are more related to media sentiment. These observations support the role of sentiment in local media. Finally, Soo (2015) points out that the sentiment measure could indeed proxy for household sentiment but could also proxy for unobserved fundamentals that would be hard to measure otherwise. Would the latter be the case, performing this research for the Dutch market would still be useful since it would provide a measure for these unobserved fundamentals. This is not the only reason demonstrating the relevance of this research.

(22)

Apart from gaining a broader understanding of the role of media sentiment in a different housing market, studying the role of sentiment in news media for the Dutch housing market is important for at least two reasons.

The first reason is that news media may mitigate informational issues in inefficient markets. As put forth by the literature, irrational households are affected by news in the sense that they use the information supplied in their decision making. Coupled with the inefficiency of a housing market, the influence of media on households is likely significant. Case and Shiller (1989, 1990) first suggested that housing markets are not informationally efficient, which was not rejected by subsequent papers (Pallakowski and Ray, 1997; Brunnermeier and Julliard, 2008; Woodward and Hall, 2011). The presence of housing market inefficiency is confirmed for the Dutch housing market (Conijn, Schilder and Englund, 2010). Conijn et al. (2010) further indicate a strong dysfunctionality of the Dutch housing market, resulting in, among others, significant welfare losses. The second reason deals with the importance of the Dutch housing sector for the overall Dutch economy. In order to maximize their utility, households accumulate and reallocate their wealth over time. Wealth is accumulated by consumption and financial assets that produce income in order to smooth consumption during years of minor- or zero income (McCollough and Karani, 2014). Reallocation of current income towards either consumption or financial assets happens as new information arrives on either the overall economy or the assets owned by these households (Deaton, 1992). Deaton (1992) argues that, from a neoclassical stance, households make these reallocation decisions in a rational fashion. Literature previously mentioned shows, however, that households exhibit irrational behavior and are prone to the influence of sentiment. History confirms that households decreased their consumption and increased their savings after receiving information about the falling values of house prices prior to the Great Recession. This, in turn, was a factor that took part in amplifying the Great Recession (McCollough and Karani, 2014).

Both aforementioned reasons outline the relevance of this research. To the best of my knowledge, this is the first study that examines both the role of sentiment in national media and the role of sentiment in online media for the Dutch national housing market. The importance of studying the relationship between different media sources and other markets is highlighted by Oberholzer-Gee and Waldfogel (2006), and Engelberg and Parsons (2011). Oberholzen-Gee and Waldfogel (2006) find that turnout of Hispanic voters increases in markets where local Spanish language news is made available. In addition, for all S&P 500 firms with earnings announcements, coverage of local media can be used to robustly estimate local trade (Engelberg and Parsons, 2011). The influence of sentiment of local press is thus important for local housing markets. The next

(23)

section, therefore, considers studies on the role of the media outside of financial markets, in order to gain a broader understanding of the role of the media.

2.3 Non-Financial Markets

Opposed to the limited amount of research published on the relationship between media sentiment and housing markets stands the amount of work focusing on the influence of media outside financial markets. Their most important findings are summarized in Table 3 and their contributions to this study are briefly discussed. Additionally, several papers are considered that investigate the biasedness of media reporting. Studies outside of financial markets are examined because they provide additional explanations on how news media may affect households.

2.3.1 Effectiveness of opinion change through media

Influential historical studies on the influence of media on the general public, also called the political economy of media, highlight some problems in this line of work. Lazarsfeld (1944) examines the impact of a U.S. presidential candidate’s campaign on the outcome of the election and notices that the role of media is minor. Here, a selection bias exists: voters that pay attention to the media are those that expose themselves only to the press releases of their political side, because they have already picked a side. No influence of media was found in the election and this could be due to the selection bias (Lazarsfeld, 1944). This may imply that ex-ante beliefs on the state of the housing market result in households ignoring news articles that contradict their beliefs. Even though households may be influenced by the media in general, they can never be influenced if they ignore the media. Fortunately, research on the housing market allows for the analysis of price changes over time, so that actions taken by households can be observed as changes in house price. In terms of participation, media does seem to affect voters. Researching U.S. press coverage on knowledge of citizens and their actions taken, Snyder and Strömberg (2010) report that if local press covers more of a local representative, voters are more likely to vote for him or her. Households’ activity in the market may thus increase if the media reports more on the housing market.

