Master’s Thesis International Economics and Business:
The effect of monetary policy on house price bubbles:
A cross-country study
By: J.R. Boer BSc.
S1078445 University of Groningen Faculty of Economics and Business
Date: July 7, 2011
Developments in the housing and mortgages markets have been pointed out as having been instrumental in setting off the recent turmoil in the international financial system. For nearly two decades prior to these events policy rates have been kept predominantly low to stimulate economic growth. This study investigates the possible link between loose monetary policy and the occurrence of house price bubbles, by constructing a basic macroeconomic model using multiple interest rates, GDP, residential construction, and mortgage interest deductions to explain house price developments. Loose monetary policy is found to be of influence on the occurrence of house price bubbles.
Abstract ... 2
Introduction ... 5
Overview ... 5
Aim and Research Questions ... 6
Literature Review ... 8
Theoretical model ... 14
Hypotheses ... 16
Data and Methods ... 18
Data sources ... 18
Data description and measures ... 19
Methodology ... 21
Empirical Results ... 24
Bubble determination ... 24
Data and Model Description ... 25
Discussion of the results ... 30
Conclusions ... 32
Limitations ... 33
References ... 35
Appendices ... 38
Appendix A: Individual OLS regressions for interest rates ... 38
Appendix B: Individual OLS regressions for control variables ... 38
Appendix C: OLS regressions for model using short interest rates ... 38
Appendix E: OLS regressions for model using policy rates ... 39
Appendix F: Panel regressions using short interest rates ... 40
Appendix G: Panel regressions using long interest rates ... 41
Overview
By 2008 the world had plunged into a financial crisis that caused a recession, which is considered by many to be the worst since the Great Depression of the 1930s. The catalyst for this situation was the burst of the subprime mortgage bubble in August 2007. In the popular media the cause for this crisis has been pointed out to be the greed of evil bankers, which in part was fed by the lack of regulation. Economists have determined this crisis to be like a stool, standing on three legs: excessive debt accumulation, debt-stimulating fiscal policy, and loose monetary policy (Steil, 2010). At the heart of the problem lies the excessive debt accumulation (Reinhart and Rogoff, 2009), which occurred in many countries from the mid-1990s onward (Allen, Chui, and Maddaloni, 2007) in terms of mortgages. The other two legs of the stool, fiscal and monetary policy, contributed to this mortgage debt accumulation. Mortgage interest deductibility has led to an incentive for consumers in many countries to purchase larger homes and home equity loan interest deductibility schemes have given consumers an incentive to increase external financing on their homes, increasing their default risk (Steil, 2010). Loose monetary policy is the third and final leg on which this financial crisis stands. Low interest rates and the resulting low cost of capital have increased the incentive to finance housing by debt, resulting in an increase in total mortgage debt.
In the 1960s Keynesian theory, which encouraged governments to steer the economy using fiscal policy, considering monetary policy to be an ineffective instrument to that end, was professed to and used by most countries. According to that theory interest rates had to be low and steady to prevent them from interfering with productive investment (Goodhart, 2010) and should not be used to steer the economy. When Keynesian theory went out of fashion the focus was increasingly on monetary policy as an economic steering device and fiscal deficits were to remain low and stable, which in Europe led to the introduction of the Stability and Growth Pact, which regulated that countries participating in the Euro should never run a deficit greater than three per cent of GDP.
this phenomenon. Although the lowered interest rates did not prevent the 1993 recession from happening, it did lead to a period of strong growth during the latter stages of that decade, which became known as the “Roaring Nineties” (Stiglitz, 2002). During this period productivity levels in the United States rose sharply and both inflation and unemployment fell to extraordinary low levels. Other countries showed a similar development. Contrary to other periods of growth the Federal Reserve and other central banks kept interest rates low, despite unemployment falling below the “non-accelerating inflation rate of unemployment” (NAIRU). This is in part why this period could become so successful in terms of economic development (Stiglitz, 2002).
Since then, interest rates have been relatively low in many countries with only little fluctuation to control inflation.
Aim and Research Questions
Given the fact that a long period of low interest rates (i.e. low cost of capital) coincided with a growing reliance on debt finance of real estate, culminating in a real estate bubble bursting and leading to a financial crisis, this is a very interesting period to take a closer look at. Intuitively, it would be very easy to blame loose monetary policy for the house price bubble and the consequences it turned out to have. The question is whether this conclusion would be justified based on factual evidence rather than intuitive reasoning. Another interesting point to make is that not all countries actually show signs of there being a house price bubble. Countries like the United States and the United Kingdom have suffered from a severe double digit price correction from 2008, whereas countries like Canada or Australia have shown little or no decline in prices after 2008 and have house price levels in 2010 well above the 2008 “bubble” prices. Could this difference in house price development be explained by differences in monetary policy? Especially this last question is an often overlooked issue in the literature.
The previous results in a main research question, which actually consists of two separate questions and will be treated as such. The research question on which this research will be based is:
To what extent did the prolonged period of low interest rates since the early 1990s contribute to the housing bubble and can differences in monetary policy explain the differences between countries?
To help answer this question a total of six specific research questions have been formulated to further specify this research. These specific research questions are:
1) What is the effect of interest rates on house prices?
2) Is the relationship between interest rates and house prices linear, as is claimed by some and disputed by others in the literature, or exponential, as would be in line with bubble theory?
3) What is the effect of GDP on house prices?
4) What is the effect of the number of building permits issued for residential dwellings on house prices?
5) What is the effect of the possibility to deduct mortgage interest payments from taxes on house prices?
6) Is there evidence to suggest that countries that were more severely hit by the house price bubble were so because of differences in their monetary policy?
After answering these questions it should be possible to determine the extent to which loose monetary policy actually could have caused any house price bubbles to occur.
It is generally accepted that house prices are very important in a well-functioning economic system. This is because a house performs both the function of a commodity and of an investment good (Iacoviello, 2000). In general, its house is a household‟s biggest source of capital accumulation (Anderson, Clemens, and Hanson, 2006), typically accounting for a much greater fraction of capital than does corporate equity (Poterba, 1991). In the United Kingdom, for instance, housing represented more than 40 per cent of total household wealth in 2001 (Aoki, Proudman, and Vlieghe, 2004). External financing in the form of mortgage finance plays a large role in the sustainability of that capital and as much as 80 per cent of household borrowing is secured on housing in the United Kingdom (Aoki et al., 2004). Furthermore, declines in real house prices in the past have been associated with declines in GDP growth and have otherwise had a significant influence on economic activity and inflation (Ahearn et al., 2005) and have at times even been associated with the occurrence of financial distress (Ahearn et al., 2005).
