The role of house prices in the monetary transmission mechanism
in the Netherlands
Master Thesis
Hubrecht Jan de Rijke 5970512
Master of Economics
Specialization: Monetary Policy and Banking Supervisor: Lex Hoogduin
1 Abstract
This paper adds insight to the lack of consensus on the role of house prices in the monetary transmission mechanism by analyzing studies with conflicting results. Besides that, understanding is provided through the construction of a vector autoregression model similar to one of the conflicting studies. However, this model is applied to a different timeframe and linked to developments in the Dutch housing market. The results reveal an increase in the forecasting value of money market fluctuations with respect to house price behavior, which can be explained by increasing loan-to-value ratios over time with respect to mortgages and a high share of interest-only mortgages prevalent in the Dutch economy. Furthermore, the counterfactual analyses reveal a significant role for house prices in the monetary transmission mechanism in the Netherlands. However, the magnitude of households‟ negative response to a decrease in housing wealth caused by money market fluctuations seems to be dependent on the established level of housing wealth. This can be explained by households behaving according to a utility function with respect to housing wealth that has a concave shape and decreasing marginal utility.
2 Table of contents
1. Introduction ... 3
2. Literature review ... 3
2.1 Monetary transmission mechanism ... 3
2.2 Housing market and the monetary transmission mechanism ... 4
2.3 Empirical evidence ... 5
2.4 Characteristics of the Dutch housing market ... 8
2.5 Consequences for the Dutch economy ... 10
3 Methodology ... 11
3.1 Data ... 11
3.2 Vector auto regression ... 11
3.3 Identification ... 13
3.4 Stationarity, cointegration, breaks and lags ... 14
4. Results ... 14
4.1 Base-line model ... 14
4.2 Forecast error variance decomposition ... 17
4.3 Base-line model extended with consumption ... 17
4.4 Counterfactual analysis ... 20
5. Conclusion ... 23
3 1. Introduction
In the aftermath of the recent financial crisis, Dutch home-owners experienced an erosion of their housing wealth, house prices declined by 4% annually on average in the post-crisis years. Furthermore, the Dutch society has a long history of political discussions regarding its mortgage debt relief program, which has led to a large share of the population currently being home-owners. This prominent role of real estate in the Dutch economy gave rise to questions regarding the dynamics involved. Therefore, this paper studies the role of house prices in the monetary transmission mechanism (MTM) in the Netherlands.
Previous studies mostly focused on comparing the role of house prices in the MTM across European countries. Unfortunately, their results did not provide consensus on the topic. Besides that, the construction of general conclusions is further complicated due to differing methodological approaches, since most studies employ a vector autoregressive model but take different assumptions into account. This paper tries to overcome the lack of consensus by comparing two conflicting studies with similar foundations and constructing a third model similar to one of the aforementioned studies and applying it to a recent time-frame.
The main results show an increasing role for credit markets with respect to their influence on house price behavior, possibly caused by higher loan-to-value (LTV) ratios and and a high share of interest-only mortgages prevalent in the Dutch economy. Furthermore, house prices play a significant role in the Dutch economy. However, households are likely to behave according to a utility function with respect to housing wealth that has a concave shape and decreasing marginal utility. Because the magnitude of a households‟ negative response to a decrease in housing wealth caused by money market fluctuations seems to be dependent on the established level of housing wealth.
The remainder of the paper is organized as follows: section 2 provides a literature review on topics related to the role of house prices in the MTM, starting with an analysis of literature on the MTM itself. Thereafter, empirical evidence on the role of house prices in the MTM is discussed and compared. Furthermore, the characteristics of the Dutch housing market are described. Finally, the characteristics of the Dutch housing market are linked to their behavior the Dutch economy.
Section 3 explains the methodology, beginning with a data description, followed by the characteristics of vector autoregressive models and their identification. Furthermore, properties with respect to stationarity, cointegration, breaks and lags are presented. Section 4 demonstrates the results and section 5 concludes.
2. Literature review
2.1 Monetary transmission mechanism
The MTM describes how changes in the money stock and/or interest rates, motivated by monetary policy objectives, affect real economic variables such as inflation, output and employment. Currently, the main objective of the European Central Bank (ECB) is to maintain price stability, which means the ECB aims at inflation rates of below, but close to, 2% over the medium term1. Price stability is thought to be beneficial since it improves the transparency of the price mechanism and it characterizes a mitigation of distortions arising from business cycles that shift output levels away from their natural rate.
A central bank‟s main „traditional‟ tool to influence interest rates is through changing the monetary base, a segment of the money supply which consists of currency in circulation and banking system reserves. The monetary base alters when the central bank purchases or sells government
1
4 securities in the open market, this process changes nominal interest rates since these are determined through a process of money supply and money demand. Besides that, the long-term real interest decreases due to future inflation expectations following expansionary monetary policy. This lowers the cost of capital, evoking higher investment spending by firms and households, which ultimately leads to an increase in aggregate demand (Mishkin, 2007b).
The reach of monetary policy is not limited to altering interest rates. Firstly, a change in interest rates generates international currency flows. For example, lower interest rates lead to depreciation, making exports cheaper, which increases aggregate demand. Secondly, expansionary monetary policy induces households to spend their „extra‟ money in the stock market, which leads to higher stock prices. High stock prices increase investment since the cost of capital is low relative to the market value of the firm, a theory known as Tobin‟s q. Furthermore, an increase in stock prices raises the financial well-being of households. Modigliani‟s life-cycle theory illustrates that households spread their wealth over their lifetime, therefore consumption increases after an increase in financial wealth (Mishkin, 2007b).
Another transmission channel, the liquidity effect, operates through household‟s desire to purchase durables and houses. Due to their illiquid nature, durables and houses are hard to sell during times of financial hardship. Therefore, if possibilities of financial distress are looming on the horizon, households prefer not to hold durables and houses, since forced sales would lead to getting a low return. Thus, expansionary monetary policy increases household‟s financial wealth, which reduces the chance of financial distress, which increases spending on houses and durables (Mishkin, 2007b). Initial reduced-form evidence for the transmission mechanism was presented in the early 1960s by the monetarist movement led by Milton Friedman. Friedman found that money growth has a lagged effect on output, on average a peak in the growth rate was followed by a peak in output 16 months later. Furthermore, he constructed a model that was supposed to be superior to the traditional Keynesian model that ruled economics at that time. This model demonstrated a correlation between money supply and output. Historical evidence can be found where money affects the real economy. For example, interest rates preceding the Great Depression were assumed to be high and an increase in reverse requirements by the Federal Reserve was followed by a severe recession of 1937-1938 in the U.S (Mishkin, 2007b).
Subsequent research was conducted by Bernanke & Blinder (1992), who employed a structural vector auto regression (SVAR) with innovations behaving as policy shocks. A shock to the federal funds rate, the difference between the observed value and the optimal forecast value, creates an impulse response in other variables, enabling one to observe the MTM. Granger-causality tests prove that rather the federal funds rate can be used than money aggregates in order to forecast economic responses to policy measures.
Furthermore, Bernanke & Blinder (1992) illustrate that the federal funds rate is not contemporaneously influenced by economic variables, but rather by their lagged values. Acknowledging this, they uncover that tight monetary policy leads to a short-run sell-off of securities, a decrease in loans and an increase in unemployment. With both the latter taking more time to respond to a policy shock than the former.