Later research by Lumsdaine (1953) studies how the degree of effectiveness of opinion change through communication depends on: (1) the credibility of the communicator from the readers’ perspective, (2) the degree of factual information used and (3) audience characteristics influencing their susceptibility and response. Applied to newspapers, these results imply that more influence is brought about by more credible newspapers and newspapers that report more factual information about the housing market. It must be noted, however, that higher credibility only affects opinion in the short-run. Also, influence may depend on household’s characteristics, such

(24)

as income or education. If household’s opinions are only reinforced by the media because of their prior beliefs, and this reinforcement depends on the factors previously mentioned, how is opinion change then brought about? This question is discussed in the next section.

2.3.2 Role of others in opinion change

Four papers that answer this question describe the role of the group in influencing the beliefs of the individual (Sherif, 1935; Weimann, 1991; Loeper, Steiner Stewart, 2013; Katz and Lazarsfeld, 1955). Sherif (1935) investigates the ‘reference group theory’: whether an individual’s values and attributes are influenced when placed in a group. After first building up his or her own norm, the individual replaces it with the group’s common norm (Sherif, 1935). In addition, Weimann (1991) concludes that a friend or family member with superior knowledge about a product is often used as reference point of information by a less informed friend or family member. Loeper et al. (2013) show that if a less informed individual observes the knowledge of its perceived expert, he or she Table 3: Literature on the political economy of media

Author(s) Study Finding(s)

On the effectiveness of opinion change through media

Lazarsfeld (1944) The influence of a presidential campaign and the role of the media.

Selection bias: voters that keep an eye on the media are those who already picked a side and expose themselves only to the media released by that specific side.

Snyder and Strömberg (2010)

Estimate the impact of U.S. press coverage on the knowledge of citizens and actions taken.

If local press covers more of a voters’ local representative, this voter is more able to recall the name of this person, more able to either describe or rate him or her and more likely to vote.

Lumsdaine (1953) How opinion changes depend on several factors.

Effectiveness of opinion change through communication depends on its publisher’s credibility, factual content and type of target audience itself.

On the role of others in opinion change

Sherif (1935) Influence of the group on values and attitudes of individual.

Individuals reject their own opinion to form a common norm.

Weimann (1991) Analyze the role of friends and relatives in the information diffusion process.

Friends and relatives with superior knowledge are viewed as most reliable for information by prospective consumers, who adopt their beliefs.

Loeper, Steiner and Stewart (2013)

Do less informed individuals adopt the knowledge of an expert?

Less informed individual will adopt the knowledge of the better informed after observing the information of the superior informed.

Katz and Lazarsfeld (1955)

Channel through which opinion changes are brought about.

Two-step flow hypothesis: media influences a specific part of society that influences another.

(25)

will adopt this knowledge. These papers provide evidence for the existence of the ‘two-step flow of communication’, as proposed by Katz and Lazarsfeld (1955). According to them, individuals that are more exposed to media are influenced by it and subsequently influence others. When households form their expectations, they are likely to be influenced by those of whom they believe possess superior information. Households that are more interested in the housing market are more exposed to the media, thereby consuming more information to subsequently alter the beliefs of less informed households. Media may thus lead the herding behavior of households (see also Shiller and Akerlof, 2009). For this reason, it is important to address literature that examines the role of potential biasedness in media reporting. If media reports unbiasedly, households may read any newspaper to form an objective belief on the housing market’s future and the articles of any newspaper can be collected for this research.