The link between house prices and household consumption has been thoroughly described in the literature, both in theory and empirically. The theoretical image is not very straightforward. Since most consumers live in the houses they own and will value their houses accordingly, any increase in the value of the property will in theory be offset by an increase in the opportunity cost of using it to live in (Aoki et al., 2004). This would indicate that an increase in house prices does not automatically shift the budget constraints of households outwards. The increased value of their houses does, however, increase the amount of collateral available to homeowners and may encourage them to borrow more, for instance by making use of mortgage equity withdrawal, to establish the desired levels of both consumption and housing (Aoki et al., 2004). Furthermore, house prices and consumption may move together as the increase in house prices may cause consumers to be more optimistic about economic prospects, which in turn may cause them to increase their consumption in both housing and non-housing goods (Aoki et al., 2004).
whether it would be different for housing wealth than for financial or stock wealth. Several surveys have suggested that consumers are not particularly responsive to changes in stock wealth when dealing with consuming or saving decisions, but any comparative evidence was lacking. As it turns out, in an international scope, a ten per cent increase in house prices can lead to an increase in household consumption of between 0.4 and 1.1 per cent, whereas a ten per cent increase in stock wealth hardly has any effect at all (Case et al., 2005). Especially renters tend to save less after an increase in house prices (Case et al., 2005), which is counterintuitive since the need for a larger down payment would require them to save more, if they want to purchase a house in the future. The positive effect of a rise in house prices tends to be a more long term effect than an immediate one, with the immediate (next quarter) marginal propensity to consume rising by 2 cents on the dollar, amounting to a total long run effect over several years of 9 cents (Carroll, Otsuka, and Slacalek, 2006).
Therefore, central banks are best advised to pay attention to house price stability when deciding on monetary policy, but there is only limited evidence that they do so (Ahearn et al., 2005). The European Central Bank (ECB) for instance states in its mission statement that its goal is to maintain the euro‟s purchasing power and thus price stability. It has also publicly stated that although it does not specifically target asset prices, it does take close notice of them because of the large potential cost to the economy and its price stability associated with strong appreciations and rapid reversals in asset prices (ECB, 2005).
bubbles cause the fundamentals to catch up with prices like in the case of a self-fulfilling prophecy or additional information concerning the fundamentals can eventually make sure the inflated prices are justified again (Stiglitz, 1990). Also, a sharp decline in prices does not need to be the result of a broken bubble (Stiglitz, 1990). These factors contribute to the complexity of identifying bubbles when they occur, even if one does have a clear definition of what a bubble is. Given these facts it is not surprising that a commonly accepted method to identify bubbles, booms and busts does not exist in the literature (Agnello and Schuknecht, 2009).
In principle, the fundamental house price is determined by two major factors besides the demand for housing: the price of the land it is built upon and the costs of construction. In all but the most extraordinary of cases the supply of land will be inelastic (Ahearn et al., 2005), especially in the short run, turning the price of land into a demand-based factor in house prices. Other than that, most fluctuations in house prices should move with the real construction cost of houses (Himmelberg et al., 2005) and any long-run growth of house prices in excess of construction costs would indicate that the land is appreciating more rapidly than the structures that are built upon it. Since the supply of land is inelastic, this must be solely due to demand, which could be an indicator of overvaluation. This would indicate that house prices could be used to identify bubbles by looking at house price growth, house price-to-rent ratios, or the house price-to-income ratio (Himmelberg et al., 2005). Various ways of identifying booms have been used by various researchers, but Detken and Smets (2004) speak of an asset price boom when real asset prices are more than 10 per cent above an estimated trend. Others have relied on the identification of peaks and troughs in house prices, which should not be exceeded for a pre-determined period of time (Ahearn et al., 2005).
living in that or a similar property, which they call the imputed rent, to the lost income from not being able to invest their capital in anything else, otherwise known as the opportunity cost of capital. This would imply that it is necessary also taking factors like differences in risk, tax benefits and liabilities associated with owning a house, expenses on maintenance, and anticipated capital gains or losses from owning the property into account.
The reliance of the user cost of housing on opportunity costs has an important implication: interest rates are a key element in determining fundamental house prices. Lower interest rates decrease the user cost of capital, because the cost of debt would be lower, which in turn would decrease the opportunity cost of house ownership, increasing its attractiveness. Himmelberg, Mayer, and Sinai (2005) go on to underline the importance of the expected growth rate of house prices in their model. A similar user cost model has previously been developed by Poterba (1991). The addition of the concept of opportunity cost in the user cost of housing model makes the inclusion of the expected growth rate of house prices inevitable and even though the concept of the expected growth rate of house prices is well understood, its practical application can appear to be somewhat nebulous, especially in terms of determining whether a bubble exists. Since expectations can play a large role in determining proper house prices and these expectations can be very different for different regions even within a country (Del Negro and Otrok, 2005), they have the potential to hugely complicate the question whether today‟s house prices can be justified or not. Furthermore, a perfectly reasonable expectation of house prices can become unreasonable with additional information about the future, spurring a correction in prices, which can give the appearance of a bubble, even when one did not exist (Stiglitz, 1990).
The previously described model to determine the correct house prices is one of many (Ahearn et al., 2005), but even if everyone agreed on one model, bubbles would still be hard to identify in real time, since it is very difficult to distinguish prices derived from economic fundamentals from those inflated by exuberance (Ahearn et al., 2005).
their view, if the phenomenon could be shown to be national rather than local, monetary policy would be a likely cause. They went on to conclude that up to and including 2000 these local trends could be confirmed according to their data, but that they started to see a clear national trend from 2001 through 2004, which was the last year included in their study. Even though they found the impact of policy shocks to be statistically significant, they also found it to be small.
A rather similar research had been conducted by Iacoviello (2000) on France, Germany, Italy, Spain, Sweden and the United Kingdom, except that he was not interested in regional versus national effects, but whether he could show the role of monetary policy shocks on house prices in an international setting in the long run. Like Del Negro and Otrok (2005) he used several monetary policy instruments and GDP in order to determine the effect of policy shocks on house prices. Different from Del Negro and Otrok he did find the role of monetary and demand shocks to play an important role in house price variations. Other important findings included that adverse monetary shocks have a negative effect on house prices and that different housing and financial market institutions influence the response of house prices on monetary policy shocks. Iacoviello has also shown that house prices can be effectively embedded in a relatively simple macroeconomic model to determine the effects of various shocks.
Jarocinski and Smets (2008) have conducted research on the role of the housing market and monetary policy in business cycles between 1987 and 2007 in the United States, with special focus on what they call “the housing boom and bust in the new millennium”. They found that house prices and other developments in the market cannot be sufficiently explained by developments in GDP, especially when looking at the developments since 2000. Furthermore, they find that even though demand shocks in the housing market can have a significant effect on house prices, the effects on the economy as a whole appear to be limited, notably in terms of growth and inflation. Finally, they conclude that even though the role of house prices has not been as significant as it had been considered by some, there are signs that loose monetary policy has been a contributing factor to the housing boom of 2004 and 2005 and that it therefore would be a wise idea to actually take house prices into account when deciding on monetary policy issues.