2.2 Housing market and the monetary transmission mechanism
In order to study the role of house prices in the transmission mechanism, one must evaluate their dynamics regarding interest rates and aggregate demand. Monetary policy shapes housing market behavior in several fashions. For example, a decrease in real interest rates reduces the cost of purchasing real estate, which leads to higher demand for real estate. The supply of houses will react to the increase in demand. Therefore, real estate investment increases, construction companies and
5 housing market intermediaries increase their activities, which fuels aggregate demand. (Mishkin, 2001).
A reduction in the cost of purchasing real estate proceeds through the transmission of official rates to mortgage rates, which occurs most rapid by countries with a large share of variable mortgage rates. According to neoclassical standards, there should be no distinction between variable or fixed rates, since the average variable rate portrays the expectations available in the economy that determined the fixed rate. However, when households face credit constraints and therefore cannot borrow in order to smooth out consumption, an increase in interest rates reduces their cash flow diminishing the opportunity to acquire real estate. Evidence reveals that house prices are more volatile in economies with a high share of variable-rate mortgages. Additionally, high transaction costs and/or a bias against housing lending might harm the transmission process (Mishkin, 2007a).
Furthermore, besides financial assets, houses are included in the financial wealth of households. Higher house prices increase the financial wealth over one‟s lifespan and therefore increase spending. However, future home-owners see a decrease in income due to higher housing costs which counteracts the former wealth effect. Nonetheless, aggregate consumption could rise when the marginal propensity to consume differs between the two groups. Additionally, beneficial tax structures, low transactions costs and increased confidence regarding the future increase the opportunity to spend an increase in housing wealth. When comparing houses to other assets, for example stocks, their effect on consumption is assumed to be relatively higher because they are perceived to be less volatile and more equally distributed among the population (Mishkin, 2007a).
Additionally, higher house prices increase the borrowing opportunities of households due to an increased value of their collateral. The size of the effect will depend on banking circumstances such as LTV ratios, availability of re-mortgaging and how eager banks are to provide mortgage equity withdrawals (MEW) (Giuliodori, 2005). Likewise, higher collateral reduces bank losses after a default, therefore more bank capital remains that can be used for lending (Mishkin, 2001). Moreover, an increased value of real estate could drive up rents for tenants and income of landlords. Therefore, the overall effect could turn negative since one can assume tenants having a higher marginal propensity to consume than landlords. The magnitude of the effect will depend on regulations regarding the rental market such as rent caps and social housing structures (Mishkin, 2007a).
Although house prices seem a vital component of monetary policy, Mishkin (2001) states that targeting of house prices by the central bank is hard to accomplish. Since central bankers have no clear advantage regarding information about the development of asset prices, they will not „outsmart‟ private investors and „pierce‟ the bubble. Thus, if asset prices are too high relative to economic fundamentals, private investors would understand and the bubble would not exist. However, determining economic fundamentals and assigning „natural‟ values to them is a complex procedure that could deem to be impossible.
Furthermore, asset prices are not solely determined by monetary policy and it would be a misinterpretation to automatically link increasing housing prices to future increases in productivity levels as one would do with other assets such as stocks. Increases in house prices can for example stem from conditions such as a restriction of supply. Therefore, reaching objectives regarding asset prices will be complicated and could possibly make the central bank lose face. Additionally, targeting asset prices might be perceived as too much interference, which could lead to a fall in central bank support (Mishkin, 2001).
2.3 Empirical evidence
Mishkin (2007a) analyses several sources regarding the elasticity of consumption with respect to housing wealth. He finds results varying between 5% and 17% and is unable to conclude on whether
6 the elasticity is different from the elasticity of other asset prices. Cho (2011) does a similar analysis preceding his own research and finds contradicting results. An analysis comprising studies where increases in housing wealth produce a large and significant impact on aggregate consumption alternated by studies that find little to no effect. The ambiguous results can be linked to the previously described difference in wealth effects between home-owners and non-home-owners.
Cho (2011) tries to disentangle the difference in consumption reactions to house wealth developments by splitting up data on the Korean housing market in five income groups. A negative effect in the low-income group and a positive effect in the high income group would explain the lack of consensus on the topic. The result of a regression on the total dataset without distinguishing between income groups provides an insignificant wealth effect. However, in a separate analysis, the low income group has a negative wealth effect, whereas the high income group has a positive effect and both effects are significant.
Different effects occur because homeowners, represented by the high income group, and non-homeowners, represented by the low income group, react differently to changes in housing wealth. Homeowners observe an increase in their wealth after a rise in house prices, this induces them to increase their consumption. Whereas non-homeowners would face an increase in rent after increased house prices, which negatively affects their consumption.
Giuliodori (2005) estimates several VARs using a database that contains variables from the 1979 – 1998 period covering nine European countries. In many studies of the MTM VARs are employed because they allow for analyzing linear interdependencies among multiple time series. The baseline model includes the following variables; consumer price index (CPI), real GDP (GDP), real house prices (RHP) and the money market interest rate (MMR). In order to solve the identification problem a Choleski decomposition2 is applied. The MMR responds contemporaneously to shocks in the other variables, whereas changes in GDP, RHP and CPI caused by a change in the MMR are delayed by one quarter, which illustrates their „sticky‟ nature. A temporary exogenous increase of the MMR does not produce a significant reaction in CPI. GDP decreases temporarily for all countries except Spain. The size of the reaction of RHP to the MMR shock is heterogeneous. Nonetheless, in most countries a significant inverted U-shaped response is apparent.
To analyze the role of house prices in the MTM Giuliodori (2005) adds consumption as a variable to the model specified above. Two simulations are carried out regarding the response of consumption to a monetary shock, differing in whether RHP is allowed to fluctuate or not. The results illustrate that the negative response of consumption to a contractionary monetary shock is significantly amplified by about 50% via a decrease in housing wealth in those countries where the housing market is more competitive and efficient, such as the U.K., the Netherlands and Spain.
Elbourne (2008) studies the role of house prices in the MTM of the U.K. using data from January 1987 until May 2003. The model contains the following variables; prices, retail sales, short term interest rate, money supply, house price index, nominal exchange rate, commodity prices and the Federal Funds rate. Instead of a Choleski decomposition, Elbourne (2008) employs a structural vector auto regression (SVAR). An SVAR incorporates specific identifying assumptions, based on economic theory, reflecting agents‟ behavior and/or policy implications. Through placing specific restrictions in the model, one does not have to rely on a recursive structure, as is the case with a Choleski
2 A Choleski decomposition relies on partial identification, which means only one of the structural shocks is
identified. In the case of a monetary shock, which is usually represented by a change in interest rates in the MTM literature, variables ordered before the interest rate do not respond immediately to monetary policy, monetary policy reacts contemporaneously to them. Variables ordered after the interest rate respond immediately to monetary policy, monetary policy does not respond contemporaneously to them. A technical description of a Choleski decomposition can be found in section 3.3.
7 decomposition. Therefore, an SVAR can allow for a contemporaneous relationship between variables. For example, policy makers incorporate current exchange rates in their decisions and exchange rates respond immediately to shifts in interest rates.
Elbourne (2008) includes more restrictions than necessary, therefore his system is over-identified, which can make it easier to estimate parameters. The restrictions are based on short and medium term assumptions with respect to economic theory. His rationale for an SVAR instead of a Choleski decomposition is because the latter proves to be problematic regarding the inclusion of jointly determined variables such as interest rates and exchange rates as described previously. Furthermore, due to partial identification one can only analyze one shock at a time. Therefore, in order to analyze the role of house prices in the MTM one would have to set up two models including the monetary and housing shock respectively. This is not optimal since “each model looks at one half of the transmission mechanism” according to Elbourne (2008).