Throughout the literature, the media has not been found to report unbiasedly (Mullainathan and Shleifer, 2005; Groseclose and Milyo, 2005; Gentzkow and Shapiro, 2010). According to Mullainathan and Shleifer (2005), newspapers segment the market and slant towards the beliefs of their readers. As for Groseclose and Milyo (2005), they are of the opinion that newspapers report biasedly towards either the left- or right political spectrum. Later research confirms both studies: newspapers report towards a political ideology and report to consumer preferences (Gentzkow and Shapiro, 2010). Households should then read as many newspapers as possible to develop an unbiased opinion. This seems rather impossible in practice. Reader heterogeneity appears to play a major role in newspaper reporting and should be taken into account.

2.4 Dutch Housing Market Background

Because this research will focus on the Dutch housing market, this section discusses the background of the Dutch housing market in order to create a broader understanding of its historical price development and characteristics. Nominal Dutch house prices have almost tripled from nearly 95.000 EUR in 1995 to roughly 265.000 EUR in 2017. Real prices also increased but at a slightly lower rate. Prices increased from about 130.000 EUR in 1995 to almost twice this value in 2017: 240.000 EUR. Three growth stages can be identified between 1995 and 2017 (see Figure 1 in Appendix B for a graph of these developments).

First, between 1995 and 2008, prices seem to grow at an increasing rate, followed by a diminishing increase and subsequently rise again until the 2008 crisis. Within the second stage, except marginal increases in 2010 and 2011, prices decline until 2014 with real prices dropping below nominal prices. Third and final, from 2014 onward average prices rise again, with the nominal level even exceeding the pre-crisis 2008 level. Average annual real growth over these three

(26)

stages is 6%, -5% and 5%, respectively. Which factors account for these changes in Dutch house prices?

Many international papers provide explanations of price developments in housing markets. Galati et al. (2011) analyze the role of micro and macro factors in Dutch house price dynamics. They state empirical studies generally categorize these factors, also called determinants, as macro variables, regional- or funding conditions. Because of the relatively scarce amount of literature for the Dutch market, the factors found per determinant are first summarized on the international level and subsequently discussed for the Dutch market. An overview of the important variables found throughout international studies is presented in Appendix A.

2.4.1 Macroeconomic variables

Several macroeconomic factors can be identified from international studies. In sum, these studies state the importance of income, real gross domestic product (GDP), the interest rate, stock prices and inflation in the determination of real house prices (Reichert, 1990; Englund and Ioannides; 1997; Sutton, 2002; Hofmann, 2003; Hofmann and Peersman, 2017). Sutton (2002), for example, finds that the three major drivers of house price fluctuations in the U.S., U.K., Canada, Ireland, the Netherlands and Australia are national income, interest rates and stock prices. It must be noted, however, that the role of the interest rate seems ambiguous. Opposite to the findings of Sutton (2002) are two studies that find a smaller role for the interest rate (Bhutta, Dokko and Shan, 2010; Gyourko, Gottlieb and Glaeser, 2010). Appendix A presents the overview of macroeconomic variables identified as house prices drivers by international research.

Turning to the housing market of the Netherlands, Verbruggen et al. (2005) estimate an error correction model to explain its price development between 1980 and 2003. Verbruggen et al. (2005) find that, in the long-run, real house prices are determined by real available income from wages, real interest rate, housing stock and financial wealth of households. In the short-run, prices are only influenced by real available income from wages and housing stock. Furthermore, prices are altered by the nominal interest rate, the consumer price index and deviations from the actual and long-term level of the previous period (Verbruggen et al., 2005). In addition, they analyze the drivers during specific time frames. In terms of Figure 1 of Appendix B, they attribute the strong growth in the beginning of the first stage (1995-2000) to two wealth-related factors. First, long-term home values increased because of the higher income and financial wealth of the households. Second, income growth has a direct impact on house price developments. The subsequent smaller growth between 2001 and 2003 is attributed to reduced wealth in combination with a relatively low interest rate reduction during the period. Later research verifies the role of income, interest and

(27)

financial assets (Kranendonk and Verbruggen, 2008). Besides the Dutch confirmation of the importance of income, interest, wealth and housing stock on house prices, it must be noted that determinants in house prices are found to vary over time. The latter is backed by a paper of Dröes and Van Der Minne (2015), who find that long-run house price determinants in the local housing market of Amsterdam change over time. Additional evidence on the explanatory power of the interest rate is provided by more studies (De Wit et al., 2013; Galati et al., 2011).

