Quantitative easing and house price exuberance: to what extent does quantitative easing cause price misalignment in the housing market?
Ian IJnzen 10659358
Msc. Real Estate Finance Supervised by Dr. M.I. Droës
Abstract
The aim of this study is to investigate the extent at which the large-scale asset purchase program of the ECB also known as Quantitative Easing influences the housing and whether this program leads to price misalignment within this market. To examine the effects of Quantitative Easing: First a hedonic price model is build. Consequently, IV-regression is conducted. Finally, a user cost model is used to examine if the observed housing prices are still in line with their fundamental values. The data consists of 446,085 transactions and macro-economic variables over the period 2000 to 2017 using quarterly data. The model shows a large and significant impact of QE on housing prices in both Amsterdam and the larger municipalities. A weakly significant impact is found for the country wide index. The results range from 0.34 average percentage point extra quarterly growth in the Netherlands to 2.35 percentage point extra quarterly in Amsterdam. The findings are robust over multiple sample periods. Moreover, when comparing current price-to-rent ratios with historic ratios the market seems to be overpriced. However, when controlling for the lower interest rates and tax deductibility of mortgage interest payments using the Himmelberg, Mayer and Sinai (2005) user cost model the outcome of the model does not yield significant overpricing.
2 Statement of originality
This document is written by Student Ian IJnzen 10659358 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.
Acknowledgements
I would like to thank Dr. M. I. Droës for his supervision and feedback on this Msc. Thesis.
Furthermore, I would like to thank the Dutch Association of Real Estate Agents (NVM) for providing transaction data of all the requested municipalities and time periods.
At last I want to express my gratitude towards dr. F.P.W. Schilder for second reading my thesis and providing feedback.
3 Table of contents
1. Introduction 4
2. Literature review 5
2.1. Quantitative Easing 6
2.2. Fundamental drivers of housing prices 7
2.2.1. Micro determinants 9
2.2.2. Macro-economic determinants 9
2.3. Bubbles 10
3. Data 12
3.1. Data sources 12
3.2. Constant quality hedonic price index 12
3.3. Macro-economic variables 14
4. Methodology 16
4.1. Measuring the effect of quantitative easing 16
4.2. Robustness checks 18
5. Results 19
5.1. Price misalignment in the housing market 23
6. Conclusion and discussion 27
Reference list 30
Appendix 33
Table of figures Tables
Table 1: Important announcements regarding quantitative easing 6
Table 2: Data sources 12
Table 3: Transactions in the hedonic index 13
Table 4: Descriptive statistics as input for the hedonic price index 13
Table 5: Descriptive statistics of macro-economic variables 15
Table 6: The four-step Barron and Kenny (1986) model. 18
Table 7: Results of the four-step model 19
Table 8: Results of the TSLS IV-regression. 20
Table 9: The effects of quantitative easing on house price growth 20
Table 10: The effects of quantitative easing on the housing market using different sample periods 21
Table 11:Input for Himmelberg et al. (2005) user cost model 24
Table 12: Input variables for land lease calculations 26
Table 13: User cost and price to rent ratios 26
Figures
Figure 1: Comparing the Hedonic index vs. the CBS price index 14
Figure 2: Comparing the nominal, real and mortgage interest rates 15
Figure 3: The impact of QE on mortgage rates vs. the purchasing amount 21 Equations
Equation 1: Hedonic Price Index 12
Equation 2: Simple OLS 16
Equation 3: First differences 16
Equation 4: Logged first differences 17
Equation 5: First step of TSLS-IV 17
4 1. Introduction
In the aftermath of the 2007 global financial crisis, Europe suffered from low growth and deflationary pressure. To stimulate both inflation and economic growth, the European Central Bank started a large-scale asset purchase program (LSAP) in March 2015. Currently, more than ten years after the start of the financial crisis, the European Central Bank is still buying assets worth of 30billion a month. By doing so the ECB has grown its’ balance sheets to 4.5 trillion euro, an amount unprecedent in history.
Several papers have investigated the effect of quantitative easing on bond prices. De Santis (2016) found “sizeable” effects on bond prices with the vulnerable euro countries benefiting the most. Furthermore, Lamoen et al. (2017) observed exuberant bond price behavior after the announcement and implementation of QE. Altavilla et al. (2015) estimated the decline in yields on 10-year government bonds to be around 30-50 basis points. Recent market interest rates and mortgage interest rates have been fluctuating around record lows (Bloomberg, 2018; DNB, 2018). Existing literature is consistent, both interest rates and high availability of capital are important factors in determining housing prices (Harris, 1989; Reichert 1990; Tsatsaronis and Zhu, 2004; McQuinn & O’Reilly, 2008; Campbell et al. 2009). From the year 2015 housing prices have indeed been increasing rapidly (CBS, 2018; NVM, 2018). However, the link between quantitative easing and the current rapid increase in housing price has not yet been made. This study aims at investigating to what extent the large-scale asset purchase program of the ECB influences housing prices and whether this leads to price misalignment within the housing market. The central question asked:
“Quantitative easing and house price exuberance: to what extent does quantitative easing cause price misalignment in the housing market? “
Three important channels though which quantitative easing affects housing prices can be distinguished. 1. The interest rates channel. This channel directly affects the price of housing through lowered mortgage payments. 2. The rebalancing of investment portfolios. 3. The risk-channel where quantitative easing is expected to have a positive effect on liquidity and thus decreases the liquidity premia. The focus of this study is foremost on the interest rate channel, the channel that ex-ante seems to be most impactful.
This research fits in the existing literature by linking the already examined effects of quantitative easing on interest rates and bond pricing to the real estate market. Quantitative easing was first introduced in Japan in 2001, but after the 2007 global financial crisis, quantitative easing is currently one of the most used unconventional tool by central banks all over the world (Oikonomou, 2017). Given the vast amounts of money and the significant impact of QE on the financial markets it is
5 important to carefully map all externalities of this purchasing program. This paper contributes by making the first steps into mapping the externalities in the real estate market.
To examine the effects of QE on housing prices three consecutive steps are taken. At first, to control for changes in the quality of a property, a hedonic price index is estimated. The input for the hedonic index consists of 446,086 transactions over the 2000-I to 2017-IV period collected from the NVM database. An index is created for both Amsterdam and an aggregate index of the largest municipality for every province in the Netherlands. Secondly, the estimated indices and a CBS index for the Netherlands as a whole are used as the dependent variables in a TSLS IV-regression. In this regression the mortgage rate is the endogenous variable and a QE dummy functions as the instrument. Controls are added to distinguish macro-economic effects from QE.
The findings show a significant effect of quantitative easing on housing prices. The effect is smallest for the aggerate country index where QE attributes to 25% of the quarterly growth in period after 2015-I. The largest effect is seen in Amsterdam. When QE is active the lowered interest rates increased quarterly growth in housing prices by 2.35% percentage points (9.7% YoY). For the larger municipalities this increase in house price growth equals on average 1.24% percentage point per quarter (53% of house price growth). These findings are robust over multiple sample periods. Furthermore, the decline in the mortgage interest rate is more severe when the purchasing amount increases.
