THE PREDICTIVE POWER OF CONSUMER EXPECTATIONS ON HOUSE PRICES IN THE NETHERLANDS WIJNAND. G.K. REIJNEVELD 10482024 MSc Business Economics UNIVERSITY OF AMSTERDAM Supervisor: prof. dr. J. (Johan) Conijn
Statement of Originality
This document is written by Student Wijnand Reijneveld, 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 completion of the work, not for the contents.
Abstract
This paper examines the predictive power of household expectations on house prices. Furthermore, it examines whether household expectations have explaining power in existing
house price models. Aforementioned is tested by using the VEH Market Indicator over a sample period of approximately 10 years with quarterly data. Using OLS, two different models are estimated. It is found that the third and fifth lag of the indicator is informative for
pseudo out of sample forecasting. Moreover, the indicator does add significant explaining power to existing house price models.
Contents
1. Introduction ... 5
2. Literature overview ... 7
2.1 International literature ... 7
2.2 Dutch literature ... 8
2.3 Household expectations and housing variables ... 10
2.4 Chapter summary ... 13
3. Households’ expectations ... 14
3.1 VEH Market Indicator ... 14
3.1.1 Chicken or Egg ... 15
3.1.2 Regional Differences ... 16
3.2 The market indicator decomposed ... 16
3.2.1 Prospective price expectations ... 17
3.2.2 Prospective general conditions ... 18
3.2.3 Prospective interest expectations ... 18
3.2.4 Aggregate prospective expectations ... 19
3.3 Consumer confidence (CBS) ... 20
3.4 Conclusion ... 21
4. Methodology and data ... 22
4.1 Methodology ... 22
4.1.1. Forecasting with household expectations ... 22
4.1.2. Households’ expectations as explanatory variable ... 22
4.2 Data 24 5. Results ... 25
5.1 Data preparation ... 25
5.2 Forecasting with household expectations ... 26
5.2.1. The estimated model ... 26
5.2.2. Forecasting with the estimated coefficients ... 29
5.3 Households’ expectations as explanatory variable ... 29
6. Conclusion ... 33
Bibliography ... 34
1. Introduction
This paper examines the predictive power of households’ expectations on Dutch house prices. Unrealistic positive or negative expectations are often referred as bubble and bust builders in house price models. Although researchers typically agree about the relationship between household expectations and house prices, research on this topic is scarce.
Dutch institutes like the OTB, CPB and Ortec Finance use economic variables like interest rates and income to evaluate, explain and predict future house prices. The variation that cannot be explained using the beforementioned variables is called over- or
undervaluation and is often referred to as a bubble or bust. Households’ expectations are not used as an explanatory variable in the cited models. Arguments on why households’
expectations are not used in these modelsis are missing as well.
International research already confirmed the importance of households’ expectations in explaining and predicting house prices. In addition, recent Dutch research confirmed that the seven months lagged household expectations can predict the quantity of houses sold (Boumeester, 2014). According to the economic theory demand should influence prices when the Dutch housing supply is inelastic, therefore we expect that household expectations do also influence house prices. We argue that household expectations should be able to (partly) explain house prices. The goal of this research is to test whether house prices can be
forecasted by household expectations and to what extend these expectations can improve the current Dutch house price models. The following research questions will be answered in this paper:
1.! What index is the best proxy for household expectations? 2.! Can we use household expectations to predict house prices? 3.! Are households’ expectations informative in house price models?
This study is relevant because the impact of housing wealth is very important for households and the economy. Housing is, at approximately €1.400 billion, the biggest asset class in the Netherlands (CBS, 2011). Development of house prices can therefore have major impacts on national wealth. Several studies confirmed that house price developments can explain
consumption and therefore house prices are increasingly used in macro economic modelling. (Case, Quigly, & Shiller, 2013); (Miles, 1993); (Johan Verbruggen, 2005).
Using households’ expectations to explain and predict houseprices is especially interesting because it can contain information about the future. This is in contrast with the common fundamental factors where the value of the future is unknown.
In order to answer the abovementioned research questions we will start in chapter two with reviewing the relevant literature. In chapter three there will be eleborated on the best proxy for households’ expectations in order to answer the first research question. In chapter four the research methodology with respect to the second and third research question will be introduced. In chapter five the results will be discussed and in chapter six we will conclude and discuss on how the outcomes can be used.
2. Literature overview
House price models can differ per country because of government regulation and cultural differences. The focus in this study is on the Dutch housing market. Therefore, we will do a quick introduction of some international literature in the first paragraph, this part will be descriptive and not discussed in detail, the estimated models will be omitted. The Dutch part in paragraph two will be discussed more extensively, the estimated models will also be included. In the third and last paragraph the literature about households’ expectations and housing variables is discussed.
2.1 International literature
Abraham & Henderschott (1993) finds that the typical house price model is to describe an (equilibrium) price where the market price adjusts to. The common variables used are income growth, construction costs, changes in real interest rate, lagged appreciation and actual versus equilibrium price. It was found that approximately 60% of the change in house prices can be explained by the before mentioned variables. This research is based on 30 cities across the United States.
One paper that focuses on the common explaining factors between countries is written by Englund & Ioannides (1997). This paper is about house price cycles and the fundamentals that influence house prices. It is based on a dataset of the period 1970-1992 and covers 15 OECD countries. They question whether house prices are predictable, which factors can explain house prices and if there is an international house price cycle. Especially the first and the second question are interesting for this paper. They find that housing markets are driven by the following variables across countries: GDP growth, real interest rate and the previous house prices. The researchers find strong (one year lag) autocorrelation, which means that values of past year are explanatory for the future. (Englund & Ioannides, 1997) states there are basically two explanations for this pattern: The autocorrelation is a result from the existence of bubbles and busts or, the asset market model is not suitable for capturing the housing market returns.
Fuss (2010) did an analysis of 15 OECD countries over a period of 30 years and found that when economic activity increases by 1%, house prices increase by 0.6% in the long run. In addition to economic activity, construction costs and interest rates show significant long term effects of 0.6% and 0.3% respectively. Furthermore, the adjustment
process to a long run equilibrium is only 16% per year which means that it can take many years before prices are adjusted to an equilibrium in the long run.
2.2 Dutch literature
In this paragraph we will focus on the house price models published by the following three institutes: CPB, CBS and Ortec Finance. The three models are based on the same
econometric Error Correction Model (ECM) but differ in details, explanatory variables and time interval. As stated before we will extensively review these three Dutch models.
The CPB, which is the Bureau for Economic Policy Analysis developed a model to explain house prices in the Netherlands. For the years 1980 to 2007 they have performed a long-run co-integrating acquaintance (Kranendonk & Verbruggen, April 2008). Their results show that house prices in the Netherlands react like expected to macro-economic supply and demand factors. Additionally, it is concluded that there is no overvaluation. The model can explain 75% off the variation in house prices. The long-term co-integration relationship (1) and the short-term relationship (2) are described below.
