Martin Kern
House Prices, Booms and Busts and the Credit Market
An Empirical Analysis on the Importance of the Housing Market for Macroeconomic Research
Dissertation
Wirtschafts-
wissenschaft
Importance of the Housing Market for Macroeconomic Research
House Prices, Booms and Busts and the Credit Market
Inauguraldissertation zur Erlangung des Doktorgrades der Fakultät für Wirtschaftswissenschaft der FernUniversität in Hagen
vorgelegt von
Martin Kern, M.Sc.
Berlin, 6. Dezember 2019
Univ.-Prof. Dr. Helmut Wagner Univ.-Prof. Dr. Hans-Jörg Schmerer Lehrstuhl für Volkswirtschaftslehre,
insb. Makroökonomie
Lehrstuhl für Volkswirtschaftslehre, insb. Internationale Ökonomie
Universitätsstr. 11 Universitätsstr. 11
58097 Hagen 58097 Hagen
Tag der Disputation 18. November 2019
ii
List of Figures v
List of Tables vii
1 Introduction 1
1.1 Bibliography . . . 15
2 Forecasting House-Price Growth in the Euro Area with Dynamic Model Averaging 23 2.1 Introduction . . . 26
2.2 Econometric Methodology . . . 29
2.2.1 Dynamic Model Averaging . . . 29
2.2.2 Evaluation of Forecasts . . . 30
2.3 Empirical Framework . . . 32
2.3.1 The Data . . . 32
2.3.2 Determining the Foregetting Factors . . . 37
2.4 Empirical Results . . . 39
2.4.1 Probability of Inclusion . . . 39
2.4.2 Model Performance . . . 42
2.4.3 Cumulative Forecast Errors . . . 48
2.4.4 Sensitivity Analsysis . . . 50
2.5 Conclusion . . . 51
2.6 Bibliography . . . 53
2.A Data Appendix . . . 59
iii
Evidence from a Varying Minimum Cycle Length 71
3.1 Introduction . . . 74
3.2 Data and Methodology . . . 78
3.2.1 Data . . . 78
3.2.2 Methodology . . . 82
3.3 Results . . . 87
3.3.1 Basic Characteristics of Booms and Busts . . . 87
3.3.2 Boom-Bust Cycles . . . 98
3.4 Conclusion . . . 105
3.5 Bibliography . . . 107
3.A Data Appendix . . . 111
4 House-prices and the credit market – Evidence from an International Panel of Industrialized Economies 129 4.1 Introduction . . . 132
4.2 The Data Set . . . 137
4.2.1 The Regressors . . . 137
4.2.2 Mortgage Market Characteristics . . . 141
4.2.3 Identification of Cycles . . . 143
4.2.4 Summary Statistics of the Data Set . . . 152
4.3 Methodology . . . 154
4.4 Results . . . 158
4.5 Conclusion . . . 168
4.6 Bibliography . . . 170
4.A Data Appendix . . . 175
iv
2.1 Housing Growth Rates . . . 37
2.2 Expected Size of the Forecast Models over Time for h= 1 (left) and h= 4 (right) . . . 39
2.3 Inclusion Probabilities: h= 1 . . . 40
2.4 Inclusion Probabilities: h= 4 . . . 41
2.5 Cumulative Forecast Errors . . . 49
2.6 Sensitivity of the Training Period for h= 1 (left) and h= 4 (right) . 50 3.1 House-Price Series and their Cyclical Components . . . 85
3.2 House-Price Cycles’ Boom and Bust Phases (1) . . . 91
3.3 House-Price Cycles’ Boom and Bust Phases (2) . . . 92
3.4 House-Price Cycles’ Boom and Bust Phases (3) . . . 93
3.5 Duration, Amplitude, Slope and Severity of Booms and Busts . . . . 101
3.6 Reaction of Macroeconomic and Financial Variables two years before and after a Peak in House Prices . . . 104
4.1 GDP Series, their Growth Cycle and Recessions . . . 147
4.2 House-Price Booms and Busts . . . 151
v
2.1 The Data Set . . . 33
2.2 Datastream Codes . . . 34
2.3 Forecast Models (1.1) - Performance (h= 1) . . . 44
2.4 Forecast Models (1.2) - Performance (h= 1) . . . 45
2.5 Forecast Models (2.1) - Performance (h= 4) . . . 46
2.6 Forecast Models (2.2) - Performance (h= 4) . . . 47
2.7 Comparison Innovations Variance Performance . . . 51
3.1 The Data Set . . . 79
3.2 Datastream Codes (1) . . . 80
3.3 Datastream Codes (2) . . . 81
3.4 Datastream Codes (3) . . . 81
3.5 Basic Characteristics of House-Price Cycles . . . 89
3.6 Booms in Short, Medium and Long-term House-Price Cycles . . . 95
3.7 Busts in Short, Medium and Long-term House-Price Cycles . . . 97
3.8 Boom-Bust Cycles in Housing Markets . . . 100
3.9 Macroeconomic and Financial Variables’ Performance over the Boom- Bust Cycle . . . 103
vii
4.2 Datastream Codes . . . 141
4.3 National Housing Finance Characteristics . . . 142
4.4 Recessions in GDP Series . . . 146
4.5 House-Price Booms and Busts . . . 150
4.6 Summary Statistics of the Data Set . . . 153
4.7 Base Model . . . 160
4.8 Boom-Bust Model . . . 163
4.9 Base Model, Corrected for Housing Finance Characteristics (1) . . . . 165
4.10 Base Model, Corrected for Housing Finance Characteristics (2) . . . . 167
viii
Introduction
1
The collapse of Lehman Brothers on September 15th, 2008 marked the preliminary climax of the United States Subprime Mortgage Crisis and the beginning of the Global Financial Crisis, whose macroeconomic and financial implications continue to this day. The crisis had its beginning in the United States subprime housing market and the securitization of mortgage debt obligations (mortgage backed se- curities) that were intermediated to advanced-economy investors in Europe, the United Kingdom and the United States, along with attractive risk ratings (Mian and Sufi, 2009; Shin, 2012; Justinianoet al., 2014). This development pushed credit supply upward, eased mortgage financing costs and hence, resulted in a pronounced house-price boom that only ended when housing affordability came to a halt and delinquency as well as foreclosure rates started to increase (Taylor, 2008; Bernanke et al., 2011).1 In the wake of these events, risk ratings of mortgage backed securities were reassessed and their prices plummeted. This resulted in substantial deprecia- tions among their investors and ultimately to banking imbalances and insolvencies all around the industrialized world, which could only be contained by large-scale public banking bailout programs (Acharya and Richardson, 2009; Acharya and Schnabl, 2010). In the ensuing European Sovereign Debt Crisis it became evident that in some countries these programs were so extensive that even their government sol- vency became threatened and could only be secured by the resolute intervention of the Eurosystem and the establishment of the European Stability Mechanism (Schu- larick, 2014; Jordàet al., 2016).