2007. The main point of contention was whether it was possible to find booms and busts while they were still developing and whether there was a relationship between the size and duration of house price booms and the size and severity of subsequent busts. Their identification method included both measures of magnitude and persistence to determine the severity of booms and busts. They found that they were able to detect booms and busts early on, that recent booms were among the most during and severe in the past 40 years, that there is a strong correlation between the duration and severity of booms and their subsequent busts, and that past economic growth and short term interest rates greatly influence the probability of booms or busts occurring.
Ahearne, Ammer, Doyle, Kole, and Martin (2005) have used a different methodology to determine the relationship between monetary policy and house prices. Different from other authors, they focus on the increasing role of speculation in house prices and they explicitly address the concept of bubbles in their research. They do that by examining house prices from 1970 through 2004 for 18 industrial countries. They find a link between easing monetary policy and house price booms in the sense that house prices generally take off at the same time interest rates reach their bottom and reach their peaks three years later on average. After the interest rates have remained at their low for about a year, they then increase until after house prices have reached their peaks, quickly reversing when GDP growth starts to fall. This research paper also acknowledges the difficulties in distinguishing between price changes for fundamental reasons and those associated with irrational expectations.
The same ambiguity can be found in discussions on the exact role of macroeconomic factors like GDP. Although all reviewed papers that addressed the issue agree on the existence of a link between GDP and house prices, the extent to which GDP and house prices influence each other is a hotly debated topic. Whereas Ahearn et al. (2005) conclude that house prices are pro-cyclical in their movements with GDP and inflation and find a clear pattern of interaction, Jarocinski and Smets (2008) conclude that the influence of house price movements on GDP and inflation is limited.
Another less widely accepted role in the movement of house prices that has been found in the literature is construction, emphasised both in terms of cost and amount of construction. Himmelberg et al. (2005) suggested that at least in theory house prices should move with construction costs, given that land supply is inelastic. Ahearn et al. (2005) take the size of residential construction into account in their analysis of house price developments. They do this by looking at residential construction as a percentage of GDP. They conclude that residential construction is highly sensitive to cyclical variations and tentatively conclude that construction is positively correlated to real house prices, although their evidence for this is not very extensive.
Finally, tax policy is generally accepted to be a factor in house price development. Many countries, like for instance the United States, use the tax system to encourage home-ownership among its households, to get them to accumulate wealth in real estate (Anderson et al., 2006), usually in the form of a system of mortgage interest deduction. This system allows taxpayers to deduct interest payments from taxable income and has a significant effect on the choice between house ownership and rental and on total housing consumption (Anderson et al., 2006). Many have written about the implications of such a system on the development of house prices and the accumulation of wealth. This system also puts extra emphasis on the effects of the income tax on the user cost of housing (Poterba, 1991). All these factors contribute to the importance of mortgage interest deduction on the development of house prices.
Theoretical model
relative importance of the various economic variables and shed light on how they are actually interconnected. In this model house prices (hpi) will be a function of interest rates (ir), national income (gdp), residential construction (rc), and mortgage interest deductions (td).
hpi = f (ir, gdp, rc, td)
Interest rates are included as the most important measure of monetary policy and the subject of the main research question. They serve as the cost of capital and are therefore instrumental for house price development. Using interest rates both as an instrument of monetary policy and as the cost of capital implies that this study will have to look at multiple interest rates, including both long term interest rates and policy rates. Since many researchers have used short interest rates, they will be included as well. Constructing a correlation matrix to verify possible mutual correlation between these interest rates will be inevitable. Many of the named studies have acknowledged the link between house prices and interest rates, but differences of opinion remained on the relative importance of monetary policy and house price booms.
Even though the relative importance of the reciprocal relationship between GDP and house prices is contested in the literature, the fact that they somehow co-move is not. A macroeconomic model that accurately depicts house prices should therefore always include GDP.
Residential construction is entered into the model as a measure of the supply of housing. The supply of housing is very important to take into account when trying to distinguish between a justified price development and a bubble, since a bubble is based on unrealistic expectations that are mostly demand induced.
Hypotheses
The hypotheses for this research have been drafted according to the expected answers to the specific research questions that were formulated in the introduction.
As mentioned before, the interest rate represents both the cost of capital and monetary policy in the theoretical model. This has certain implications for which interest rates should be used and will be discussed in greater detail in the data description. For now, it is assumed that interest rates act as the cost of capital. Higher costs of capital will decrease the demand for housing, which will, ceteris paribus, decrease house prices. Previous studies have confirmed this and have in many cases also found the relationship to be linear, although this has been contested by some. Hypotheses H1 and H2 are therefore:
H1: Since mortgage rates serve as the cost of capital determining house prices, the effect of interest rates on house prices is expected to be negative.
H2: Although the issue is fiercely contested in the literature, many recent papers find a linear relationship between interest rates and house prices, which is therefore also the expectation in this case.
Most papers agree and show that house prices are pro-cyclical, moving with GDP and inflation. The importance of this effect has been point of contention, but that the effect exists has not. This leads one to draft hypotheses H3 as follows:
H3: Since the literature has found house prices to be pro-cyclical, the effect of increasing GDP is expected to be positive.
H4: The number of building permits issued is entered in the model as a measure of supply of housing and is therefore expected to have a negative effect on house prices.
Mortgage interest deductions decrease the cost of capital and therefore the user cost of housing decreases as well. Lower costs will, ceteris paribus, increase demand for housing and therefore house prices will increase. Hypothesis H5 is consequently formulated as:
H5: Since mortgage interest deductions decrease the cost of capital incurred by the interest rates, they are expected to have a positive effect on house prices.
The previous five hypotheses should give an insight into the appropriate answer to the last question, which will serve as the last hypothesis. Given the results found in the literature it seems likely that looser monetary policy has an effect on whether a housing bubble will occur. Hypothesis H6 will therefore be as follows:
H6: Differences in monetary policies between countries are expected to have contributed to the occurrence or non-occurrence of house price bubbles in various countries.
This study will not include any member of the Euro area, since those countries will all have the same monetary policy from January 1999, controlled by the ECB rather than their own respective central banks. Given the time frame of this study, this could lead to an odd transition in monetary policy between 1998 and 1999, which could prevent one from drawing accurate conclusions. Furthermore, given the goal of the study, which is to look for the potential effects of differences in the stance of monetary policy, it would appear to be counterproductive to use countries that have no differences between them in terms of monetary policy.