1 ""ℎ$= " &'+ "&)"*$+ "&+"",$-+ "&
.""/$+ "&0"1$+ "2$""345"6 = 1980, … . ,2007
(2) "∆ℎ$= " @)"∆*$A+ "@+""∆,$A+ "@."∆∇C$+ "@0"∆1$+ "@D"E+'''+ "@F"GHI$J)+ "@KGHI$J)L + 2$""
Var. Description Var. Description
H House price index ℎ$ MN(P$/R$)"""
Y Disposable aggregate labour income *$ MN(T$/R$)"""
I Long-term interest rates ,$- ,$"− ∇R$
W Other financial assets of households /$" ln"((X$/X$J))"/ 2 / R$)
S Total housing stock Y$ ln"((Y$/Y$J))"/ 2)
P Consumer Price Index ∇P$ R$"/"R$J)"-1
Income, real interest rate, wealth and housing stock have respectively an elasticity of 1.5, -6, 1.6, and -3. Although the research by the CPB is clear and the outcomes are reasonable, there are also some weaknesses. Some variables used by the CPB are not stationary, the Durbin Watson (DW) value is low and the CPB used a dummy for the year 2000 but did not exactly explain why they came up with this dummy.
In the research published by the OTB (Boelhouwer, Haffner, Neuteboom, & de Vries, 2001) it is argued that there are basically five factors which can explain house price developments. First the supply side can be important. Here price developments depend on the elasticity of supply and building costs. Second, speculative or psychological effects (households’
expectations). These effects can explain the price increases in the short term. It is argued that when people experience price increases this will influence the demand (and therefore prices) in the future. This is due to people willing to buy assets that are increasing in value, expecting prices to increase for some time to come. The third factor is the general economic
developments like income, prosperity and interest rates, which are important in explaining house prices. The fourth important factor is the development in demographics. Fifth is institutional policy like the availability and conditions for lending money. Also the
availability of new land and the fiscal policy is depended on government decisions. In the end the OTB came up with the below model to explain house prices in the short-term. The model is based on half yearly data from the period 1978 – 2000. Using a short-run ECM where only demand variables are used to explain house prices.
3 ""∇P$= "\'+ ""\)∇P$J)+ "\+,]]$J++ "\.E$+ "\0∇T$+ "\D∆,$-+ "2$
Variable: Description:
∇P$ Percentage change of real house prices, nominal house prices deflated by CPI
,]]$ After tax interest-to-income ratio defined by P$,$(1 − ^)/T$ ,(F is fiscal advantage) E$ Dummy variable for seasonal effects
∇T$ Percentage change of real household income
∆,$-"" (,$-− ",$J)- ) Absolute change in real interest rate
The lagged own house price index, interest-to-income, real interest and real income leads to respectively the following elasticity’s: 0.55, -0.19 -2.1, 0.56. On top of that it is found that house prices are 2.7% higher in the first half year, this is known as a seasonal effect. The OTB is using the percentage change of house prices instead of log differences. Furthermore, it does not include private wealth and housing stock. Housing stock is not included by arguing that prices are determined within the current stock, since the market is strongly regulated and the supply of land is scarce.
Francke (2009) argues that there are two types of models: These are the short-term models (demand models) and the medium- to long-term models (supply and demand models). In the
demand models the housing market is based on the existing houses and the only factors that can have an impact on the price in the short term are on the demand side. This because the supply of housing is inelastic in the short run. In the short run an affordability model can also be informative. In this model the focus is on price/income ratio or
mortgage-payments/income ratio. In the long run, the macroeconomic demand and supply model should be better in explaining house prices. Demand variables are: Interest rates, disposable income and borrowing capacity. Supply variables are: housing stock, construction costs, number of households and wealth. Francke (2009) used the models of the OTB and the CPB as a starting point to come with an improved model. The improved model is applied to yearly data over the period 1965Q1 until 2009Q1. Furthermore, in contrast to the models by OTB and CPB, supply variables are not included.
4 "∆ℎ$= " 0.6142∆ℎ$J)− 0.3149GHI$J)− "0.3002∆T$J)+ "0.0149∆,I$J)
5 "ℎ$∗= " −0.1360 + 0.9534*$− 0.0859,I$+0.0190Trend
Variables used are:
P$ Log real house prices
T$ Log real modal labor income per employee ,I$ Mortgage interest rate minus inflation
Only the real log labor income is found to be stationary. By estimating the normal ECM (with trend in long-run relation) the following results are found: Income elasticity = 0.95, interest elasticity = 1% extra interest leads to 8.6% lower prices. Furthermore, the linear trend shows a yearly increase of around 2% and is included to capture omitted fundamental factors. Predicted overvaluation using the ECM: 2006 = 11.9%, 2007 = 11.5% and 2008 = 2.4%. Francke argues that the variables are not stationary and the ECM with trend should be replaced by a random walk (with drift). When one sees overvaluation as the sum of Error Correction Term (ECT) and the random walk, then the overvaluation in the last decade is 10% to 38% (2007 and 2008). If only the Error Correction Term is used than the
overvaluation is between -13% (2001) and 15% (2008). In the end Francke (2009) argues that the ECM with random walk is the preferred model for explaining house prices.
2.3 Household expectations and housing variables
In this paragraph, we will review literature regarding household expectations and housing variables. Housing variables can be house prices and quantity of houses sold. As stated
before, households’ expectations are not used as explanatory variables in the current Dutch house price models. This in contrast to some international research, which find that
households’ expectations are informative when forecasting and explaining house prices. Therefore, we will discuss the literature regarding house prices and housing variables.
In the first three US studies by (Huang, 2013); (Roberto M. Croce, 2009); (Schneider, 2009); (Dua, 2005), the Good Time To Buy (GTTB) index is used to test for predictive power. This index is simply quantified data based on the question: ‘do you think now it is a good time or a bad time to buy a house?’ Respondents can only answer ‘bad’ or ‘good’, which is used in the following formula: 100 + %good -%bad. As a result, the index ranges between 0 and 200 where zero is very negative, 100 is neutral and 200 is very positive. This monthly time series data is being published since 1978.
In research by Huang (2013) it is argued that the housing booms and busts can be explained by households’ expectations. It is found that expectations are useful at forecasting house prices outside the sample. Huang defines bubbles as extreme price increases that are the result of expectations that are over-optimistic and diverge from fundamental macro economic values. These periods of price increases are usually followed by periods of price decreases as a result. In this study it is found that households’ expectations are informative for forecasting these bubbles and busts. By analysing the data, the researchers found a structural break which divides the data into two separate datasets with their own characteristics. Furthermore, the GTTB index is used as an threshold. Combining these elements resulted in a ‘Structural Break Treshold Vector Autoregression’ (SBTVAR). By Estimating this model, it is found that the GTTB can be used to forecast house prices outside the sample period.