At last, to evaluate whether house prices are still in line with their fundamental values a user cost model is used. This model is an extension of the conventional price-to-rent ratio and incorporates among others the mortgage rates to control for the effects of quantitative easing and the tax deductibility of interest rate. This user cost model is first introduced by Himmelberg, Mayer and Sinai in 2005. By simply comparing the observed price-to-rent ratio with the long-term average, the housing market does indeed show to be overvalued. However, when using the extended Himmelberg et al. (2005) model, the allowed price-to-rent ratio as estimated by the model is lower than the observed price-to-rent ratio. This suggest that when allowing for the current low interest rate environment, the relatively low property taxes in the Netherlands and the tax deductibility of mortgage interest payments, the market does not seem to be mispriced. Nonetheless a slight increase in interest rates or the revoking of the tax deductibility of interest payments could shift the market from being fair priced to a state over overpricing.
The remainder of this thesis is structured as follows: First existing literature on quantitative easing and its effect on housing prices as well as the literature on asset prices bubbles will be examined. In section 3 the data is specified. Thereafter the methodology to estimate the exact effects of QE on housing prices will be introduced, as well as a model to identify price misalignment. Section 5 describes
6 the results and robustness checks are introduced. This paper ends with a summary of the most important results including a discussion on the possibilities of further research.
2. Literature Review
In this section the existing literature will be examined. First quantitative easing will be described in the most general form. Thereafter the channels through which QE can affect housing prices are examined. In section 2.3 other determinants of housing prices are described. Section 2.4 outlines the relevant literature on asset prices bubbles and bubble detection.
2.1 Quantitative easing
Before the 2007-2008 global financial crisis, monetary policy makers mainly targeted inflation by adjusting the short-term interest rates. The three main rates that the European Central Bank could adjust are the interbank interest rates, the rate of the ECB deposit facility and the rate for main financing operations (ECB, 2018). However, the 2007-2008 global financial crisis led to the deepest recession since the great depression of 1929. With interest rate already at and below zero, conventional monetary policy was no longer enough to deal with the deflationary pressure (Joyce et al. 2012). In January 2009 the Federal Reserve was the first of the world’s largest central banks to directly buy assets from financial markets since the adoption of QE in 2001 by the Bank of Japan. In the years thereafter, quantitative easing was used as an instrument by central banks all over the world. On the 9th of March 2015 the European Central Bank first started to buy European sovereign bonds
and securities from European institutions and national agencies as part of their Public Sector Purchasing Program (Claeys et al. 2015). Initially the purchasing would consist of 60 billion euro per month lasting till September 2016. However, only one day after the start of the program, the ECB announced to further increase the amount of assets bought from 60bn to 80bn euros. On the 8th of
Table 1: This table presents the important announcement of the ECB regarding quantitative easing
Date Announcement Press release
04-09-2014 QE announced (buying of covered bonds and ABS only) Link QE 1
22-01-2015 Buying of government bonds announced Link QE 2
03-12-2015 Increased buying to 60bn per month Link QE 3
10-03-2016 Increased buying to 80bn per month Link QE 4
02-06-2016 Buying of corporate bonds announced Link QE 5
08-12-2016 Decreased buying to 60bn Link QE 6
09-03-2017 Decreased buying to 30bn from Jan 2018 onwards Link QE 7
7 December 2016 it was announced that the assets purchasing program would be extended till December 2017 but with a lowered amount of 60bn euro a month following from March 2017 (ECB, 2016). At the 9th of March 2017 the ECB announced to lower the amount of asset purchased to 30bn
euro onwards from January 2018. Currently, more ten years after the start of the financial crisis, the European Central Bank is still buying assets worth of 30billion a month and inflated its balance sheets to 4.5 trillion euro, an amount unprecedent in history. However, on June the 14th 2018 the ECB
announced to stop the LSAPP after December 2018. Table 1 summarizes the important announcements regarding quantitative easing from the ECB.
The way quantitative easing is used to target inflation is explained by the European Central Bank (2018) as follows: The European central bank buys bonds from banks. The increased prices of those assets create money in the banking system. A wide range of interest rates fall and lending becomes cheaper. Consumption and investments increase. This increase in consumption and investments lead to economic growth and the creation of jobs. As a consequence, prices start to rise and thus inflation will increase. Next to the interest rate channel described above, two other channels through which QE affects asset prices are described in the literature. These channels are the rebalancing of investment portfolios and the decrease in risk premia. All three channels will be discussed in more detail in section 2.2 on the effects of the LSAPP on real estate prices.
2.2. The effects of quantitative easing on real estate prices
The long-term effects of the European Central Banks’s asset program on real estate prices have not yet been investigated. However, existing literature on the effects of QE on other asset classes specify three channels through which the large-scale assets purchasing program (LSAPP) could affect real estate prices: Through the interest rate channel, due to the rebalancing of investment portfolios and through a decrease in liquidity premia.
Several studies found that quantitative easing has driven down market interest rates. De Santis 2017 concluded that the purchasing program has a sizeable negative impact on bond yields where most of the impact on yields took place before the actual start of the program. Their econometric model estimates that the decline on the average 10-year euro area government bonds amounts to 63 basis points up to October 2015. Altavilla et al. (2015) estimated the decline in yields on 10-year government bonds to be around 30-50 basis points. Findings from Claeys et al (2015) are consistent as they report a rapid decline in yields in anticipation of the APP mid-2014.
Furthermore, Lamoen Mattheussens & Droës (2017) investigated bond pricing behavior after the announcement of QE. They used both a BSAF and GSDAF test to evaluate bond prices in 10 Euro area countries. They find evidence of exuberant bond pricing behavior mostly between the announcement of QE in 2014 and the start of the actual purchasing program in March 2015. With a
8 confidence interval of at least 99% their results are highly significant. To look at the above findings from a market perspective: currently the Dutch government pays negative interest rates on bonds up to five-year maturity and mortgage interest rates are on a record low (Bloomberg, 2018; DNB, 2018). Given the consistent results in the literature on the effect of quantitative easing on interest rates, it is useful to evaluate how these interest rates could affect housing prices. Literature on house price determinants are coherent, studies show a negative correlation between real interest rates and housing prices (Harris, 1989; Reichert 1990; Tsatsaronis and Zhu, 2004; McQuinn & O’Reilly, 2008; Campbell et al. 2009). The reason for this is twofold. On the one hand a decline in interest rates is a direct decrease in borrowing costs. On the other hand, as stated by Joyce et al. (2011), lower bond yields increase the present value of future dividends, thus raising equity prices (considering the Dividend Discount Model). Since the Dividend Discount Model is also used to value real estate (Direct Capitalization) a lower discount rate could substantially increase the real estate valuations, which in turn increases its price. Considering the above and given that the large-scale assets purchasing program has driven market interest rates down, it is expected that LSAPP positively affects housing prices through the interest rate channel.
The second channel through which quantitative easing could affect housing prices is the rebalancing of investment portfolios (substitution effect). The portfolio balance channel reflects the direct impact on asset prices of investors rebalancing their portfolio in response to a central bank’s QE-related asset purchases (Joyce et al. 2011). Given the favorable risk-return characteristics of bonds, institutional investors hold a vast amount of their assets as bonds. However, because of the suppressed yields partly caused by the ECB’s LSAPP, investors might substitute part of their bond portfolio by other asset classes. Real estate’s risk-return characteristics are on average between that of bonds and stocks and could thus be a favorable substitute. Indeed Joyce et al. (2011) argue that the decrease in the relative expected returns of bonds could increase the demand for other long-term assets. Gupta et al. (2010) concurs and states that a large part of the variation in the 2005-2007 house price appreciation was explained by investments and the buying of second homes. On the contrary, Nguyen et al. (2017) did not find significant portfolio rebalancing effects by investors in the United States after the implementation of the Federal Reserve’s LSAPP.