The Financial Crisis of 2007/2008 shed a new light on the destructive forces of house-price bubbles on macroeconomic and financial stability. Yet the link between house-price busts and banking distress was not entirely unknown. In a sample of 15 industrialized countries (1970−1998), Bordo and Jeanne (2002) identify that every fourth house-price boom is associated with a banking crisis.2 Their result
1The house-price boom was fuelled by the low long-term interest rate environment at that time.
This can mainly be attributed to two reasons. First, low policy rates after the bursting of the dot-com bubble and the terrorist attacks in September, 2001 and second, sticky long-term interest rates due to a global savings glut (Taylor, 2008; Bernankeet al., 2011).
2Bordo and Jeanne (2002) refer to the definition on banking crises on Eichengreen and Bordo (2003). Hence, an episode is classified as banking crisis, if bank runs, widespread bank failures
becomes even more striking, when reversing this relationship. Reinhart and Rogoff (2008) compare the development of early-warning indicators prior to the Financial Crisis of 2007/2008 relative to earlier banking and financial crises and identify that profound house-price upswings are a recurrent characteristic of postwar financial crises. Moreover, they find that in the run-up of the five most severe financial crises, house-price increases were stronger than for the remaining ones.3 One reason for this observation can be attributed to the real estate market’s credit financed nature.
IMF (2003) show that housing market busts are associated with a higher volume of non-performing loans, lower pre-tax profits and a decreased capital-to-asset ratio.
Thus, they affect the banking sector’s capacity and willingness to lend and at worst may even evolve into a credit crunch with severe consequences for the real economy.
These have inter alia been analyzed by Jordà et al. (2015b). They study output reactions during house-price busts that have been preceded by booms, using a panel of 17 industrialized countries from 1870 until 2013. Furthermore, they distinguish between low and high credit house-price bubbles. For low credit house-price bub- bles, they find output effects to be comparable to normal recessions. But, if the bubbles are highly leveraged, they pose a substantial risk for macroeconomic and financial stability with GDP losses of up to 30% over a period of five years.4 This negative impact can be further amplified in countries with a large and labor inten- sive construction sector and a high labor market flexibility, as these will intensify the unemployment rate and, hence, output reactions via decreasing consumption, declines in tax revenues and rising government expenditure (Girouard et al., 2006;
IMF, 2008; Reinhart and Rogoff, 2013). Moreover, GDP reactions are also stronger in countries where instruments such as mortgage equity withdrawal are available, which allow households to tap their housing wealth for consumables (Catte et al., 2004; Mian and Sufi, 2011; Calza et al., 2013).
or banking imbalances and insolvencies are observed.
3Reinhart and Rogoff (2008) classify Spain (1977), Norway (1987), Finland (1991), Sweden (1991) and Japan (1992) to be the five most severe financial crises (for further information, see Kamin- sky and Reinhart, 1999; Gerardet al., 2005).
4Jordàet al.(2015b) define a high credit house-price boom, if bank lending exceeds the historical mean, otherwise they classify it to be a low credit boom.
In the evolution of house-price booms and busts, a key role can be attributed to the monetary authorities, as their decisions influence the affordability of mortgage borrowing. For example for a data set of 17 industrialized countries between 1870 and 2012, Jordàet al. (2015a) show that a decrease in short-term interest rates by 1 percentage point leads to a decline in long-term interest rates of 0.5 percentage points after 2 years. This again favours borrowing of mortgage loans and thus their demand, which subsequently increases house prices by 4 percentage points within 3 years.5 A similar result is obtained by Agnello and Schuknecht (2011), who identify in a probit approach, that an increase in domestic credit demand and a drop in interest rates improve the probability of a build up of house-price booms, while for busts the opposite holds true (for the impact of monetary policy on house-price growth, see also Del Negro and Otrok, 2007; Goodhart and Hofmann, 2008; Jarocinski and Smets, 2008; Glaeseret al., 2012). Yet, not only central banks can influence long-term interest rates, but also regulatory authorities via financial liberalization (Diamond and Lea, 1992). This can either be achieved by opening the mortgage market to nontraditional lenders to intensify competition or by relaxing credit constraints, e.g. the abolition of interest rate and credit controls or the application of reduced risk weights on mortgages secured by residential property (IMF, 2008; Jordà et al., 2016). Both of these measures are aimed at achieving the same goal: an increase in credit supply and thus a fall in mortgage rates to enable more households to participate in the housing market. E.g. Rajan and Ramcharan (2015) show for farm land prices in the United States in the 1920s that in counties with higher banking density, competition for deposit and lending rates is greater, credit supply is larger and thus, land prices are higher.6 For a more recent time period this relationship is also analyzed by Favara and Imbs (2015).
They compare differences in regulatory barriers for interstate bank branch expansion
5For the run-up of house prices prior to the Financial Crisis, the relationship between short and long-term interest rates is less clear. Although the federal funds rate increased between 2004 and 2006 by 425 basis points, long-term interest rates remained low (Bernankeet al., 2011).
6For a panel of 14 industrialized countries between 1870 and 2012, Knollet al.(2017) find that the main driver of house-price increases is rising land prices (from 1950 onwards). They can show this by decomposing house-price series into two components: a land price and a construction costs component.
across the United States and find between 1994 and 2005 mortgage lending to be higher and house-price increases to be larger in deregulated relative to regulated states. Moreover, they identify that dependent on the elasticity of the housing supply, a rise in credit either leads to higher prices or to a larger housing stock (for the influence of an increased credit supply due to financial market liberalization on house prices, see also Adelino et al., 2012; Favilukis et al., 2012; Justiniano et al., 2015).7
However, deregulation and thus the greater availability of mortgage loans comes at a price: an increased risk of hazardous house-price busts. These have inter alia been studied by Ceruttiet al.(2017). They focus in a first step on different characteristics of housing finance and to which extent these affect the predictive power of observing a house-price boom. Among these, they find that the maximum observed loan-to- value ratio, the presence of full recourse and securitization as well as wholesale as banks’ funding types, significantly increase a country’s probability of experiencing a house-price boom.8 In a second step, they repeat this set of exercises for bad booms (booms that ended in recession) and show that those are less likely in countries where banks’ funding mainly relies on retail deposits. Their interpretation of these results is that non-retail deposit funding is more scalable in boom times, but can also easily be drained, when housing markets turn into busts. This interpretation is also in line with Mian and Sufi (2009) for the most recent financial crisis. In their analysis, they use United States ZIP code-level data for mortgage loans and default rates and identify between 2002 and 2005 a strong increase in securitization and a drop in credit denial rates in subprime relative to prime areas. In addition, they find that from 2006 onwards, these trends reversed, leading to a sharp increase in credit defaults and thus to severe contractions in the housing market. Pavlov and Wachter (2011) develop this link between an easing of credit standards and increases in house
7The elasticity of housing supply is mainly determined by geographical restrictions (Saiz, 2010) or by land use regulations (Hilber and Robert-Nicoud, 2013).