Roberto (2009) Tested whether turning points in the housing market can be
forecasted. Roberto was especially interested in the following: New home permits, new home sales and home starts. In the research the NAHB Housing Market Index (HMI) and the Michigan survey, which is known as the GTTB index, are used as predicting variables. The found that the GTTB and the HMI as well, are causing the housing market variations as mentioned above. In the end is was concluded that the GTTB index (not the HMI) must be added to the leading indicators of explain housing market variables.
Dua (2005) examined the determinants of households’ expectations regarding the housing market and found that the GTTB is caused by prospective and retrospective interest rates, prospective and retrospective real disposable income and house prices.
Schneider (2009) found that only a small part of (optimistic) buyers can impact house prices very significantly. In his US study he used the Michigan Survey to measure the amount of ‘traders’ (buyers) which are optimistic about the housing market and future house prices. By taking the amount of optimistic buyers together he found that, while optimistic buyers only form a small part of the housing market, they can have a big impact on house prices. Three features / assumptions are important in this case: The price has to be the result of a negotiation process; where the price is the valuation of the (optimistic) buyer. Second the (optimistic) buyers needs to buy a huge amount / have a big share in the housing market; so they can drive (up) transaction prices. Third, there need to be costs (transaction costs) to keep the quantity of transactions low.
Lung (2011) used the turnover of assets as a proxy for confidence. By using this proxy, he was able to predict the bubbles and busts. The result is that households’ expectations can explain house prices in the US that are not correctly priced.
Saiz (2008) tested for the impact of supply variables on house price models. In the research it is argued that when supply is inelastic, optimistic buyers have more effect on prices than on the supply of new houses.
Shiller (2007) did a case study on trends in house prices and argued that
(psychological) expectations are the major cause of big (momentum) effects in housing. In the study it is found that people’s expectations can result in big price differences even while the fundamental macro economic variables have not changed.
Boumeester (2014) Tested for the relationship between household expectations and quantity of houses sold. He used the VEH (Vereniging Eigen Huis) Market indicator as a proxy for household expectations. This indicator is based on a monthly questionnaire filled in by 1600 respondents. He found interesting relationships between the VEH Market Indicator and the quantity of houses sold. In the article it is argued that the 7 months lagged indicator should be informative. In these seven months searching for a suitable house, the process of buying and the time to get the house available are accounted for. The correlation between the 7 months lagged indicator and the quantity of real transactions is 0.93. This means that the two variables are strong correlated. Regressing the quantity of houses sold on the 7 months lagged indicator, it shows highly significant results. But one very important basic assumption is not satisfied: since the regression is based on the nominal values, the data is non-stationary. This is one of the basic assumptions for time series regressions, which means that the results are not valid. In addition, he has used monthly data but did not control for seasonal effects.
2.4 Chapter summary
Depending on the purpose, house price models are based on OLS or ECM techniques. International factors that can explain house prices are very similar to the variables in the Netherlands. From the internationally reviewed literature the following factors are found to have explaining power: GDP, interest, previous house prices, construction costs, lagged appreciation, income and the actual vs the equilibrium price. The discussed Dutch models are all ECM where the following variables are used: house prices, income, interest, financial assets of households, housing stock and the consumer price index. Household expectations are used in none of the before mentioned models. Although there is no research where household expectations are used as explanatory variable in house price models, there are some studies regarding expectations and housing variables. Using household expectations Huang (2013) found that bubbles and busts are predictable. Boumeester (2014) found a significant relation between the VEH Market Indicator and the quantity of houses sold. But since the basic assumptions of time series regression are violated, there can be doubts regarding the validity of these outcomes.
3. Households’ expectations
To test the relationship between household expectations and Dutch house prices an
appropriate proxy for households’ expectations with respect to the housing market is needed. This proxy is preferably quantified monthly time series data, which measures the
‘temperature’ of the Dutch housing market and is useful in predicting Dutch house prices. We analysed two available indicators. The first is the VEH Market Indicator developed by the Technical University (TU) delft. The second is the consumer confidence published by the CBS, this index is not focused on housing but it might be interesting because the available time period is much longer than the market indicator.
3.1 VEH Market Indicator
Vereniging Eigen Huis (VEH), which is the association for (coming) homeowners, initiated the VEH market indicator. The indicator is developed because the VEH wanted to measure the ‘temperature’ on the housing market. The indicator is developed and maintained by the OTB from the Technical University of Delft (TU Delft). The data ranges from 2004 until now and is based on a questionnaire filled in by 1600 respondents. The questionnaire examines what households think of the following three factors:
1.! The general buying conditions (housing supply, regulation and lending standards) 2.! The development of selling prices
3.! The development of mortgage interest rates.
These three factors have to be answered in two scenarios: Based on the last 12 months (retrospective) and based on the coming 12 months (prospective). Answering is possible in the following way: Very negative, negative, neutral, positive and very positive. Where the scores are relatively 0, 50, 100, 150, 200. The average of these six questions determines the market indicator where 0 is very negative, 100 is neutral and 200 is very positive. It is interesting to see that this index has the same range as the GTTB index but is based on six questions instead of one.
In figure 1 the market indicator and the house price index are graphically presented, red for the price index, blue for the indicator. Note that the left y-axis represents the indicator values and the right y-axis represents the price values.
Figure 1. VEH Indicator and the house price index.
Looking at the indexes, we see that the development of the indicator is more volatile than the house price index. This is in line with Boumeester (2014) which found that people in general over-react on changes of the housing market. When looking at the indicator it shows some small declines, going from 100 points in 2005 to around 65 points mid-2008. Thereafter the confidence seems to recover but during the period 2010-2012 it drops to it’s lowest level, which is around 50 points. From 2012 to now, the confidence has been increasing again and has reached, with 110 points, its all-time high since the start in 2004. When we look into the house price index, we see an upward trend up to 2008. After 2008, it starts to decline until it reaches 85 points in 2013. This is 25% lower compared to the price level of 2008. After 2013 prices are not declining anymore, they are even rising slightly. The correlation between the indicator and the house price index is with -0.0412 negative and very low.
3.1.1 Chicken or Egg
When households see prices and/or transactions are increasing, their view (retrospective) and expectations (prospective) will most likely also increase. The question is whether the
indicator follows the real variables or the real variables follow the indicator. In the analysis by Boumeester (2014) it was found that expectations, especially in retrospective view, are connected with the real variables. But prospective expectations seem to diverge from the actual variables. This is also confirmed by the low and slightly negative correlation between the house price index and the indicator (-0.0412). When looking at graphed indicator in figure 1, we see that both indices behave quite different. It looks like the real variables follow
for out of sample forecasting because it seems to contain information that is not accounted for in the housing market yet. Although it becomes clear that house prices follow households’ expectations we will do some statistical robustness checks in chapter five to confirm this.
