PRICE TURNOVER
CORRELATION IN THE
DUTCH HOUSING MARKET
A research into the macro-economic factors
influencing the price turnover correlation, high-
lighting the differences between the Amsterdam
housing market and the housing market in a rural
area within the Netherlands.
Bachelor thesis
Wietske Voorhoeve
Statement of Originality
This document is written by Wietske Voorhoeve who declares to take full responsibility for
the contents of this document.
I declare that the text and the work presented in this document is 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 comple-
tion of the work, not for the contents.
1. Introduction 1.1 Introduction
The Dutch housing market was heavily influenced by the global financial crisis and the collapse of the housing market in the United States in 2008. Despite the fact that the real income in the Netherlands increased in 2008 and 2009 and that the government kept the interest rate at a fairly low level, the Dutch housing market responded immediately to the credit crunch. The consumer confidence in financial institutions was at an absolute low point. Households postponed moving, the number of transactions decreased and the average housing price dropped (Badcock, 2012). The housing market remained depressed until 2013. In 2014, the housing market started its recovery. In general, the prices started rising and the number of transactions was higher. However, the recovery of the housing market seems to have gone at a different pace in different areas of the country. According to recent research from the housing sector research group Calcasa (2015), 37% of Amsterdam housing is currently selling above the asking price. The Union Bank of Switzerland (2016) even rated Amsterdam to be in the danger-zone and concluded that the houses in Amsterdam were significantly overpriced, warning for a possible upcoming bubble. In the Netherlands as a whole, just 7% of properties are sold for more than the asking price. In the more rural areas of the country it is still very common for houses to be sold below the asking price (Calcasa, 2016). This seems to highlight significant differences between urban areas and rural areas.
An interesting angle from which we can investigate the dynamics within the housing market, is the price-turnover correlation. A ‘boom’ in the housing market seems to be characterized by higher housing prices and higher liquidity. In contrary, when the housing market is in a ‘bust’, fewer houses are being sold and the prices decrease. This positive correlation is strongly supported by empirical evidence and has been demonstrated in studies by Stein (1995), Berkovec and Goodman (2006), Andrew and Meen (2003), Ortalo-Magné and Rady (2004b), Clayton, Miller and Peng (2010) and De Wit, Englund and Francke (2012). A negative relation has only been found by Follain and Velz (1995) for the US in the 1990s and Hort (2000) for Sweden. In Sweden, this was explained by the finding that a shock to fundamentals have an immediate impact on sales, but a gradual impact on prices. For the US in the 1990s this was due to the easy access of mortgages. This led to price changes not affecting the number of transactions.
Housing markets play an important role in the economy. In the Netherlands, on average, people spend 29,4% of their disposable income on housing costs (Eurostat, 2014) and about 57% of Dutch
the housing market on the economy as a whole, examining the various factors that influence the housing market is extremely useful.
1.2 Research objective
This thesis will investigate the influence of several macro-economic factors on the correlation between prices and the number of transactions to get a better understanding on what market fundamentals are accountable for the dynamics in the Dutch housing market. We will be able to see which macro-eco- nomic factors determine the difference between the Amsterdam housing market and the rest of the Netherlands. First, we will establish whether there is a positive correlation between prices and number of transactions in the different regions of the Netherlands. Then, we will develop a bivariate VAR- model to explain the linkage between house prices and number of transactions. We chose this model, since it seems to be the most advanced model.
1.3 Practical and scientific contribution
The positive correlation between prices and turnover in housing markets has been the subject of multi- ple empirical researches. Most previous literature on this topic was focused on US data, e.g. researches from Stein (1995), Berkovec and Goodman (2006), Andrew and Meen (2003), Clayton et al. (2010), but there have also been studies focusing on European countries such as the United Kingdom and Sweden. The correlation between the rate of price increase and the number of transactions for the Dutch housing market for the period 1985-2007 has been documented by de Wit et al. (2012). There has also been a very recent cross-country European study on this topic by Droes and Francke (2016) from the Tinbergen Institute, in which they discussed the key determinants of prices and turnover for European countries as a whole and highlighted some differences across countries. However nationally, there has not been any cross-regional research into this topic, comparing the Amsterdam housing mar- ket with the other regions within the Netherlands. This paper aims to fill this gap. Since Calcasa (2016) research pointed out that there are big differences between the Amsterdam housing market and rural housing markets within the Netherlands, it will be useful to know which macro-economic factors cause these differences.
We will investigate the mechanism, giving rise to the correlation and highlight the differences between the urban areas and the rural areas. It is important to realize that, in contrary to the researches men- tioned above, the housing markets within the Netherlands are open markets. It is easy to move from one region to another within the Netherlands, whereas a household would not easily choose to move to another country when the economic circumstances are more favorable there compared to their own country. Another important difference is that, in contrary to the Cross-European research from Droes and Francke, there are not much fixed institutional differences between the regions, since policy like
0,09 0,08 0,07 nte) 0,06 0,05 0,04 0,03 0,02 0,1 0,15 0,2 0,25 0,3 0,35 0,4
Correlation( delta log price, delta log transactions')
Droes and Francke (2016) connect the importance of their research to the volatility of the housing market. They find that for European countries in general, the lower the correlation, the lower the risk and vice versa. We connected the standard deviation of the percentage change in house prices for vari- ous counties and cities in the Netherlands versus the correlation between the logarithm of house price changes and the logarithm of changes in the number of transactions (turnover). This is shown in Fig- ure 1. A region such as Zeeland has a low correlation and the lowest risk. However, Amsterdam is also characterized by a low correlation, but has a high risk.
Amsterdam Noord-Brabant (PV) Noord-Holland (PV) Ov Drenthe ( Geld Friesland (PV) erijssel(PV) PV) erland (PV) Utrecht (PV) Nederland Flevoland (PV) Groningen Utrecht (ge (PVZ)ui d-Holland (PV) meente) 's -Gravenhage (gemee Zeeland (PV) Rotterdam Limburg (PV)
Figure 1. Housing market volatility and the price-turnover relationship 1.4 Thesis outline
In order to answer the question, which macro-economic factors influence the positive correlation be- tween prices and turnover and what the main differences are between Amsterdam and rest of the Neth- erlands, this thesis will be arranged in the following manner. After this introductory chapter, the sec- ond chapter will discuss the relevant literature concerning the main topics. The third chapter describes the data and research approach. The fourth chapter describes the methodology. The fifth chapter de- scribes the results. The sixth chapter draws conclusion and suggests some further research.
st an dar d de via tio n (de lta lo g pr ic e)
2. Literature study 2.1 Introduction
The correlation between prices and turnover is a well-known puzzle in the housing market. Nowadays, real estate agents take this as a given: when the market is ‘hot’, prices are high and the number of transactions increases; when the market is ‘cold’, prices are low and the number of transactions de- creases. In this thesis, we will investigate what macro-economic factors influence this correlation. First, we will look at the level of efficiency of the housing market, since this determines whether shocks in the macro-economic factors are incorporated in the behavior of the actors on the housing market. Then we will discuss a few studies regarding the price-volume correlation to provide an over- view. The economic theories used and constructed in these studies, will be useful for our own study. 2.2 A review of the literature
2.2.1. The (in)efficiency of the Dutch housing market
Traditional financial theory is based on the assumption that investors act rationally, correctly consider- ing all currently available information in the decision making process (Kishore, 2006). This leads to the Efficient Market Hypothesis (EMH): on an informationally efficient asset market without frictions, a shock to macro-economic factors should have an immediate price effect. Fama (1970) defined an ef- ficient market as “a market in which prices always ‘fully reflect’ available information” (p. 381). This means that all information and expectations are incorporated in the price. Relevant information for ac- tors on the housing market would include income, housing stock, demographic changes, interest rates, credit availability and the tax structure (Farlow, 2004a). A change in one of these factors should influ- ence the pricing of the housing market immediately according to the Efficient Market Hypothesis. Actual markets approach the theory of an efficient market, when actors on the market have low-cost access to all information, transaction costs are low, the market is liquid and investors are rational (Quiry, Dallocchio, Le Fur & Salvi, 2014).
Firstly, the actors on the market should have low-cost access to all information. A lot of the infor- mation regarding the housing market is widely available. Interest rates and tax structure is publicly known. The prices of comparable houses in the neighborhood can be obtained cheaply. However, a research from Levitt and Syverson (2008) showed that real estate agents sold their homes for 3,7% more than amateurs sell their homes and leave their houses on the market roughly ten days larger. This could be caused by an informational advantage for experts.