In addition to the interest rate channel and the portfolio rebalance channel a third possible channel though which quantitative easing could affect asset prices is the risk channel. Liquidity is one of the biggest risks associated with real estate. According to Joyce et al. (2011) the presence of a significant buyer in the bond market could improve the functioning of the market and reduce the risk premium for illiquidity. When this decreased liquidity premia depresses bond yields even further it might increase the capital flow to other assets classes and thus decrease the liquidity premia in real estate assets as well. A decrease in risk premia could, all else equal, lead to an appreciation of real
9 estate price. Joyce et al. (2011) expect this effect to be temporary and limited to the duration of the LSAPP.
Another implication of the LSAPP is the effect on exchange rates. Georgiades and Gräb (2016) investigate a sample of 39 Euro Area trading partners. Based on their dataset containing daily data points in the period 2007-2015 they found a significant depreciation in the euro against other currencies. Given that this would mainly influence the profitability of Euro based exporting companies and that literature on the effect of exchange rates on real estate prices isn’t consistent, these effects are not considered in this research.
2.3 Fundamental drivers of housing prices
In section 2.2 the channels through which quantitative easing could affect real estate prices are described. For a complete analysis of the effect of QE on housing prices it is needed to carefully map all other determinants that influence housing prices. The determinants of housing prices can be split into two main categories. One contains the micro determinants or the characteristics of the property itself. The second contains macro-economic factors.
2.3.1. Micro determinants of housing prices
Micro determinants are characteristics that directly influence the value of a property. Examples of such characterises are size, location, number of rooms, parking space and maintenance. To control for these characteristics a hedonic price index in estimated. Houben, Lamoen and Droës (2017) use a similar approach to control for observable housing characteristics in their paper on house price appreciation in Amsterdam the Netherlands. Section 3.1. describes the hedonic model to estimate a constant quality price index.
2.3.2 Macro-economic determinants of housing prices
Many studies have investigated the effects of interest rates and housing prices. Tsatsaronis and Zhu (2004) emphasize the strong link between declining interest rates and house price growth. They find that these effects are more severe in a time of lower inflation and specifically state that monetary policy should be wary of the possible misalignment between housing prices and their fundamental value during such a period. McQuinn and O’Reilly (2008); Harris (1989); Reichert (1990) all found negative correlations between interest rates and housing prices.
However, to correctly assess the effects of interest rates on housing prices, Campbell et al. (2009) argue that instead of using real interest rates, a housing premium over these interest rates should be considered. Sandor and Sosin (1975) examine the risk premium of mortgage rates. An important determinant of the risk premium is the Loan-Appraisal ratio (similar to a LTV ratio). Other
10 variables that have an impact on the risk premium are: the existence of secondary financing, the loan term, the quality of the neighbourhood and the quality of the house itself. A simple proxy that includes this risk premium would be mortgage interest rates.
Another important determinant of housing prices is the rent. According to Shiller (2005) the price-rent ratio can be an indicator of the relationship between the observed housing prices and their fundamental value. This mechanism is mostly equivalent to that of a price-earnings ratio in the stock market. Campbell et al. (2009) concur with these findings and state that explanations of housing prices that do not take into account rents can be misleading. According to campbell et al. (2009) the price of a property should be equal to the discounted future cashflows, also known as the Gordon-Growth model. Other studies that press the importance of rents in determining housing prices: Himmelberg et al. (2005); Engsted (2014); Ghysels et al. (2013).
The last of the four foremost fundamental determinants of housing prices is the real income. According to Holly and Jones (1997) and Tsatsaronis and Zhu (2004) income influences demand for housing over the long run. A widely used proxy for income is the GDP.
Next to the main factors of influence stated above, several other macro-economic variables do affect housing prices. Although these determinants are generally accepted in the literature, the econometric models are not always able to capture the full significance of all determinants. These factors of influence are: unemployment, shifts in demographics, construction, exchange rates, inflation, investments and taxes (Harris, 1989; Reichert, 1990; Holly and Jones, 1997; Jacobsen and Nuag, 2005; McQuinn & O’Reilly 2008; Campbell et al. 2009; Gupta et al 2010, Zhang et al. 2011). Given that both inflation and mortgage rates are found to be the most important factors in determining housing prices, Zhang et al. (2011) argue that loose monetary policy could be a good explanation of the recent housing boom in China.
In the Netherlands mortgage interest rate payments are tax deductible1. According to Dutch
Central Bank (2017), the tax deductibility of interest rate could lead to a higher overall mortgage debt and inflated housing prices. Therefore, when comparing Dutch housing prices to other countries it is useful to incorporate a correction for the presence of tax deductibility.
2.4 Bubbles
After the implementation of the European Central Bank’s largescale assets purchasing program, housing prices in the Netherlands have indeed been increasing rapidly. For example, year-on-year price increases of over 20% can be examined in Amsterdam (CBS, 2018). But when exactly can one speak of price misalignment? According to Case & Shiller (2003) the mere fact that there have
11 been rapid price increases is in itself no evidence of a bubble. Lamoen Matheussens and Droës (2017) add to that and argue that a distinction must be made between intrinsic and speculative bubbles.
A speculative bubble is mostly caused by mass psychology and extraneous events (Wu, 1997). During a speculative bubble current asset prices are sustained by the expectation of future price increases in that asset class. Scheinkman and Xiong (2003) investigate investor behavior during the 1998-2000 tech bubble. They found that agents paid stock prices that exceeded their own valuation because of the expectation to be able to find a future buyer who is willing pay even more. Case & Shiller (2003) concur and state that: “during a housing price bubble, home buyers think that a home that would normally be considered too expensive for them is now an acceptable purchase because they will be compensated by significant further price increases”. A speculative bubble is generally characterized by rapid price increases followed by a heavy correction.
An intrinsic bubble on the other hand is a bubble where the variation in prices are all derived from exogenous economic fundamentals (Froot & Obstfield, 1991). Wu (1997) finds that much of the variation in U.S. stock markets can be explained by intrinsic bubbles coming from changing fundamentals. Charemza & Deadman (1995) argue that intrinsic bubbles are of a less explosive nature and have a lower probability to burst. Quantitative easing is an exogenous shock which affects interest rates. Interest rates, among others, are fundamental determinants of housing prices. Therefore, it is more likely that QE causes an intrinsic bubble than a speculative one. Gürkaynak (2008) adds to the above discussion and states that both speculative bubbles and intrinsic bubbles can be rational.
A lot of research is done on the detection of asset price bubbles, most notably by Homm and Breitung (2012) and Gürkaynak (2008). Both studies investigate measures to test for bubbles in the stock market. Homm and Breitung (2012) mainly focus on speculative bubbles and find that the GSDAF test initially developed by Phillips et al. (2015) does well in detecting speculative bubbles. Gürkaynak (2008) focuses on rational bubbles which can both have a fundamental component or a speculative one. Since quantitative easing is most likely to cause an intrinsic price bubble it is most useful to look at the findings of Gürkaynak (2008). Gürkaynak’s (2008) findings show that a commonly used way to test for intrinsic bubbles in the stock market is the model used by (Froot & Obstfield, 1991). However, his general conclusion is that all the investigated models have a hard time distinguishing between misspecified fundamentals and actual bubbles. For this reason, it is better to look at a way to specifically identify bubbles in real estate.
An example of such a study that focuses on an asset price bubble in real estate is Himmelberg (2005). They use a user cost model to investigate price misalignment in the US housing market. One of the advantages of the Himmelberg et al. (2005) model is that it allows for deviations in interest rates and incorporates the tax deductibility of interest rates. According to Pavlidis et al. (2014) the biggest issue with models based on price-rent ratios like the one in Himmelberg (2005) is the unavailability of
12 rental data. Given that data on rents is available in the period of interest a user cost model can be used. The exact model is specified in section 5.1.