3.1.2 Regional Differences
The indicator is published on a national level, which can result in biased results on a disaggregate regional level. This notion is also supported by regional differences in house price developments and number of transactions. At this moment we see that prices and quantity of houses sold are quite stable and recently increasing in urban area’s. On the other hand, rural area’s still seem to suffer from decreased demand and low but stable or slightly decreasing prices. Ideally we would test regional prices versus the regional indicator to control for regional differences. But unfortunately the indicator is not published per region.
The probability of different expectations was also seen by the OTB, which resulted in a report (Boumeester, 2013) about the regional differences in the market indicator. To test for the regional differences the OTB almost tripled the respondents to approximately 5000 in the months October, November and December in 2013. In the analysis it was found that the indicator behaves differently per region. People in the more urban area’s seem to be more positive than people in the less urban and rural area’s. In some cases, the differences between the highest and the lowest score was even about 18 points (72,3 in Zeeland vs 90,6 in
Flevoland). From this information it seems clear that the indicator differs per region, but by t-testing the indicator, Boumeester (2013) found significant differences in only 18% of the regions, which means 82% is not significantly different from the average. Although 82% of the regional differences appear not to be significant, there has also been tests on factors which possible causes differences between the different regions. By testing for differences in
characteristics of respondents and differences in the housing market conditions it was found that these differences could only explain small parts of the variance between regions. Based on above mentioned, Boumeester (2013) argues that the indicator based on different regions would not lead to better or more accurate results.
3.2 The market indicator decomposed
By describing the VEH Market Indicator it becomes clear that it is built on three different factors and with two different scenarios (prospective and retrospective). On the aggregate it would probably be a very complete indicator with all-important aspects accounted for. But by using the indicator as proxy for household expectations it could be biased. First because we
want to focus on the coming months (prospective). Second because the underlying factors can cancel each other out. The question is whether the separate factors can bring us information that we cannot find in the real fundamental values and whether underlying factors can be better used separately or not. In order to test for this, we will decompose the indicator and analyse the sub questions separately. Since we are looking for consumer expectations with respect to the future, we will not analyse the retrospective parts. This results in the following parts to be analysed: interest rate expectations, expectations with respect to general
conditions, price expectations and interest, general conditions and price expectations together (prospective expectations).
3.2.1 Prospective price expectations
In figure 2 below, the expected house price (expected price), the indicator and the price index are shown.
Figure 2. ‘expected prices’ together with the VEH indicator and the house price index
The expected price shows an upward trend in the period April 2004 – Dec 2006, at this time the value of the index is 120, which means people expect prices to increase. After this long period of increasing expectations, the factor seems to stabilize. In the last quarter of 2007 it starts to decline and reaches its lowest point in December 2008 with a level of approximately 20 points. We can conclude that consumer expectations with respect to house prices
decreased dramatically during one year. It is very interesting to explore a turning point in the expected price between August 2007, and when the price index reaches its all-time high (107) in August 2008. This means that households are already expecting decreasing house prices in August 2007. When we compare house price expectations with the aggregate indicator, we
find some important differences. It seems that the aggregate starts to decline in May 2008, which is ten months later than the individual price indicator. Further the strength of the decrease is less extreme. In August 2007 the aggregate index is at 89 and in December 2008 is has declined to 67, which means it declined by 22 points. The correlation between the indicator and the price expectations is 0.84. The correlation between the price index and the price expectations is with 0.29 quite low.
3.2.2 Prospective general conditions
The factor ‘general buying conditions’, measures households’ expectations with respect to housing supply, regulation and lending standards. The data is included in figure 3 together with the indicator and price index. When looking at the index it seems that indicator and expected conditions do not move perfectly together
Figure 3. ‘expected conditions together with the VEH indicator and the house price index
We see stable expectations in the period 2004 – 2008, increasing conditions from December 2008 until August 2011, slightly decreasing expectations in 2012 and recovery in 2013. It looks like expected general buying conditions drives the aggregate index up in periods where price expectations are reaching very low levels. The correlation is between the indicator and expected conditions is 0.31 and the correlation between the price and expected conditions is -0.44
3.2.3 Prospective interest expectations
One can argue that interest rates expectations can explain house prices. In the analysis by Francke (2009) it was found that one percent higher interest rates results in approximately eight percent lower house prices. So when households’ expectations are in line with real
interest rates, this factor should be useful in predicting house prices. Below the expectations with respect to interest rates, the indicator and the house price index are shown in figure 4. The relation between the interest expectations and the indicator or house price index is hard to find. After declining slightly in the end of 2005, it looks quite stable for two years. This is remarkable because prices heavily increased during that period. In the end of 2008 it starts to increase to above 100 points while prices are already declining. In the beginning of 2009 it starts to decline slightly to approximately 70 points, which is in line with the price pattern. But after 2011 it starts to increase and at the end of 2014 it reaches the point where it was five years earlier.
Figure 4. ‘expected interest’ together with the VEH indicator and the house price index
The correlation between expected interest and indicator is -0.24. And the correlation between the expected interest and the house price index is -0.53
3.2.4 Aggregate prospective expectations
When taking expectations with respect to interest, price and conditions together we have the ‘aggregate prospective expectations’. In figure 5. we plotted the aggregate expectations together with the VEH Market Indicator and de house price index. We see that both indices follow the same route. The correlation coefficient of 0.97 confirms this notion. Since the two indices are almost perfectly correlated we will not consider the aggregate prospective
Figure 5. ‘Prospective expectations’ together with the VEH indicator and the house price index
3.3 Consumer confidence (CBS)
The CBS, which is the Central Bureau for Statistics in the Netherlands measures how people think about the current economic situation. It is based on five questions with respect to the general economic situation, whether it is a good time to buy durable goods and about the financial situation of households. The data is available for the period 1972 until now and published monthly. The consumer confidence published by the CBS is commonly used as the proxy of how consumers think about the economy. One could argue that when people are optimistic about their own financial situation and the economy, they would be more likely to buy a house. On the other hand, buying a sofa or television is still something very different from buying a house. One big advantage of the Consumer Confidence in relation to the VEH Market Indicator, is the available time period. The Consumer confidence contains almost 43 years of monthly observations. When looking at figure 6. we see that the CBS confidence is highly upward trending from 2006 until 2008. Where the VEH Market Indicator is slightly decreasing. This means that both indices diverge. The correlation coefficient between the CBS and the VEH Market Indicator of 0.56 confirms that both indices are not multicollineair and are not exchangeable.
Figure 6. The CBS consumer confidence and the VEH Market Indicator
3.4 Conclusion
In this chapter we where looking for the best proxy of households’ expectations with respect to the housing market. As mentioned, the CBS consumer confidence covers an extensive time period and is available for almost 43 years. But unfortunately it does not directly measure housing market expectations. The correlation coefficient of 0.56 between the CBS and the VEH Market Indicator confirms that both are not exchangeable. The Market Indicator measures the temperature on the housing market but is unfortunately only available for ten years. This research will therefore be based on a limited time period.