Secondly, transaction costs should be low. However, the housing market is typified by high transac- tion costs (Ommeren, 2008). These transaction costs include search costs, legal and administrative costs, adjustment costs and financing costs (Quigley, 2002). Search costs are those costs devoted to viewing different houses, before ultimately deciding which house to buy. The purchase of a house is one of the largest and most important decisions for a household and since houses are highly heteroge- neous goods (Clayton et al., 2010), people tend to visit a lot of houses before they decide. The average searching time for a house in the Netherlands is not known. However, research in the United Kingdom found that the average duration to find a house is 16 weeks, in which 8 houses are visited (E. ON, 2014). This requires a substantial expenditure of effort. Although Internet services, e.g. the Dutch site www.funda.nl, enable people to view a house online, search costs remain high. Technology helps eliminate some alternatives, but in the end a physical inspection is required (Dixon, Thompson, McAl- lister, Marston & Snow, 2008). In addition, if you engage a real estate agent in the searching process, you pay 2% estate agent fee, which can also be classified as search costs. Legal and administrative costs include several fees and the costs of the transfer tax, which is 2% of the purchase price. Other legal and administrative costs include the costs for the pre-sale agreement, the costs for the transfer contract, the mortgage arranging costs, the costs for the mortgage contract, the costs for the valuation report and the deemed rental value (Dutch Association of Real Estate Brokers and Real Estate Experts [NVM], 2016). Adjustment costs include the costs of moving your possessions. Finally, there are purely financial costs associated with the purchase of a house. These are the costs associated with bor- rowing money. Thus, when you buy a house with a mortgage, you pay interest rate. Taking the search costs, legal and administrative costs, adjustment costs and financing costs into account, there is a sub- stantial amount of transaction costs (Ommeren, 2008; Quigley, 2002).
Thirdly, the more liquid a market is, the more efficient it is. If a good is traded frequently, new infor- mation will quickly be incorporated in the price (Quiry et al., 2014). The housing market is a famously illiquid market (Anglin & Wiebe, 2013). On average, Dutch people move seven times in their entire life (Planbureau voor de Leefomgeving [PBL], 2015). This includes owning a home and renting a home. It is clear that houses are not traded frequently.
Fourthly, and finally, investors should act rational. A rational investor makes decisions logically weighing up the respective costs and benefits before acting and taking all information into considera- tion. According to Ackert, Charupat, Church & Deaves (2006), house-buyers fail to assimilate all in- formation in their decision. Smith and Smith (2006) also argue that most people involved in housing transactions are inexperienced amateurs. A few researches have showed that actors in the housing mar-
could be explained by the fact that people buy and sell houses on a very infrequent base. They have little or no experience determining the value of the house they are buying or selling (Hong, 2007). This makes it highly unlikely that the price they are paying or asking for a house is equal to the present value of the expected cash flow of the house (Meese & Wallace, 1994). Furthermore, there is a sub- stantial emotional aspect involved in buying or selling a house. Khoo, Thyne & Harris (2007) empha- size that human emotion can significantly disrupt the ability to rationally valuate a house. This re- search also points out that it is over-simplistic to assume that actors on the housing market are infor- mation-processing machines who base their decision purely on an assessment of the costs and benefits. To conclude our discussion regarding the EMH and the housing market, we discuss a widely used technique to test the efficiency: examining the correlation of daily returns. The existence of a correla- tion implies that the returns of one year are influenced by the returns of the previous year. This rejects the efficiency of a market (Quiry et al., 2014). In 1990, Case and Shiller conclude that price changes in one year influence the price changes in the following year. The influence of the relevant macro-eco- nomic factors remains low. This indicates that prices are not based on all available information, but on historical prices. Case and Shiller based their research on U.S. data. This is in line with the research from Brown and Matysiak (2000), who find that returns of previous years can explain 80% of the cur- rent returns. Clayton (1998) comes to a same conclusion for the Canadian housing market.
From the above, it is safe to conclude that the housing market is characterized by several factors, such as high transaction costs, low liquidity and the absence of rational investors, which undermine its effi- ciency. On top of that, several researches into the efficiency of the housing market have pointed out that the housing market is not efficient in the theoretical economical view.
2.2.2 Possible explanations for the price-turnover correlation
Several theories have been developed explaining the positive price-turnover correlation. One theory is based on the principle that the housing market can be characterized as a search market (Diaz & Jerez, 2013). Because of the heterogeneously of houses, the market is populated by households looking for a match. Buyers look for a house they like to buy, while sellers look for a household that is willing to pay the price for the house. The match occurs when the price set by a buyer equals or exceeds the price of the seller. When this happens, a transaction will occur (Geltner, Miller, Clayton & Eichholtz, 2007). The way buyers and sellers react to shocks in macro-economic variables may be different, leading to a lagged feedback mechanism. Research from Genesove and Han (2011) show that sellers gradually ad- just their prices when a shock occurs. For instance, when a macro-economic shock causes an increase in demand, the number of transactions will increase. Since sellers did not immediately adjust their prices when the shock occurred, the prices will rise as a consequence of the increase in demand, ex-
Another theory explaining the positive price-turnover correlation concerns the credit constraints (Stein, 1995). The theory is based on the idea that the actors on the housing market usually act as a buyer and seller simultaneously. A lot of households buy a house as a substitute for the house, they’re living in at that moment. Thus, when a positive shock in the macro-economic variables occur, the house they’re selling will increase in value, allowing them to spend more on a new house. This leads to more people buying and selling houses, which consequently lead to higher prices.
Finally, the loss aversion of homeowners could explain the positive price-turnover correlation (De Wit et al., 2013). If homeowners are loss averse, they will not be willing to sell their houses for less than what they paid. When demand is low, the sellers will keep the asking price too high, leading to less transactions. Eventually, when the transaction volume keeps decreasing, prices will decrease as well, explaining the positive price-turnover correlation.
2.2.3. Researches into the price-turnover correlation
In this sector the most recent and influential researches into the price-turnover correlation will be dis- cussed.
In 2010, Clayton et al. has conducted a research into the price-volume dynamics of the U.S., using a panel data set, consisting of housing markets in 127 metropolitan statistical areas in the U.S. He built a panel VAR model and estimated how exogenous variables, such as conditions in the labor market, fi- nancial market and mortgage market affect the price-turnover rate in housing markets. Clayton finds that home prices affect the number of transactions, but it’s not the driving factor behind the positive- volume correlation. Clayton found that changes in employment, average income, the unemployment rate and the mortgage rate had a significant influence on both prices and the number of transactions. However, the transaction volume reacted more dramatically to shocks in the macro-economic varia- bles then home prices do. The effect is on the highest level just one quarter after the shock, meaning that home prices and transaction volume responds very quickly.
In 2011, de Wit et al. conducted a similar research for Dutch data, using a VEC-model. In this research he investigated the influence of shocks in the macro-economic variables on the price-turnover correla- tion for the Netherlands as a whole. Contrary to the study by Clayton et al., de Wit et al. found that the effect of shocks in the macro-economic variables only came gradually. This is consistent with the the- ory that housing markets are populated by agents who learn gradually about changed market condi- tions.
Population growth, inflation and loan-to-GDP ratio do not have a significant impact. This research also confirms that buyers and sellers adjust gradually to changes in fundamentals.
3. Data
This section discusses our choice of variables in the autoregressive panel VAR model. Several statistical publications by CBS have been combined to create a dataset on house prices and turnover. The dataset contains information about macro-economic factors and housing market indicators, both pre-crisis and post-crisis period. We simultaneously model prices and turnover as two interdependent endogenous variables. We took the twelve counties and the four major cities in The Netherlands, since the CBS database StatLine contains most relevant data for those counties and cities. We completed data with the data from the HypoStat reports 2007 and 2015. Because HypoStat is based upon CBS data, we checked mutual consistency.
3.1 Endogenous variables 3.1.1. House prices
The house price indicator is the nominal house price index, using 2010 as base year. We use the percentage change in this variable, not the level of the index itself. For the entire sample period, the average house price change has been relatively stable. As shown in Table 1, the only two areas that go beyond the 3% are Amsterdam and Utrecht, both cities. The rest of the areas all have an average house price change between 2.1% and 2.9%. As a result of the financial crisis, the house prices have
decreased in every part of the Netherlands for the period 2009-2015. Gelderland, a rural area, has had the biggest price decreases, with an average of -3%. Amsterdam, a city, has had the smallest price decreases, with an average of -0.5%. The regional differences in house prices have been generally accepted in empirical research (Capozza, Mack & Mayer, 1997).