A 1% shock to real house prices leads to a rise in consumption by 0.07%, an interest rate shock of 100 basis points decreases house prices by 0.75%. Ultimately, about 15% of the decline in consumption after a monetary shock is transmitted via changing house prices. Furthermore, variance decomposition illustrates that house prices are hardly affected by other variables considering low horizons. As the horizon extends to about three years, the Federal Funds rate determines about 33% of the variation in house prices, which illustrates the leading role of U.S. rates in setting monetary policy.
Jarocinski & Smets (2008) examine the housing market from a wider perspective by studying its role in U.S. business cycles using an identified Bayesian VAR, which adopts Bayesian methods for estimation. Contrary to standard VAR models, parameters are treated as random variables, which have prior probabilities assigned to them. Jarocinski & Smets (2008) conclude that a shock with respect to housing demand significantly affects housing investment and house prices. Besides that, loose monetary policy during 2002-2004 supported the housing boom in 2004 and 2005. However, the effect of both aforementioned phenomena on the U.S. economy regarding aggregate growth and inflation is limited. Inflation due to housing shocks was about 0.25%, which could indicate a small role of house prices in the MTM.
Carstensen et. al (2009) find similar responses as Giuliodori (2005) using a panel-VAR model, which adds a cross-sectional dimension to a standard VAR model, covering 12 Euro-countries over the 1995 – 2006 period. Identification is achieved through sign restrictions, which means some variables‟ behavior is limited to respond in one „direction‟ only. Carstensen et. al (2009) find that output and inflation move in the same (negative) direction as real house prices after a contractionary monetary shock. Similar to Giuliodori (2005), they find a significant difference in the response of key macroeconomic variables depending on the institutional nature of the housing market with respect to competitiveness and flexibility. Therefore they suggest that real house prices affect the impulse response functions. This result is backed by Calza et. al (2013) who find that the impact of a monetary shock on house prices and residential investment is influenced by the flexibility of the mortgage market and the interest rate structure that is dominant in a country. However, consumption seems to be only affected when equity release is common and when most mortgage contracts are variable.
Bjornland & Jacobsen (2010) do not apply a Choleski decomposition nor solely short and medium term restrictions in their SVAR. Identification is also attained through restrictions on the long-run multipliers of shocks, while allowing for contemporaneous reactions of asset prices and the interest rate. The real exchange rate and real GDP are assumed not to be affected by monetary policy in the long run, hence the long run restriction. However, long-run restrictions are not applied to real house prices, since it is hard to distinguish whether a „natural‟ level of house prices exists since demographic changes could change housing demand and therefore house prices. Furthermore, central bankers are assumed to utilize all information available when setting policy, including house prices.
8 which contradicts Giuliodori‟s (2005) restriction on contemporaneous effects. Furthermore, a positive shock to house prices significantly immediately raises interest rates in Sweden and the UK. A raise in interest rates could stem from the central bank anticipating on expected inflation in the future. Assuming the central bank does correct predictions, an increase in inflation after a positive housing shock militates in favor of a wealth effect of house prices, since it means this wealth is being spent, driving up inflation.
Lastly, Milcheva and Sebastian (2012) adopt a counterfactual VAR approach, similar to the one used by Giuliodori (2005), covering 14 European countries examining data from the 1990 – 2008 period. They find significant house price responses in countries with fixed rate mortgage contracts and low mortgage market flexibility, contrary to what Calza et al. (2013) found. However, the majority of countries with significant house price responses do have developed mortgage markets. Furthermore, rather than just analyzing house prices, residential investment is taken into account. Residential investment is affected significantly after a monetary shock in half of the countries, where the nature of the mortgage market or Euro-area membership does not seem to play a role. In the Euro zone, the effect of housing on consumption through house prices is only positively significant in Denmark. Changes in consumption seem to stem mostly from shocks to residential investment, which is the case in half of the countries examined.
The counterfactual analysis reveals a minor role for the housing market (prices and investment) in most euro-area countries. Except for France and Ireland, where the shock to consumption is large and transmits mostly through residential investment, which could be due to high labor-intensity in the construction sector. Nordic countries outside the euro-area exhibit significant housing effects (prices and investment). There is no clear evidence for stating that housing effects are only prevalent in countries with flexible mortgage markets. The only common factor of significantly affected countries is high LTV ratios in the mortgage market, which could be due to the effect that those mortgage markets are highly leveraged and the housing value serves as collateral.
After examining the above mentioned literature, one can state that the housing market does play a role in the MTM. House prices and residential investment are affected by official rates and it seems the speed of adjustment of these two variables might have been underestimated in the past. Nonetheless, no consensus has been reached on the total impact of altered house prices and investment plans on consumption. Firstly, different groups in society have opposite reactions to changes in the housing market. Secondly, the magnitude of the effect seems to be affected by characteristics of the housing market. Thirdly, most research is conducted through VAR analysis where the identification of the VAR differs per study, which can be a cause for differing results.
2.4 Characteristics of the Dutch housing market
The Dutch housing market is characterized by public involvement and intervening policy measures in order to stimulate home ownership and provide access to affordable housing for low income households. Two pivotal subsidy instruments supporting this policy are a monthly income-based grant for renters and income-deductable mortgage relief for homeowners. Furthermore, the Dutch government affects the housing market through spatial planning, land policy, regulation, supervision of housing associations and rent policies. The Dutch housing approach seems to have been effective, today 60% of Dutch houses are owner-occupied compared to 42% in 1980 (Vandevyvere & Zenthöfer, 2012).
A pioneering step in regulating the Dutch housing market was the formation of the Housing Act (Woningwet) in 1901, which required municipal governments to develop zoning plans accompanied by the provision of water and sewerage. After the Second World War the Dutch housing market experienced pressure due to a housing shortage caused by the war combined with population
9 growth. The government responded by subsidizing rents below their free market levels and constructing large social housing projects, by the 1980s the main shortages were overcome. Currently, social housing is mostly dealt with by municipalities and housing associations, the former can develop their own regulations as long they comply with the 1965 law for spatial planning (Wet voor Ruimtelijke Ordening) (Vandevyvere & Zenthöfer, 2012).
Today, around 93% of the rental sector is regulated, 79% of these rental homes are coordinated by non-profit social housing corporations. In the 1990s most of the subsidies for housing construction were abandoned, which made the social housing corporations financially independent and decreased the construction of social housing. Currently, social housing makes up 33% of the total housing stock. Rent regulation consists of a maximum rent level, annual rent adjustments capped by the government and tenant protection. About 15% of the Dutch people receive grants, determined by their income and rent3, for renting a house. These grants stand at the the third highest value in the OECD (Vandevyvere & Zenthöfer, 2012).