From the analysis it seems that the VEH Market Indicator reveals additional information on the common fundamental housing variables. By testing the underlying prospective parts of the indicator it is found that all of them are informative. When we take the prospective parts together and compare this with the indicator we see that both are highly correlated. Therefore, we conclude that the VEH Market Indicator is the best proxy for households’ expectations with respect to the housing market.
4. Methodology and data
In the first paragraph the methodology and models in line with the second and third research question will be specified. Paragraph two will present the data and sources in further detail.
4.1 Methodology
The first subparagraph will focus on the model which will be used to forecast house prices with the VEH Market Indicator. In the second subparagraph the model will be specified in order to test the additional explaining power of the VEH Market Indicator in existing house price models.
4.1.1. Forecasting with household expectations
Below the OLS specification to test the forecasting power of households’ expectations on house prices is presented. This will be used to answer the second research question. The model will be estimated on a subsample of the complete dataset to test the forecasting power on real data. The dependent variable is the log difference of the real House Price Index and the independent variable is the log difference of the indicator. Log differences are taken to control for the non-stationarity of the original time series.
(1) ∆P$= " @'+ "@)∆G$Jc+ "2$ where j= 1,2,3…,8
Where:
∆P$ log difference of the real House Price Index ∆G log difference of the indicator
Different lags of the indicator will be tested as explanatory variable in order to get the best fitted model.
4.1.2. Households’ expectations as explanatory variable
Households’ expectations are often referred to as bubble and bust builders. Therefore, we would expect that adding households’ expectations as explanatory variable will improve house prices models. The specifications below will be estimated in order to test this
hypothesis. The variables in the specification by Francke (2009) are taken as a starting point. In contrast with the specification by Francke (2009) the model will be estimated using OLS as the log differences are taken to obtain stationarity. We are not looking for over- or
undervaluation and one could also argue that the analysed period is to short to find under- or overvaluation. Another difference is the analysed period and frequency of the data (quarterly versus yearly), but we will further elaborate on this in the second paragraph.
Below one can see three different OLS specifications. The first specification (1) will be estimated with the variables used by Francke (2009): price, interest, and income.
In the second specification (2) some other variables from national and international literature will be tested in addition to the variables used in the first specification. This will lead to the current ‘best’ model given the available period and frequency (quarterly data).
In the third (3) and last specification we will ad the VEH Market Indicator to the second specification (2). This will test whether the indicator can add explaining power to the current ‘best’ model. The above models will be estimated on a subsample of the complete dataset to test the forecasting power on real data.
The models:
(1) ∆P$= " @'+ @)∆P$Jc− "@+∆T$Jc+ "@.∆,I$Jc+"@0Y + "2$"
(2) ∆P$= " @'+ @)∆P$Jc− "@+∆T$Jc+ "@.∆,I$Jc+ "@0∆d$Jc+ "@D∆e$Jc+ "@F∆H$Jc+ "@0Y"+"2$"
(3) ∆P$= " @'+ @)∆P$Jc− "@+∆T$Jc+ "@.∆,I$Jc+ "@0∆d$Jc+ "@D∆e$Jc+ "@F∆H$Jc+ @K∆G$Jc+ "@0Y + "2$"
where j= 1,2,3…,8
∆P$ log difference of the real House Price Index
∆T$ log difference of real disposable income per household
∆,I$ log difference of real mortgage interest rate ∆d$ log difference unemployment
∆e$ log difference of real GDP
∆H$ log difference of real construction costs ∆G$ log difference of the indicator
4.2 Data
Due to the limited availability of the VEH Market Indicator, the sample period of the data is 2004Q2 until 2014Q4. Summary statistics can be found in appendix 1 & 2
Indicator The proxy for households’ expectations is the VEH Market indicator (Lamain & Boumeester, 2015) which is monthly quoted. To obtain quarterly data the data from months 3, 6, 9, and 12 are taken.
Interest The interest rate is the Annualised Agreed Rate (AAR) and is used as interest rate of loans for house purchases, 10 years fixed. These rates are reported monthly and available at the ECB
(http://sdw.ecb.europa.eu). To obtain quarterly data the data from months 3, 6, 9, and 12 are taken.
The macro economic data was downloaded from the CBS via Statline
(http://statline.cbs.nl/Statweb/). The CBS is the Central Bureau of Statistics in the
Netherlands and provides yearly, quarterly or monthly data. The available frequency and time period depends on the index.
House prices The index provides aggregate monthly and quarterly house prices. Prices are quoted in relation to the price level in 2010.
Income Average disposable household income. Prices are annually quoted in thousands and available until 2013. The data is interpolated to quarterly data by simply dividing the yearly growth by four.
BBP Real Gross Domestic Product in the Netherlands. Quarterly quoted in thousands.
Unemployment This is the percentage unemployment regarding the working population and is quoted quarterly.
Construction costs The output index is taken, which is available in annual and quarterly data as well.
CPI The CPI is the Consumer Price Index and used to deflate price levels to ‘real prices’.
5. Results
In this chapter the specified models will be estimated and there will be elaborated on the outcomes to answer the research questions. Paragraph one focuses on the data preparation. Paragraph two focuses on the second research question: Can we use household expectations to predict house prices? The third and last paragraph focuses on the third question: Are households’ expectations informative in house price models?
5.1 Data preparation
To get real price levels, the financial data (prices, construction costs and disposable
household income) is deflated by the Consumer Price Index (CPI). For the mortgage interest rate, we took the percentage growth of the CPI and subtracted this from the interest rate. See appendix 1 & 2 for the analysed series in nominal and real values. From now on, we will proceed with the acquired real variables from the previous step. In the next two
subparagraphs we will elaborate on the assumptions regarding time series regression and the lag selection.
5.1.1 Assumptions Time Series Regression
In order to get unbiased and reliable estimation results, the following assumptions have to be satisfied: The expected value of the error term has a conditional mean of zero, the data may not have large outliers, there should be no perfect multicollinearity and the data has to be stationary. The first assumption will be tested after regressing, the second is confirmed by the graphs in appendix 2, since there are no outliers.
Third, there are no problematic correlations which are much higher than 0.8 (see appendix 6). So the existence of multicollinarity can be rejected.
The fourth, all (real) variables (except the GDP series) are non-stationary. This is confirmed by testing with the Dicky Fuller (DF) test. In order to make the time series variables stationary, we took the log difference. The DF test confirms that all the time series are stationary after transforming. See the table in appendix 2 & 3 for the Dicky Fuller statistics and a graphical presentation of the data before and after the log difference transformation.
5.1.2 Lag selection
The change in values of explaining housing variables have a typically delayed effect on house prices. The reviewed models in the literature are based on yearly data while this research is based on quarterly data. Therefore, the delay (lag) in this research can be different. We used theory, previous research, the Bayes Information Criterion (BIC) and the Akaike Information Criterion (AIC) to select the lags. Only the VEH Market Indicator and the variables used by Francke (2009) are explained below. The statistics for the other variables are displayed in appendix 4.