3.1.2. Turnover
The other key endogenous variable is the turnover rate. Instead of the actual transaction volume as dependent variable we have used the housing stock per region to normalize turnover. To determine the turnover rate, the number of transactions is divided by the housing stock. We have to take into account that the housing stock, includes rental housing. If we would ignore this, our research would be
considerably biased. Unfortunately, owner-occupancy rate is not very well known in European countries and neither in The Netherlands. However, CBS administered those data quite intensively between 2006 and 2013. The development of the owner-occupancy rate seemed to be quite linear, so we used the 2000-2012 data to extrapolate for the other years. Finally, we checked the outcomes with some single observations, especially for Amsterdam. We found the owner-occupancy rate to be much
Amsterdam Nederland Rotterdam 's-Gravenhage (gemeente) Utrecht (gemeente) 60 50 40 30 20 10 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Flevoland (PV) Nederland Noord-Holland (PV) Utrecht (PV) Zuid-Holland (PV) 70 60 50 40 30 20 10 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 Drenthe (PV) Friesland (PV) Groningen (PV) Limburg (PV) Noord-Brabant (PV) Overijssel (PV) Gelderland (PV) Nederland Zeeland (PV) 70 60 50 40 30 20 10 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
extremely low. Mid-nineties only 10% of the Amsterdam houses was owner-occupied. Nowadays, it is about 30%, which is still not very high. This is shown in Figure 2-4.
Figure 2. Average owner occupancy rate, Cities
Figure 3. Average owner occupancy rate, populous counties
Figure 4. Average owner occupancy rate, less populous counties
ow ne r o ccu pa tio n rat e [% ] ow ner oc cu pa tion ra te [% ] ow ne r o ccu pa tio n ra te [% ]
For the house price changes, we have seen big differences between the pre-crisis period and the post- crisis period. As shown in Table 1, only slight differences occur for the average turnover rate changes between the two periods.
Table 1. House price changes and turnover rate
Average house price change (%) Average turnover rate change (%) Average turnover rate (%) region 1995- 1995- 2009- 1995- 1995- 2009- 1995- 1995- 2009-2015 2008 2015 2015 2008 2015 2015 2008 2015 Neder- land 2,5 5,3 -2,5 -0,038 -0,013 -0,086 4,7 5,4 3,2 Gronin- gen (PV) 2,5 5,2 -2,6 -0,048 -0,010 -0,118 4,4 5,1 2,9 Friesland (PV) 2,4 5,4 -3,1 -0,063 -0,059 -0,070 4,1 4,9 2,7 Drenthe (PV) 2,4 5,2 -2,7 -0,065 -0,047 -0,097 4,4 5,2 2,8 Overijs- sel (PV) 2,5 5,1 -2,5 -0,041 -0,011 -0,098 4,2 4,9 2,9 Flevo- land (PV) 2,2 4,7 -2,4 -0,026 0,055 -0,176 5,4 6,7 3,0 Gelder- land (PV) 2,4 5,3 -3,0 -0,044 -0,019 -0,091 4,2 4,9 2,9 Utrecht (PV) 2,7 5,4 -2,2 -0,035 -0,012 -0,078 5,2 6,0 3,6 Noord- Holland (PV) 2,9 5,5 -2,0 0,015 0,032 -0,016 5,3 5,9 4,1 Zuid-Hol- land (PV) 2,6 5,1 -2,2 -0,080 -0,044 -0,146 5,4 6,4 3,5 Zeeland (PV) 2,7 5,2 -1,8 -0,019 0,011 -0,076 4,1 4,7 3,0 Noord- Brabant (PV) 2,5 5,4 -2,9 -0,039 -0,023 -0,068 4,2 4,9 2,9 Limburg (PV) 2,1 4,6 -2,6 -0,024 -0,004 -0,060 3,8 4,4 2,6 Amster- dam 3,6 5,9 -0,5 0,102 0,184 -0,050 9,7 10,4 8,3 's-Gra- venhage (ge- meente) 2,7 5,2 -1,9 -0,084 -0,022 -0,199 6,9 8,1 4,4 Rotter- dam 2,9 5,0 -1,1 -0,237 -0,246 -0,221 7,9 9,6 4,3 Utrecht (ge- meente) 3,2 5,5 -1,1 -0,033 0,026 -0,143 7,3 8,3 5,3
yearly turnover rate 1995-2015
12,0 10,0 8,0 6,0 4,0 2,0 0,0Figure 5. Average turnover rate for different regions. Based on data from 1995-2015. Turnover rate is the number of trans- actions normalized by the housing stock.
3.2 Exogenous variables
There are several factors that we include in this analysis as exogenous variables. We would expect changes in local demographic conditions and changes in the mortgage market to affect housing demand and supply. Therefore, we include two mortgage market indicators: the GDP as a measure of wealth and the interest rate on new mortgage loans. Other factors taken into account are population and share of young population.
For the factors GDP/GDP per capital, population and the share of young population, we managed to distill local data out of StatLine. The GDP has been determined per region by distributing the national GDP to income ratio per region. In this paper a region is a county or a city. Counties are including the big cities, which are located in the county. We did not adjust GDP to real GDP.
Interest rate on new mortgages are assumed to be the same all over the country and not discriminating between counties and cities.
outstanding mortgages/BNP HICP (2010=100) 120 100 80 60 40 20 0
Inflation interest rate
8 7 6 5 4 3 2 1 0
Amsterdam Nederland Rotterdam 's-Gravenhage (gemeente) Utrecht (gemeente) 30,0% 25,0% 20,0% 15,0% 10,0% 5,0% 0,0% 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Figure 6. Outstanding mortgages/GDP and HICP
Figure 7. Inflation and interest rate Share of young population
We chose to include the share of young population in our model, since this varies greatly among the urban and rural areas within the Netherlands. Urban areas have a much higher share of young population than rural areas. Since young people are usually starters on the housing market and have little accumulated wealth, they tend to be more affected by house price shocks. Unfavorable
circumstances on the housing market could delay them to leave the parental home. As you can see in the graphs below, young people tend to move to the cities.
Figure 8. Sum of share 18-30, Cities
ra te s [% ] ra te s [% ] % p op ul at ion 18 -30 y ear s
Drenthe (PV) Friesland (PV) Gelderland (PV) Groningen (PV) Limburg (PV) Nederland Noord-Brabant (PV) Overijssel (PV) Zeeland (PV) 25,0% 20,0% 15,0% 10,0% 5,0% 0,0% 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Flevoland (PV) Nederland Noord-Holland (PV) Utrecht (PV) Zuid-Holland (PV) 25,0% 20,0% 15,0% 10,0% 5,0% 0,0% 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Figure 9. Sum of share 18-30, Less populous counties
Figure 10. Sum of share 18-30, Populous counties Autocorrelation
Research from Capozza et al. (1997) shows that house prices are auto correlated in the short run. De Wit et al. (2013) also finds autocorrelation for the housing prices in the Dutch housing market, in con- trary to Clayton et al. (2013). To test whether the positive price-turnover correlation is caused by auto-
correlation, we include the price-indext-1 and the turnoverratet-1. Table 2 shows the correlation between
the current price index and the price index of the previous period and the correlation between the cur- rent turnover rate and the turnover rate of the previous period. These outcomes also show that housing prices are auto correlated. This autocorrelation is much stronger than the autocorrelation within the turnover rates. Secondly, we notice that the autocorrelation of the price index in popular cities like Amsterdam and Utrecht is a bit lower than in the rest of the country.
% p op ul at ion 18 -30 y ea rs % p op ul at ion 18 -30 y ea rs
Table 2. Autocorrelation
Region Correlation
Delta log price index
Delta log price
index t-1
Correlation Delta log turno- ver rate
Delta log turno-
ver rate t-1 The Netherlands 0,87 0,24 Groningen (county) 0,88 0,21 Friesland (county) 0,87 0,18 Drenthe (county) 0,87 0,27 Overijssel (county) 0,87 0,00 Flevoland (county) 0,85 0,42 Gelderland (county) 0,88 0,19 Utrecht (county) 0,83 0,09 Noord-Holland (county) 0,81 0,22 Zuid-Holland (county) 0,86 0,35 Zeeland (county) 0,90 0,12 Noord-Brabant (county) 0,88 0,22 Limburg (county) 0,85 0,26 Amsterdam (city) 0,73 0,14 's-Gravenhage (city) 0,83 0,36 Rotterdam (city) 0,87 0,43 Utrecht (city) 0,72 0,12 Mean correlation 0,85 0,22
3.3 The price-turnover correlation
The correlations between the price index and the turnover rate are shown in Table 3. It is clear that the housing market in Amsterdam behaves differently from other parts of the Netherlands. Amsterdam has a much lower correlation between the price index and turnover rate. We will try to explain this differ- ence by performing the autoregressive panel VAR analysis, since this is the method to test the influ- ence of variables on a correlation.