3. Data
In this section the data and data sources will be described. Paragraph 3.1. outlines the sources of the data. Thereafter the descriptive statistics for the hedonic constant quality price index are
presented. Finally, the macro-economic variables are described.
3.1 Data Sources
The data on housing prices and their respective characteristics are provided by the NVM which is the Dutch Association of Realtors. Their database consists of over 70% of all transaction occurring in the Netherlands and their data is used in a wide variety of economic research. Data on macro-economic variables and quantitative easing are collected from the CBS, DND and ECB. The Central Bureau of Statistics is the national statistics agency of the Netherlands. The DNB and ECB are the Dutch Central bank and European Central Bank respectively. The Dutch tax authority provides the data on marginal tax rates and the investment figures are collected from existing literature and Real Capital Analytics. Table 2 sums the data sources per category.
Table 2: Data sources
Source Variables
Dutch Association of Realtors (NVM) Housing prices and characteristics
CBS/DNB/ECB Macro-economic variables including rents
Dutch National Tax Authority Taxation rates
Bloomberg Risk-free interest rates
RCA Investment volumes Amsterdam
3.2 Constant quality hedonic price index
To prepare the data for the IV-regression some modifications need to be made. At first, to control for changing house characteristics a hedonic constant quality price index is created. The index is constructed from NVM data using the following model:
𝐿𝑛(𝑃𝑟𝑖𝑐𝑒
𝑖,𝑡) = 𝛽
0+ 𝛽
1𝑆𝑖𝑧𝑒
𝑖,𝑡+ 𝛽
2𝑅𝑜𝑜𝑚𝑠
𝑖,𝑡+ 𝛽
3𝐶𝑜𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑖𝑜𝑛𝑦𝑒𝑎𝑟
𝑖,𝑡+
13 Where β1 – β7 represent the housing characteristics and τt represents the index. The index consists of
quarterly time dummies for the period 2001-1 till 2017-IV. Where 2000-I = 100. This index is used as the dependent variable in the IV-regression and the four-step Baron and Kenny (1986) model. These models will be described in section 4.1.
Table 3 describes the municipalities that are in the price index. These are the largest municipalities for every province in the Netherlands. As expected the municipality with the highest average housing prices is Amsterdam. Rotterdam has the highest normalized standard deviation (0.712). In total 446,085 transactions over a period of 72 quarters are incorporated in the index. Initially the dataset contained slightly more than 500.000 observations. However, the dataset needed to be cleaned from incomplete and flawed observations. An example of such a deleted observation is a 15m2
Table 3: Transaction prices of the data collected from the Dutch Association of Realtors (NVM).
Municipality Province N Mean Std. Dev.
Almere Flevoland 33,165 200,279 85,353 Amsterdam Noord-Holland 119,165 300,425 208,139 Eindhoven Noord-Brabant 32,398 224,489 118,611 Emmen Drenthe 15,230 169,229 76,710 Enschede Overijssel 20,275 173,616 96,591 Groningen Groningen 39,334 171,596 89,300 Leeuwarden Friesland 17,379 150,034 77,752 Maastricht Limburg 7,135 226,890 121,230 Nijmegen Gelderland 24,673 218,182 111,114 Rotterdam Zuid-Holland 67,294 198,259 141,234 Terneuzen Zeeland 6,071 153,970 84,056 Utrecht Utrecht 63,966 244,099 63,966 Total 446,085 202,589 153,335
appartment with 73 rooms and multiple parking facilities. Furhtermore all observation that either had a size of 0m2 or 0 rooms are omitted from the final dataset.
In the Table 4 the descriptive statistics for the hedonic constant quality price index can be observed. The variables are separated into two different types, ratio and ordinal. The average trans action price of the combined muncipalities equals 202,589.-. The lowest single transaction is a 45,000 euro appartment. The maximum price of 6,195,000 is paid for a large monumental property in the city centre of Amsterdam. The ordinal variable of maintenance is constructed using the mean of the outside and inside mainentance of the respective property and is ranked on scale from 1 to 9.
14 Table 4: Descriptive statistics for the hedonic price index are described.
N=446,085
Ratio
Variable Mean Std. dev. Min Max
Transaction price 202589 153335 450000 6195000
LnPrice 12.20 0.50 10.71 15.64
Size (m2) 105 41 35 536
Number of rooms 5 1.36 2 125
Ordinal
Construction year Ranked 0-9 where 0 is oldest
City centre Ranked 0-3 where 0 is furthest from centre Maintenance Ranked 1-9 where 1 is worst level of maintenance
Parking Ranked 0-8 where 0 is no parking and 8 means multiple parking spots Monument 0 = no ; 1 = yes
Now that the input variables have been discussed it is useful to look at the difference between the estimated constant quality hedonic price index (CQPI) and the countrywide CBS price index (CBSI). Figure 1 plots both indices against time. The first thing to note is the different base year of the indices. The hedonic price index has 2000-I as the indexation date whereas the CBS index takes 2004-1 as the base. Up until the end of 2013 the hedonic index and CBS index show the same pattern. However, the
15 larger municipalities have a slightly larger volatility than the countrywide CBS index. After the implementation of QE, which is marked by the black vertical line, a more rapid increase in the CQPI can be observed. This suggests that the more urban areas as represented by the CSQPI index are affected more by the LSAPP. The hedonic price index varies from 100 in its origination quarter of 2000-I and almost doubles to 192.01 in the 2017-2000-IV quarter. This growth is significantly higher than the 53% growth of the CBS index over the same period.
3.3 Macro-economic variables
Table 5 presents the descriptive statistics of the macro-economic variable’s raw input. The table separates between dependent, independent and dummy variables. The estimated constant quality hedonic price index is modelled as the dependent variable. It is important to note that GDP per capita is the quarterly GDP divided by the population in that quarter. The annual income per capita ranges from somewhat over 28,000 in the year 2000 to almost 43,000 in 2017. Unemployment was highest in 2014-I at 8.1%, thereafter the unemployment rate declined rapidly as a consequence of the recovering economy. The unemployment rate in 2017-IV equals 4.3%. The inflation growth is the QoQ growth of inflation. For the analysis these growth figures are turned into an index with base 2000-I as 100.
Table 5: Descriptive statistics of the macro-economic variables for the TSLS IV-regression
N=72 Dependent
Variable Mean Std. dev. Min Max
Hedonic price index 141.95 18.95 100 192.01
Independent
Variable Mean Std. dev. Min Max
Mortgage rates 4.73 0.60 3.33 6.21
Nominal interest rates 3.09 1.58 0.05 5.60
Real interest rates 1.21 1.30 -1.23 3.63
GDP per capita 9075 1070 6828 11183 Inflation growth 0.45 0.93 -1.30 2.70 Unemployment 5.27 1.13 3.40 8.10 Rent Index 127 17 100 158 Dummy Variable Description
Quantitative easing 1 = quantitative easing in quarter; 0 = otherwise
Important differences can be observed between the different interest rates. Mortgage interest rates contain a risk premium over the risk-free interest rate and are naturally highest. Figure 2 plots
16 the three interest rates against time. All three interest rates seem to respond to quantitative easing. A small but steep rise in the risk-free interest rates was followed by a rapid decrease in when there first was speculation on quantitative easing in the beginning of 2014. The declining trend continuous after the actual announcement in 2014-III which is marked by the vertical line. In the period between 2014
Figure 2: Comparing the nominal, real and mortgage interest rates
and 2017 nominal and real interest rates converge when inflation is near zero. Furthermore, as opposed to the expectation of Joyce et al. (2011) no significant decline in the risk premium of mortgage rates over the risk-free rate can be observed after the implementation of QE.