8Theloan-to-value ratiocharacterizes the maximum share of mortgage loan relative to the housing value that is granted to finance real estate. Full recourse indicates the degree borrowers are liable for their mortgages in addition to the collateral andbanks’ funding typespecifies the main sources of banks to refinance their mortgage loans (Ceruttiet al., 2017).
prices even further by considering, besides subprime mortgages, also other aggressive lending instruments, such as interest-only, negative-amortization or zero-equity loans and show that in areas with a higher concentration of these loan types, house-price upturns are more pronounced, but their downturns are also deeper. They attribute this finding to a stronger demand of these lending instruments in less affordable markets, where households are credit constrained for traditional mortgage contracts (for the expansion of credit supply to riskier borrowers and its impact on house price growth, see also Landvoigt et al., 2015; Adelino et al., 2016; Di Maggio and Kermani, 2017).
This short introduction is aimed at outlining the importance of house-price move- ments as early-warning indicator for emerging macroeconomic and financial imbal- ances. However, this is only the most prominent and vivid reason why the study of housing markets has recently attracted a rapidly growing interest. There are also reasons apart from periods of banking distress in which the analysis of house prices contributes invaluably to macroeconomic research: residential investment and con- struction account for a large share of GDP. Hence, depending on the size of the construction sector as well as the labor market flexibility and the labor intensity, residential investment provides an important early warning signal for oncoming busi- ness cycle downturns (see e.g. Catte et al., 2004; IMF, 2008). For example Leamer (2007) reports for the United States that residential investment has been the primary leading indicator for the past eight recessions since World War II, excluding the ones after the Korean War and the bursting of the dot-com bubble. In addition, hous- ing also constitutes a major component of households’ wealth and is generally more evenly distributed than, for instance, stock market wealth. Accordingly, changes in house prices can lead to significant consumption effects via the consumption wealth channel (Ludvigson et al., 2002; Case et al., 2005, 2012). This channel can even be intensified in countries, where households are legally entitled to consume out of their housing wealth via equity release products, such as mortgage equity with- drawal (Girouardet al., 2006; Mian and Sufi, 2011). In these countries, house-price variations also represent an important factor in the monetary policy transmission (Calza et al., 2013). Finally, the purchase of residential property is usually the
largest transaction in most households’ lives and generally highly leveraged (Jordà et al., 2015b). Thus, mortgage credits represent a large position on the assets side of banks’ balance sheets and vice versa on the liabilities side of households’ balance sheets. Consequently, house-price changes, which lead to revaluations of these bal- ance sheet positions, influence the ability to borrow (balance sheet channel) as well as the willingness to lend (bank capital channel) and ultimately via their spending behavior also aggregate demand (Iacoviello and Minetti, 2008; Jorda et al., 2013;
Jordà et al., 2016).
This dissertation extends the current literature on the role of real estate prices in macroeconomic research in three dimensions. In the first part, we study the informa- tional content of various macroeconomic, monetary and demographic fundamentals for forecasting house-price growth in the six largest countries of the European Mon- etary Union, namely Belgium, France, Germany, Italy, the Netherlands, and Spain.
To do this, we use the newly developed dynamic modeling averaging and selection approach. This econometric methodology has recently been introduced into the housing market literature by Bork and Møller (2015) and has the advantage of be- ing able to control for fundamental shifts such as the end of the Cold War, the start of the European Monetary Union, and the recent financial crisis. This is done by al- lowing the coefficients as well as the forecasting model to change over time (Raftery et al., 2010; Koop and Korobilis, 2012). As a result, we obtain for every regressor a probability of inclusion over the evaluation period and can assess whether there exist heterogeneities across the countries’ main drivers of house-price growth. Our set of regressors includes the following variables:
– Personal disposable income, gross domestic product, and industrial production to capture the interrelation of house prices and the business cycle (Iossifov et al., 2008; Agnello and Schuknecht, 2011).
– Residential investment to measure shifts in housing supply, based on the q theory of investment (IMF, 2008; Igan and Loungani, 2012).
– Effective exchange rate to cater for foreign housing demand (Abelson et al., 2005; Calza et al., 2013).
– Term spread and consumer price inflation to seize information about the di- rection of future short-term interest rate movements (Tsatsaronis and Zhu, 2004).
– Stock price index to assess business prospects and, thus, future income growth (Sutton, 2002).
– Credit supply to the non-financial sector andbroad money as proxies for mort- gage credit and the deepness of the countries’ financial markets (Otrok and Terrones, 2004; Iossifov et al., 2008).
– Total labor force and the unemployment rate to control for the number of potential buyers in the housing market (Iossifov et al., 2008).
Naturally, all the variables have already been found in the economic literature to also influence house prices. Our main finding is that house prices in our country sample are influenced heterogeneously. Hence, we do not identify any variable that is important for all countries over the whole forecasting period. In Belgium, France, and Germany, we find house-price growth to be mainly driven by macroeconomic fundamentals such as the unemployment rate, the consumer price inflation, gross domestic product and residential investment, whereas in the Netherlands and Spain, financial variables such as the term spread and the credit supply are more important.
Furthermore, the Italian market seems to be rather unaffected by our predictors.
We attribute these results to the heterogeneous development of our country sample’s mortgage markets. However, as we do not consider mortgage credit or the volume traded in secondary markets and mortgage equity withdrawal, this interpretation can not be verified. Nevertheless, this article provides essential insights on the role of macroeconomic and financial fundamentals for European house-price movements.
The second part of this dissertation concerns itself with housing market boom and bust phases. In the analysis of house-price cycles, one distinguishes between theclas- sical and the growth cycle (Harding and Pagan, 2002). The former defines house- price booms and busts via turning points in the (logarithmized) series, while the latter considers them as deviations from their trend (for theclassical cycle analysis, see e.g. IMF (2003), Helbling (2005) and Claessenset al. (2012) and for thegrowth
cycle analysis, see e.g. Borio and Lowe (2002), Detken and Smets (2004) and Good- hart and Hofmann (2008)).9 Furthermore, both approaches use censoring rules to avoid the identification of spurious cycles. Yet, these rules differ substantially in the housing cycle literature. This section contributes to this literature by highlight- ing to which extent the cyclical patterns of housing markets change, when varying the minimum distance between two turning points. We are doing this by applying three different minimum cycle lengths on the cyclical component ofHodrick Prescott filtered house-price series using quarterly data for 18 industrialized countries from 1975/1 to 2016/2. One method that has proven suitable for this purpose is the Harding and Pagan (2002) algorithm. It is based on a three step approach:
1. Definition of the minimum cycle and phase length, q and r.
2. Identification of all observations within a period of q quarters that satisfy the following criteria:
– for a peak: (yt−r, . . . , yt−1)< y+t >(yt+1, . . . , yt+r) and – for a trough: (yt−r, . . . , yt−1)> y−t <(yt+1, . . . , yt+r)
with (yt−r, . . . , yt−1)≥r; (yt+1, . . . , yt+r)≥rand (yt−r, . . . , yt+r)≥qquarters.