Market indicator According to Boumeester (2014), the lag is 7 months. Within our quarterly dataset the third lag (which includes 7 months) and the fifth are significant.
Price index The second and third lag are significant. Significance of lags more than one year is not found. This is in accordance with Francke (2009) where the one year lagged value was used Interest Only the first lag is significant. This is in accordance with
Francke (2009) who used one year lagged values.
Income The fourth lag seems to reveal the most significant explaining power. This is in accordance with Francke (2009) who used the one year lagged value.
5.2 Forecasting with household expectations
In this paragraph, the forecasting power of the VEH indicator will be examined. The model will be estimated as specified in 4.1.1. This will be based on the data until the second quarter of 2013 (2013Q2). Consequently, these estimation results are used to (pseudo out of sample) forecast the changes in the house price index (subparagraph two). This forecast will be projected on the period covering third quarter of 2013 (2013Q3) to the fourth quarter of 2014 (2014Q4).
5.2.1. The estimated model
See appendix 5 for the estimation results of the model as specified in paragraph 4.1.1. The third and the fifth lag (9 months and 15 months) are significant with coefficients of 0.0621 and 0.0818 respectively. Both lags are positive, which is in line with the theory. The significance of the third lag can be explained by theory and is confirmed by Boumeester
(2014). The significant fifth lag is equal to 15 months and cannot be explained by theory. On the other hand, it is not unthinkable that prices lag 15 months to changing expectations. This because of the typical inertia of house prices. House prices are often referred as lagging and smoothing. In addition the process of buying a house can take some time, which can even exceed the 7 months as stated by Boumeester (2014). The F-value of the model is 7.55 and the R2 is 0.39 which means that approximately 39% of the variation in the price index can be explained by the Market Indicator. Because the constant is not significant and the dependent variable is a percentage growth we also estimated the model without a constant. This model seems to fit even better with an R2 of 0.46 and a F-value of 12.38.
The ARCH LM test does not reject the H0 of no conditional heteroskedasticity
(p=0.0661). This indicates homoscedastic standard errors which means that all OLS
assumptions are satisfied. However, the Durbin Watson (DW) statistic is with 1.05 low which indicates autocorrelation that is the result of an incomplete model. The model can therefore be considered as weak and other variables should be included to get a model that fits better.
The final estimation results from model without a constant (1) are shown below. See table 1. For the complete specification.
(1) ∆P$= "0.1187291∆G$J.+ 0.1532047∆G$JD
Where:
∆P$ log difference of the real House Price Index ∆G log difference of the indicator
Since the dependent and independent variables are the differences of the logs it can be seen as an elasticity. When the three or five quarters lagged indicator goes up with 1% the price will increase by approximately 0.12% or 0.15% respectively. The coefficients look small but one has to consider two important facts: First the indicator is very volatile and has shown
quarterly percentage changes between -14% and +27%. Second one has to keep in mind that these growth rates are quarterly, multiplying the growth rates by four, places the outcome in a complete different perspective. When the third and fifth lag both decrease by 14%, the
resulting price drop will be 3.8%. When the indicator increases by 27% in the third and fifth lag, the resulting price growth will be 7.3%. Considering an average priced house (ca €222K), the impact would be between ca. -8K and +€16K.
Table 1. Regression results Price (1) Price (2) L3.indicator 0.104** 0.119** (0.009) (0.003) L5.indicator 0.134** 0.153*** (0.003) (0.001) _cons -0.00437 (0.099) N 31 31 F 8.769 12.38 df_m 2 2 df_r 28 29 r2 0.385 0.46 DW 1.04 1.05 ARCH LM (p-value) 0.5278 0.0661 p-values in parentheses * p<0.05, ** p<0.01, *** p<0.001
5.2.2. Forecasting with the estimated coefficients
The estimated model (1) with the three and five quarters lagged VEH Market Indicator is used to predict the house price changes. The forecast is applied to the period 2013Q3 until 2014Q4. In figure 7 one can see the percentage change in the price index (blue) and the forecast (red).
Figure 7. Forecasted versus Real Figure 8. Forecast including confidence interval vs real
It seems that the forecast fits the real data quite well. All the real data lies between the confidence interval of the prediction. The correlation between the forecast and the real price changes is 0.45. The real house prices are within the forecast confidence interval. Therefore, we can carefully conclude that the VEH Market Indicator has some forecasting power.
5.3 Households’ expectations as explanatory variable
In this paragraph the models as specified in paragraph 4.1.2. will be estimated. The first specification (2) (see model 2 in table 2) is based on the variables used by Francke (2009). It is found that the second lag of the own price index and the fourth lag of income are
significant (p=0.00 and p=0.006, respectively). The resulting coefficients for the model without a constant are: 0.602 for price (lag 2) and 0.598 for income (lag 4). The R2 of the model is 0.58 and the DW statistic is 1.45. We could not find evidence of a significant relation between the interest rate and the price index. This can be caused by the atypical behaviour of the price index compared to the interest rate in the analysed time period. Normally prices would increase when interest rates are going downward. In the analysed period we don’t see such a behaviour.
In the second specification (3) we add Construction costs, GDP, unemployment and income to the model (see model 4 in table 2). Only unemployment can significant explain some variation but the significance of the whole specification declines slightly. The resulting model has an R2 of 0.65 and the DW statistic is 1.41.
-. 0 2 0 .02 .04 .06 2013q3 2013q4 2014q1 2014q2 2014q3 2014q4 date
price Fitted values
-. 0 5 0 .05 -.04 -.02 0 .02 .04 .06 Fitted values 95% CI Fitted values price
In the third specification (4) we add the VEH Market Indicator as an explanatory variable (see model 6 in table 2). Adding the third and fifth lag of the Indicator yields a model that explains 74% of the variation in price. By using quarterly dummy variables, we also tested for seasonality but we did not find any significant effects. The DW statistic rejects the presence of autocorrelation (DW=1.75). The ARCH LM test cannot reject the H0 of no conditional heteroskedasticity (p=0.78). As a result, the conditional variance of the error term is constant and an ARCH or GARCH model (to correct for conditional variance) model does not converge. The estimated models are shown below. For a complete specification see table 1.
(2) ∆P$= " 0.602∆ℎ$J++ "0.685∆T$J0"
(3) ∆P$= " 0.592∆ℎ$J++ "0.460∆T$J0− 0.0611∆d$J.""