4. Methodology
This section describes the methodology used. Paragraph 4.1. presents the model for estimating the effects of quantitative easing on housing prices using the TSLS IV-regression. Section 4.2. outlines the Himmelberg et al. (2005) user cost model for bubble detection.
4.1. Measuring the effect of quantitative easing
The simplest method to estimate the effect of quantitative easing would be to regress the quantitative easing variable on the estimated constant quality hedonic price index using multivariate OLS, where Wi are the macro-economic control variables:
𝐼𝑛𝑑𝑒𝑥𝑡 = 𝛽0+ 𝛽1𝑄𝐸𝑡 + 𝜕1𝑊1,𝑡+ ⋯ + 𝜕𝑗𝑊𝑗,𝑡+ 𝜀𝑡 (2)
However, some issues arise from doing this. At first, given that the data are time series, autocorrelation might be present in several variables. Autocorrelation could lead to insignificant results and/or
17 incorrect significant results. To test for serial correlation a Durbin-Watson test (1951) for first order autocorrelation and a Breusch-Godfrey test (1978) for higher-order serial correlation are conducted. Both are significant at the 1% level with test scores of 0.768 (DW) and 26.216 (BG) respectively. Using the above model would thus lead to incorrect results. One way to deal with autocorrelation is taking the first differences. This is shown in Formula (3). After the modification the Durbin-Watson (1.638) and Breusch-Godfrey (2.361) tests for autocorrelation do no longer yield significant results. Appendix Table A.1. summarizes these findings.
(𝐼𝑛𝑑𝑒𝑥𝑡− 𝐼𝑛𝑑𝑒𝑥𝑡−1) = 𝛽0+ 𝛽1𝑄𝐸𝑡+ 𝜕1(𝑊1,𝑡− 𝑊1,𝑡−1) + ⋯ + 𝜕𝑗∆𝑊𝑗,𝑡+ 𝜀𝑡 (3)
Secondly, it is important to look at the trends within the variables. One way to do this is by looking at the so-called unit root. This can be tested by regressing the first lag of the variable on the current value (AR1). If the estimated coefficient equals 1 then a stochastic trend is present. This test is done for all the economic variables including the estimated hedonic price index. All the macro-economic variables do have a unit root expect for QE and inflation growth (the inflation index does have a unit root). According to Bun (2017) a way to eliminate stochastic trends is to take the first differences. If the variables are cointegrated another way of dealing with unit roots is to use a vector error correction model. However, given the fact we need to do an IV-regression, using an VECM will needlessly complicate the analysis. To build upon the previous equation we will we add logs for interpretation purposes.
𝐿𝑛(𝐼𝑛𝑑𝑒𝑥𝑡) − 𝐿𝑛(𝐼𝑛𝑑𝑒𝑥𝑡−1) = 𝛽0+ 𝛽1𝑄𝐸𝑡+ 𝜕1∆𝐿𝑛(𝑊1,𝑡) + ⋯ + 𝜕𝑗∆𝐿𝑛 (𝑊𝑗,𝑡) + 𝜀𝑡 (4)
Continuing from equation (4), quantitative easing is expected to not directly affect housing prices but is expected to do so through the interest rate channel. A good method to estimate indirect effects is using instrumental variable two-staged least squares regression. To test whether mortgage rates are endogenous a Durbin-Hausman-Wu test is conducted. This is done by regressing the residuals of the by QE estimated mortgage rates on the hedonic prices index. The test score is significant at the 5% level (P<0.0131).
TSLS consists of two steps. In the first step the instrument (QE) is regressed on the endogenous variable mortgage rates:
𝐿𝑛(𝑀𝑅𝑡) − 𝐿𝑛(𝑀𝑅𝑡−1) = 𝛾0+ 𝛾1𝑄𝐸𝑡+ 𝛾2∆𝑙 𝑛(𝐺𝐷𝑃𝑡) + 𝛾3∆𝑙 𝑛(𝑢𝑛𝑒𝑚𝑝𝑡) +
18 In the second step the predicted mortgage rates are regressed on the hedonic price index:
∆𝐿𝑛(𝐼𝑛𝑑𝑒𝑥𝑡) = 𝛽0+ 𝛽1∆𝑀𝑅̂ + 𝛽𝑡 2∆𝑙 𝑛(𝐺𝐷𝑃𝑡) + 𝛽3∆𝑙 𝑛(𝑢𝑛𝑒𝑚𝑝𝑡) +
𝛽4∆𝐿𝑛(𝑖𝑛𝑓𝑙. 𝑖𝑛𝑑𝑒𝑥𝑡) + 𝛽4∆𝑙𝑛(𝑟𝑒𝑛𝑡𝑡) + 𝜀𝑡 (6)
In both steps macro-economic control variables are added. Note that the logged first differences are used for variables that have autocorrelation and unit roots. The outcomes of the TSLS regression are presented in the results section 5.
At last, it is useful to check whether the effect of quantitative easing is a full indirect or a partial indirect effect where the other part of QE directly affects housing prices. The most well-known method to estimate indirect effects is established by Barron and Kenny in their 1986 research on mediation effects in social psychological research. They propose a four-step model consisting of several regression analyses to filter the indirect effects of X on Y. This four-step model is widely used in existing literature. According to Zhao et al. (2010), by 2009 more than 12,688 studies cited the Baron and Kenny model. The model consists of four steps:
Table 6: The four-step Barron and Kenny (1986) model.
Method Visual2
Step 1 𝑌𝑡= 𝛽0+ 𝛽1𝑋𝑡+ 𝜀𝑡
Step 2 𝑀𝑡= 𝛽0+ 𝛽1𝑋𝑡+ 𝜀𝑡
Step 3 𝑌𝑡= 𝛽0+ 𝛽1𝑀𝑡+ 𝜀𝑡
Step 4 𝑌𝑡= 𝛽0+ 𝛽1𝑋𝑡+ 𝛽2𝑀𝑡+ 𝜀𝑡
Mediation is present when step 1-3 yield significant results. Step 4 tests whether the effect is an effect of full-mediation (completely indirect) or partial-mediation. Table 6 shows the regression output of the four-steps. Indeed step 1 to 3 have significant coefficients which means that there is a mediation effect. In step 4 the effect of mortgage rates is significant whereas the effect of QE is no longer significant when controlling for mortgage rates. According to Barron and Kenny (1986) this means that there is full-mediation in which the effect of quantitative easing is a complete indirect effect through the interest rate channel.
19
Table 7: The four-step model regression output.
y Index Mortgage rates Index Index
step (1) (2) (3) (4) QE 30.31*** -1.15*** - 0.40 (4.83) (0.13) - (4.81) Mortgage rates - - -26.28*** -26.09*** - - (2.08) (3.00) Constant 136.90*** 4.92*** 266.22*** 265.31*** (1.97) (0.05) (9.94) (14.86) R-squared 0.36 0.51 0.69 0.69 N = 72 ***p<0.01, ** p<0.05, * p<0.1
The purpose of this regression is solely to examine whether the effect is a full indirect effect or partial effect not the quantify the effects of QE on housing prices. This will be done in the IV-TSLS regression in section 5. The coefficients of step 1 to 3 of the model yield significant results. This mean that there is a mediation effect. Furthermore, in the step 4 regression the effect of QE is no longer present when controlling for interest rates. Therefore, QE is a full indirect effect mediated through mortgage interest rates.