Figure 1: Tax relief on debt financing cost of homeownership as a % of GDP (2009)
Full deductibility of interest payments from one‟s income benefits higher income households mostly since they experience the highest marginal tax rates. Figure 1 illustrates the mortgage tax relief situation in Europe in 2009, where the Netherlands is leading the ranks. Interest payments can be deducted up to a period of 30 years maximum. This has shifted households‟ preferences to buying a house completely financed with borrowed capital and postponement of paying off mortgages until the loan has matured. In 2010, 56% of the Dutch mortgage market consisted of „interest-only‟ mortgages, which enables households to benefit fully from their interest deductibility, whilst maintaining the funds that would have been used for a gradual mortgage pay-off (Vandevyvere & Zenthöfer, 2012). Furthermore, in exchange for paying an insurance fee of 0.7% of the mortgage, the Dutch government insures the home-owner against default of a property value up to 350000. Apart from the reliefs stated above, an increase in property value is taxed by 0.6% for values up to 1 million euro, higher values are taxed by 1.30%. The property value is calculated through an estimate of the imputed rent, which is based on the rent of a similar house. The Netherlands is among the few countries that employ such a tax (Vandevyvere & Zenthöfer, 2012).
The favorable tax structure regarding mortgage payments in the Netherlands has caused an increasing trend in leverage against property values. The average LTV ratio at the moment of a residential purchase rose from 79% in 1970 to 100% around the 2000s to 120% in 2009. Furthermore, this increase can be linked to the re-mortgaging of increases in property values and a slow decrease of the existing mortgage stock due to postponement of mortgage pay-offs. Mortgage interest payments
3
Grants can be received up to a certain income threshold of around €21000 for a single household and €28000 for households consisting of more than one person.
10 are biased towards a more fixed nature in the Netherlands. 13% of the payments are variable, 3% fixed up to 5 years, 56% fixed between 5 and 10 years and 28% fixed for a period of 10 years or more (Vandevyvere & Zenthöfer, 2012).
2.5 Consequences for the Dutch economy
Favorable macroeconomic conditions in the Dutch economy, such as economic growth, an increasing labor participation rate and depressed real interest rates, led to household expectations of economic prosperity including growing asset values. Therefore, house prices have been surging from 1970 onwards, except for a drop at the beginning of the 1980s and an annual average house price decrease of 4.2%4 since the financial crisis. Analyses on whether this price increase was based on fundamentals produced varying results. In 2008, the IMF stated that the price increases between 1997 and 2007 had a share of 30% which could not be explained by fundamentals.
However, the CPB Netherlands Bureau for Economic Policy Analysis (CPB) found that the IMF analysis was flawed since it did not account for country specific elements of the Dutch housing market. According to the CPB any overvaluation gap that existed in 2003 had evaporated by 2007. This evaporation can be explained by a lower real house price growth from 2004 onwards compared to the long-term equilibrium price (Vandevyvere & Zenthöfer, 2012).
The net costs of the Dutch housing policy with respect to home-ownership hovered around 13.5 billion euro‟s, which boils down to about 2.3 % of GDP between 2006 and 2011. Furthermore, the guaranteed mortgages by the Dutch government have a value of around 24% of GDP. Besides that, Dutch households have the highest household debt related to housing as a ratio to GDP (109% in 2010) and the highest households‟ interest payments relative to disposable income.
Nonetheless, the increasing household debt is surpassed by the growth of household wealth, which was fuelled by gains on house prices in particular. As capitalization on housing wealth remained prevalent, households shifted their portfolios, the share of housing wealth of total household wealth rose from 31% to 39% between 1993 and 2010. Furthermore, savings and investments in financial products are taxed at 1.2%, about twice as high as the aforementioned tax on increases in property values up to 1 million euro‟s (Vandevyvere & Zenthöfer, 2012).
Apart from affecting tax schemes and household portfolio‟s, the Dutch housing market policy has steered the redistribution of wealth in the Netherlands. Firstly, gains on house prices create intergenerational redistribution. First-time buyers pay a high price and thereby realize the returns on houses for the elder. Secondly, income is redistributed among income categories. High-income households have a higher marginal tax rate, therefore they can deduct a larger share of their mortgage interest payments than lower-income households. Furthermore, the bridge between regulated rents and unregulated property sales prices is hard to cross for many households. Thirdly, income thresholds are not monitored after a household benefits from a social-rent-dwelling. When income increases, social rent is still applicable. This creates unnecessary benefits for existing tenants compared to new households eligible for social rents (Vandevyvere & Zenthöfer, 2012).
Furthermore, the gains of housing corporations are not monitored properly, balance sheets often portray houses at historic cost prices. High returns on housing can be liquidated and used for other means. Lastly, social housing rules make it unattractive to move, which affects labor mobility. For example, a waiting list system based on your current location and small incentives to increase working hours due to decreased housing benefits distort the market (Vandevyvere & Zenthöfer, 2012).
11 3 Methodology
The counterfactual analyses of Giuliodori (2005) and Milcheva and Sebastian (2012) exhibit contradicting results with respect to the role of house prices in the MTM in the Netherlands and other countries such as the UK and Spain. Nonetheless, the focus of this analysis will be on the Netherlands. Giuliodori (2005) finds the negative response of consumption to a monetary contraction to be weakened by 40 to 50 per cent when house prices are not allowed to influence decisions with respect to consumption. Whereas Milcheva and Sebastian (2012) find no difference between the two scenarios with respect to house prices and a small difference when besides house prices, residential investment is not allowed to fluctuate.
The aim of this analysis is to dismantle the root of the conflicting results in the aforementioned studies. Therefore, a VAR analysis similar to the one employed by Giuliodori (2005) will be constructed. However, the time-frame will be identical to the one chosen by Milcheva and Sebastian (2012). Besides exposing the origin of the differing results, the analysis should give an indication on how the role of house prices in the MTM has developed over time. Firstly, a base-line model will be constructed, extended with a consumption variable afterwards and compared to Giuliodori‟s (2005) base-line model and extension. Furthermore, a counterfactual analysis will be carried out, which illustrates the role of house prices in the MTM. This counterfactual analysis is compared to the counterfactual results of both Giuliodori (2005) and Milcheva and Sebastian (2012).
3.1 Data
In order to keep the analysis concise and thereby save degrees of freedom in the estimation process, the proposed base-line model will only capture the variables necessary for illustrating the monetary transmission process. The choice of variables is based on the base-line model of Giuliodori (2005), who uses real house prices (RHP), consumer price index (CPI), real GDP (GDP) and the money market interest rate (MMR). The data entries run from 1990:Q1 until 2013:Q4, this time-frame differs from Giuliodori‟s (2005) and is the same as the one chosen by Milcheva and Sebastian (2012), providing space for comparison.
GDP values were taken from Eurostat, in millions of euros at market prices, seasonally adjusted and adjusted by working days. In order to convert GDP to real GDP, the values are divided by the GDP deflator presented by Eurostat. The consumer price index was taken from the OECD statistic database. The house price index is taken from the NVM (Nederlandse Vereniging van Makelaars o.g. en Vastgoeddeskundigen) and divided by CPI to create the real identities. The money market rate is taken from Eurostat, based on historical interbank three-month rates for the Netherlands until 1998:Q4 and on Euro-area interbank rates onwards. All variables are presented as their natural log, except for the money market rates.