Table 3. Price-turnover correlation for different regions
Region Correlation
Delta log price index
Log turnover rate
Correlation Delta log price index
Delta log turno- ver rate The Netherlands 0,86 0,26 Groningen (county) 0,90 0,27 Friesland (county) 0,90 0,21 Drenthe (county) 0,91 0,18 Overijssel (county) 0,82 0,19 Flevoland (county) 0,85 0,25 Gelderland (county) 0,84 0,23 Utrecht (county) 0,87 0,25 Noord-Holland (county) 0,74 0,27 Zuid-Holland (county) 0,86 0,30 Zeeland (county) 0,88 0,14 Noord-Brabant (county) 0,88 0,21 Limburg (county) 0,81 0,35 Amsterdam (city) 0,35 0,18 's-Gravenhage (city) 0,86 0,38 Rotterdam (city) 0,86 0,28 Utrecht (city) 0,81 0,32 Mean correlation 0,82 0,25
To do so, we will compare the outcomes for Amsterdam with the outcomes of a moral rural region and the country as a whole. We chose Drenthe being a representative of a rural region. Therefore, we will first present the descriptive statistics for these different regions. Since inflation and HICP are only available on the country level, we present them only once in the country table (Table 4).
Table 4. Descriptive Statistics on the levels and log differences of the variables, Netherlands
N Minimum Maximum Mean Std. Deviation
Housepriceindex (2010=100) 357 30,0 107,2 80,643 21,3234 #Transactions 357 2508,0 209767,0 21652,359 39784,7504 Housingstock (OwnerOccupied) 357 34192,30 4354462,49 457703,54 848380,37 Turnover rate 357 2,15% 13,56% 5,3722% 2,13758% Population 357 232718,000 16900726,000 2031850,513 3665511,574 Share (18-30yr) 357 ,113931098 ,273676393 ,164097817 ,031484655 GDP (mln euro) 357 3886,082550 678572,0000 66068,0743 126908,71908 GDP per capita 357 14814,0000 56259,0000 30406,0261 8226,5778 Inflation 21 ,6 4,5 2,029 ,8742 HICP 20 74,7768687 108,646337 92,8774637 11,1521255 Valid N (listwise) 20
Table 5. Descriptive Statistics on the levels and log differences of the variables, Amsterdam
N Minimum Maximum Mean Std. Deviation
Housepriceindex (2010=100) 20 30,0 106,60 78,03 24,3 #Transactions 20 2864 12494 7109 2701 Housingstock (OwnerOccupied) 20 34192 119981 74052 26140 Turnover rate 20 6,39% 13,56% 9,67% 1,90% Population 20 715148 821752 751360 31530 Share (18-30yr) 20 ,1800 ,2185 ,1967 ,0101 GDP (mln euro) 20 16218 27025 27786 7340 GDP per capita 20 22456 49277 36660 8464 Valid N (listwise) 20
Table 6. Descriptive Statistics on the levels and log differences of the variables, Drenthe
N Minimum Maximum Mean Std. Deviation
Housepriceindex 20 38,0 105,10 80,36 21,40 (2010=100) #Transactions 20 3264 7028 5344 1247 Housingstock (OwnerOccupied) 20 105779 143277 124159 11217 Turnover rate 20 2,31% 6,32% 4,40% 1,30% Population 20 454864 491411 479254 12045 Share (18-30yr) 20 ,1139 ,1602 ,1262 0,0151 GDP (mln euro) 20 7738 14142 11446 2073 GDP per capita 20 17011 28945 23795 3796 Valid N (listwise) 20 3.3 Stationarity
Stationarity is a common assumption for a VAR analysis. A stationary process has the property that the mean, variance and autocorrelation structure are constant over time. Many macro-economic
variables are characterized to be non-stationary. For instance, real income, employment, the price level and the population have been growing over almost all developed countries since the Second World War (Wasserfallen, 1986). We have tested for stationarity, using the Fischer test based on the augmented Dickey-Fuller test.
ﰀ0 = 𝐴𝐴𝑙𝑙𝑙𝑙 ﰀ𝑎𝑎ﰀ𝑒𝑒𝑙𝑙ﰀ 𝑎𝑎ﰀ𝑒𝑒 ﰀﰀﰀ − ﰀﰀ𝑎𝑎ﰀ𝑖𝑖ﰀﰀ𝑎𝑎ﰀﰀ
ﰀ1 ≥ ﰀﰀ𝑒𝑒 ﰀ𝑎𝑎ﰀ𝑒𝑒𝑙𝑙 𝑖𝑖 ﰀﰀ𝑎𝑎ﰀ𝑖𝑖ﰀﰀ𝑎𝑎ﰀﰀ
belief that many macro-economic variables are characterized by non-stationarity. A possible
explanation for this is that the sample period is too short for the variables to be non-stationary. Based on the results of this test, we chose to include the levels of the variables instead of the differences. Table 7. Stationarity: Panel Unit Root Test
Variable Inverse Chi-sq. p-val.
LogPriceIndex 201.75 0.0000 LogTurnoverRate 111.49 0.0000 LogPriceIndex (t-1) 68,54 0.0004 LogTurnoverRate (t-1) 171,13 0.0000 InterestNewMortgageLoans 54.86 0.0132 LogInterestRateNewMortgageLoans 34.65 0.4368 LogGDPperCapita 335,62 0.0000 LogPopulation Sample period: 1995-2015 Number of regions: 17 136.78 0.0000
Null hypothesis: All panels contain unit roots; Alternative hypothesis: At least one panel is stationary Fisher test based on the augmented Dickey-fuller test. All tests include one lag and a trend. Table 8. Stationarity: Panel Unit Root Test Differences
Null hypothesis: All panels contain unit roots; Alternative hypothesis: At least one panel is stationary
Variable Inverse Chi-sq. p-val.
DeltaLogPriceIndex 25.50 0.8528 DeltaLogTurnoverRate 55.16 0.0123 DeltaLogPriceIndex (t-1) 22.08 0.9426 DeltaLogTurnoverRate (t-1) 58.41 0.0057 DeltaInterestNewMortgageLoans 100.58 0.0000 DeltaLogGDP 95.69 0.0000 DeltaShareofPop 18-30 26.80 0.8055 DeltaLogPopulation Sample period: 1995-2015 Number of regions: 17 56.60 0.0088
2
4. Methodology
We estimate the model, using a bivariate panel vector autoregressive. This is a way to summarize the dynamics of macroeconomic data, with the intention to provide empirical evidence on the response of house prices/turnover to various exogenous impulses. With this regression two variables can be used as endogenous variables. This is a method to test the influence of variables on a correlation.
𝑙𝑙ﰀ𝑔𝑔 𝒑𝒑𝒊𝒊𝒕𝒕 𝝉𝝉1𝒕𝒕 ﰀ1
ﰀ1
𝑙𝑙ﰀ𝑔𝑔𝒑𝒑𝒊𝒊𝒕𝒕−1 𝛽𝛽1′ 𝜀𝜀1,𝑖𝑖𝑡𝑡
𝑙𝑙ﰀ𝑔𝑔 𝒕𝒕𝒓𝒓𝒂𝒂𝒕𝒕𝒆𝒆𝒊𝒊𝒕𝒕 = 𝝉𝝉2𝒕𝒕 + [ﰀ2 ﰀ2] [𝑙𝑙ﰀ𝑔𝑔𝒕𝒕𝒓𝒓𝒂𝒂𝒕𝒕𝒆𝒆𝒊𝒊𝒕𝒕−1] + [𝛽𝛽 ′] ﰀ𝑖𝑖𝑡𝑡 + [𝜀𝜀2,𝑖𝑖𝑡𝑡 ]
𝝉𝝉𝑖𝑖,𝒕𝒕 is used to describe the time fixed effects. ﰀ𝑖𝑖𝑡𝑡 is used to describe the macro-economic variables. To
simplify, we restricted our model to log house price index and log turnover rate as dependent variables and log GDP per capita, log population, share of young population and log interest rate on new mort- gages as independent variables.