This research will be carried out over the period from 1990 through 2010. The theoretical model will be applied to data from a total of seven nations: Australia, Canada, Norway, Sweden, Switzerland, the United Kingdom, and the United States. These countries were selected because they can be easily grouped in accordance with the research aim. The first four mentioned countries can be described as having had an upward trend in their respective house price indices throughout the period 1990-2010 with possibly only a small dip in house prices after 2007, whereas the latter three show no such trend or show an upward trend followed by a sharp decline.
Data sources
increase comparability. Eurostat is the statistical office for the European Union, situated in Luxembourg.
The house prices for Canada were retrieved from Teranet‟s Composite-6 index, an index containing house prices for the six largest cities in Canada. Teranet manages Ontario‟s Electronic Land Registration system after building it and in that capacity offers information on real estate to its customers.
Since all seven countries have been members of the Organisation for Economic Co-operation and Development (OECD) for many years, this organisation has been instrumental in obtaining the macroeconomic figures needed for this research. Data on GDP and interest rates was derived from their research, as well as information on residential construction for Australia, Canada, Norway, Sweden, Switzerland, and the United Kingdom. This organisation, which was brought into life to promote world trade and stimulate economic progress, was founded in 1961 by 20 countries and has 34 member states today. Its work is based on continued monitoring of events both inside and outside its member states. The organisation collects and analyses data and uses that information to give advice on policy matters.
Data on residential construction for the United States was obtained from the United States‟ Census Bureau. This organisation offers statistics on mining, manufacturing and construction as well as demographic and economic census statistics. It has a mission statement to serve as the United States‟ leading source of quality data about the nation‟s people and economy. The Census Bureau has been a permanent agency since 1902, but its research on census statistics dates back far further than that.
The information on the existence or non-existence of mortgage interest deductions has been obtained from various national sources. These sources include, but are not limited to the Internal Revenue Service (IRS) for the United States, the Canada Revenue Agency (CRA) for Canada, and the Australian Taxation Office (ATO) for Australia.
Data description and measures
question posed the introduction to this paper. They are the dependent and the main independent variable, respectively:
hpi – Real House Prices; this is the dependent variable in the model. It has been indexed
(Q4-2004=100) for all nations and deflated against it respective national consumer price deflators.
ir – Interest Rate; This variable serves as an indicator of monetary policy and is used to
determine the effect of monetary policy on house prices. One can really choose between three different types of interest rates when dealing with this issue. Given the nature of the research question it would be most reasonable to do the model with policy rates. Alternatively, the long interest rate is probably a better estimator for house price developments since mortgage rates are long interest rates, but the correlation of long interest rates to policy rates is lower than the correlation of short interest rates to policy rates (see appendix J). The third option, which is used in many studies, is the short interest rate. It is more closely linked to policy rates and would likely link better to house prices than the policy rate (appendix J). This research will look at all three of these variables to see if there are any large differences in outcomes.
The theoretical model further consists of three macroeconomic control variables, which are GDP, residential construction and mortgage interest deduction. These variables serve to turn this model into a basic macroeconomic model, which is supposed to have enough explanatory power of the many factors determining house price development:
gdp – Gross Domestic Product (GDP); Research has shown house prices to be pro-cyclical,
co-moving with e.g. GDP. Therefore, GDP would be a very important control variable in this research. This variable has also been indexed (Q4-2004=100) for all nations and the figures have been seasonally adjusted.
rc – Residential Construction; residential construction is included as a measure of supply of
measure to be negatively correlated with prices, which is why this issue deserves some extra attention. This variable has been indexed (Q4-2004=100) for all nations.
td – Mortgage Interest Deduction; included in the model in the form of a dummy variable,
since this is a tax policy instrument. The reason for this variable to be a dummy is that the effects of tax deduction instruments are highly dependent upon the taxation policies as well as revenue and other institutions, which will more than likely differ across various countries, and this research is on monetary policy rather than tax policy. This variable is included, however, since such a policy was abolished in the United Kingdom during the time of this research, possibly making it a factor of interest in house prices in at least one of the investigated countries. However, being a dummy variable, any adjustments made to a national tax system other than the introduction or abolition of a mortgage interest deduction tax instrument will not give visible results. This specifically happened in Sweden in 1991 when the maximum deductible amount was changed from 90 per cent of interest paid to 30 per cent. This policy change, even though it may have had a major impact on house prices in Sweden is of no significance in this model because of the way the model is built.
Methodology
One of the goals of this research is to construct a basic macroeconomic model as was suggested to be possible in a study by Iacoviello (2000). In its simplicity it should be an accurate description of how house prices move and react to various macroeconomic variables. This is needed to on the one hand prove the relationship between interest rates and house prices, while on the other hand determining what that effect looks like.
The methodology of this study consists of two consecutive steps. The first step is to determine whether it is reasonable to assume that each country is suffering from a house price bubble. The second step is to determine as accurate a description of the house price model as possible, given the variables presented earlier in this paper, which are all assumed to play their part in the movement of house prices.
Step 1: to bubble, or not to bubble
peaks in which the ascent lasts for at least three years. Detken and Smets (2004) define an asset price boom to be a period in which real asset prices exceed an estimated trend by at least 10 per cent. Agnello and Schuknecht (2009) look at both the duration and magnitude of above- and below-trend house prices to determine booms or busts.
This study will use a method to identify bubbles, which will borrow elements from all three schools of thought described above. First of all, both duration and severity of a peak should be considered to pinpoint a bubble, in accordance with the methodology used by Agnello and Schuknecht (2009). Secondly, for a peak to qualify as a bubble the house price level may not have been exceeded in the same series for at least three years. This way, one can ensure that only clear and exceptional peaks will be able to qualify as a bubble, without running the risk of pointing at every deviation from trend as being a house price bubble. This aspect will be referred to as the duration rule and has been derived and adjusted from the methodology used by Ahearn et al. (2005). Thirdly, the peaks that have been identified using the above method should deviate at least 10 per cent from the 1990-2010 trends, as determined by fitting trend lines. This aspect will be referred to as the severity rule and has been derived and adapted from the methodology used by Detken and Smets (2004). And finally, since bubbles were identified by Stiglitz (1990) as high prices, without the possibility of explaining those prices by using fundamentals, any potential bubble that may be the cause of a significant change in the fundamentals will be discarded as a bubble in this study. This rule will be referred to as the fundamentals rule.
In order for a peak to qualify as a bubble it should abide by the duration rule, the severity rule, and the fundamentals rule.