3. A full cycle is determined, if it consists of an upturn (trough to peak) and a downturn (peak to trough) phase or vice versa.
The identified cycles are then studied further with respect to their Duration, Am- plitude,Slope and Severity:
– Duration: Length between a trough (peak) and its successive peak (trough).
– Amplitude: Percentage deviation of the logarithmized index between a trough (peak) and its successive peak (trough).
9In academics, filters that have proven suitable for detrending macroeconomic time series are the band pass or the Hodrick Prescott filter (see e.g., Hodrick and Prescott, 1997; Christiano and Fitzgerald, 2003).
– Slope: Quotient of the Amplitude and the Duration (in years), denoting the yearly change.
– Severity: Triangle area of the upturn (downturn), with the Duration and Amplitude as the two legs.
The three minimum cycle lengths that we compare are inspired by the literature. We refer to them as short-term (q = 5, r= 2), medium-term (q = 13, r = 6) and long- term (q= 19,r= 9) cycle (see e.g., IMF, 2003; Borio and McGuire, 2004; Drehmann et al., 2012). Moreover, we define a boom (bust), if theSeverity falls above (below) the 75th (25th) percentile of all observed house-price upturns (downturns) across all countries with a minimum cycle and phase length of q = 5 and r = 2. We then compare our obtained house-price booms and busts across the three cycle lengths and use the results in a second step to derive boom-bust cycles. Our main findings can be summarized as follows: First, the long-term cycle definition captures housing market boom and bust cycles best. Second, average boom-bust cycles have a length of roughly 60 quarters, half split across both phases. Furthermore, they rise in the upswing by 60 percent, fall in the downswing by 40 percent and are associated with GDP losses of roughly 25 percent. Third, we can not verify the finding of Agnello and Schuknecht (2011) that the longer and stronger the boom, the longer and sharper will also be the price corrections in the subsequent bust. Hence, the article depicts nicely that interrelations between house-price booms and busts with respect to the Duration, Amplitude, Slope and Severity are sensitive to the chosen housing cycle definition.
The third and final part of this dissertation links the results of the preceding two chapters, by analyzing to what extent the influence of housing market determinants and especially the credit market vary across countries and time. We are doing this by controlling for periods of house-price booms and busts relative to normal phases and a structural break in 1985 (Great Moderation). Moreover, we consider selected characteristics of housing finance, to account for differences in financially regulated and deregulated countries. From the economic housing market literature it is known, that in periods of house-price booms the positive impact of credit
growth on house prices is intensified, but insignificant in bust phases (Goodhart and Hofmann, 2008; IMF, 2011) and, that countries with high loan-to-value ratios and developed secondary credit markets are more likely to experience a house-price boom, yet these are also more likely to end in recession (Cerutti et al., 2017). Our contribution is to connect these two strands of literature by proceeding as follows.
First, we start our analysis on the basic relationship between the housing market and its determinants by estimating acountry and time-fixed panel model, using quarterly data for 18 industrialized countries between 1975/01 and 2017/02. Our determinants include macroeconomic, monetary and demographic fundamentals and are inspired by chapter 2. In the second and third step, we then gradually extend our base model to assess how our regressors vary between financially regulated and liberalized countries and whether the credit market’s influence differs over the housing cycle.
For the housing cycle, we refer to the long-term cycle definition from chapter 3, that we found to best capture house-price booms and busts. Our results show inter alia that a high loan-to-value ratio (LTV ≥ 100) has a large and positive impact on house-price growth, while the most severe contractions can be found in countries whose banks primarily rely on non-retail funding. Furthermore, these are also the countries where a change in credit growth is transmitted strongest to the housing market in normal and boom times. We interpret this finding to shed some new light on the role of the LTV ratio and banks’ funding type in shaping house-price movements. In normal times, a high LTV ratio simplifies households to access the real estate market, due to the lower downpayment. In addition, in countries where secondary credit markets are more developed, credit supply can be easier adjusted.
Hence, even in boom times when credit demand is spurring, mortgage rates will be favorable. However, as could be observed in the Financial Crisis of 2007/2008, this additional credit supply can also be drained more easily, and thus intensify the housing market downturn.
This dissertation extends the economic housing market literature in several aspects.
First, we show that international housing markets are heterogeneously linked to the rest of the economy. Second, we compare different definitions of house-price booms and busts and reveal that interrelations between the upswings and downswings are
sensitive to the chosen definition and third, we highlight that mortgage market characteristics are capable of intensifying the impact of the credit market on house- price growth innormal and boom times. However, our empirical approach can only determine whether and to which extent our variables explain house-price growth, but is not capable of identifying the transmission channels through which they exert their influences. This would require a more theoretical approach. Nevertheless, our results also provide essential insights for theoretical analyses. In particular that the development of mortgage markets is crucial for the determination of international house prices and that the influences of housing market fundamentals vary over the house-price cycle.
However, there are also some caveats with respect to the data quality when analyz- ing international housing markets as the construction of house-price indexes is not internationally harmonized. These caveats exist in three dimensions: First, housing is a heterogeneous good. Hence, its quality varies due to the location, the year of construction (e.g. new vs. existing) or the size of the property (e.g. single- family houses vs. dwellings). In addition, housing quality differs from the extent of refurbishment and maintenance work completed over the life of the property. Sec- ond, house-price series are generally constructed from actual transactions within a given sample period. Yet, these transactions may be distributed infrequently and unequally across areas, and third, the data coverage of the house-price series may be diverse. For example some house-price series measure nationwide developments while others only refer to metropolitan areas. Besides, as most national house-price indexes are constructed from several sources, these caveats may not only exist be- tween but also within countries (see e.g., Mack and Martìnez-Garcia, 2011). In the further course of this dissertation, we use the house-price series from the Interna- tional House Price Database of the Federal Reserve Bank of Dallas. They have been designed to be most consistent with the quarterly U.S. house price index for exist- ing single-family houses provided by the Federal Housing Finance Agency. Hence they guarantee at least for the house-price series’ conceptual and methodological framework (e.g. data aggregation, seasonal adjustment and release of the data) a
sufficient level of international comparability.10
Finally, this dissertation can be extended to further essential questions regarding the importance of house prices in macroeconomic research. First, a valuable ex- tension would be to integrate the size and regulation of the rental market into our analyses. On the one hand, a large rental market reduces the demand in the oc- cupier’s market and on the other hand, only less credit constrained households will demand residential property. Hence, it can be assumed that in countries with large and strongly regulated rental markets house-price booms are less pronounced and less likely to end in recession. Second, it would be of great value to apply our anal- ysis on micro level data to assess the dynamics of structural change in explaining house-price movements. A suitable model could capture the manufacturing sec- tor to demand production space on the outskirts of cities, where large areas are available and transport links are advantageous. On the contrary, the service sector rather demands production space in the city centre where land is scarce, but where it is closest to its customers. Hence, as opposed to the manufacturing sector, the service sector competes with households for space. Therefore, it can be expected that in cities where production is more service relative to manufacturing orientated, house prices are higher. Third, younger households, due to their shorter period of employment, are normally more credit constrained than older households. Thus, residential property is generally owned by older and rented by younger households.