4.2 Hypotheses
Based on our review of the literature and our primary analysis of the data, we formulated the following hypotheses on the difference between the price-turnover correlation in Amsterdam and the rest of the Netherlands:
I. The prices and turnover rate will be positively correlated in Amsterdam, however less strong than in other parts of the Netherlands.
II. Autocorrelation within the housing prices and the turnover rate will be stronger in Amsterdam than in other parts of the Netherlands.
III. There is a connection between the share of young population and the price-turnover rate. IV. There is a connection between the GDP/Capita and the price-turnover rate.
The first hypothesis is based on the knowledge of sharp increasing prices in Amsterdam the last couple of years, while a relatively low owner-occupation rate and a small amount of new building limit supply. As a consequence, we assume in the second hypothesis a stronger autocorrelation with the price index a year earlier. The third hypothesis is based on the relatively high share of young
population living in Amsterdam. This might explain the price index being an autonomous variable not much correlated with the turnover rate. The last hypothesis is based on the economic idea that GDP might be an explanatory factor for the price-turnover rate. As a side note, we do not base any
hypotheses on the mortgage interest rate nor the inflation. Francke et al. (2016) showed that neither the mortgage interest rate or the inflation was of great influence on the price-turnover correlation. Besides that, on a regional level they are not distinguishable. A possible explanation might be that in the
5. Results
It proved to be very hard to actually perform a bivariate panel VAR analysis, as this function is not supported by SPSS and a regular version of STATA. STATA can only perform a regression for panel time series data with one endogenous variable. A bivariate analysis is possible in STATA, however unfortunately not for panel time-series data. So finally we performed our panel VAR analysis in EView. However, in EView it proved not to be possible to test the (auto) correlation with the prices- index and the turnover rate a year earlier because of the ‘singular matrix error’. So for practical reasons we had no other choice than to limit ourselves to a panel VAR evaluation with LOG price index and LOG turnover rate as dependent variables and LOG GDP/capita and the population as independent variables. In order to explain the differences between the city of Amsterdam and the rest of the Netherlands we performed the test on the whole dataset, Amsterdam and the county of Drenthe as a representative of the rural counties. The results of the three tests are shown in Table 9, Table 10 and Table 11.
Table 9 Bivariate Panel VAR analysis Netherlands
6.62E+35
The standard errors are between parentheses (..). The t-statistics are between the hook [..]
LOGPRICEINDEX LOGTURNOVERRATE LOGPRICEINDEX(-1) 0.033202 -0.017570 (0.05435) (0.05614) [ 0.61091] [-0.31295] LOGPRICEINDEX(-2) 0.026631 -0.068361 (0.05113) (0.05282) [ 0.52086] [-1.29434] LOGTURNOVERRATE(-1) -0.000864 0.077034 (0.05482) (0.05662) [-0.01577] [1.36045] LOGTURNOVERRATE(-2) -0.024333 0.091546 (0.05333) (0.05509) [-0.45630] [1.66185] C 4.01E+09 -1.82E+09 (3.9E+08) (4.0E+08) [ 10.2562] [-4.50025] LOGGDPCAPITA -0.065784 -0.012840 (0.02139) (0.02210) [-3.07537] [-0.58111] LOGPOPULATION 0.004680 -0.069955 (0.12202) (0.12604) [ 0.03835] [-0.55500] R-squared 0.040388 0.025490 Adj. R-squared 0.021196 0.006000
Sum sq. resids 2.37E+20 2.52E+20
S.E. equation 8.88E+08 9.17E+08
F-statistic 2.104398 1.307850
Log likelihood -6757.683 -6767.648
Akaike AIC 44.06959 44.13451
Schwarz SC 44.15457 44.21949
Mean dependent 4.22E+09 -2.72E+09
S.D. dependent 8.98E+08 9.20E+08
Determinant resid covariance (dof adj.)
Determinant resid covariance 6.32E+35
Log likelihood -13524.97
Akaike information criterion 88.20174
Table 10. Bivariate Panel VAR analysis Amsterdam LOGPRICEINDEX LOGTURNOVERRATE LOGPRICEINDEX(-1) -0.087842 0.038625 (0.30182) (0.22148) [-0.29104] [ 0.17439] LOGPRICEINDEX(-2) -0.103946 0.075821 (0.30276) (0.22217) [-0.34333] [ 0.34127] LOGTURNOVERRATE(-1) 0.035402 -0.530292 (0.38308) (0.28111) [ 0.09242] [-1.88641] LOGTURNOVERRATE(-2) -0.124207 -0.435251 (0.41243) (0.30266) [-0.30116] [-1.43810] C 6.06E+09 -5.64E+09 (3.0E+09) (2.2E+09) [ 2.02263] [-2.56649] LOGGDPCAPITA -0.483753 0.811239 (0.82183) (0.60308) [-0.58863] [ 1.34515] LOGPOPULATION -0.677951 0.421372 (1.25224) (0.91893) [-0.54139] [ 0.45855] R-squared 0.051880 0.388289 Adj. R-squared -0.465276 0.054629
Sum sq. resids 1.51E+19 8.11E+18
S.E. equation 1.17E+09 8.59E+08
F-statistic 0.100318 1.163726
Log likelihood -396.9584 -391.3879
Akaike AIC 44.88427 44.26532
Schwarz SC 45.23053 44.61157
Mean dependent 4.15E+09 -1.96E+09
S.D. dependent 9.67E+08 8.83E+08
Determinant resid covariance (dof adj.) 1.00E+36
Determinant resid covariance 3.74E+35
Log likelihood -788.2567
Akaike information criterion 89.13963
Table 11. Bivariate Panel VAR analysis Drenthe LOGPRICEINDEX LOGTURNOVERRATE LOGPRICEINDEX(-1) -0.156961 -0.035981 (0.30593) (0.02782) [-0.51307] [-1.29333] LOGPRICEINDEX(-2) -0.137377 -0.034245 (0.29054) (0.02642) [-0.47283] [-1.29611] LOGTURNOVERRATE(-1) -0.019511 0.628791 (1.40426) (0.12770) [-0.01389] [ 4.92388] LOGTURNOVERRATE(-2) -0.142498 -0.042178 (0.67115) (0.06103) [-0.21232] [-0.69106] C -3.54E+11 1.13E+11 (6.0E+11) (5.5E+10) [-0.58471] [ 2.06020] LOGGDPCAPITA 0.096481 -0.018966 (0.16987) (0.01545) [ 0.56796] [-1.22770] LOGPOPULATION 273.8232 -87.36548 (465.085) (42.2946) [ 0.58876] [-2.06564] R-squared 0.102119 0.941570 Adj. R-squared -0.387634 0.909699
Sum sq. resids 1.42E+19 1.17E+17
S.E. equation 1.14E+09 1.03E+08
F-statistic 0.208511 29.54330
Log likelihood -396.4256 -353.2695
Akaike AIC 44.82507 40.02995
Schwarz SC 45.17132 40.37620
Mean dependent 4.19E+09 -3.18E+09
S.D. dependent 9.65E+08 3.44E+08
Determinant resid covariance (dof adj.)
1.36E+34
Determinant resid covariance 5.06E+33
Log likelihood -749.5436
Akaike information criterion 84.83818
Schwarz criterion 85.53069
Table 12. Influence of GDP and Population on Price index and Turnover Rate
Price index Turnover rate
Netherlands GDP - -Population + -Amsterdam GDP - + Population - + Drenthe GDP + -Population +
-In Table 12, we have summarized the influence of the factors GDP and Population on Price index and Turnover rate. The results are inconclusive. We can see that the housing market in Amsterdam reacts in opposite direction to changes in GDP and population than the housing market in Drenthe. In Amsterdam, GDP and Population both have a negative influence on the Price index and a positive influence on the Turnover rate. In Drenthe, GDP and Population both have a positive influence on the Price index, whereas they have a negative influence on the Turnover rate. This is not what we would have expected. We would sooner expect the other way around, since the supply of houses in
Amsterdam is limited, in contrary to the supply of houses in Drenthe.