4.3. Robustness checks
To investigate the consistency of the results, different robustness checks will be carried out. To test the robustness of the estimated hedonic index and see differences between the larger municipalities, the country as a whole and Amsterdam, three different analysis will be done. One with the hedonic price index, one with CBS index and the last with a hedonic index of Amsterdam only. Furthermore, different time periods of both quantitative easing and control period are taken to see the robustness of the results. At last, we look at whether including Amsterdam population growth and investment figures to the analysis would alter the findings. In the next section, first the results from the IV-regression will be discussed. Thereafter the robustness checks are presented.
5. Results
Table 7 summarizes the results of the TSLS IV regression using equations (5) and (6) from the methodology section. The significant effects of both QE on mortgage rates and mortgage rates on housing prices in the first two regressions indicate that QE does affect housing prices. More precise: QE has negative effect on mortgage interest rates of around 6 basis points. Moreover a 1 percentage point decrease in mortgage interest rates has a positive effect of 18 percent in house price growth of the constant quality price index. Combining both effects, for the municipality price index QE has a positive effect on house price growth of -0.064*-18.4 = 1.17 percentage point in a quarter where QE is active.
20 Table 8: Results of the TSLS IV-regression.
N = 71 Expected
sign
(1) (2) (3)
Step 2 ΔLn(Municipalities) ΔLn(Amsterdam) ΔLn(Netherlands)
ΔMortgage rates predicted - -0.184** -0.371*** -0.053¥
(0.084) (0.125) (0.032) ΔLn(GDP per Capita) + 0.331*** 0.227 0.045 (0.124) (0.217) (0.046) ΔLn(Unemployment) - -0.095 -0.019 -0.018 (0.196) (0.137) (0.023) ΔLn(Inflation index) + 0.794 0.523 0.345 (0.630) (1.212) (0.211) ΔLn(Rent index) + -0.003 -0.019 -0.005 (0.009) (0.016) (0.003) Constant O 0.000 -0.009 -0.218
Step 1 Δ(Mortgage rates) Δ(Mortgage rates) Δ(Mortgage rates)
QE - -0.064*** -0.064*** -0.064***
(0.018) (0.018) (0.018)
Constant O -0.053* -0.053* -0.053*
Macro-economic controls Yes Yes Yes
***p<0.01, ** p<0.05, * p<0.1
In step 1 the quantitative easing dummy is regressed on the logged first differences of the mortgage rate variable. In the second step three different dependent variables are used: The hedonic constant quality price index for the larger municipalities, a hedonic index for Amsterdam only and the CBS index for the whole of the Netherlands. ¥ weakly significant (p<0.104). Because of readability purposes the exact sign and significance of the macro-economic controls in step 1 are left out of the table.
To make these findings more intuitive: If QE is active in a certain quarter the house price growth for that quarter is increased by 1.17 percentage point on average due to the depressed interest rates. In 2017-IV housing prices in the hedonic index increased by 3.9% that means that according to
Table 9: The effects of quantitative easing on quarterly house price growth
2017-IV Average 2015-I till 2017-IV
Growth QE growth (%) attributable to QE Growth QE growth (%) attributable to QE
Netherlands 1.39% 0.34% 24% 1.39% 0.34% 25% Municipalities 3.90% 1.17% 29% 2.12% 1.17% 55% Amsterdam 4.43% 2.35% 57% 3.20% 2.35% 73%
21 this model set up almost, one third of that growth can be attributed to lowered mortgage interest rates caused by quantitative easing. As expected this effect is more severe for Amsterdam where QE leads to an additional 2.4%-point growth. Table 9 summarizes these findings.
In Figure 3 the QE purchasing amount is plotted against the impact of quantitative easing on the mortgage rates. The general picture shows an increasing impact of QE when the purchasing amounts are higher. In the first half of 2015 the ECB purchased assets worth a total of 240bn euro from the financial markets. This amount was highest in the first half of 2016 with a total purchasing amount
Figure 3: The impact of QE on mortgage rates vs. the purchasing amount
of 480bn euro. In the first half year of 2017 the average mortgage interest was 3.51 percent. A decrease of 10 bps translates in a decrease in mortgage interest rates of almost 3%. This effect is lower than is found by Altavilla et al. (2015) who estimate the impact of QE on bond prices to be around 30-50 bps.
At last, robustness checks are carried and possible alternative explanations for the significant results are given. First, the findings over different sample periods are examined. Table 10 states the
Table 10: The effects of quantitative easing on the hedonic index (CQPI) using different sample periods
Period Step 1 Step 2 Combined
2000-I till 2017-IV -0.064*** -0.184** 1.17%
2004-I till 2017-IV -0.069*** -0.187** 1.28%
2008-I till 2017-IV -0.075*** -0.193** 1.43%
2010-I till 2017-IV -0.067*** -0.220** 1.46%
22 quarterly effects of quantitative easing on the estimated hedonic price index of the larger municipalities using different sample periods. The impact ranges between 1.17 percentage point and 1.46 percentage point starting form 2010. The effects of quantitative on mortgage rates differ by at most 1 bps when taking shorter run-up periods. All findings are significant at either the 0.01 or 0.05 level. Given that there are no large deviations in both magnitude and significance the effects are robust over different sample periods.
In the above model controls are added for both changing micro variables, by the usage of the estimated constant quality price index, and for changes in macro-economic variables in the IV-regression. However, due to lack of data over the entire sample period some variables that could affect housing prices are omitted from the model. This could lead to an overstated effect of quantitative easing on housing price. Especially the effects for Amsterdam seem to be rather large where the model explains 73% of the growth in housing prices from the year 2015 by the existence of quantitative easing.
One reason for the differences in magnitude between the countrywide index and Amsterdam could be due to the rebalancing of investment portfolio effect. As explained in the literature section the lower interest rate might push investors from bonds towards real estate. The municipality of Amsterdam has the highest investment volumes both from national as foreign investors. The purchase of investment properties may have inflated prices more than the country average. Due to lack of free available data on investment volumes, investments volumes have not been added as a control variable. However, for the period 2008 till 2017 investment volumes are available solely for Amsterdam. A separate regression is done to investigate the impact of investment volumes on the model outcomes. This regression output can be found in Appendix Table A.3. Including investment volumes as a variable indeed decreases the impact of mortgage rates on housing prices by 2.5 percentage points and is a significant explainer of housing prices.
Furthermore, according to the DNB (2017) cities are becoming more popular. This increase in popularity is especially high amongst young and well-educated individuals. In the 2002 and 2017 the population of Amsterdam has increased by 16.45% for the city of Amsterdam and 15.59% of the municipality of Amsterdam (CBS Statline, 2018). In the initial model both income and population were added as control variables. However, due to large correlation between these two variables (>0.85) the model suffered from multicollinearity. This has led to the incorporation of GDP per capita variable instead of two separate variables. In order to see whether the incorporation of population growth does alter the findings of the analysis another regression is carried out. The results can be found in appendix Table A.5. Although some slight changes in the main explanatory variables do occur (-0.41 vs. -0.37 for mortgage interest rates and -0.064 vs -0.053 for the QE variable) they do not alter the overall conclusion of the results. The best way to control for the increasing popularity of the city amongst
23 higher educated and higher income individuals would have been to add a separate income per capita variable for Amsterdam. Unfortunately, no reliable source for Amsterdam income is available over the sample period.