(4) ∆P$= " 0.506∆ℎ$J.+ "0.569∆T$J0+ 0.103∆G$J.+ 0.0973∆G$JD""
∆P$ log difference of the real House Price Index
∆T$ log difference of real disposable income per household
∆,I$ log difference of real mortgage interest rate
∆d$ log difference unemployment
∆e$ log difference BBP (GDP)
∆H$ log difference Construction costs ∆G$ log difference of the indicator
∆Y$ Dummy variable per quarter
Since the variables are in logarithms one can read the outcomes as elasticities. Using the third model (4) (see model 6 in table 2) the interpretation is as follows: When the three quarters lagged house price goes up with 1% the house price index will increase with 0.506%. When the four quarters lagged income increases with 1% the resulting house price change is 0.57%. When the third and fifth lagged VEH Market Indicator increases with 1% the resulting house price change is 0.10% and 0.10% respectively.
Table 2. Regression results
Price (1) Price (2) Price (3) Price (4) Price (5) Price (6) L2.price 0.459** 0.602*** 0.456** 0.592*** (0.003) (0.000) (0.001) (0.000) L3.price 0.433** 0.506*** (0.001) (0.000) L4.income 0.685** 0.598** 0.549** 0.460* 0.640** 0.569** (0.001) (0.006) (0.005) (0.025) (0.001) (0.003) L.unemployed -0.0585* -0.0611* (0.015) (0.020) L3.indicator 0.0836* 0.103** (0.012) (0.002) L5.indicator 0.0844* 0.0973** (0.011) (0.004) _cons -0.00547* -0.00525* -0.00333 (0.015) (0.012) (0.0116) N 32 32 32 32 31 31 F 19.16 20.27 17.56 17.83 15.6 19.46 df_m 2 2 3 3 4 4 df_r 29 30 28 29 26 27 r2 0.569 0.575 0.653 0.648 0.706 0.742 DW 1.56 1.4497 1.6266 1.4063 1.8811 1.7498 ARCH LM (p-value) 0.3372 0.8971 0.3634 0.5646 0.5464 0.7820 p-values in parentheses * p<0.05, ** p<0.01, *** p<0.001
The estimated model (4) with the three and five quarters lagged VEH Market Indicator is used to forecast the house price changes. The forecast is applied to the period 2013Q3 until 2014Q4. In figure 9 one can see the percentage change in the price index (blue) and the forecast (red).
Figure 7. Forecasted versus Real Figure 8. Forecast including confidence interval vs real
It seems that the forecast fits the real data quite well. All the real data lies between the confidence interval of the prediction. The correlation between the forecast and the real price changes is 0.62. The real house prices are within the confidence interval of the forecast.
The VEH Market Indicator performs well as an explanatory variable in reviewed house price model. The R2 and the F-value increases significantly when the VEH Market Indicator is added as explanatory variable. The DW statistic does not indicate autocorrelation and the ARCH LM test cannot reject the hypothesis of no heteroskedasticity. Also the pseudo out of sample forecast on the data seems to fit well. Therefore, we can conclude that the VEH Market indicator is informative in house price models and can be used to predict house prices. -. 0 4 -. 0 2 0 .02 .04 2013q3 2013q4 2014q1 2014q2 2014q3 2014q4 date
Fitted values price
-. 0 6 -. 0 4 -. 0 2 0 .02 .04 -.04 -.02 0 .02 .04 Fitted values 95% CI Fitted values price
6. Conclusion
The purpose of this study was to test whether households’ expectations can be used to predict house prices. Furthermore, it is also tested whether households’ expectations are informative in house price models.
We started with looking for the best proxy according to households’ expectations and found that the VEH Market Indicator is the best proxy to use. One disadvantage is the limited available time period because the data is only available from 2004Q2 to 2014Q4.
By using OLS on the time series data it is found that the third and fifth lagged values of the VEH Market Indicator are significant in explaining house prices. The real price changes are within the confidence interval of the forecast and the correlation between the price index and the forecast is 0.45. We can therefore carefully conclude that the VEH Market Indicator has forecasting power but the DW statistic indicates that other explanatory variables need to be included.
By adding the VEH Market Indicator in the reviewed house price model it is found that the third and fifth lagged values are significant. The explained variance increases with ca 10 percentage points to 74% and the significance does also increase. This indicates that the VEH Market Indicator is informative in house price models and can, together with income, lagged and house prices be used to predict house prices in the short term.
The above results are significant and the size of the impact is substantial. But one has to consider that the models are based on a short period which includes the credit crisis, the flash crash and the recession. Further research could be done on the existence of bubbles and burst and how households’ expectations can predict or explain these.
Bibliography
Abraham, J., & Henderschott, P. (1993). Patterns and determinants of metropolitan house
prices 1977-1991. Boston: Federal reserve bank of Boston.
Berg, L. (1995). Housing and financial wealth. Sweden: Scandinavian Journal of economics. Boelhouwer, P., Haffner, M., Neuteboom, P., & de Vries, P. (2001). Koopprijsontwikkeling en
de fiscale behandeling van het eigen huis. Delft: OTB research institute TU Delft.
Boumeester, H. (2014). Nadere analyse verloop en betekenis van de Eigen Huis
Marktindicator. Delft: OTB - TU Delft.
Boumeester, H. (2013). Regionale verschillen in de Eigen Huis Marktindicator. Delft: OTB - TU Delft.
Brodin, P., & Nymoen, R. (1992). Wealth Effects and Exogeneity: The Norwegian
Consumption Function 1966(1)-1989(4)’,. Oxford Bulletin of Economics and
Statistics 54(3): 431-54.
Case, K., Quigly, J., & Shiller, R. (2005). Comparing wealth effects.
Case, K., Quigly, J., & Shiller, R. (2013). Wealth effects revised. Cambridge: The national bureau of economic research.
CBS. (2011). Woningwaarde tweemaal zo hoog als hypotheekschuld.
Chung E., H. Consumer sentiment and housing market activities: Home sales and housing
starts in US . Housing studies review, volume18 (4).
Diaz, A., & Luengo-Prado, M. (2010). Homeownership: Economic Benefits. Madrid: Universidad Carlos III de Madrid.
Dua, P. (2005). Analysis of Consumers’ Perceptions of Buying Conditions for Houses. Journal of real estate finance and economics.
Englund, P., & Ioannides, Y. (1997). House Price Dynamics: An International Emperical
Perspective. Journal of housing economics 6 .
Francke. (2009). Evaluation of house price models using an ECM approach: the case of the
netherlands. Amsterdam: Ortec finance.
Fuss. (2010). Macroeconomic Determinants of International Housing Markets.
Huang, M. (2013). Bubble-like housing boom–bust cycles: Evidence from the predictive
power of households’ expectations. US: The Quarterly Review of Economics and
Finance.
Johan Verbruggen, H. K. (2005). Welke factoren bepalen de ontwikkeling van de
Karl E. Case, R. J. (1990). Forecasting Prices and Excess Returns in the Housing Market. US: Real Estate Economics, vol. 18, no. 3, pp. 253-273,.
Kranendonk, H., & Verbruggen, J. (April 2008). Are houses overvalued in the netherlands. Den Haag: CPB.
Kuzmanovic, M. (2012). Can consumer confidence data predict real variables. London: European Bank for reconstruction and development.