Step 2: the regression model
Although the model has been previously described as a simple function of interest rates, GDP, residential construction, and mortgage interest deduction, this may well turn out to be a little more complicated than first assumed. As reported before, linearity between interest rates and house prices has been contested in the literature, so it may turn out that the actual end result looks quite a bit different than a simple linear function of the proposed model:
hpi = f (ir, gdp, rc, td)
They have done so to determine the effect of policy shocks on house prices. Since this study does not concern itself with the effects of sudden changes, but rather with a period of predominantly stable, low interest rates, this methodology would be less useful in this case. Another reason not to choose for a vector autoregressive model in this study is the fact that according to the literature one cannot a priori go by the assumption of a linear relationship as was pointed out by MacLennan, Muellbauer and Stephens (1998).
This research will be treating the data as a panel data set and will carry out the regressions accordingly. This will allow for distinguishing between countries or groups of countries which show distinctly differing trends and also allows for dealing with the possible institutional differences between countries as they were pointed out by MacLennan, Muellbauer and Stephens (1998). In its fixed effects form a panel data analysis can control for factors that:
- Vary across entities (countries or groups of countries), but do not vary over time. - Could cause omitted variable bias if omitted.
In accordance with the methodology description in the previous section the first thing that needs to be done is to determine whether house price bubbles have occurred in any of the countries in this study. Once that is done the model will be finalised using the appropriate tools.
Bubble determination
The first issue to deal with is the matter of whether bubbles do or do not occur in the given time period in either of the countries. To qualify as a bubble three requirements must be met. The first rule is the duration rule, stating that a peak must not be exceeded by another peak within three years. The second rule is the severity rule, stating that a peak must deviate from the long term trend by at least ten per cent in order to qualify as a bubble. The third rule is the fundamentals rule, which states that a potential bubble cannot be an actual bubble if its appearance can be explained by a substantial change in the fundamentals of house prices. Appendix K presents a graphic depiction of the results of this method this method, in which the actual house prices are plotted against an exponential trend line.
Applying the duration rule to the data first, results in a total of three peaks that may qualify as a bubble. One peak is to be found in Sweden in the first quarter of 1991, one in the United Kingdom in the third quarter of 2007, and one in the United States in the first quarter of 2007.
These three deviations from the trend were then subjected to the severity rule. As it turns out all of these deviations diverge by at least 10 per cent from their respective trends. Each of the three potential bubbles therefore passes the severity test.
The situation in Sweden in 1991 is slightly different though. Since that potential bubble actually coincided with a change in the maximum percentage of mortgage interest that one is allowed to deduct from 90 to 30 per cent and as was conceded before this change cannot be accounted for by this macroeconomic model, it seems reasonable to assume that this peak was likely caused by a change in fundamentals rather than a bubble. Therefore, Sweden is not considered to be a bubble country.
One can also see a rather large anomaly in the house prices in Switzerland in the fourth quarter of 1998, but this has not been identified as a bubble for two reasons. First of all, it would be rather misleading to identify an extraordinary low point in house prices as a bubble given the definition of a bubble that is used for this study, which explicitly states that a bubble occurs when prices are higher than can be explained by fundamentals. This particular anomaly does not abide by that condition. Secondly, when studied in more detail, it turns out that this anomaly is actually the result of the burst of a potential bubble, which had its highest point in the fourth quarter of 1989. This potential bubble would qualify as a bubble if the three rules of this research are applied, but since this bubble occurred outside the scope of this research, it will not be taken into account as such. Another point to be made is that potential bubbles occurring in 1989 (in the case of Switzerland) or 1991 (in the case of Sweden) are not very likely to be caused by the low interest rates occurring since 1992.
In conclusion, only the United Kingdom and the United States are considered to be countries in which a bubble occurred and Australia, Canada, Norway, Sweden, and Switzerland are qualified as non-bubble countries. This distinction will later be used in the panel regression to see whether a distinct set of institutions can be found to exist between these two groups.
Data and Model Description
rates or short interest rates. The same methodology was used to determine the individual relationships of the control variables with house prices and the results are presented in appendix B. The relationship of GDP with house prices is positive and has a very high explanatory value (R-squared = 0.8101). The relationships of residential construction and mortgage interest deductions with house prices are positive and negative respectively, which is opposite to expectations in both cases, although with low explanatory power (for a quick overview of these expectations see appendix I). All the control variables are significant to the 99-percent level, except mortgage interest deductions, which is significant to the 95-percent level
The simple scatter plot of short interest rates and house prices in figure 1 made clear that linearity would probably not the best way to describe the relationship between these variables. The same observation can be made with even more certainty when using long interest rates, as is done in figure 2.
Figure 1: Relationship of house prices to short interest rates
confirmed by the first ordinary least squares (OLS) regressions on the model for short interest rates (appendix C).
Figure 2: Relationship of house prices to long interest rates
Even though a negative linear relationship is supported at the 95-percent level in the model, a model in which both a quadratic and a linear term are included is supported at the 99-percent level for both. Since the adjusted R-squared for the model also slightly improves, the model appears to be working better with squared short interest rates, which would indicate that the basic macroeconomic model that was proposed earlier would take the following form:
hpi = β₀ + β₁ ir² + β₂ ir + β₃ gdp + β₄ rc + β₅ td
This image is confirmed using the model with long interest rates. The results of this regression, both with and without squared long interest rates are presented in appendix D. The non-linearity of the interest rate is much clearer in this instance than it was with the short interest rate. The other variables behave in more or less the same way as in the model using the short interest rate.
To complete the image appendices C, D, and E also include the results of the OLS regressions that try to establish a possible logarithmic relationship between house prices and interest rates. These regressions give a more accurate view of the model than do the linear regressions, but the results do not in general show a better fit than those of the quadratic model. Furthermore, since this methods requires a relatively large amount of manipulation of the data compared to the quadratic model, the decision to reject the logarithmic model in favour of the quadratic model appears to be well justified.
Two general remarks need to be made before moving on. First of all, even though the long interest rate is individually a better estimator for house prices, the respective OLS models do not show an improved or deteriorated R-squared, when using short or long interest rates or policy rates. Therefore, the explanatory power of the model does not change substantially when changing the type of interest rate used. And secondly, the mortgage interest deductions variable appears to have lost its statistical significance when used in the model.
The next step is to divide the countries into two groups according to the analysis made in step 1 and do another panel regression using the fixed effects method. A different division will not affect the results of a random effects panel regression and therefore this regression will not have to be repeated. Group 1 will consist of the countries that were labelled as “bubble countries”, which were the United Kingdom and the United States, and group 2 will consist of Australia, Canada, Norway, Sweden and Switzerland. The results of this regression can be found in Appendix F.
The results of this regression are mostly in line with the results from the previous panel. The Hausman test finds a p-value of 0.2334 in this case, which would indicate that in this particular division the groups are not sufficiently different from one another in terms of institutional differences that the use of the fixed effects method is justified with 90 per cent certainty. This result therefore does not support the notion that there are substantial institutional differences between bubble countries and non-bubble countries.