Consequently, rising house prices will decrease younger household’s consumption and increase older household’s savings. This in turn will lead to falling interest rates, even less attractive conditions for younger households to save and an intensification of this intergenerational wealth redistribution. In the medium term, this could lead to serious macroeconomic distortions. For example, Kumhofet al.(2015) show that rising inequality encourages poorer households to over-levarage, which eventually causes financial crises. Finally, housing is not only an important good for living but also an asset. Hence, changes in the return on housing will also affect other asset
10see, https://www.dallasfed.org/institute/houseprice for further information on the database.
classes. For this reason, understanding the links between different asset classes gives valuable insights on the cyclical pattern of the financial cycle.
1.1 Bibliography
Abelson, P.,Joyeux, R., Milunovich, G.and Chung, D. (2005). Explaining House Prices in Australia: 1970-2003. Economic Record, 81 (255), 96–103.
Acharya, V. V. and Richardson, M. (2009). Causes of the Financial Crisis.
Critical Review,21 (2-3), 195–210.
— and Schnabl, P. (2010). Do Global Banks Spread Global Imbalances? The Case of Asset-Backed Commercial Paper During the Financial Crisis of 2007-09.
IMF Economic Review, 58 (1), 37–73.
Adelino, M., Schoar, A. and Severino, F. (2012). Credit Supply and House Prices: Evidence from Mortgage Market Segmentation, NBER Working Paper No. 17832.
—, — and —(2016). Loan Originations and Defaults in the Mortgage Crisis: The Role of the Middle Class. Review of Financial Studies, 29 (7), 1635–1670.
Agnello, L.andSchuknecht, L.(2011). Booms and Busts in Housing Markets:
Determinants and Implications.Journal of Housing Economics,,20(3), 171–190.
Bernanke, B. S., Bertaut, C., DeMarco, L. P. and Kamin, S. (2011).
International Capital Flows and the Returns to Safe Assets in the United States 2003–2007. Banque de France Financial Stability Report,15, 13–26.
Bordo, M. D. and Jeanne, O.(2002). Monetary Policy and Asset Prices: Does
‘Benign Neglect’ Make Sense? International Finance,5 (2), 139–164.
Borio, C. and Lowe, P. (2002). Asset Prices, Financial and Monetary Stability:
Exploring the Nexus. BIS Working Papers, No. 114.
— and McGuire, P. (2004). Twin Peaks in Equity and Housing Prices? BIS Quarterly Review, March, pp. 79–93.
Bork, L. and Møller, S. V. (2015). Forecasting House Prices in the 50 States using Dynamic Model Averaging and Dynamic Model Selection. International Journal of Forecasting,31 (1), 63–78.
Calza, A.,Monacelli, T.and Stracca, L.(2013). Housing Finance and Mon- etary Policy. Journal of the European Economic Association, 11 (S1), 101–122.
Case, K. E., Quigley, J. M. and Shiller, R. J. (2005). Comparing Wealth Effects: The Stock Market versus the Housing Market. Advances in Macroeco- nomics, 5 (1), 1–34.
—,—and—(2012). Wealth Effects Revisited 1978–2009.Critical Finance Review, 2, 101–128.
Catte, P., Girouard, N.,André, C. andPrice, R. (2004). Housing Markets, Wealth and the Business Cycle, OECD Economics Department Working Papers, No. 394.
Cerutti, E., Dagher, J. and Dell’Ariccia, G. (2017). Housing Finance and Real-Estate Booms: A Cross-Country Perspective.Journal of Housing Economics, 38, 1–13.
Christiano, L.and Fitzgerald, T.(2003). The Band Pass Filter.International Economic Review, 44 (2), 435–465.
Claessens, S., Kose, M. A. and Terrones, M. E. (2012). Financial Cycles:
What? How? When? In R. H. Clarida and F. Giavazzi (eds.), NBER Interna- tional Seminar on Macroeconomics 2010, University of Chicago Press, pp. 303–
343.
Del Negro, M.and Otrok, C.(2007). 99 Luftballons: Monetary Policy and the House Price Boom across U.S. States. Journal of Monetary Economics, 54 (7), 1962–1985.
Detken, C.andSmets, F.(2004). Asset Price Booms and Monetary Policy. ECB Working Paper, No. 364.
Di Maggio, M. and Kermani, A. (2017). Credit-Induced Boom and Bust. The Review of Financial Studies,30 (11), 3711–3758.
Diamond, D. B. and Lea, M. J. (1992). The Decline of Special Circuits in De- veloped Country Housing Finance. Housing Policy Debate, 3 (3), 747–777.
Drehmann, M., Borio, C. and Tsatsaronis, K. (2012). Characterising the Financial Cycle: Don’t Loose Sight of the Medium-Term! BIS Working Papers, No. 380.
Eichengreen, B. and Bordo, M. D. (2003). Crises Now and Then: What Lessons from the Last Era of Financial Globalization? In Monetary History, Exchange Rates and Financial Markets, 3, Edward Elgar Publishing.
Favara, G.andImbs, J.(2015). Credit Supply and the Price of Housing.American Economic Review, 105 (3), 985–992.
Favilukis, J., Kohn, D., C. Ludvigson, S. and Van Nieuwerburgh, S.
(2012). Housing and the Financial Crisis, University of Chicago Press, chap. In- ternational Capital Flows and House Prices: Theory and Evidence, pp. 235–299.
Gerard, C.,Klingebiel, D.,Laeven, L.andNoguera, G.(2005). Appendix:
Banking Crisis Database. In P. Honohan and L. Laeven (eds.),Systemic Financial Crises: Containment and Resolution, Cambridge, U.K.: Cambridge University Press, pp. 341–360.
Girouard, N., Kennedy, M., van den Noord, P. and André, C. (2006).
Recent House Price Developments: The Role of Fundamentals, OECD Economics Department Working Papers, No. 475.
Glaeser, E.,Gottlieb, J. andGyourko, J.(2012). Can Cheap Credit Explain the Housing Boom? In Housing and the Financial Crisis, National Bureau of Economic Research, Inc, pp. 301–359.
Goodhart, C. and Hofmann, B. (2008). House Prices, Money, Credit, and the Macroeconomy. Oxford Review of Economic Policy, 24 (1), 180–205.
Harding, D. and Pagan, R., Adrian (2002). Dissecting the Cycle: A Method- ological Investigation. Journal of Monetary Economics,49, 365–381.
Helbling, T. F.(2005). Housing Price Bubbles – A Tale Based on Housing Price Booms and Busts. InReal Estate Indicators and Financial Stability, vol. 21, Bank for International Settlements, pp. 30–41.