However, as seen in the tables above, the R2 is very low. The R2 is approximately 0.04 for the model
regarding the Netherlands, 0.05 for the model regarding Amsterdam and 0.1 for the model regarding Drenthe. This means that only a very minor part of the behavior of the price turnover rate is described by this model. For this reason, we don’t expect the results to be reliable. To find some more insight why the analysis above did not lead to the expected results, we made an additional analysis in the statistics program R. In R, we made a panel VAR analysis, using the package PLM (Croissant & Millo, 2008), based on the following two linear regressions:
𝑙𝑙ﰀ𝑔𝑔 𝒑𝒑𝒊𝒊𝒕𝒕 = 𝛽𝛽0 + 𝑙𝑙ﰀ𝑔𝑔 𝒑𝒑𝒕𝒕−𝟏𝟏 ∗ 𝛽𝛽1 + 𝑙𝑙ﰀ𝑔𝑔 𝒕𝒕𝒓𝒓𝒂𝒂𝒕𝒕𝒆𝒆𝒕𝒕−𝟏𝟏 ∗ 𝛽𝛽2 + 𝑙𝑙ﰀ𝑔𝑔ﰀ𝐷𝐷ﰀ𝑐𝑐𝑎𝑎ﰀ𝑖𝑖ﰀ𝑎𝑎 ∗ 𝛽𝛽3 + 𝑆𝑆ℎ𝑎𝑎ﰀ𝑒𝑒𝑌𝑌ﰀ𝑔𝑔ﰀﰀﰀ ∗ 𝛽𝛽4
+ 𝜀𝜀1,𝑖𝑖𝑡𝑡
𝑙𝑙ﰀ𝑔𝑔 𝒕𝒕𝒓𝒓𝒂𝒂𝒕𝒕𝒆𝒆𝒊𝒊𝒕𝒕 = 𝛽𝛽0 + 𝑙𝑙ﰀ𝑔𝑔 𝒑𝒑𝒕𝒕−𝟏𝟏 ∗ 𝛽𝛽1 + 𝑙𝑙ﰀ𝑔𝑔 𝒕𝒕𝒓𝒓𝒂𝒂𝒕𝒕𝒆𝒆𝒕𝒕−𝟏𝟏 ∗ 𝛽𝛽2 + 𝑙𝑙ﰀ𝑔𝑔ﰀ𝐷𝐷ﰀ𝑐𝑐𝑎𝑎ﰀ𝑖𝑖ﰀ𝑎𝑎 ∗ 𝛽𝛽3 + 𝑆𝑆ℎ𝑎𝑎ﰀ𝑒𝑒𝑌𝑌ﰀ𝑔𝑔ﰀﰀﰀ
∗ 𝛽𝛽4 + 𝜀𝜀1,𝑖𝑖𝑡𝑡
During this analysis we noticed that adding the lag variables 𝑙𝑙ﰀ𝑔𝑔 𝒑𝒑𝒕𝒕−𝟏𝟏 and 𝑙𝑙ﰀ𝑔𝑔 𝒕𝒕𝒓𝒓𝒂𝒂𝒕𝒕𝒆𝒆𝒕𝒕−𝟏𝟏 led to a
singular matrix situation which could not be resolved. Because of the central limit theorem, we are not sure whether we can compare the results for Amsterdam and Drenthe (N=20) with the overall results
The results of the linear regressions are shown in Table 13. The complete regressions are included in the Appendix.
Table 13, Linear regression results
Whole dataset
Country level
Amsterdam
Drenthe
House price Turnover
rate
House price Turnover
rate
House price Turnover
rate
House price Turnover
rate Log House price 0.86 (0.027) 0.19 (0.096) 0.76 (0.408) -1.08 (1.166) 0.88 (0.160) -0.70 (0.257) 0.83 (0.244) -1.98 (0.801) Log Turnover rate -0.008 (0.017) 0.55 (0.060) 0.15 (0.084) 0.76 (0.240) 0.19 (0.115) 0.34 (0.184) 0.18 (0.065) 0.64 (0.213) Log GDP capita 0.17 (0.031) -0.40 (0.112) -0.13 (0.684) -2.08 (1.954) -0.06 (0.605) -1.02 (0.972) -0.49 (0.378) 0.043 (1.239) Log Share Young Population -0.48 (0.208) -1.44 (0.746) -3.41 (7.133) -39.8 (20.388) 1.98 (3.392) -21.2 (5.446) -4.91 (6.044) -50.9 (19.809) R2 0.9374 0.47 0.99 0.91 0.96 0.77 0.99 0.92 Sample size 355 20 20 20
This Table shows the estimators. The standard errors are between parentheses (..).
Although these regressions do not have the price-turnover correlation as one dependent variable, we can use these outcomes to describe some differences between the housing markets.
Firstly, we see that this model fits the data much better than the bivariate panel VAR analysis, as the
R2 and adjusted R2 show high values, while the analysis of the residuals shows low values. Hypothesis
1, the house prices and turnover rate being correlated, was already proven in Table 2 but is less clear from this regression. Apart from that, we can see that house prices and turnover rates are auto correlated (hypothesis 2). In Amsterdam the autocorrelation of the house prices is somewhat stronger than on a country level or in Drenthe. That might support our second hypothesis. Turnover rates are also auto correlated, however less strong than the house prices. This is where we see a rather big difference between Amsterdam and Drenthe. Amsterdam only has a autocorrelation of 0.34, whereas Drenthe has an autocorrelation of 0.64 and the Netherlands as a whole even find an autocorrelation of 0.76. A possible explanation is that the turnover rate is very much influenced by external factors, like the supply of houses for sale or new building while the housing prices can develop more
market while Drenthe can be characterized as a buyer’s market. The influence of the GDP/Capita and the share of young population show mixed results and even high negative factors where we do not expect them. For instance, we would expect a higher GDP/Capita to have a positive influence on the house price, which is not confirmed in above results. This might be a result of the fact that we used the house price index (100=2010) instead of real house prices. For the variable ‘share of young
population’, we find a negative influence on all turnover rates and all house price indexes, except for the house price index in Amsterdam. However, together with the high observed autocorrelation rates, we cannot substantiate our hypotheses 3 and 4.
6. Conclusion
Prices and turnover rates have a positive correlation in all parts of the Netherlands. However, while studying the price-turnover correlation, we immediately see that Amsterdam behaves differently from other parts of the Netherlands. While the price-turnover correlation is around the 0,8 in other parts of the country, in Amsterdam this correlation is 0,35. This paper was aimed to investigate the relationship between the house price index and the turnover rate for Amsterdam, Drenthe and the Netherlands as a
whole. However, for all three bivariate panel VAR analyses, we noticed a very low R2 and a high sum
of square residuals. For a model that fits the data well this should be quite the opposite. This means that the model does not fit very well. We also notice a large constant factor c. This might be the result of using the level of each variable instead of the difference between two subsequent instances of a variable. The models for Amsterdam and Drenthe seem to fit slightly better.
We found that price indexes in the housing market are auto correlated in Amsterdam, as well as in Drenthe and in the Netherlands. However, we notice a difference when looking at the autocorrelation of the turnover. The turnover rate of the Netherlands as a whole and the turnover rate of Drenthe is more auto correlated than the turnover rate of Amsterdam. This could be explained by the fact that the supply of houses in Amsterdam is limited.
In this model, we find a negative influence of the GDP on the housing prices and turnover rate. Although previous research also pointed out that GDP is no key determinant in explaining house prices and turnover rates, it seems strange that the GDP has a negative influence. This outcome might be caused by the high autocorrelation in the house prices and turnover rates.
For the variable ‘share of young population’, we find a negative influence on all variables except for the house price index in Amsterdam. This is probably because the prices in Amsterdam and the share of young population have both been rising during our sample time period. However, the high negative
website, making it possible for individuals to earn money by putting their house up for rent for a limited amount of periods per year. Since Amsterdam is a very popular destination for a holiday or city trip, citizens of Amsterdam can earn a significant amount of money with Airbnb, making a house in Amsterdam more valuable. It would be interesting to investigate what the effect of Airbnb would be on the price-turnover correlation in Amsterdam.
References
Ackert, L. F., Charupat, N., Church, B. K., & Deaves, R. (2006). An experimental examina-
tion of the house money effect in a multi-period setting. Experimental Economics, 9(1),
5–16. http://doi.org/10.1007/s10683-006-1467-1
Andrew, M., & Meen, G. (2003). Housing Transactions and the Changing Decisions of
Young Households in Britain: The Microeconomic Evidence. Real Estate Economics.
http://doi.org/10.1111/j.1080-8620.2003.00060.x
Anglin, P. M., & Wiebe, R. (2013). Pricing in an Illiquid Real Estate Market. The Journal of
Real Estate Research, 35(1), 83–102. Retrieved from http://login.library.sherid-
anc.on.ca/login?url=http://search.proquest.com/docview/1348127125?accountid=3455
Badcock, B. (2012). Beyond Home Ownership. Housing, Welfare and Society. Urban Policy
and Research, 30(3), 341–344. http://doi.org/10.1080/08111146.2012.702075
Berkovec, J. a, & Goodman Jr., J. L. (1996). Turnover as a measure of demand for existing
homes. Real Estate Economics, 24(4), 421–440. http://doi.org/10.1111/1540-6229.00698
Brown, G. R., & Matysiak, G. a. (2000). Sticky Valuations, Aggregation Effects, and Property
Indices. The Journal of Real Estate Finance and Economics, 20(1), 49–66.
http://doi.org/10.1023/A:1007879805481
Campbell, J. Y., & Cocco, J. F. (2007). How do house prices affect consumption? Evidence
from micro data. Journal of Monetary Economics, 54(3), 591–621.
http://doi.org/10.1016/j.jmoneco.2005.10.016
Capozza, D., Mack, C., Mayer, C., 1997. The dynamic structure of housing markets. Working
Paper. University of Michigan, Ann Arbor.