3.2 Vector auto regression
In this analysis a VAR will be applied, VARs are useful for examining the effects of shocks to the economy and they allow for interaction between multiple endogenous variables. Consider the following first-order VAR:
𝛾𝑡= 𝑏10− 𝑏12𝑍𝑡 +𝜃11𝛾𝑡−1+𝜃12𝑍𝑡 −1+𝜖𝛾𝑡 (1)
12 Where it is assumed that 𝛾𝑡 and 𝜒𝑡 are stationary and 𝜖𝛾𝑡 and 𝜖𝜒𝑡 are uncorrelated white noise
disturbances. One can see from the equations above, when 𝑏12 and 𝑏21 are non-zero, it allows for
contemporaneous effects among endogenous variables. After transforming (1) and (2) and putting them in matrix-form one obtains the following system:
1 𝑏12 𝑏21 1 𝛾𝑡 𝑍𝑡 = 𝑏10 𝑏20 + 𝜃11 𝜃12 𝜃21 𝜃22 𝛾𝑡−1 𝑍𝑡−1 + 𝜖𝛾𝑡 𝜖𝑍𝑡
Which is the same as:
𝐵𝑥𝑡= 𝜑0+ 𝜑1𝑥𝑡−1+ 𝜖𝑡 (3)
(3) shows the structural form, where: 𝐵 = 𝑏1 𝑏12 21 1 𝑥𝑡 = 𝛾𝑡 𝑍𝑡 𝜑0= 𝑏10 𝑏20 𝜑1= 𝜃11 𝜃12 𝜃21 𝜃22 𝑒𝑡 = 𝜖𝛾𝑡 𝜖𝑍𝑡 Multiplying (3) by 𝐵−1 gives: 𝑥𝑡 = 𝐴0+ 𝐴1𝑥𝑡−1+ 𝑒𝑡 (4)
(4) shows the reduced form, where:
𝐴0= 𝐵−1𝜑0 𝐴1= 𝐵−1𝜑1 𝑒𝑡 = 𝐵−1𝜖𝑡
(3) and (4) can be generalized to the multivariate systems (5) and (6):
𝐵𝑥𝑡 = 𝜑0+ 𝜑1𝑥𝑡−1+ ⋯ + 𝜑𝑝𝑥𝑡−𝑝+ 𝜖𝑡 (5)
𝑥𝑡 = 𝐴0+ 𝐴1𝑥𝑡−1+ ⋯ + 𝐴𝑝𝑥𝑡−𝑝+ 𝑒𝑡 (6)
Where
𝑥𝑡= an (n x 1) vector depicting the variables included in the VAR
𝜑0= an (n x 1) vector of intercept term
𝜑𝑝= an (n x n) matrix of coefficients
𝑒𝑡= an (n x 1) vector of error terms
Equation (6) can be estimated through ordinary least squares (OLS), where one assumes 𝑒𝑡 to be
serially uncorrelated. However, the coefficients can be unreliable since the model is built up of endogenous interacting variables and therefore multicollinearity issues arise. Furthermore, 𝑒𝑡 does not
reflect the „true‟ shocks one tries to analyze after the estimation of the VAR. The underlying structure is:
𝑒𝑡 = 𝐵−1𝜖𝑡 (7) or 𝜖𝑡 = 𝐵𝑒𝑡 (8)
If one can disentangle how the system (6) reacts to shocks represented by vector 𝜖𝑡 one can produce
13 unknown, except the diagonal row of 1‟s. Nonetheless, the identification problem can be solved by adding additional restrictions. Firstly, transposing (8) on both sides gives:
𝐵Ʃ𝑒𝐵′ = Ʃ𝜖
Where, Ʃ𝑒 = covariance matrix of 𝑒𝑡 and Ʃ𝜖 = covariance matrix of 𝜖𝑡.
Ʃ𝑒 can be estimated and thus provides 𝑛(𝑛+1)
2 known values since it is symmetric. In Ʃ𝜖 the covariance
elements are known to be zero, since the variables are mutually orthogonal. Therefore, 𝑛 unknown variables remain, the variances. In B only the diagonal 1‟s are known, thus 𝑛(𝑛−1)
2 unknown variables
remain. In total this boils down to 𝑛2 unknown variables and 𝑛(𝑛+1)
2 known variables. The order
condition reveals that at least 𝑛(𝑛−1)
2 restrictions have to be applied in order to achieve identification
(Enders,2008: ch. 5).
The shocks to the system as described above are exogenous and therefore they represent elements omitted from the model. However, these omitted elements are possibly correlated with the endogenous variables in the model. If this is the case, the VAR results such as IRFs will be infected by omitted variable bias. This should be taken in mind when analyzing the shocks to the money market rate in this analysis since monetary policy is usually not subject to sudden shocks (Stock and Watson, 2001). On the contrary, central bankers are forward looking and take a multitude of variables into account when setting interest rates.
3.3 Identification
In the literature review restrictions are usually applied by using a recursive Choleski decomposition or a non-recursive decomposition based on economic theory. A Choleski decomposition relies on partial identification, which means only one of the structural shocks is identified. In the MTM literature this shock is usually a monetary shock with respect to interest rates. Regarding the identification problem 𝐵Ʃ𝑒𝐵′ = Ʃ𝜖, it means 𝐵 is lower triangular when a Choleski decomposition is applied. Since 𝐵 is
lower triangular the ordering of variables determines the interacting behavior of variables. In the case of a monetary shock, variables ordered before the interest rate do not respond immediately to monetary policy, monetary policy reacts contemporaneously to them. Variables ordered after the interest rate respond immediately to monetary policy, monetary policy does not respond contemporaneously to them (Enders, 2008: ch. 5).
If one does not rely on a Choleski decomposition, shocks to the economy are allowed to affect each other contemporaneously. Every non-diagonal element of 𝐵 is considered specifically and decided upon whether it should be restricted or whether it should be deliberately estimated, which could be to illustrate a contemporaneous relationship. Similar to a Choleski decomposition, one can decide to assign the number 0 to a parameter. Besides that, elasticities, sign restrictions and long-run relations can be used for identification (Enders,2008: ch. 5).
In this analysis a Choleski decomposition will be applied. A Choleski decomposition is appropriate because the point of interest is the economy‟s reaction to a single shock (Milcheva and Sebastian, 2012). Furthermore, an uncontemporaneous relationship with the MMR variable will not be problematic because GDP, CPI and RHP are assumed to be sticky and therefore will not react within the same quarter to monetary policy shocks. Variables are ordered as in Giuliodori‟s (2005) model, (CPI,GDP,RHP,MMR). Applying the Choleski decomposition results in the following equation regarding the identification:
14 𝜖𝑐𝑝𝑖 𝜖𝑔𝑑𝑝 𝜖𝑟ℎ𝑝 𝜖𝑚𝑚𝑟 = 1 0 𝑏21 1 0 0 0 0 𝑏31 𝑏32 𝑏41 𝑏42 1 0 𝑏43 1 𝑒𝑐𝑝𝑖 𝑒𝑔𝑑𝑝 𝑒𝑟ℎ𝑝 𝑒𝑚𝑚𝑟
Policymakers are assumed to use all information available when setting interest rates, therefore the interest rates are affected contemporaneously by the price level, GDP and house prices. Prices are sticky and do not react to any variable in the same quarter. GDP is allowed to be affected contemporaneously by the price level. RHP are affected contemporaneously by GDP and CPI.
Besides Giuliodori (2005), Milcheva and Sebastian (2012) also use the above structure for GDP,MMR and RHP. The fact that RHP is not affected by interest rates is plausible even in countries with a majority of variable rates, since these rates are usually annually adjusted (Milcheva and Sebastian, 2012). Furthermore, one can assume that finding a house and buying it after a change in interest rates takes longer than one quarter.