Hilber, C. A.andRobert-Nicoud, F.(2013). On the Origins of Land Use Regu- lations: Theory and Evidence from US Metro Areas.Journal of Urban Economics, 75, 29–43.
Hodrick, R. J.and Prescott, C., Edward (1997). Postwar U.S. Business Cy- cles: An Empirical Investigation. Journal of Money, Credit and Banking, 29 (1), 1–16.
Iacoviello, M.andMinetti, R.(2008). The Credit Channel of Monetary Policy:
Evidence from the Housing Market. Journal of Macroeconomics, 30 (1), 69–96.
Igan, D.and Loungani, P. (2012). Global housing cycles. IMF Working Paper, No. 217.
IMF(2003). When Bubbles Burst.World Economic Outlook, April 2003, pp. 61–94.
IMF(2008). Housing and the Business Cycle.World Economic Outlook April 2008, pp. 103–132.
IMF(2011). Durable Financial Stability Getting from There to Here. Global Finan- cial Stability Report April 2011, pp. 111–157.
Iossifov, P., Cihák, M.ˆ and Shanghavi, A. (2008). Interest Rate Elasticity of Residential Housing Prices, IMF Working Paper No. 08/247.
Jarocinski, M.and Smets, F. (2008). House Prices and the Stance of Monetary Policy. Federal Reserve Bank of St Louis Review, 90 (4), 339–365.
Jorda, O., Schularick, M. and Taylor, A. (2013). When Credit Bites Back.
Journal of Money, Credit and Banking, 45 (S2), 3–28.
Jordà, O., Schularick, M. and Taylor, A. M. (2015a). Betting the House.
Journal of International Economics,96, S2–S18.
—, — and — (2015b). Leveraged Bubbles. Journal of Monetary Economics, 76, S1–S20.
—, — and — (2016). Sovereigns Versus Banks: Credit, Crises, and Consequences.
Journal of the European Economic Association,14 (1), 45–79.
Jordà, O.,Schularick, M.andTaylor, A. M.(2016). The Great Mortgaging:
Housing Finance, Crises and Business Cycles.Economic Policy,31(85), 107–152.
Justiniano, A., Primiceri, G. E. and Tambalotti, A. (2014). The Effects of the Saving and Banking Glut on the U.S. Economy. Journal of International Economics, 92, S52–S67.
—,—and—(2015). Credit Supply and the Housing Boom, NBER Working Paper No. 20874.
Kaminsky, G. L.andReinhart, C. M.(1999). The Twin Crises: The Causes of Banking and Balance-Of-Payments Problems.American Economic Review,89(3), 473–500.
Knoll, K.,Schularick, M.andSteger, T.(2017). No Price Like Home: Global House Prices, 1870–2012. American Economic Review,107 (2), 331–353.
Koop, G.andKorobilis, D. (2012). Forecasting Inflation Using Dynamic Model Averaging. International Economic Review, 53 (3), 867–886.
Kumhof, M., Rancière, R. and Winant, P. (2015). Inequality, Leverage, and Crises. American Economic Review,105 (3), 1217–1245.
Landvoigt, T., Piazzesi, M. and Schneider, M. (2015). The Housing Mar- ket(s) of San Diego.American Economic Review, 105 (4), 1371–1407.
Leamer, E. E.(2007). Housing is the Business Cycle, NBER Working Paper, No.
13428.
Ludvigson, S., Steindel, C. and Lettau, M. (2002). Monetary Policy Trans- mission through the Consumption-Wealth Channel. Economic Policy Review, (May), 117–133.
Mack, A. and Martìnez-Garcia, E. (2011). A Cross-Country Quarterly Database of Real House Prices: A Methodological Note, Federal Reserve Bank of Dallas Globalization and Monetary Policy Institute Working Paper No. 99.
Mian, A.and Sufi, A. (2009). The Consequences of Mortgage Credit Expansion:
Evidence from the U.S. Mortgage Default Crisis. The Quarterly Journal of Eco- nomics, 124 (4), 1449–1496.
—and—(2011). House Prices, Home Equity-Based Borrowing, and the US House- hold Leverage Crisis. American Economic Review, 101 (5), 2132–2156.
Otrok, C. and Terrones, M. E. (2004). The Global House Price Boom. World Economic Outlook, 2004 (September), 71–89.
Pavlov, A.and Wachter, S. (2011). Subprime Lending and Real Estate Prices.
Real Estate Economics, 39 (1), 1–17.
Raftery, A. E., Kaŕný, M. and Ettler, P. (2010). Online Prediction Under Model Uncertainty via Dynamic Model Averaging: Application to a Cold Rolling Mill. Technometrics, 52 (1), 52–66.
Rajan, R. and Ramcharan, R. (2015). The Anatomy of a Credit Crisis: The Boom and Bust in Farm Land Prices in The United States in the 1920s.American Economic Review, 105 (4), 1439–1477.
Reinhart, C. M.andRogoff, K. S.(2008). Is the 2007 US Sub-Prime Financial Crisis so Different? An International Historical Comparison. American Economic Review, 98 (2), 3.
— and Rogoff, S., Kenneth (2013). Banking Crises: An Equal Opportunity Menace. Journal of Banking & Finance, 37, 4557–4573.
Saiz, A.(2010). The Geographic Determinants of Housing Supply. Quarterly Jour- nal of Economics, 125 (3), 1253–1596.
Schularick, M. (2014). Public and Private Debt: The Historical Record (1870- 2010). German Economic Review, 15 (1), 191–207.
Shin, H. S.(2012). Global Banking Glut and Loan Risk Premium. IMF Economic Review, 60 (2), 155–192.
Sutton, G. D.(2002). Explaining Changes in House Prices.BIS Quarterly Review, pp. 46–55.
Taylor, J. B.(2008). The Financial Crisis and the Policy Responses: An Empirical Analysis of What Went Wrong. A Festschrift of David Dodge’s Contributions to Canadian Public Policy, Bank of Canada, pp. 1–18.
Tsatsaronis, K. and Zhu, H. (2004). What Drives Housing Price Dynamics:
Cross-Country Evidence. BIS Quarterly Review, 2004 (March), 65–78.
Forecasting House-Price Growth in the Euro Area with Dynamic Model Averaging
Marian Rissea and Martin Kernb
Published in: North American Journal of Economics and Finance, 38, pp. 70–85 Presented at: 14th Annual Conference of the European Economics and Finance So-
ciety (EEFS), Brussels (Belgium), June 2015
a Department of Economics, Helmut Schmidt University, Holstenhofweg 85, P.O.B.
700822, 22008 Hamburg
b Department of Economics, University of Hagen, Universitätsstraße 11, 58084 Ha- gen, Germany
23
We use a dynamic modeling and selection approach for studying the informational content of various macroeconomic, monetary and demographic fundamentals for forecasting house-price growth in the six largest countries of the European Mone- tary Union. The approach accounts for model uncertainty and model instability.