De Wit et al. (2013) find both an indirect and direct effect of the interest rate. Direct is the influence of the mortgage interest rate on the frequency of housing sales and more gradual is its effect on housing prices, which implies households may gradually adapt to changing market conditions (De Wit et al., 2013). Galati et al. (2011) confirm that especially the long-term interest rate has strong explanatory power. They add another macro variable explains house prices: the dependency ratio. This ratio is defined as the population aged above 65, divided by the population aged between 15 and 64 (Galati et al., 2011). Their results indicate a significant positive relationship, so higher relative amounts of elder people result in higher house prices. House prices in regions or cities of which the population consists more of elder people may thus be higher.

Finally, Tu, De Haan and Boelhouwer (2017) study long-run equilibrium in the Dutch housing market, incorporating the degree of regulation in their model. Their findings indicate a long-run relationship between house prices, strict regulation, income, interest rate and inflation. Multiple macroeconomic variables are thus important for Dutch house price developments and their influence can vary over time.

2.4.2 Regional conditions

The stance of the literature on the influence of regional variables affecting house price development is discussed in this section. From the international literature, it becomes evident house price dynamics exhibit geographical differences, based on a number of factors. Factors found to influence these dynamics are local: investor speculation, income, population mutation, construction costs, regulation, supply elasticity, cyclicality of the market, ageing and neighborhood effects (Case and Shiller, 2003; Glaeser and Gyourko, 2005; Capozza, Hendershott and Mack, 2004; Green, Malpezzi and Mayo, 2005; Himmelberg, Mayer, Sinai, 2005; Gao, Lin and Na, 2009; Takáts, 2010; Kiefer; 2011; Glaeser, Gyourko, Morales and Nathanson, 2014). Capozza et al. (2004) estimate serial correlation in combination with mean reversion for a set of U.S. metropolitan statistical areas (MSAs) from 1979 to 1995 and find that serial correlation is more present in areas with higher income, population growth, and real construction costs. They also document that larger MSAs, faster-growing cities and cities with lower construction costs exhibit more mean reversion. Case

(28)

and Shiller (2003) add that speculation may even explain price changes better than income in some geographical areas. Again, an overview of the market-specific conditions identified as house prices drivers by international studies is provided in Appendix A.

Studies that examine the influence of regional variables on house price development in the Dutch market are scarce and present a slightly different view on the influence of some variables. Glaser and Gyourko (2005) note a more significant decline in house prices in the U.S. due to a decrease in population than an increase by a similar amount. On the contrary, research by Rouwendal and Vermeulen (2007) suggests demographic factors may be of less influence on the Dutch market. The latter is confirmed by Francke (2010), who finds no empirical evidence for a relationship between Dutch price changes and changes in population or mutations to the number of households. Ebner (2013) documents that the low supply elasticity of housing supply can partially be used to explain the Dutch house price boom of the 1990s. Another study omits housing supply due to its inelastic nature (Mrkaic, Hassine, and Saksonovs, 2005). Therefore, the overall usage of housing supply when modelling Dutch house prices appears inconsistent.

Regarding local markets, housing supply is found to play a different role throughout different locations (De Vries and Boelhouwer, 2005). They show that newly built housing around Amsterdam, Rotterdam, The Hague and Utrecht results in a decrease in prices. These results do not hold in other areas, indicating that increases in the number of houses are market-compliant (De Vries and Boelhouwer, 2005). Influence of supply thus appears to differ across areas. Galati and Teppa (2017) further investigate the degree of heterogeneity across Dutch market segments and its potential drivers for house prices from 2003 to 2016. Their results point out heterogeneity across different market areas. They confirm the results of the study by Gao et al. (2009) for the Dutch market: some markets are more cyclical and thereby exhibit more serial correlation. In particular, they posit that the speed of convergence, the mean reversion and the efficiency of the housing market depends on the house’s geographical location, urbanization, year- and type of construction.