A third explanation could be in the supply shortage in the housing market. During the financial crisis construction of new housing partly ceased (CBS, 2018). The lowered construction and the increasing popularity of the cities has increased the supply shortage (DNB, 2017). Although this supply shortage has increased after the construction delays during the financial crisis the shortage has been structural and was present before the start of quantitative easing (Rabo Research, 2018). The actual supply shortage thus might not be reason for the increase after the start of quantitative easing. However, the delay of home purchasing during the financial crisis might be the reason why the supply shortage becomes more apparent now than before the start of quantitative easing. Further research on this effect could test the robustness of the stated findings.
At last Houben et al. (2017) and the DNB (2017) mention several other factors that could influence housing prices especially in Amsterdam. These factors have not been controlled for in the stated analysis and could thus partly be captured by the QE dummy variable. Those include the presence of AirBnb, the large social housing sector in Amsterdam and increasing popularity of Amsterdam after the Brexit.
To sum up the findings. Quantitative easing has a significant positive effect on housing prices. The effect is larger for more urban areas. The impact of quantitative easing increases when the purchasing amount is higher. The findings are robust over different sample periods. However, when comparing the findings to existing literature, the results seem to slightly overstate the impact of mortgage rates on housing prices and understate the impact of quantitative easing on the mortgage interest rate. Reasons for this could be the lack of control for both investment volumes, increasing popularity of cities and supply shortages, as well as the omittance of controls for AirBnb, the social housing sector and the increasing popularity of Amsterdam after the Brexit.
5.1. Price misalignment in the housing market
To check if a misalignment between the fundamental value of housing and their observed values is present, a user cost model as described in Himmelberg et al. (2005) is used. This model equates the price-to-rent ratio with the inverse of the user cost of housing:
𝑃𝑡 𝑅𝑡
=
1
𝑢𝑡 (7)
24
𝑢
𝑡= 𝑟
𝑡 𝑟𝑓+ 𝑤
𝑡− 𝜏
𝑡(𝑟
𝑡𝑚) + 𝛿
𝑡− 𝑔
𝑡+1+ 𝛾
𝑡+ ∀
𝑡(8)
The risk-free rate 𝑟𝑡𝑟𝑓represents the opportunity cost of capital. When a home-owner would not have bought the house, he could have invested this money risk-free at this rate. 𝑤𝑡 equates to the property tax rate. To account for the deductibility of mortgage interest payments the marginal tax rate is multiplied by the mortgage rate 𝜏𝑡(𝑟𝑡𝑚). Note that in the original model the property taxes are also tax-deductible, in the Netherlands this is not possible so are taken out of the deducting equation. 𝛿𝑡 reflexes the maintenance costs of the property and depreciation on the property whereas 𝑔𝑡+1 represents the long-term real appreciation rate of housing prices (net of inflation). At last a risk premium 𝛾𝑡 is added to account for the riskiness of buying as opposed to renting a property.
The advantage of this model over a general comparison of the price-to-rent ratio is that it allows for fluctuations in interest rates. Himmelberg et al. 2005 describe the effects of interest rates in the model as follows: If, based on the input variables (in their example: risk-free rate 4.5%; mortgage rates 5.5%; depreciation 2.5%; taxes 25%; prop. tax. 1.5%; risk premium 1.8%), the user costs are equal to 5%. Then a prospect buyer should pay a maximum of 20 times the rent for a property. If interest rates where to decline to 4% and 5%, the new user costs equal 4.6% and a prospect buyer should now pay a maximum of 21.9 times the rent to acquire a property. Considering interest rates in this way would give good insight in whether current prices are in line with the fundamental costs of living taking into account the lower interest rates caused by QE.
Table 12 describes the input variable for both the Netherlands and Amsterdam. The risk-free Table 11: Input for the Himmelberg et al. (2005) user cost model
Name Character Netherlands Amsterdam
Risk-free rate
𝑟
𝑡
𝑟𝑓 0.70% 0.70%
Marginal tax rate
𝜏
𝑡 40.85% 40.85%Property taxes
𝑤
𝑡 0.14% 0.05% Mortgage rate𝑟
𝑡𝑚 3.33% 3.33% Real appreciation𝑔
𝑡+1 0.49% 1.14% Depreciation𝛿
𝑡 2.50% 2.50% Risk premium𝛾
𝑡 3.00% 3.00% Land lease ∀𝑡 0.00% 0.48%rate is set equal to the 10-year government bond yields in the Netherlands and is collected from Bloomberg (2018). The marginal tax rates are retrieved from the Dutch Tax Authority whereas the the property tax rates are collected from the municipality. The real long-term appreciation comes from
25 the appreciation rates after WW-II collected by Eichholtz (1997) complemented by the NVM database and CBS Statline (2018) netted for long-term inflation of 2%. Furthermore, Harding et al. (2006) estimate the long-run depreciation rate of housing to be equal to 2.5%. Existing literature does is not consistent regarding the risk premium of owning a house over renting. Estimates range from a negative risk premium Sinai and Souleles (2005) to 2% Himmelberg et al. (2005) and 3% Diamond (1980). The recent declines in housing prices (2009-2013) have shown that homeownership is not riskless. Therefore the 3% as stated by Diamond is maintained.
For Amsterdam a separate land lease cost needs to be added. In 1896 the municipality of Amsterdam introduced ground leases. These ground leases where introduced to facilitate affordable housing, allocate the increasing land value to the community instead of individual landlords and to control the usage of land (Ploeger and Bounjoh, 2017). Amsterdam still owns the majority of the land and homeowners have to pay ground leases to the municipality. Three different land lease contracts can be agreed upon. A variable contract which has a 3% land lease costs that are indexed on a yearly basis. A ten-year contract with a fixed percentage of 2.90% and a 25-year fixed contract with a cannon percentage of 3.25% (Municipality of Amsterdam, 2018). The average land lease equals about 3% of the land value. To calculate the cannon payment, one should multiple the cannon percentage with the land value:
𝐿𝑎𝑛𝑑 𝑙𝑒𝑎𝑠𝑒 𝑝𝑎𝑦𝑚𝑒𝑛𝑡𝑡 = 𝐶𝑡∗ 𝐿𝑉𝑡 (1)
However, land lease payments are tax deductible. Therefor the land lease payment should be multiplied by one minus the marginal tax rate.
𝑁𝑒𝑡 𝑙𝑎𝑛𝑑 𝑙𝑒𝑎𝑠𝑒 𝑝𝑎𝑦𝑚𝑒𝑛𝑡𝑡 = (1 − 𝜏𝑡)(𝐶𝑡∗ 𝐿𝑉𝑡) (2)
At last we need this cost as a percentage cost of the total value of the property to match the other costs in the user cost model.
𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑙𝑎𝑛𝑑 𝑙𝑒𝑎𝑠𝑒 𝑝𝑎𝑦𝑚𝑒𝑛𝑡𝑡 = (
∀
𝑡) = (1 − 𝜏𝑡)(𝐶𝑡∗ 𝑃𝑡∗ %𝐿𝑉𝑡)/𝑃𝑡 (3)Where P is the average price of a residential property in Amsterdam. %LVt equals the average
percentage of land value of Amsterdam housing properties. In Table 3 the input variables are presented.