3.4 Stationarity, cointegration, breaks and lags
Firstly, the variables are tested for being stationary through the Augmented Dickey-Fuller test. The results illustrate all variables being non-stationary. Giuliodori (2005) and Milcheva and Sebastian (2012) also encountered non-stationary variables. Nonetheless, both estimated the VAR in levels, since taking first differences leads to a loss of information contained in the levels. Furthermore, Sims et al. (1990) found that a system can be estimated in levels when co-integration exists among the variables. This is the case since the Johansen trace test for cointegration among all variables presents a significant degree of cointegration of 1 among the variables. Therefore, all variables except for the MMR remain in levels.
The Hannan-Quinn criterion and Schwarz criterion produce an optimal lag of one. The Chow-test reveals a break around the financial crisis, with p-values increasing from 2008:Q4 onwards. In order to examine the effects of this break on the role of house prices in the MTM a dummy variable is placed affecting the period of 2009:Q1 – 2013:Q4. Comparing these results to the original model covering the 1990:Q1 – 2008:Q4 period can provide insights on the consequences of the financial crisis. Nevertheless, it must be noted that a dummy variable might not be sufficient for capturing the complexity of a financial crisis.
4. Results
4.1 Base-line model
Figure 2 depicts the responses of CPI, GDP, RHP and MMR to a money market shock in the base-line model. The price level hovers around the zero-bound and does not display a significant negative or positive reaction, which similar is to the reaction found by Giuliodori (2005). This could be due to the fact that a country such as the Netherlands that faces wage rigidities, especially downwards, due to unionization. Therefore, the price level does not show major swings even though the economy is facing negative pressures reflected by real GDP and real house prices. GDP has a significant U-shaped pattern, of which the bottom level is reached after 17 quarters at -0.51%. RHP has a similar pattern, however, the impact is larger compared to GDP. The bottom of the U-shaped curve illustrates a price decrease of -1.5% after 15 quarters. The MMR response depicts the initial shock to interest rates dying out after 16 quarters.
15
Figure 2: Impulse responses to a one standard deviation money market rate shock (1990-2008 period) with +- 2 standard error confidence intervals depicted in red.
The U-shaped responses of RHP and GDP demonstrate the lagged effect of the shock to the economy and the recovery of the economy afterwards. This shape is also present in the results of Giulodori (2005) and the impact levels reached at the bottom are similar. However, the impulse response functions of GDP and RHP of Giuliodori (2005) take about half as long to reach the bottom compared to this analysis. Furthermore, the responses in figure 2 have a more gradual nature, without sudden drops or rises, contrary to Giulodori‟s (2005) results.
This difference could indicate a „maturing‟ Dutch economy, which has more stability. This stability could have several causes, possibly reinforcing each other. For example, a transparent central banking policy on price stability promotes a steady wage market. Furthermore, higher diversification due to global interconnectedness leads to smaller vulnerability regarding internal shocks. Besides that, a better distribution of housing wealth among society eases the absorption of housing shocks for the economy as a whole. Lastly, the addition of 10 years of data characterized by an increase in overall economic prosperity could enhance the results in a positive manner with respect to economic stability.
-.006 -.004 -.002 .000 .002 2 4 6 8 10 12 14 16 18 20 22 24 Response of CPI to MMR -.010 -.008 -.006 -.004 -.002 .000 2 4 6 8 10 12 14 16 18 20 22 24 Response of GDP to MMR -.03 -.02 -.01 .00 2 4 6 8 10 12 14 16 18 20 22 24 Response of RHP to MMR -.4 -.2 .0 .2 .4 2 4 6 8 10 12 14 16 18 20 22 24 Response of MMR to MMR
16
Figure 3: Impulse responses to a one standard deviation money market rate shock (1990-2013 period) with +- 2 standard error confidence intervals depicted in red.
Figure 3 illustrates the impulse response functions produced by a one standard deviation money market rate shock after including the post-financial crisis data. The impulse response functions of RHP, GDP and MMR are different from those in figure 2. The final decrease in house prices is -2% compared to -1.5% in the pre-crisis scenario, GDP decreases until -0.63% compared to -0.51% in the pre-crisis scenario and the MMR does not die out as quick as before. Furthermore, the U-shaped pattern is absent in this time-frame, indicating a slower recovery process.
GDP decreases further since aggregate demand experienced substantial downward pressure during the post-crisis period. The shock to MMR does not die out as before since interest rates severely dropped after the crisis for reasons not captured in this model. The alteration of the response of house prices makes sense since a malfunctioning housing market was at the base of the 2008 financial crisis. House prices dropped 4.2% annually on average after the crisis, despite decreasing interest rates.
The alteration of the impulse response function of house prices illustrates house prices behaving differently than in the set without post-crisis data. After analyzing multiple crises through history Reinhart and Rogoff (2009) illustrated that post-crisis behavior of economic variables is different than before the crisis. This behavior is characterized by a slow recovery process among other things. In this manner, the addition of post-crisis data causes an alteration of the IRFs, which could indicate a (temporary) shift in housing market behavior due to disturbances caused by the crisis.
-.005 -.004 -.003 -.002 -.001 .000 .001 .002 2 4 6 8 10 12 14 16 18 20 22 24 Response of CPI to MMR -.012 -.010 -.008 -.006 -.004 -.002 .000 2 4 6 8 10 12 14 16 18 20 22 24 Response of GDP to MMR -.05 -.04 -.03 -.02 -.01 .00 .01 2 4 6 8 10 12 14 16 18 20 22 24 Response of RHP to MMR -.2 -.1 .0 .1 .2 .3 .4 .5 2 4 6 8 10 12 14 16 18 20 22 24 Response of MMR to MMR
17 4.2 Forecast error variance decomposition
Forecast error variance decomposition (FEVD) illustrates the proportion of each exogenous shock in explaining the forecast error variance of a particular endogenous variable (Enders,2008: ch. 5). Figure 4 portrays the FEVD of RHP for the periods of 1990-2008 and 1990-2013. Initially, house price volatility is mainly driven by shocks to real house prices themselves. However, when taking a broader horizon of 24 lags into account, money market rate innovations embody 40% and 39% of house price volatility in the 1990-2008 and 1990-2013 period respectively. The forecasting value of house price shocks to real house prices diminishes as time passes by.
One should be careful when interpreting these results. Firstly, the forecasting values are constructed relatively to the other variables captured by the model. Therefore, if variables with high predicting potential are omitted from the model forecast values can become misleading. Secondly, since the forecasting values are averages it is difficult to apply them when the fundamentals of the economy do not behave according to „moderate‟ times. For example, post-crisis interest rate cuts did not cause a rise in house prices. On the contrary, house prices kept decreasing despite further loosening of monetary policy. However, one could state that house prices would have diminished even further without intervention. Such counterfactual problems complicate the specification of forecasting relations between variables.
Figure 4: Forecast error variance decomposition of RHP in the 1990-2008 and 1990-2013 period
The pattern of MMR shocks gaining more significance over time represents the small portion of short-term variable mortgage contracts prevalent in the Dutch economy. When a large share of households renegotiate their contracts every 5 to 10 years, it takes several years for MMR shocks to affect house prices, which is illustrated in figure 4. Contrary to this FEVD analysis, Giuliodori (2005) found money market rate shocks accounting for roughly 15% of real house price volatility. The increase of relevance of MMR in two FEVDs in figure 4 compared to Giuliodori‟s (2005) result indicates a stronger role for credit markets with respect to mortgage markets. This can be explained by increased LTV ratios, 79% in 1970 and 120% in 2009, and the increasing prevalence of interest-only mortgages,
which stood at 56% in 2010.