We find superior performance compared to various alternative forecasting models.
Plots of cumulative forecast errors visualize the superior performance of our ap- proach, particularly after the recent financial crisis.
JEL classification: C32; C53; R30
Keywords : House prices; Dynamic model averaging; Forecasting; Europe
Acknowledgements:
We thank Michael Berlemann, Marc-André Luik, Christian Pierdzioch, Max Friedrich Steinhardt, and two anonymous reviewers for invaluable feedback. We are grateful to the participants of the Annual Conference of the European Economics and Finance Society in Brussels (2015) for helpful comments. The usual disclaimer applies.
25
2.1 Introduction
House price busts can substantially harm an economy’s financial and economic sta- bility, as has become evident during the Subprime Mortgage Crisis of 2007/2008.
The IMF (2003) analyzed 20 house price busts in 14 countries from 1970-2002, and found that any of the analyzed countries experiences a collapse in house prices ev- ery 20 years with a mean price correction of 30% over a period of 4 years. Leamer (2007) reports for the U.S. market that, excluding periods of the Korean War and the burst of the New Economy Bubble in 2001, the housing market has been a primary leading indicator for recessions since 1949. Plakandaras et al. (2015) validate these findings for the breakdown of the U.S. real estate market during the Global Finan- cial Crisis from 2006 to 2009. Hence, forecasting house prices may contain valuable information for analyzing business-cycle movements (see Stock and Watson, 2003;
Gupta and Das, 2010; Gupta and Hartley, 2013). In times of economic growth, rising demand for housing pushes upward residential investment and construction employment and thereby strengthens aggregate demand. In contraction phases, in turn, falling income and job uncertainty decrease housing demand, reduce prices and weaken the attractiveness of residential investment. As a result, the construction sector cuts production and lays off employees, further dampening business prospects and accelerating economic downturns.
Throughout the last years, improvements in econometric forecasting techniques spurred researchers to study forecasting models with a wide range of possible pre- dictors for house prices.1 However, most of the studies focus on the U.S. market.
Rapach and Strauss (2007, 2009) use an autoregressive distributed lag model to investigate real house prices for various U.S. states and districts. They show that their approach outperforms a simple autoregressive model using a medium-sized set
1Early studies on predicting house prices try to improve forecast precision by testing for serial correlation in past returns (see Gau, 1985; Case and Shiller, 1989, among others) or including macroeconomic and monetary variables (see Linneman, 1986; Case and Shiller, 1990; Muell- bauer and Murphy, 1997). Ghysels et al.(2012) provide a detailed survey of the house-price forecasting literature over the past 30 years.
of macroeconomic and financial regressors. Jarociński and Smets (2009) examine the boom and bust cycles in the U.S. housing market before and after the financial crisis. Their results, derived from a Bayesian vector autoregressive model, suggest that the federal funds rate is a good indicator for house prices in 2000 and their drop in 2006. Miles (2008) argues that non-linear models increase the forecasting perfor- mance of housing markets, especially if they are subject to great volatility. Gupta and Miller (2012) find recursive models to provide a superior fit in forecasting the house price index in eight metropolitan areas in Southern California. Kouwenberg and Zwinkels (2014) utilize a smooth-transition model to analyze the deviation of U.S. house prices from their fundamental values. They show that, between 1992 and 2005, house prices were mainly driven by positive autocorrelation. Plakandaras et al.(2015) develop a forecasting methodology which is based on a machine-learning approach and predict the downfall in the U.S. real estate market (2006-2009) up to two years ahead.
Moving the focus away from the U.S. market towards an international view, re- searchers applied vector autoregressive (VAR) models for examining influential de- terminants of house-price movements in a cross-country framework. Sutton (2002) observes that fluctuations of the stock market play a similar role in explaining changes in house prices as do gross national products and real short-term interest rates for Australia, Canada, Ireland, the Netherlands, the UK, and the U.S. Tsat- saronis and Zhu (2004) examine house-price dynamics in 17 industrialized countries and find that 50% of the price variation can be explained by inflationary shocks.
One of their interpretations of this finding is that inflation directly affects mortgage financing costs and hence, housing demand, through the nominal interest rate. Al- gieri (2013) estimates a structural VAR for France, Germany, Italy, the Netherlands, Spain, the UK, and the U.S. Her results indicate that income, labor force, and the stock market positively influence house prices, while long-term interest rates and private residential investment have a negative effect.2
2Other researchers who analyze house prices in an international context are, among others, Ter- rones and Otrok (2004); Belkeet al.(2008); Beltratti and Morana (2010); Bagliano and Morana (2012).
We contribute to this growing and significant literature by applying a dynamic model averaging and selection approach on a data set consisting of macroeconomic, mone- tary, and demographic fundamentals to predict house-price growth in the six largest countries of the Euro area (Belgium, France, Germany, Italy, the Netherlands, and Spain).3 In particular, we are interested in identifying the main drivers of national house-price movements in a pseudo out-of-sample forecasting exercise. We expect a diverging link between housing markets and the rest of the economy across the countries in our data set because their mortgage markets are heterogeneously dereg- ulated, with the highest degree of liberalization being observed for the Netherlands and Spain (see Iacoviello, 2005; Calza et al., 2013; Cesa-Bianchi et al., 2015).4 The dynamic model averaging and selection approach was recently introduced into the housing literature by Bork and Møller (2015) to predict house-price growth for individual U.S. federal states.5 The approach has the advantage that it accounts for model and parameter uncertainty which are likely to be characteristic features of the data given recent crises. Therefore, the econometric model controls for fundamental shifts in the European economy, such as the end of the Cold War, the start of the EMU, and the recent financial crisis. Our results show that the approach leads to a substantial improvement in forecasting performance as compared to a simple autoregressive process. Moreover, we show that our model performs particularly well in uncertain times of the recent financial crisis, when accounting for various alternative forecasting models.
We organize the remainder of this paper as follows. In Section 2.2, we describe the econometric methodology. In Section 2.3, we characterize our data. In Section 2.4, we summarize the results. In Section 2.5, we conclude.
3The choice of these countries is primarily driven by the availability of data. For a similar choice of countries representing the Euro area, see Forniet al.(2003).
4Mortgage markets in the Netherlands and Spain stand out due to their ability for mortgage equity withdrawal and the size of their secondary markets (see IMF, 2008). Both indicators have been found in the literature to influence house prices (see Mian and Sufi, 2009; Ebner, 2013).
5The methodology was introduced by Raftery et al. (2010). For recent applications to other economic research, see Guptaet al.(2014); Buncic and Moretto (2015); Beckmann and Schüssler (2016), among others.
2.2 Econometric Methodology
2.2.1 Dynamic Model Averaging
We begin explaining the econometric framework with a short presentation of the dynamic model averaging and selection approach.6 For a more in-depth discussion, we refer the interested reader to the paper by Rafteryet al. (2010).