Case, K. E., & Shiller, R. J. (1989). The efficiency of the market for single-family homes.
American Economic Review. http://doi.org/10.2307/1804778
Clayton, J. (1998). Further Evidence on Real Estate Market Efficiency. Journal of Real Estate
Research, 15(1/2), 41–58.
Clayton, J., Miller, N., & Peng, L. (2010). Price-volume correlation in the housing market:
Causality and co-movements. Journal of Real Estate Finance and Economics, 40(1), 14–
40.
http://doi.org/10.1007/s11146-008-9128-0
Croissant, Y., & Millo, G. (2008). Panel data econometrics in R: The plm package. Journal of
Statistical Software, 27(2), 1-43.
Díaz, A., Jerez, B., 2013. House prices, sales, and time on the market: A search
‐ theoretic
framework. International Economic Review 54, 837-872.
Dixon, T., Thompson, B., McAllister, P., Marston, A., & Snow, J. (n.d.). Real Estate and the
New Economy: The Impact of Information and Communications Technology.
Droes, M., & Francke, M. K. (2016). What Causes the Positive Price-Turnover Correlation in
European Housing Markets. Tinbergen Institute Working Paper.
Fama, E. F. (1970). Efficient Capital Markets - A Review of Theory and Empirical Work.pdf.
The Journal of Finance, 25(2), 36. http://doi.org/10.2307/2329297
Follain, James R. Velz, O. (1995). Incorporating the Number of Existing Home Sales into a
Structural Model of the Market for Owner-Ocuppied Housing. Journal of Housing Eco-
nomics. http://doi.org/10.1006/jhec.1995.1005
Fraser, J., & Kick, E. (2007). The Role of Public, Private, Non-profit and Community Sectors
in Shaping Mixed-income Housing Outcomes in the US. Urban Studies, 44(12), 2357–
2377. http://doi.org/10.1080/00420980701540952
Farlow, A. (2004a). UK house prices: a critical assessment. Paper presented at the Credit 32
Suisse First Boston Housing Market Conference.
Geltner, D. M., Miller, N. G., Clayton, J., Eichholtz, P., 2007. Commercial real estate analysis
and investments. OnCourse Learning, Mason.
Genesove, D., & Han, L., 2012. Search and matching in the housing market. Journal of Urban
Economics 72, 31-45.
Hong, H. (2007). Behavioural finance: Introduction. European Financial Management, 13(3),
389–393. http://doi.org/10.1111/j.1468-036X.2007.00365.x
Hort, K. (2000). Prices and turnover in the market for owner-occupied homes. Regional Sci-
ence and Urban Economics, 30(1), 99–119. http://doi.org/10.1016/S0166-
0462(99)00028-9
Kishore, R. (2006). Theory of behavioral finance and its application to property market: a
change in paradigm. Australian Property Journal, (1979), 1–17. Retrieved from
http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:Theory+of+Behav-
ioural+Finance+and+its+Application+to+Property+Market:+A+Change+in+Para-
digm.#0
Khoo, C., Thyne, M., & Harris, P. (2007). Making Cents: The Role of Consumers Emotion in
Property Valuation. International Journal of Business and Management, 2(5), 84-90.
Langevoort, D. C. (2003). Foreward: Revisiting Gilson and Kraakman’s Efficiency Story.
Journal of Corporation Law, 28(4), 499–502. Retrieved from http://search.ebsco-
host.com/login.aspx?direct=true&db=bth&AN=12189985&authtype=sso&cus-
tid=s8993828&site=eds-live&scope=site
Levitt, S. D., & Syverson, C. (2008). Market Distortions When Agents Are Better Informed:
The Value of Information in Real Estate Transactions. Review of Economics and Statis-
tics, 90(4), 599–611. http://doi.org/10.1162/rest.90.4.599
Meese, R., & Wallace, N. (1991). Nonparametric Estimation of Dynamic Hedonic Price Mod-
els and the Construction of Residential Housing Price Indices. Real Estate Economics,
19(3), 308–332. http://doi.org/10.1111/1540-6229.00555
Meese, R., & Wallace, N. (1994). Testing the Present Value Relation for Housing Prices:
Should I Leave My House in San Francisco? Journal of Urban Economics.
http://doi.org/http://dx.doi.org/10.1006/juec.1994.1015
Ommeren, J., & Van Leuvensteijn, M. (2005). New evidence of the effect of transaction costs
on residential mobility. Journal of Regional Science, 45(4), 681 702.
http://doi.org/10.1111/j.0022-4146.2005.00389.x
Ortalo-Magné, F., & Rady, S. (2004). Housing transactions and macroeconomic fluctuations:
A case study of England and Wales. Journal of Housing Economics, 13(4 SPEC.ISS.),
287–303. http://doi.org/10.1016/j.jhe.2004.09.005
Quigley, J.M. (2002), Transaction costs and housing markets, Housing Markets and Public
Policy, O’Sullivan, A. and K. Gilb, eds, 54-64, Blackwell Publishing
Smith, M. H., & Smith, G. (2006). Bubble, Bubble, Where’s the Housing Bubble? Brookings
Papers on Economic Activity, Spring (1), 1–67. http://doi.org/10.1353/eca.2006.0019
Stein, J. C. (1995). Prices and Trading Volume in the Housing Market: A Model with Down-
Payment Effects. Quarterly Journal of Economics, 110(May), 379–406.
http://doi.org/10.2307/2118444
Verbruggen, J., & Kranendonk, H. (2005). (2005). CPB Document. Retrieved from www.cen-
traalplanbureau.nl
Vernimmen, P., Quiry, P., Dallocchio, M., Le Fur, Y., & Salvi, A. (2014). Corporate Fi-
nance: Theory and Practice (Fourth Edi).
Wasserfallen, W. (1986). Non-Stationarities in Macro-Economic Time Series—Further
Evidence and Implications. Canadian Journal of Economics, 498-510.
Whittle, R., Davies, T., Gobey, M., & Simister, J. (2014). Behavioural Economics and House
Prices: A Literature Review. Business and Management Horizons, 2(2), 15.