Following the findings above, the influence of demographical factors on Dutch house prices seems minor, supply may only play a regional role and heterogeneity is present throughout market segments. These results stress the need to examine these variables carefully when modelling either national or regional house prices.

2.4.3 Funding conditions

Finally, the role of funding conditions as a potential determinant of house prices is discussed (Galati et al., 2011). From the body of international literature that exists on the role of funding

(29)

conditions, three dimensions are identified that influence the housing market. First, the adaptability and depth of the national mortgage market (Warnock and Warnock, 2008). Second, the structure of mortgage markets (Tsatsaronis and Zhu, 2004; Jung and Lee, 2017). Third, the different effect of lending on house price dynamics across market areas (Mian and Sufi, 2009; Damianov and Escobari, 2016). The factors that belong to these dimensions that influence house prices through funding conditions are summarized in Appendix A.

Swank, Kakes and Tieman (2002) find that price volatility of owner-occupied houses in the Netherlands is driven by favorable tax regulations for homeowners and high loan-to-value ratios. Here, especially the mortgage interest deduction plays a major role in house price growth (Swank et al., 2002). On the contrary, Galati et al. (2017) do not find that distinct mortgage finance structures result in different price dynamics across housing markets. Arguing from the findings that price dynamics in local real estate markets behave differently may provide an explanation for these divergent results. This could be explained by the fact that the research of Swank et al. (2002) is based on aggregated house prices in the Netherlands and that of Galati et al. (2017) on the regional level.

Francke, van de Minne and Verbruggen (2014) analyze the impact of credit conditions on the Dutch housing market. Credit conditions are commonly defined as the supply of credit on the mortgage market, apart from the interest rate and income (Fernandez-Corugedo and Muellbauer, 2006; Francke et al., 2014). Francke et al. (2014) find that their credit condition index, representing changes in the supply of mortgage credit, can explain almost half of the Dutch real house price decline from 2009 to 2012. These results show that, besides the role of income and interest, credit conditions can also explain house price changes.

Evident from the stance of the presented literature is that explaining price dynamics of the Dutch housing market is difficult and may depend on many factors which could vary over time. The final set of these explanatory variables for the Dutch housing market are discussed in the methodological section. The rejection of the existence of a purely rational investor, however, does appear to be clear-cut. Seminal work, several articles and the behavioral model confirm the existence of an irrational investor that acts not only on factual information but also on his or her non-factual beliefs; beliefs that are prone to the influence of the sentiment entailed in news media. Here, the amateurism of households is a potential predictor of a greater influence of sentiment in the housing market (Cumming et al., 2016). Case and Shiller (2003) argue that this gives rise to price movements and increases in trading activity, especially during times of financial distress. Among others, Baker and Wurgler (2007) confirm the view of Case and Shiller (2003), adding that assets hard to value or arbitrage are more prone to the influence of sentiment. Although two studies

(30)

reject this view, their conclusions may be driven by methodological issues (Campbell et al., 2012; Bhattacharya et al., 2009).

Most findings indicate a role for media sentiment in the housing market. Therefore, it is puzzling why only Walker (2014) and Soo (2015) empirically analyze the relationship between media sentiment and housing markets. Both studies confirm the role of media sentiment. Walker (2014) finds only sentiment Granger-causes real house price changes and not the number of published articles. Nonetheless, the number of published articles does affect trading activity (Walker, 2014). Soo (2015) adds that the effect is higher within cities that experience unstable price appreciation. Studies that examine the influence of media on the general public outside of financial markets add that market participation increases after press coverage and that the media’s influence depends on household characteristics such as income or education (Snyder and Strömberg, 2010; Lumsdaine, 1953). Trading activity in the Dutch housing market may thus increase if media reports more on the market and the influence of sentiment may differ throughout Dutch cities with different household characteristics. Based on the literature, hypotheses are formed in the next section.