4.3 Base-line model extended with consumption
The variable real consumption (CONS) was constructed by dividing the absolute value of final consumption of households by the price index of final consumption of households and the data was obtained from Eurostat. The new ordering of variables after adding consumption is: (CPI, GDP,
18 CONS, RHP, MMR), which is the same ordering as employed by Giuliodori (2005). Through this ordering consumption is not affected contemporaneously by house prices, for which Giuliodori (2005) provides two motivations. Firstly, house price fluctuations are usually not available within the same quarter. Secondly, consumption is assumed to react sluggishly towards house price fluctuations. Furthermore, real GDP and the price level affect consumption contemporaneously, which is plausible since prices are faced within the same quarter and fluctuations in GDP can be spent in the same quarter.
Figure 5: Impulse responses to a one standard deviation money market rate shock (1990-2008 period) in the base-line model extended with consumption, +- 2 standard error confidence intervals depicted in red.
Figure 5 depicts the impulse response functions of a one standard deviation shock to the money market rate in the 1990 – 2008 period. The reaction of the variables besides CONS has not changed compared to figure 2. Similar to the other variables CONS portrays a Ushaped pattern. It reaches a bottom of
--.005 -.004 -.003 -.002 -.001 .000 .001 .002 2 4 6 8 10 12 14 16 18 20 22 24 Response of CPI to MMR -.010 -.008 -.006 -.004 -.002 .000 .002 2 4 6 8 10 12 14 16 18 20 22 24 Response of GDP to MMR -.012 -.008 -.004 .000 2 4 6 8 10 12 14 16 18 20 22 24 Response of CONS to MMR -.025 -.020 -.015 -.010 -.005 .000 .005 2 4 6 8 10 12 14 16 18 20 22 24 Response of RHP to MMR -.3 -.2 -.1 .0 .1 .2 .3 .4 2 4 6 8 10 12 14 16 18 20 22 24 Response of MMR to MMR
19 0.66% after 15 quarters. This result is similar to the impulse response function found by Milcheva (2012). The impulse response function of Giuliodori (2005) is similar in terms of impact and shape. However, Giuliodori‟s (2005) IRF experiences an irregularity in the first few quarters.
The response of consumption after a monetary contraction is in line with what one would expect from economic theory. Aggregate demand decreases, which decreases household income. Modigliani‟s life-cycle theory illustrates that households spread their wealth over their lifetime, therefore consumption decreases. Figure 6 depicts the impulse responses to a one standard deviation money market rate shock with the post-crisis period included in the dataset. Similar to the inclusion of the real consumption variable to the 1990 - 2008 period, the reaction of other variables remains unchanged compared to figure 4. CONS initially behaves similar in figure 5 and 6. However, as observed with the behavior of other variables after the addition of post-crisis data to the 1990-2008 period dataset, the recovery process from the shock is absent and the bottom point is lower, -0.79% compared to -0.66%.
20
Figure 6: Impulse responses to a one standard deviation money market rate shock (1990-2013 period) in the base-line model extended with consumption, +- 2 standard error confidence intervals depicted in red.
4.4 Counterfactual analyses
The results above illustrate that real household consumption is affected by changes in interest rates. The remaining question is what role house prices played in this response of consumption. In order to solve this matter, a counterfactual analysis is constructed where house prices are withdrawn from the MTM. This boils down to transforming the RHP variable and its one lag counterpart into an exogenous variable, therefore it is not affected by other variables and remains at its pre-shock level. RHP at lag 0 is in the reaction function of the central bank and should therefore be included. RHP at lag 1 is in the reaction function of the households and should therefore be included.
Figure 7 depicts the counterfactual analyses produced by Giuliodori (2005) and Milcheva and
-.005 -.004 -.003 -.002 -.001 .000 .001 .002 2 4 6 8 10 12 14 16 18 20 22 24 Response of CPI to MMR -.012 -.010 -.008 -.006 -.004 -.002 .000 2 4 6 8 10 12 14 16 18 20 22 24 Response of GDP to MMR -.0150 -.0125 -.0100 -.0075 -.0050 -.0025 .0000 2 4 6 8 10 12 14 16 18 20 22 24 Response of CONS to MMR -.04 -.03 -.02 -.01 .00 2 4 6 8 10 12 14 16 18 20 22 24 Response of RHP to MMR -.2 -.1 .0 .1 .2 .3 .4 .5 2 4 6 8 10 12 14 16 18 20 22 24 Response of MMR to MMR
21 Sebastian (2012).The set up of the analyses of Giuliodori (2005) and Milcheva and Sebastian (2012) have two differences. Firstly, the time-frame, Giuliodori (2005) collects data for the 1979 - 1998 period. Milcheva and Sebastian (2012) cover the 1990 - 2008 period with their data selection. Secondly, the choice of variables is different. Both studies include the price level, real private consumption, real house prices and the money market rate. Giuliodori (2005) adds GDP to this selection. Milcheva and Sebastian (2012) add real residential gross fixed capital formation in housing and the oil price index as an endogenous and exogenous variable respectively.
Figure 7: Counterfactual analyses of Giuliodori (2005) and Milcheva and Sebastian (2012)
The black dotted line in Giuliodori‟s (2005) analysis represents the counterfactual analysis where real house prices are not included in the MTM. Differences between consumption levels range from 40% to 50%. The red dotted line in the analysis of Milcheva and Sebastian (2012) depicts the counterfactual analysis where real house prices are not included in the MTM, the difference is negligible. The blue dotted line represents the counterfactual analysis where real house prices and residential investment are not included in the MTM. The difference seems to be around 15% maximum.
Figure 8 depicts the result of the counterfactual analysis stemming from the application of Giuliodori‟s (2005) analysis to the 1990 – 2008 time period. The result illustrates a weakened response of consumption with the familiar U-shaped pattern. The bottom of this U-shape stands at -0.4% in the counterfactual scenario compared to -0.66% in the original scenario. Unlike the counterfactual analysis of Milcheva and Sebastian (2012) with respect to house prices only, the result illustrates a role for house prices in the MTM. However, Milcheva and Sebastian (2012) did find a role for residential investment in the MTM. Therefore, since no distinction is made between residential investment and house prices in figure 8, the difference could be partly explained by a decrease in residential investment, stemming from a decrease in house prices due to a monetary contraction, instead of solely being a reaction to a decrease in housing wealth.
22
Figure 8: Counterfactual analysis of the 1990 – 2008 period
Nonetheless, in this analysis households mitigate their negative consumption response by 24% when their housing wealth remains intact. However, the mitigation is not as strong as in Giuliodori‟s (2005) analysis, who found a mitigation of 40% to 50%. The differing degree of importance assigned to the role of house prices in the MTM between the two counterfactual analyses can be explained by the fact that the share of housing wealth of total household wealth rose from 31% to 39% between 1993 and 2010.
Consumption behavior can be characterized by a concave utility function with decreasing marginal utility. This function would be legitimate when consumption decisions respond to fluctuations in housing wealth, as was described in the literature review. Therefore, when the size of a loss in utility reflects the magnitude of the adjustment to consumption, the difference between this analysis and Giulodori‟s (2005) can be explained. For example, if households have capitalized on their houses, their housing wealth is large and they would find themselves on the far right of the curve. Therefore, their loss in utility after a loss in housing wealth would be smaller compared to when they would find themselves somewhere in the middle, which would be the case when their total housing wealth is smaller. One can assume the overall capitalization on housing wealth being smaller in the time-frame used by Giuliodori (2005), therefore the consumption response is firmer in this period. An example of such a utility function and the dynamics described are displayed in figure 10.