26 Table 12. Land lease inputs
Name Character Amount Source
Land lease percentage C 3.00% Municipality of Amsterdam
Percentage land value of property value %LV 27.00% Municipality of Amsterdam BSQ
Marginal tax rate τ 40.85% Dutch Tax Authority
Average price of property P 409.000 NVM data
To calculate the average percentage of land value the “buurtstraatqoute” (BSQ) from the municipality of Amsterdam is used. This is the same measure that the municipality uses to calculate the land lease payments (Municipality of Amsterdam, 2018). The BSQ ranges from 5% in the outskirt of Amsterdam to 49% within the city center. No exact figures on the amount per percentage is available. Therefore, as an approximation the simple average is taken which equals 27%. Using these input variables to calculate the percentage of land lease using equation (4) results in a cost of 0.48%.
Table 13 presents the outcomes of the model. Given the current low interest rate environment the model allows for a price to rent ratio of 21.74 in the Netherlands and 23.98 in Amsterdam. These ratios are higher than those currently observed in the market. The rents are collected from Pararius the largest rental broker of the Netherlands. It has to be carefully noted that these are private sector rents and therefore the observed price-to-rent ratio is slightly deflated. Based on these data the model estimates that current valuations are justified considering the low interest rate environment. If, however we compare the current price-rent ratio to the long-term average price-rent ratio for Amsterdam over the period 1650-2001 constructed by Ambrosa, Eichholtz and Lindenthal (2013)
Table 13: User costs and price-to-rent ratios according to the user cost model
Netherlands Amsterdam
User cost 4.60% 4.17%
Allowed price-to-rent ratio user cost model 21.74 23.98
Observed price-to-rent ratio 17.98 21.68
Allowed historic average price-to-rent ratio - 16.67
Overvaluation (user cost model) -17.3% -9.6%
Overvaluation (LT average) - 30.1%
current price to rent ratios and thus current valuations are high. Yet this long-term average does not allow for deviations in user financing as does the Himmelberg (2005) model.
One reason for the relatively high allowed price-rent ratio is the possibility to deduct mortgage interest payments. The Dutch Central Bank is in favor of revoking the tax deductible of interest payments. If the government would follow this advice the allowed price to rent ratio as estimated by
27 the model would drop to 16.78 for the Netherlands and 18.08 for Amsterdam. These ratios are lower than the ratios currently observed. Another important note is that a rise in interest rates could shift the market to being overpriced. A 1%-point increase in interest rates holding other factors constant would decrease the allowed price-to-rent ratio below the ratios currently observed.
To summarize the findings. When allowing for deviations in interest rates and the tax deductibility of mortgage interest payments using the modified Himmelberg (2005) user cost model, the model does not show evidence that the housing market is overpriced. However, the market seems to be overvalued considering the simple long-run perspective collected from Amrosa, Eichholtz and Lindenthal (2013). However,
6. Conclusion and discussion
The purpose of this research has been to investigate the effect of quantitative easing on the Dutch housing market. In March 2015 the European Central bank first started to directly buy assets from the financial markets. The aim of the program is to enhance economic growth and combat deflationary pressure. Given that quantitative easing is the most used unconventional tool for central banks all over the world and the ECB currently is still buying assets worth of 30bn euro each month. It is useful the map all externalities of this program. Therefore, the following question is asked:
“To what extent does quantitative easing cause price misalignment in the housing market? “
To answer the research question, the paper built upon two pillars.
At first, the effect of quantitative easing on the Dutch housing market was investigated. This was done by using instrumental variable regression. The controls for this regression where separated into housing characteristics which where estimated by a hedonic price index and macro-economic variables that influence housing prices. Quantitative easing was expected to influence housing prices by a depressing mortgage interest rates and thus functioned as the instrument for mortgage rates. The results show a significant effect of the large-scale asset purchasing program on housing prices through the interest rate channel. The fact that a significant impact of QE on mortgage rates is found supports and expands on the existing literature where significant impact of quantitative easing on bond yields have been observed. However, when comparing the results, the model that is used seems to slightly overstate the effect of mortgage rates on the housing market and seems to underestimate the effect of QE on mortgage interest rates. Therefore, it is most useful to interpret the findings in significance and order of magnitude instead of the actual magnitude itself. These effects can be described as follows: Quantitative easing depresses mortgage interest rates and thus indirectly increases house
28 price growth. The effect is more severe for urban areas. The effect increases when the purchasing amount is higher. The findings are robust over multiple sample periods.
Secondly, it is investigated whether the supplementary price growth leads to price misalignment within the housing market. When using simple price-to-rent ratio comparison the current market indeed implies overvaluation. However, when using a Himmelberg et al. (2005) user cost model, which controls for lower interest rates and the tax deductibility of interest rates, the outcomes of the model do not imply overvaluation. This does not mean that no price correction can occur. Two important factors that influence the outcomes of the model are interest rates and the tax deductibility of mortgage interest rate payments. Either a one percentage point increase in mortgage interest rates or the inability to deduct interest payments from income taxes do shift the market from being fair priced to being overpriced holding everything else constant.
Thirdly, it is useful to look at the implication of the findings. The most important implication is to create awareness of the externalities of quantitative easing. According to the model quantitative easing does have a significant impact on housing prices. Policymakers of the European Central Bank should consider whether the externality is one that can be accepted. Another way to look at the findings is from a viewpoint of inflation. The main goal of the LSAP is to create inflation, increasing house prices are in itself a form of inflation. This can be interpreted that the program is successful in the Dutch housing market. One thing to note is that the European Central Bank is an independent institution. In this sense it is hard for other policymakers to use the outcomes of this research in a way to alter the LSAP. However, it is possible for policymaker to make changes in the housing market now that housing prices are supported by quantitative easing. One such a measure could be to restrict the tax deductibility of interest rates more rapidly. As stated throughout this research the Dutch Central Bank is in favor of restricting the tax deductibility mortgage interest rate payments. In their view the current possibilities are too generous and increase the cyclicality of the market due to higher amounts of debt (DNB, 2018). According to the user cost model a reduction in the tax ductility would shift the market to a state of overpricing and could thus take some pressure of the market if a price correction would follow.
At last we look at the shortcomings of the analysis and the possibilities for future research. One shortcoming in the analysis is the lack of data on investment volumes for the stated municipalities over the time frame of interest. As can be seen from appendix Table A.4 investment volumes do have a significant impact on housing prices in Amsterdam. In upcoming research on this topic, it is therefore interesting to examine the effects of quantitative easing when investment volumes can be added over the entire sample period and width. From the analysis we can conclude that different effects between the more urban areas and the county as whole are found. A good addition to this research would be to examine where these differences come from and look at these findings from a less aggregate level.
29 As well to expand the research to all the Eurozone countries affected by the program. As well as how the supply shortages, the delay of home purchases during the financial crisis and the increasing popularity of the cities have affected the house price growth. At last it is important to mention the omittance of control variables for the presence of AirBnb, the social housing sector and the increasing popularity of Amsterdam after the Brexit.
In the user cost model, the risk premium on purchasing a home versus renting one, does have a significant effect on the allowed price-to-rent ratio. However, existing literature has not yet reached consensus over the exact magnitude of this risk premium. One topic for future research could be investigate this premium and examine where the deviations in estimates come from.
Furthermore, to test the robustness of this study an interesting variation would be to use a Vector-Error Correction Model (VECM) to estimate the effects of quantitative easing on housing prices. This could be combined with a General Gup Augmented Dickey Fuller to test for overpricing. A study that uses a GSDAF test to examine overpricing in the Amsterdam housing market is the study from Houben, Droës and Lamoen (2017). However, in their study they do not directly control for the effects of quantitative easing.