The same methodology has been applied using long interest rates and policy rates, the results of which are presented in appendices G and H, respectively. The image using long interest rates is very similar to the one using short interest rates with only minor differences in the coefficients. The signs are all the same, as are the significance levels. The results of the Hausman tests do not allow for a distinction either on country level, or on bubble versus non-bubble countries. Again, the existence of institutional differences cannot be proven to exist in this case.
One additional interesting remark that can be made when looking at the data is that the interest rates reach a minimum three to four years prior to the peak of the bubble in both the United Kingdom and the United States, which is in line with the prediction made based on the study by Ahearn et al. (2005) and that in both cases this low point is in fact the lowest over the scope of this study prior to 2009.
Discussion of the results
To sum up, some interesting results were found in this study. The first thing to be pointed out is that the regression results turned out to be rather strong in terms of statistical significance with all variables except mortgage interest deductions mostly at the 99-percent significance level. These results support the drawing of conclusions in terms of the proposed hypotheses, which is encouraging.
Four countries were determined to have had anomalous house price developments, but only two of those were actual house price peaks. The anomaly in Switzerland turns out to be the trough that followed a house price peak in the fourth quarter of 1989, but since that peak is just outside the scope of this research, it will not be researched any further. It does, however, explain the peculiar house price development in Switzerland between 1990 and 1998, which is unlike any other country included in this study.
In terms of the different interest rates that can be used to determine their influence on house prices, it turns out that the differences are not very substantial. Individually the long interest rate may be a better estimator for house prices than short interest rates or policy rates, but in terms of explanatory power of the entire model, this difference does not hold up. Especially when comparing the results of the short interest rate to those of the long interest rate, the differences are inconsequential. The policy rates behaved very similarly in the OLS environment as the other interest rates, but there were some differences in the panel environment, especially when looking at country-level differences.
The mortgage interest deductions have not been found to have a generalised effect on house prices since significance levels were generally very low.
One can now feel adequately equipped to answer the specific research questions. Interest rates are negatively correlated with house prices as was expected and the doubts with respect to the linearity of this relationship in the literature were justified. The data support a quadratic relationship, which was contrary to the expectations with which this study was ventured.
House prices were shown to be highly pro-cyclical, which is in line with both the literature on the subject and the expectations at the start of this study. GDP can therefore be said to have a positive effect on house prices.
The effect of residential construction, which in this research was measured by the number of building permits issued for residential dwellings, on house prices turns out to be positive. On the one hand this is a somewhat odd result for a measure of supply and is for that reason contrary to expectations, but it is in line with the findings by Ahearn et al. (2005), who got the same result. It would appear that the issuing of building permits is a demand driven feature, as was the case with residential construction as a percentage of GDP in the study by Ahearn et al. (2005).
The question of the effect of mortgage interest deductions on house prices cannot be sufficiently answered. In most cases the effect was not statistically significant, which prevents one from drawing a reasonable conclusion. The very nature of this variable caused it to have very little variance, which leaves it at risk of being statistically insignificant or switching its sign from time to time.
The last specific research question is a very important one. Given the dataset, the evidence for the effect of interest rates on house prices is very clear. There can be no doubt that low interest rates have contributed to the sharp rise in house prices since the second half of the 1990s. The results do not, however, support a possible difference in institutions between those countries that were affected by a bubble and those that were not.
occasional diminished statistical significance is no reason to doubt the role of interest rates in the model, nor of the model itself.
In conclusion, one can state with a high degree of certainty that low interest rates will have driven up house prices, albeit that the relationship is not a linear one. Rising house prices can easily cause increased speculation and therefore create a bubble-like effect. This research did not, however, succeed in pointing towards a distinct institutional difference between those countries that showed signs of a housing bubble and those that did not. Overall, the evidence points towards a substantial role for monetary policy in house price development and it would therefore appear to be very unlikely that monetary policy did not contribute to the development of house price bubbles.
Bearing the results of this study in mind, loose monetary policy appears to bear at least some of the responsibility of the events that took place in 2007 and therefore central banks would be well advised to give asset price stability a more central role in their monetary policy decisions, especially when considering the effects of house price deviations on household consumption. This does not necessarily mean, however, that monetary policy can be pointed out to be the only or most important reason for the credit crunch based on this research.
Limitations
The macroeconomic model that was constructed for this study performs very well in terms of explanatory value. The model is able to account for 85 per cent of variation with only a handful of variables. Some may argue that the limited number of variables included in the model is its weakness, since many more factors to influence house price development can be thought up. Anyone making the claim that other factors, like for instance temporal changes in the size of the average housing unit, also play a part in changing house prices over time would be right. This does not, however, have to be a weakness of the model. The basic model presented in this study can easily be expanded upon or altered to suit specific research objectives, knowing that in its core this model will perform well in explaining variations in house price development. In this respect the simplicity of the model can be considered a strength rather than a weakness.
study generally has not been successful in finding statistically significant institutional differences between the countries. It could be argued that one would have been more likely to have found those differences had a selection of less similar countries been used. With growing levels of globalisation, however, it seems probable that country-level institutional differences are likely to have diminished in the past and will likely continue to diminish in the coming years.
Most of the variables work well together to account for house price variations and have a clear and distinct effect on price development, albeit with differing effects in terms of importance. One of the variables in this model cannot be said to have a clear and statistically significant effect. The role of mortgage interest deductions in house price development was for the most part not found to be clear, nor statistically significant. The goal of the way this variable was constructed in the research model was to find any generalised effect this tax instrument may have on house prices in an international setting. Given the results of the various regressions carried out in this study, if such a general effect exists for this tax instrument, this research model has not been able to unequivocally identify it.
Future research using this model or a variation of this model should at least look into mortgage interest deductions and how it can be improved upon. The role of tax policy on house price development and in this capacity also on the development of house price bubbles would be an interesting field to expand upon. A variation of this model could be developed to look into that question with only minor differences, but future researchers should focus on how best to measure this effect. One would have to develop a variable, which takes the average financial gains from these deductions into account to get a more accurate image of the influence of tax policy on house price developments. It would not be sufficient to simply look at the maximum percentage of interest that could be deducted, since differences in income tax also play a role.
Agnello, L., and L. Schuknecht (2009). Booms and Busts in Housing Markets – Determinants and implications. European Central Bank – Working paper series no. 1071.
Ahearn, A.G., J. Ammer, B.M. Doyle, L.S. Kole, and R.F. Martin (2005). Monetary policy and house prices: A cross-country study. Federal Reserve Bank - International Finance
Discussion Paper nr. 841.