23
Figure 9: Counterfactual analysis of the 1990 – 2013 period
The counterfactual analysis in figure 9 presents the two IRFs of real consumption for the 1990 – 2013 period in one frame. One can observe a bigger difference than in figure 8, the bottom of the original analysis is reached at -0.79%, whereas the bottom of the counterfactual IRF stands at -0,36%. Furthermore, contrary to the original IRF, the counterfactual IRF is U-shaped. This could indicate that an exclusion of house prices in the MTM causes the economy to recover faster from a monetary shock. This is plausible since the housing market was hit hard in the years subsequent to the crisis, possibly dragging down other elements of the economy with it.
Figure 8 and 9 illustrate that removing house prices from the MTM has different magnitudes regarding the mitigating effect this removal has on the decrease in consumption enacted by households. In the 1990 – 2008 and 1990 - 2013 period, the negative consumption effect is diminished by 24% and 54% respectively. Overall housing wealth was smaller in the period 1990 – 2013 than in the period 1990 – 2008 since house prices annually dropped by 4,2% in the post-crisis years. Therefore the addition of the post-crisis data to the original dataset can have transformed the counterfactual IRFs due to the negative change in housing wealth the economy experienced. This change in the reaction of households to a change in housing wealth would advocate for their behavior being characterized by a utility function as in figure 10.
24
Figure 10: Proposed utility function describing the behavior of households with respect to housing wealth
5. Conclusion
Former research on the role of house prices has not been able to form a unified conclusion on the role of house prices in the MTM. Difficulties in the comparison of different studies have arisen due to diverging methods of research. Besides that, varying time-frames and countries subject to inquiry further complicated the formation of a united answer. Nonetheless, this study discovered two conflicting studies that allowed for comparison by constructing a third study. Revealing the root of the opposing answers was attained through setting up an identical study that analyzed a different time-period than its predecessor.
The behavior of the base-line model portrayed economic variables behaving in line with economic theory. IRFs were characterized by a U-shaped evolvement over time, indicating variables response to a shock and their equilibrating recovery thereafter. The addition of post-crisis data depicted an augmented (negative) response of the variables examined. Furthermore, the recovery process was incomplete in the examined time horizon. This alteration of IRFs illustrates the disrupting effect the financial crisis had on the fundamentals of the economy. However, conclusions based on this addition are hard to prove since a single dummy variable cannot capture the complete dynamics of a complexity such as the financial crisis.
The results of the FEVD revealed a stronger role for money market fluctuations in forecasting house price behavior compared to the 1979 – 1998 period. This stronger role for credit markets can be explained through an increase in LTV ratios over time with respect to mortgages and a high share of interest-only mortgages. Based on the counterfactual analyses one can conclude a significant role for house prices in the MTM. A negative consumption response to a monetary contraction is weakened when house prices are treated as exogenous variables and therefore remain at their pre-shock levels. However, it should be noted that this weakened response can be (partly) driven by decreases in residential investment following a decrease in house prices.
The counterfactual analysis of the 1990 – 2008 period compared to the 1979 – 1998 period saw the role of house prices in the MTM being halved. The source of the conflicting results in the counterfactual analyses can be explained by different levels of housing wealth prevalent in the economy. The share of housing wealth in the total wealth of Dutch households increased by over a quarter compared to the earlier time-period. Furthermore, housing wealth experienced severe blows due to falling house prices in the post-crisis era. The addition of this post-crisis data to the original 1990 – 2008 dataset resulted in the role of the house prices returning to its behavior as observed in the 1979 – 1998 period.
25 The varying size of the role of house prices in the MTM can be explained by households‟ reaction towards a monetary shock being dependent upon their level of housing wealth, under the assumption that households‟ consumption decisions respond to fluctuations in utility derived from housing wealth, as described in the literature review. The utility function with respect to housing wealth would have a concave shape and decreasing marginal utility. Therefore, when households find themselves on the far-right of the curve, having a high level of housing wealth, their loss in utility after a negative shock to housing wealth is relatively small. According to these dynamics, when households find themselves on the middle of the curve, having a medium level of housing wealth, their loss in utility after a negative shock to housing wealth is larger. Therefore, their consumption response to a monetary shock varies according to their established level of housing wealth.
Lastly, it should be noted that one must interpret the results presented in this study with care. VAR estimation relies on exogenous shocks, therefore they symbolize events occurring outside the scope of the model. However, central bankers are forward looking and are usually not exposed to sudden shocks. Furthermore, due to high interconnectedness in the economy and the multitude of variables relevant to monetary policy exogenous shocks can be correlated with endogenous variables, resulting in omitted variable bias. Therefore, one should be cautious interpreting the results of a VAR estimation applied to monetary policy.
6. Bibliography
BERNANKE, B.S. and BLINDER, A.S., 1992. The federal funds rate and the channels of monetary transmission. American economic review, 82(4), pp. 901-921.
BJØRNLAND, H.C. and JACOBSEN, D.H., 2010. The role of house prices in the monetary policy transmission mechanism in small open economies. Journal of financial stability, 6(4), pp. 218-229. CALZA, A., MONACELLI, T. and STRACCA, L., 2013. Housing finance and monetary policy. Journal
of the European Economic Association, 11(s1), pp. 101-122.
CARSTENSEN, K., HÜLSEWIG, O. and WOLLMERSHÄUSER, T., 2009. Monetary policy transmission
and house prices: European cross-country evidence.
CHO, S., 2011. Housing wealth effect on consumption: Evidence from household level data.
Economics Letters, 113(2), pp. 192-194.
ELBOURNE, A., 2008. The UK housing market and the monetary policy transmission mechanism: An SVAR approach. Journal of Housing Economics, 17(1), pp. 65-87.
ENDERS, W., 2008. Applied econometric time series. John Wiley & Sons.
GIULIODORI, M., 2005. THE ROLE OF HOUSE PRICES IN THE MONETARY TRANSMISSION MECHANISM ACROSS EUROPEAN COUNTRIES. Scottish Journal of Political Economy, 52(4), pp. 519-543.
JAROCIŃSKI, M. and SMETS, F.R., 2008. House prices and the stance of monetary policy. Federal
Reserve Bank of St.Louis Review, 90(July/August 2008).
MILCHEVA, S. and SEBASTIAN, S., Monetary Policy Transmission Through Housing. MISHKIN, F.S., 2007a. Housing and the monetary transmission mechanism.
MISHKIN, F.S., 2007b. The economics of money, banking, and financial markets. Pearson education.
26 MISHKIN, F.S., 2001. The transmission mechanism and the role of asset prices in monetary policy. REINHART, C.M. and ROGOFF, K., 2009. This time is different: eight centuries of financial folly. princeton university press.
SIMS, C.A., STOCK, J.H. and WATSON, M.W., 1990. Inference in linear time series models with some unit roots. Econometrica: Journal of the Econometric Society, , pp. 113-144.
STOCK, J.H. and WATSON, M.W., 2001. Vector autoregressions. Journal of Economic perspectives, , pp. 101-115.
VANDEVYVERE, W. and ZENTHÖFER, A., 2012. The Housing market in the Netherlands.