2.5 Hypotheses

After having discussed the stance of the literature on sentiment analysis and the background of the Dutch housing market, this section will form hypotheses in order to answer the research question ‘What is the relationship between news media coverage and the Dutch housing market?’. It must be noted that it is beyond the scope of this thesis to establish true causality. The Granger-causal relationship between news media reporting and the housing market is examined (Granger, 1969). Granger-causality implies that a particular variable contains predictive content for another variable. Therefore, if causality is referred to throughout this research, Granger-causality is referred to unless explicitly mentioned otherwise.

Hypothesis I: Dutch news media coverage and Dutch house prices are correlated.

Dutch media reports extensively on the Dutch housing market during both excellent and poor months. Correlation is expected between the media and housing market because it is the housing market that the media covers. News media coverage is defined as both the sentiment entailed in news articles, as well as the number of articles published. Because correlation is expected, Hypotheses II and III describe the expected effect of two components of media coverage. Expectations on the influence of sentiment entailed in online news are discussed in Hypothesis IV. Finally, Hypothesis V provides expectations on the interplay between house price segments and

(31)

sentiment. This segment influence is analyzed because research outside of financial markets shows how different household characteristics influence the susceptibility to media sentiment (Snyder and Strömberg, 2010; Lumsdaine, 1953).

Hypothesis II: Changes in sentiment can predict changes in both house prices and trading volume.

For the hypothesis, this implies that a ‘predictive causality’ of sentiment on house prices and changes in trading volume is expected. As housing markets have been found to be informationally inefficient (Case and Shiller, 1989; Verbruggen et al., 2005), sentiment is expected to affect house prices by serving as a tool of value determination for households (Cumming et al., 2016; De Long et al., 1990). Households gradually learn more about the state of the market and trade on the basis of this information, indulging prices movements and changes to trading volume (Case and Shiller, 2003). Studies for other markets, using different sentiment measures, find that sentiment contains predictive content for average house prices and trading volume (Walker, 2014; Soo, 2015).Here, the effects of increases and decreases in sentiment on house price changes and changes in trading volume need not be symmetric. Expectations regarding the symmetry of the effect are considered in the next hypothesis.

Hypothesis III: Decreases in sentiment have a different effect on the housing market than increases in sentiment.

This expectation also follows from Case and Shiller (2003) who argue the influence of sentiment in the housing market is especially persistent during times of financial distress. According to them, it is primarily explained by the presence of many amateur investors in the housing market. These are mostly first-time buyers, lack expertise and are more likely to overreact to an otherwise simple story (Case and Shiller, 2003). Although several studies do not find a more prominent role for sentiment during financial distress, the results of those studies are likely driven by the empirical difficulties accompanied with sentiment analysis (Campbell et al., 2012; Bhattacharya et al., 2009). It is thus expected that sentiment declines have a different effect in the media coverage-housing market interrelationship than rises in sentiment.

Hypothesis IV: Online published articles predict larger changes to house prices and trading volume than the coverage of traditional newspapers.

No literature exists on the relationship between the Dutch housing market and its reporting by online media. This hypothesis aims to shed light on the expected connection between the two.

Media coverage and the Dutch housing market

Media Coverage and the Dutch Housing Market

Master of Science Thesis

June, 2018

Author

M.W. (Martijn) Schaaf

University of Amsterdam

Faculty of Economics and Business

MSc Finance

Finance and Real Estate Finance

Supervisor

Statement of Originality

Media Coverage and the Dutch Housing Market

M.W. (Martijn) Schaaf

10360557

dhr. prof. dr. M.K. (Marc) Francke

Amsterdam Business School, Faculty of Economics and Business

Ortec Finance

dhr. J.T. (Jesse) Groenewegen

Rabobank

June, 2018

Finance

Finance and Real Estate Finance

Economics and Business

Acknowledgements

Abstract

Table of Contents

List of Tables

List of Figures

1

Introduction

2

Literature review

_{Media Coverage and the Dutch Housing Market}

_{M.W. (Martijn) Schaaf}

_10360557

_{dhr. prof. dr. M.K. (Marc) Francke}

_{June, 2018}

_Finance

_{Finance and Real Estate Finance}

_{Economics and Business}