We consider a time-varying coefficients model in state-space representation
yt=x(k)t−1θ(k)t +(k)t , (2.1a) θt(k)=θt−1(k) +νt(k), (2.1b) where yt denotes the growth rate of real house prices, x(k)t−1 denotes a vector of ex- planatory variables and θt(k) denotes a vector of coefficients. (k)t is N(0, Ht(k)) and νt(k) is N(0, Q(k)t ). The superscript k= 1, ..., K denotes the models to be recursively updated using the Kalman filter.7 We compute forecasts for all possible combina- tions of explanatory variables (K = 2N for N predictors) in every period of time t. The final one-step ahead prediction is then computed by averaging all K models according to their probability of occurrence, πt(k), or by selecting the model which leads to the best performance (according to its probability). The first approach is known as dynamic modelaveraging (DMA), whereas the second approach is known as dynamic model selection (DMS). As in Raftery et al. (2010), we use forgetting factors α and λ to achieve feasible computation of πt(k) and Q(k)t . We addition- ally follow Koop and Korobilis (2012) and estimateHt(k) assuming an Exponentially Weighted Moving Average (EWMA) process that includes a third forgetting factor κ.
6The econometric model in this research has been estimated using the free R programming envi- ronment (R Development Core Team, 2015), based on Matlab code for the core routines written by Koop and Korobilis (2012).
7In contrast to other model selection algorithms such as least angle regressions, the updating of the time-varying posterior distribution accounts for model uncertainty. Moreover, the inclusion of the full model space,K, allows the “true” model to change over time.
The one-step-ahead out-of-sample forecasts for period t+ 1 are estimated using all data up to time t and the updated coefficients and model probabilities. The DMA forecast is computed as
ˆ
yt|t+1DM A=
K
X
k=1
π(k)t|tx(k)t 0θˆ(k)t|t, (2.2) and for the case of model selection as
ˆ
yDM St|t+1 =
K
X
k=1
π∗t|tx(k)t 0θˆ(k)t|t, (2.3)
with1(πt|t∗ =πt|t(k)) being an indicator which characterizes the largest model probabil- ity. As an additional robustness check, we implement the dynamic cluster estimator (DMC) that is applied in Bork and Møller (2015) and originally goes back to Aiolfi and Timmermann (2006). The DMC estimator is computed by subdividing the set of probabilities in descending order in equal-sized groups and form a forecast from the group including the highest probabilities using the model averaging approach.
The forecast is then computed as
ˆ
yDM Ct|t+1 =
Kc
X
k=1
πt|t(k)x(k)t 0θˆ(k)t|t, (2.4) where Kc determines the last observation in the cluster. The only difference com- pared to the full DMA estimator is that the DMC estimator exclusively relies on the best subset of estimated models.
2.2.2 Evaluation of Forecasts
We use standard out-of-sample evaluation methods to compare forecasts resulting from the DMA/DMC/DMS model with a forecast from a benchmark model. We choose a simple autoregressive process of order one (AR1) and a constant as our preferred benchmark. To assess the performance of the DMA/DMC/DMS forecasts, we use the following specifications. We consider two baseline settings of the DMA approach including the case of forgetting (α = λ = 0.99) and no-forgetting (α = λ= 1). The latter can be interpreted as a special case of Bayesian Model Averaging
(BMA). We also use an automated selection procedure over a predefined grid to select the best values forα and λ. We lay out the procedure in Section 2.3.2.
In order to assess the performance in comparison to alternative forecasting ap- proaches, we additionally evaluate forecasts derived from an autoregressive process of order two, an OLS regression including all predictors and an equally weighted OLS forecast, derived from all possible combinations of the predictors. For the latter, we consider a recursive and a rolling data window. It is also interesting to evaluate forecasts from a model selection algorithm as it is the case for least angle regressions. Therefore, we add the widely studied least absolute shrinkage and selec- tion operator (LASSO) as a last benchmark to our forecasting models. In principle, the LASSO equals a regression model, where the absolute number of predictors is constrained by a penalty term (see Tibshirani, 1996; Elfron et al., 2004).8 In line with the DMA/DMS/DMC forecasts, we also compare the performance of these alternative forecasting approaches with the AR1 benchmark.
We compute the Clark and West (2007, CW) test statistic for nested forecast mod- els. The CW test rests on the assumption that the classical Diebold-Mariano test does not have a standard distribution and includes a correction term to approxi- mately achieve normality.9 It can be computed by using the series of forecasts from forecasting modelI,yt+hI , from the benchmark model, ˆyAR,t+h, and the actual values, yt+h, to form
ft+1= (yI,t+h−yˆAR,t+h)2−[(yt+h−yI,t+h)2−(yI,t+h−yˆAR,t+h)2], (2.5) where the CW test statistic equals the heteroscedasticity and autocorrelation cor- rected t-statistic of a linear regression of ft+1 on a constant. p-values are then computed based on a one-sided test. Rejection of the null hypotheses implies a sig- nificant improvement of the individual forecasting model over the AR1 benchmark.
8The LASSO regression approach was estimated using thepenalized package developed by Goe- man (2010).
9For an application in context of the DMA analysis, see Bork and Møller (2015) and Buncic and Moretto (2015).
Aside from the full sample evaluation, we also study the cumulative forecast errors over time. Cumulative forecast errors were used by Goyal and Welch (2003) and are also applied in the DMA context by Bork and Møller (2015) and Buncic and Moretto (2015). We compute the cumulative forecast errors,cf et, as
cf et=
Tep
X
t=Ttp+1
(ˆ2AR,t−ˆ2I,t), (2.6)
where ˆ2AR,t denotes the forecast error of the benchmark AR1 model, ˆI,t+1 the fore- cast error of the underlying forecasting model, and Ttp (Tep) marks the last obser- vation of the training period (evaluation period). A positive cumulative error in periodt indicates a superior performance over the benchmark AR1 process.
2.3 Empirical Framework
2.3.1 The Data
The data set consists of quarterly observations for Belgium, France, Germany, Italy, the Netherlands, and Spain from 1975/01 to 2015/03. We downloaded most of the data from Datastream. House prices and personal disposable income were taken from the International House Price Database of the Federal Reserve Bank of Dal- las.10 We deflated the macroeconomic variables using the national CPIs and applied transformations to approximately achieve stationarity. We follow Bork and Møller (2015) and Plakandaras et al. (2015) and leave the monetary variables in nominal amounts. A review of the data, their transformation and sources are given in Tables 2.1 and 2.2.
10The International House Price Database of the Federal Reserve Bank of Dallas provides in- ternational house-price indices that are comparable to the U.S. house-price index for exist- ing single-family houses. The French series refers to all types of existing dwellings. The Spanish index is similarly constructed, although it bases on square meter prices. The Bel- gian, Dutch, and German series are constructed of single family homes. Finally, the Italian index only considers properties that are located in the 13 main metropolitan areas. http:
//www.dallasfed.org/institute/houseprice/ gives a detailed description of the database and the house-price series’ methodologies.