Lamain, C., & Boumeester, H. (2015). Eigen Huis marktindicator 4e kwartaal 2004 tot en
met 4e kwartaal 2014. Delft: OTB - Onderzoek voor de gebouwde omgeving, TU
delft.
Lung, D. K. (2011). Explaining Asset Mispricing Using the Resale Option and Inflation Illusion. Real Estate Economics V39 , 313–344.
Mankiw, N. G. (1989). The baby boom, the baby bust and the housing market. US: Eur. Econ. Rev 36, 549-558.
Miles, D. K. (1993). House Prices, Personal Sector Wealth and Consumption: Some
Conceptual and Empirical Issues. London: The Manchester school of economic and
social studies 01/1993; 61(Suppl.):35-59.
Poterba, J. (1991). House price dynamics. US: brookings pap. econ. activity, 2143-203. poterba, J. M. (1984). Tax subsidies to owner-occupied housing: An asset market approach.
Quart. J. Econ. 99, 729-752.
Prakken, E. B. (2004). Housing Wealth Effects. Harvard & supported by National association of Realtors.
Roberto M. Croce, D. R. (2009). Predicting turning points in the housing market. Journal of
housing economics .
Saiz, E. L. (2008). Housing supply and housing bubbles. Cambridge MA: National bureau of economic research.
Schneider, M. P. (2009). Momentum traders in the housing market. Cambridge, MA: National bureau of economic research.
Shiller, R. J. (2007). Understanding Recent Trends in house prices and Homeownership. US: Housing finance.
Sinai, T., & Souleles, N. S. (2005). Owner-Occupied Housing as a hedge against rent risk.
Quarterly journal of economics .
Skiba, B. a. (2011). Momentum in residential real estate. United states: Journal of Real Estate Finance and Economics, Vol. 43, No. 3, 2011 .
Appendices
Appendix 1. Descriptive statistics
Nominal values Obs Mean Std. Dev. Min Max
price 43 96.06 6.77 84.70 106.70 interest 43 4.37 0.61 3.16 5.56 income 39 32.03 1.73 28.65 33.60 bbp 43 155613.10 6394.67 138327.00 164761.00 bcosts 43 125.64 6.94 112.70 139.10 unemployed 43 5.38 1.20 3.40 8.10 indicator 43 79.13 15.29 50.40 104.70 real values real price 43 88.09 9.48 71.37 99.28 real interest 43 3.27 0.63 1.97 4.48 real income 39 29.50 0.87 28.17 31.06 real bbp 43 142249.20 7019.39 128293.90 156971.80 real bcosts 43 115.09 10.06 94.41 130.00 log differences ln_price 42 -0.005 0.015 -0.053 0.014 ln_interest 42 -0.012 0.052 -0.129 0.097 ln_income 38 0.000 0.011 -0.019 0.018 ln_bbp 42 -0.001 0.047 -0.073 0.076 ln_bcosts 42 -0.005 0.026 -0.071 0.056 ln_unemployed 42 0.005 0.080 -0.158 0.182 ln_indicator 42 0.003 0.081 -0.136 0.267
Appendix 2. Analyzed series
1. Nominal and real prices 2. Log differences of real prices
3. Nominal and real interest 4. Log differences of real interest
5. Indicator 6. Log differences of indicator
7. Nominal and real income 8. Log differences of real income
70 80 90 100 11 0 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date price realprice -. 0 6 -. 0 4 -. 0 2 0 .02 ln _ p ri ce 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date 2 3 4 5 6 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date interest realinterest -. 1 5 -. 1 -. 0 5 0 .05 .1 ln _ in te re st 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date 40 60 80 100 in d ica to r 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date -. 1 0 .1 .2 .3 ln _ in d ica to r 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date 28 30 32 34 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date income realincome -. 0 2 -. 0 1 0 .01 .02 ln _ in co me 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date
9. Real bbp 10. Log differences of real bbp
11. Nominal and Real Construction costs 12. Log differences of real Construction costs
13. Nominal and Real unemployment 14. Log differences of unemployment
140000 145000 150000 155000 160000 165000 bbp 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date -. 0 5 0 .05 .1 ln_bbp 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date 90 100 11 0 120 130 140 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date constructioncosts realconstructioncosts -. 1 -. 0 5 0 .05 ln _ co n st ru ct io n co st s 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date 3 4 5 6 7 8 u n e mp lo ye d 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date -. 2 -. 1 0 .1 .2 ln _ u n e mp lo ye d 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 date
Appendix 3. Dickey Fuller statistics (rejection region is +/- 3.563)
Series log difference log difference, trend
Indicator -0.784 -5.420 -5.622 Interest 0.337 -3.611 -3.984 Price 1.054 -4.108 -4.567 Income -1.044 -7.789 -9.019 BBP -3.357 -18.617 -18.862 Building costs -0.075 -9.22 -10.153 Unemployed -0.673 -5.406 -5.483
Appendix 4. AIC, BIC and single regression results
AIC BIC Significant lags (coefficients between brackets)
Price 2 2 2 + 4 (0,55 + 0.47) Indicator 8 0 3 + 5 (0.07 + 0.08) Interest 3 1 1 (0.101) Construction costs 4 1 4 (0.22) BBP 5 4 4 (0.144) Unemployment 6 2 1 (-0.077) Income 8 0 4
Appendix 5. Autocorrelation and Partial Autocorrelation of the residuals Model 1 as estimated in sub paragraph 5.2.1:
Autocorrelation Partial autocorrelation
Model 3 as estimated in subparagraph 5.3:
Autocorrelation Partial autocorrelation
-0 .5 0 0.00 0.50 Au to co rre la tio n s o f re s 0 5 10 15 Lag
Bartlett's formula for MA(q) 95% confidence bands
-0 .4 0 -0 .2 0 0.00 0.20 0.40 0.60 Pa rt ia l a u to co rre la tio n s o f re s 0 5 10 15 Lag
95% Confidence bands [se = 1/sqrt(n)]
-0 .4 0 -0 .2 0 0.00 0.20 0.40 Au to co rre la tio n s o f re s2 0 5 10 15 Lag
Bartlett's formula for MA(q) 95% confidence bands
-0 .4 0 -0 .2 0 0.00 0.20 0.40 0.60 Pa rt ia l a u to co rre la tio n s o f re s2 0 5 10 15 Lag
Appendix 6. Correlation matrix log differences
Price Interest Income BBP C - Costs Unemployment Indicator CBS
Price 1.00 Interest 0.38 1.00 Income 0.70 0.24 1.00 BBP 0.49 -0.01 0.81 1.00 Construction Costs 0.54 0.05 0.66 0.59 1.00 Unemployment 0.51 -0.49 -0.44 0.11 -0.31 1.00 Indicator 0.27 -0.40 -0.15 0.24 -0.36 0.14 1.00 CBS 0.51 0.65 0.37 0.12 0.13 -0.43 -0.14 1.00