price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume
Pr
ice
tur
no
ve
r c
or
rel
at
ion
D
utch
ho
us
in
g m
ar
ke
t
21
Ju
ly 2
01
6
A p pe ndi x 1 G ra phs of the d eve lopm ent of ho us e p rice inde x a nd tur no ve rs Ne de rla nd Am ste rd am 120 250000 120 100 14000 12000 10000 8000 6000 400 0 200 0 0 100 80 200000 80 60 150000 60 100000 40 20 40 20 50000 0 0 0 ye ar ye ar Ro tte rd am 's G ra ve nh ag e 120 100 8000 7000 6000 500 0 4000 3000 2000 100 0 0 120 100 9000 8000 7000 6000 5000 4000 3000 2000 100 0 0 80 60 80 60 40 20 40 20 0 0 ye ar ye ar Ut re ch t (g em een te) Dr en th e ( PV ) 120 100 6000 5000 12 0 80 00 70 00 60 00 500 0 40 00 30 00 20 00 100 0 0 10 0 80 80 60 4000 3000 60 40 40 20 2000 1000 20 0 0 0 ye ar ye ar Fle ve la nd (PV ) G eld erla nd (PV ) 120 100 8000 7000 6000 500 0 4000 3000 2000 100 0 0 120 25000 100 80 200 00 80 60 15000 60 40 10000 40 20 5000 20 0 0 0 ye ar ye ar 33price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume
Pr
ice
tur
no
ve
r c
or
rel
at
ion
D
utch
ho
us
in
g m
ar
ke
t
21
Ju
ly 2
01
6
Frie sla nd (PV ) Ge lde rla nd (PV ) 120 100 10000 9000 800 0 7000 6000 5000 4000 300 0 2000 1000 0 120 100 25000 20000 80 60 80 60 15000 10000 40 20 0 40 20 0 5000 0 ye ar ye ar G ro ni ng en (PV ) Lim bur g (PV ) 120 100 9000 8000 7000 6000 5000 4000 3000 2000 100 0 0 120 100 14000 12000 10000 8000 6000 400 0 200 0 0 80 60 80 60 40 20 40 20 0 0 ye ar ye ar N oor d-B ra ba nt ( PV ) N oor d-H olla nd (PV ) 120 100 35000 30000 25000 20000 15000 100 00 500 0 0 120 100 40000 35000 30000 250 00 20000 15000 10000 500 0 0 80 60 80 60 40 20 40 20 0 0 ye ar ye ar O ver ijs sel (PV ) Ut re ch t (PV ) 120 14000 12000 10000 8000 6000 400 0 200 0 0 120 18000 16000 14000 12000 10000 8000 6000 4000 200 0 0 100 80 100 80 60 40 60 40 20 0 20 0 ye ar ye ar 34price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume price index 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 transaction volume
Pr
ice
tur
no
ve
r c
or
rel
at
ion
D
utch
ho
us
in
g m
ar
ke
t
21
Ju
ly 2
01
6
Zeel an d ( PV ) Ziud -H olla nd (PV ) 120 7000 6000 5000 4000 3000 200 0 100 0 0 120 50000 45000 400 00 35000 30000 25000 20000 150 00 10000 5000 0 100 80 100 80 60 40 60 40 20 0 20 0 ye ar ye ar 35Appendix 2
Panel VAR analyses
Results for the Netherlands (whole dataset) VAR Estimation Results:
=========================
Endogenous variables: LogHousePrice, LogTurnoverRate, LogGDPCapita, Share1830 Deterministic variables: both
Sample size: 355
Log Likelihood: 1452.666
Estimation results for equation LogHousePrice:
==============================================
𝑙𝑙ﰀ𝑔𝑔 𝒑𝒑𝒊𝒊𝒕𝒕 = 𝛽𝛽0 + 𝑙𝑙ﰀ𝑔𝑔 𝒑𝒑𝒕𝒕−𝟏𝟏 ∗ 𝛽𝛽1 + 𝑙𝑙ﰀ𝑔𝑔 𝒕𝒕𝒓𝒓𝒂𝒂𝒕𝒕𝒆𝒆𝒕𝒕−𝟏𝟏 ∗ 𝛽𝛽2 + 𝑙𝑙ﰀ𝑔𝑔ﰀ𝐷𝐷ﰀ𝑐𝑐𝑎𝑎ﰀ𝑖𝑖ﰀ𝑎𝑎 ∗ 𝛽𝛽3 + 𝑆𝑆ℎ𝑎𝑎ﰀ𝑒𝑒𝑌𝑌ﰀ𝑔𝑔ﰀﰀﰀ ∗ 𝛽𝛽4
+ 𝜀𝜀1,𝑖𝑖𝑡𝑡
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.07917 on 349 degrees of freedom
Multiple R-Squared: 0.9374, Adjusted R-squared: 0.9365
F-statistic: 1044 on 5 and 349 DF, p-value: < 2.2e-16
Estimator Standard Error t-value Value (Pr>(>|t|)
Log House Price 0.8596810 0.0267084 32.188 < 2e-16 ***
Log Turnover rate
-0.0081020 0.0168903 -0.480 0.631751
Log GDP/Capita 0.1796441 0.0311885 5.760 1.85e-08 ***
Share 18-30 yr -0.4853353 0.2084587 -2.328 0.020472 *
cons -1.1631726 0.3018849 -3.853 0.000139 ***
Estimation results for equation LogTurnoverRate:
================================================
𝑙𝑙ﰀ𝑔𝑔 𝒕𝒕𝒓𝒓𝒂𝒂𝒕𝒕𝒆𝒆𝒊𝒊𝒕𝒕 = 𝛽𝛽0 + 𝑙𝑙ﰀ𝑔𝑔 𝒑𝒑𝒕𝒕−𝟏𝟏 ∗ 𝛽𝛽1 + 𝑙𝑙ﰀ𝑔𝑔 𝒕𝒕𝒓𝒓𝒂𝒂𝒕𝒕𝒆𝒆𝒕𝒕−𝟏𝟏 ∗ 𝛽𝛽2 + 𝑙𝑙ﰀ𝑔𝑔ﰀ𝐷𝐷ﰀ𝑐𝑐𝑎𝑎ﰀ𝑖𝑖ﰀ𝑎𝑎 ∗ 𝛽𝛽3 + 𝑆𝑆ℎ𝑎𝑎ﰀ𝑒𝑒𝑌𝑌ﰀ𝑔𝑔ﰀﰀﰀ
∗ 𝛽𝛽4 + 𝜀𝜀1,𝑖𝑖𝑡𝑡
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.2832 on 349 degrees of freedom
Multiple R-Squared: 0.4751, Adjusted R-squared: 0.4676
F-statistic: 63.17 on 5 and 349 DF, p-value: < 2.2e-16
Covariance matrix of residuals:
LogHousePrice LogTurnoverRate LogGDPCapita Share1830
LogHousePrice 0.0062680 0.0007398 -0.001580 -0.0003187
LogTurnoverRate 0.0007398 0.0802215 0.026615 0.0049875
LogGDPCapita -0.0015800 0.0266147 0.027430 0.0031664
Share1830 -0.0003187 0.0049875 0.003166 0.0008275
Correlation matrix of residuals:
LogHousePrice LogTurnoverRate LogGDPCapita Share1830
LogHousePrice 1.00000 0.03299 -0.1205 -0.1400
LogTurnoverRate 0.03299 1.00000 0.5674 0.6121
LogGDPCapita -0.12049 0.56736 1.0000 0.6646
Share1830 -0.13996 0.61214 0.6646 1.0000
Estimator Standard Error t-value Value (Pr>(>|t|)
Log House Price 0.1927971 0.0955495 2.018 0.044380 *
Log Turnover rate 0.5495682 0.0604252 9.095 < 2e-16 *** Log GDP/Capita -0.3969973 0.1115771 -3.558 0.000425 *** Share 18-30 yr -1.4408719 0.7457622 -1.932 0.054159 cons 2.2580214 1.0799952 2.091 0.037272 * trend -0.0006890 0.0003647 -1.889 0.059692
Results for the Netherlands (country level) VAR Estimation Results:
=========================
Endogenous variables: LogHousePrice, LogTurnoverRate, LogGDPCapita, Share1830 Deterministic variables: both
Sample size: 20
Log Likelihood: 250.178
Estimation results for equation LogHousePrice:
==============================================
𝑙𝑙ﰀ𝑔𝑔 𝒑𝒑𝒊𝒊𝒕𝒕 = 𝛽𝛽0 + 𝑙𝑙ﰀ𝑔𝑔 𝒑𝒑𝒕𝒕−𝟏𝟏 ∗ 𝛽𝛽1 + 𝑙𝑙ﰀ𝑔𝑔 𝒕𝒕𝒓𝒓𝒂𝒂𝒕𝒕𝒆𝒆𝒕𝒕−𝟏𝟏 ∗ 𝛽𝛽2 + 𝑙𝑙ﰀ𝑔𝑔ﰀ𝐷𝐷ﰀ𝑐𝑐𝑎𝑎ﰀ𝑖𝑖ﰀ𝑎𝑎 ∗ 𝛽𝛽3 + 𝑆𝑆ℎ𝑎𝑎ﰀ𝑒𝑒𝑌𝑌ﰀ𝑔𝑔ﰀﰀﰀ ∗ 𝛽𝛽4
+ 𝜀𝜀1,𝑖𝑖𝑡𝑡
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.03228 on 14 degrees of freedom Multiple R-Squared: 0.9898, Adjusted R-squared: 0.9862 F-statistic: 271.8 on 5 and 14 DF, p-value: 2.012e-13
Estimator Standard Error t-value Value (Pr>(>|t|)
Log House Price 0.763450 0.407893 1.872 0.0823
Log Turnover rate 0.151954 0.083871 1.812 0.0915 Log GDP/Capita -0.127806 0.683533 -0.187 0.8544 Share 18-30 yr -3.412387 7.133372 -0.478 0.6398 cons 3.329305 5.019910 0.663 0.5180 trend 0.005265 0.017352 0.303 0.7661