Allen, F, M.K.F. Chui, and A. Maddaloni (2008). Financial Structure and Corporate Governance in Europe, the USA and Asia, Handbook of European Financial Markets and
Institutions, Oxford University Press, pp.31-67.
Anderson, J.E., A. Hanson, and J.P. Clemens (2006). Tax Reform and Incentives to Encourage Owner-Occupied Housing: Analysis of the President's Tax Reform Panel Recommendation to Convert the Mortgage Interest Deduction to a Tax Credit, University of
Nebraska Working paper, September 2006.
Aoki, K., J. Proudman, and G. Vlieghe (2004). House prices, consumption, and monetary policy: a financial accelerator approach. Journal of Financial Intermediation, no.13 (2004), pp. 414-435.
Carroll, C.D., M. Otsuka, and J. Slacalek (2006). How large is the housing wealth effect? A new approach. National Bureau of Economic Research, Working Paper No. 12746.
Case, K.E., J.M. Quigley, and R.J. Shiller (2005). Comparing wealth effects: The stock market versus the housing market. Advances in Macroeconomics, Vol. 5, Issue 1, Article 1. Del Negro, M., and C. Otrok (2005). Monetary policy and the house price boom across U.S. States. Federal Reserve Bank of Atlanta; Working paper 2005-24.
Detken, C., and F. Smets (2004). Asset price booms and monetary policy. European Central
Bank – Working paper series no. 364.
European Central Bank Article (2005). Asset price bubbles and monetary policy. European
Central Bank Monthly Bulletin, April 2005, pp. 47-60.
Goodhart, C.A.E. (2010). Lessons from the financial crisis for monetary policy. Daedalus, fall
Himmelberg, C., C. Mayer, and T. Sinai (2005). Assessing high house prices: Bubbles, fundamentals, and misperceptions. Journal of Economic Perspectives, Vol. 19, No. 4, pp. 67-92.
Iacoviello, M. (2000). House prices and the macro economy in Europe: Results from a structural VAR analysis. European Central Bank – Working paper series no. 18.
Jarocinski, M., and F. Smets (2008). House prices and the stance of monetary policy.
European Central Bank – Working paper series no. 891.
Kohn, D. (2007). Success and failure of monetary policy since the 1950s. Speech at
“Monetary policy over fifty years”, a conference to mark the fiftieth anniversary of the Deutsche Bundesbank, Frankfurt, Germany.
Leamer, E.E. (2007). Housing and the business cycle. Paper presented at the Annual
Economic Symposium organised by the Kansas City Federal Reserve in Jackson Hole.
MacLennan, D., J. Muellbauer, and M. Stephens (1998). Asymmetries in housing and financial market institutions and EMU. Oxford review of economic policy, vol. 14, no. 3, pp. 54-80.
Poterba, J. (1991). House price dynamics: The role of tax policy and demography. Brookings
Papers on Economic Activity. No. 2, pp. 143-183.
Reinhart, C.M., and K.S. Rogoff (2009). This time is different: Eight centuries of financial folly. Princeton University press,
Shell, K, M. Sidrauski, and J.E. Stiglitz (1969). Capital gains, income and savings. Review of
Economic Studies, No. 36, pp. 15-26.
Shell, K, and J.E. Stiglitz (1967). Allocation of investment in a dynamic economy. Quarterly
Journal of Economics, No. 81, pp. 592-609.
Steil, B. (2010). Debt and systemic risk: The contribution of fiscal and monetary policy. Cato
Journal, Vol. 30, No. 2, pp. 391-396.
Stiglitz, J.E. (2002). The roaring nineties. The Atlantic monthly, October 2002, Vol. 290, No.
Appendix A: Individual OLS regressions for interest rates
hpi Coef. Std. Err. t P>t 95% Conf. Interval R-Squared n sir -564.662 *** 38.256 -14.76 0.000 -639.811 -489.514 0.2875 542 constant 110.913 *** 2.157 51.41 0.000 106.674 115.151 lir -982.293 *** 42.560 -23.08 0.000 -1065.895 -898.690 0.4966 542 constant 139.056 *** 2.576 53.98 0.000 133.996 144.116 pir -501.352 *** 37.893 -13.23 0.000 -575.789 -426.915 0.2455 540 constant 106.3969 *** 2.057 51.72 0.000 102.356 110.438
*** = 99-percent significance level; ** = 95-percent significance level; * = 90-percent significance level
Appendix B: Individual OLS regressions for control variables
hpi Coef. Std. Err. t P>t 95% Conf. Interval R-squared n gdp 2.017 *** 0.042 47.46 0.000 1.933 2.100 0.8101 530 constant -99.217 *** 3.914 -25.35 0.000 -106.906 -91.527 rc 0.469 *** 0.060 7.89 0.000 0.352 0.586 0.1112 499 constant 45.627 *** 5.069 9.00 0.000 35.666 55.587 td -6.506 ** 2.675 -2.43 0.015 -11.761 -1.251 0.0108 542 constant 87.306 *** 1.964 44.46 0.000 83.449 91.164
*** = 99-percent significance level; ** = 95-percent significance level; * = 90-percent significance level
Appendix C: OLS regressions for model using short interest rates
hpi Coef. Std. Err. t P>t 95% Conf. Interval Adj. R-sq. n sir -68.606 ** 26.817 -2.56 0.011 -121.295 -15.917 0.8457 499 gdp 1.935 *** 0.052 37.26 0.000 1.833 2.037 rc 0.095 *** 0.027 3.55 0.000 0.042 0.147 td -0.545 1.202 -0.45 0.650 -2.906 1.816 constant -97.301 *** 5.796 -16.79 0.000 -108.688 -85.914 sir² 2108.938 *** 407.133 5.18 0.000 1309.009 2908.867 0.8533 499 sir -332.687 *** 57.293 -5.81 0.000 -445.256 -220.119 gdp 1.909 *** 0.051 37.53 0.000 1.809 2.008 rc 0.100 *** 0.026 3.86 0.000 0.049 0.151 td -1.860 1.199 -1.55 0.121 -4.215 0.494 constant -88.640 *** 5.892 -15.04 0.000 -100.217 -77.064lnhpi Coef. Std. Err. t P>t 95% Conf. Interval Adj. R-Sq. n sir -1.830 *** 0.338 -5.41 0.000 -2.495 -1.165 0.8570 499 lngdp 2.092 *** 0.056 37.19 0.000 1.981 2.202 lnrc 0.073 *** 0.024 3.08 0.002 0.026 0.119 td -0.010 0.015 -0.67 0.503 -0.039 0.019 constant -5.297 *** 0.265 -19.95 0.000 -5.818 -4.775
Appendix D: OLS regressions for model using long interest rates
hpi Coef. Std. Err. t P>t 95% Conf. Interval Adj. R-Sq. n lir -8.363 49.483 -0.17 0.866 -105.587 88.860 0.8436 499 gdp 2.005 *** 0.073 27.59 0.000 1.862 2.147 rc 0.094 *** 0.027 3.49 0.001 0.041 0.146 td 0.403 1.230 0.33 0.744 -2.014 2.820 constant -106.725 *** 9.132 -11.69 0.000 -124.666 -88.783 lir² 4570.572 *** 878.653 5.20 0.000 2844.205 6296.938 0.8515 499 lir -647.011 *** 131.907 -4.91 0.000 -906.181 -387.842 gdp 1.943 *** 0.072 27.08 0.000 1.802 2.084 rc 0.084 *** 0.026 3.22 0.001 0.033 0.136 td -1.163 1.236 -0.94 0.347 -3.591 1.266 constant -80.264 *** 10.251 -7.83 0.000 -100.405 -60.124
lnhpi Coef. Std. Err. t P>t 95% Conf. Interval Adj. R-Sq. n lir -2.556 *** 0.647 -3.95 0.000 -3.828 -1.284 0.8532 499 lngdp 2.012 *** 0.082 24.64 0.000 1.851 2.172 lnrc 0.073 *** 0.024 3.07 0.002 0.026 0.120 td -0.007 0.015 -0.44 0.660 -0.037 0.024 constant -4.884 *** 0.397 -12.29 0.000 -5.664 -4.103
*** = 99-percent significance level; ** = 95-percent significance level; * = 90-percent significance level
Appendix E: OLS regressions for model using policy rates
hpi Coef. Std. Err. t P>t 95% Conf. Interval Adj. R-Sq. n pir -59.647 *** 22.921 -2.60 0.010 -104.683 -14.612 0.8457 497 gdp 1.955 *** 0.050 38.93 0.000 1.856 2.053 rc 0.094 *** 0.027 3.54 0.000 0.042 0.146 td -0.409 1.202 -0.34 0.734 -2.770 1.952 constant -99.767 *** 5.403 -18.47 0.000 -110.382 -89.151 pir² 559.928 *** 183.044 3.06 0.002 200.283 919.573 0.8483 497 pir -163.595 *** 40.882 -4.00 0.000 -243.919 -83.270 gdp 1.908 *** 0.052 36.61 0.000 1.805 2.010 rc 0.108 *** 0.027 4.05 0.000 0.056 0.161 td -1.165 1.217 -0.96 0.339 -3.556 1.226 constant -93.356*** 5.753 -16.23 0.000 -104.659 -82.052
lnhpi Coef. Std. Err. t P>t 95% Conf. Interval Adj. R-Sq. n pir -1.593 *** 0.287 -5.54 0.000 -2.157 -1.028 0.8572 497 lngdp 2.126 *** 0.054 39.29 0.000 2.020 2.232 lnrc 0.074 *** 0.024 3.13 0.002 0.027 0.120 td -0.008 0.015 -0.57 0.569 -0.038 0.021 constant -5.471 *** 0.252 -21.73 0.000 -5.966 -4.976
Appendix F: Panel regressions using short interest rates
Fixed Effects by country
hpi Coef. Std. Err. t P>t 95% Conf. Interval Hausman n sir² 2313.466 *** 395.319 5.85 0.000 1536.725 3090.207 0.1198 499 sir -284.411 *** 58.641 -4.85 0.000 -399.632 -169.189 gdp 1.889 *** 0.049 38.40 0.000 1.792 1.986 rc 0.137 *** 0.027 5.17 0.000 0.085 0.189 td -10.320 *** 2.719 -3.80 0.000 -15.662 -4.978 constant -88.376 *** 5.700 -15.51 0.000 -99.574 -77.177 Random Effects
hpi Coef. Std. Err. z P>z 95% Conf. Interval n sir² 2108.938 *** 407.133 5.18 0.000 1310.973 2906.904 499 sir -332.687 *** 57.293 -5.81 0.000 -444.979 -220.395 gdp 1.909 *** 0.051 37.53 0.000 1.809 2.008 rc 0.100 *** 0.026 3.86 0.000 0.049 0.151 td -1.860 1.199 -1.55 0.121 -4.210 0.489 constant -88.640 *** 5.892 -15.04 0.000 -100.188 -77.092
Fixed Effects by group
hpi Coef. Std. Err. t P>t [95% Conf. Interval] Hausman n sir² 1984.432 *** 410.451 4.83 0.000 1177.979 2790.885 0.2334 499 sir -306.370 *** 58.563 -5.23 0.000 -421.435 -191.305 gdp 1.910 *** 0.051 37.66 0.000 1.810 2.009 rc 0.095 *** 0.026 3.68 0.000 0.044 0.147 td -0.849 1.294 -0.66 0.512 -3.392 1.694 constant -89.688 *** 5.896 -15.21 0.000 -101.273 -78.104
Appendix G: Panel regressions using long interest rates
Fixed Effects by country
hpi Coef. Std. Err. t P>t 95% Conf. Interval Hausman n lir² 5645.264 *** 924.992 6.10 0.000 3827.796 7462.731 0.3613 499 lir -711.478 *** 151.778 -4.69 0.000 -1009.699 -413.258 gdp 1.926 *** 0.070 27.34 0.000 1.787 2.064 rc 0.126 *** 0.026 4.80 0.000 0.074 0.177 td -10.113 *** 2.703 -3.74 0.000 -15.424 -4.802 constant -77.666 *** 10.778 -7.21 0.000 -98.843 -56.489 Random Effects
hpi Coef. Std. Err. z P>z 95% Conf. Interval n lir² 5735.244 *** 912.294 6.29 0.000 3947.180 7523.308 499 lir -757.050 *** 146.924 -5.15 0.000 -1045.016 -469.084 gdp 1.918 *** 0.070 27.37 0.000 1.781 2.056 rc 0.117 *** 0.026 4.49 0.000 0.066 0.168 td -6.733 *** 2.222 -3.03 0.002 -11.088 -2.377 constant -75.433 *** 10.750 -7.02 0.000 -96.503 -54.363
Fixed Effects by group
hpi Coef. Std. Err. t P>t 95% Conf. Interval Hausman n lir² 4140.427 *** 898.958 4.61 0.000 2374.157 5906.697 0.2016 499 lir -570.690 *** 136.323 -4.19 0.000 -838.537 -302.843 gdp 1.955 *** 0.072 27.25 0.000 1.814 2.096 rc 0.080 *** 0.026 3.06 0.002 0.029 0.132 td -0.049 1.340 -0.04 0.971 -2.682 2.584 constant -84.230 *** 10.386 -8.11 0.000 -104.637 -63.823
