Interest rates and house prices in the United States and around the world

BIS Working Papers

No 665

 

Interest rates and house prices in the United States and around the world

by Gregory D Sutton, Dubravko Mihaljek and Agne Subelyte

 

Monetary and Economic Department

October 2017

 

JEL classification: E39, E43, E58, G12, R31, R32.

Keywords: interest rates, house prices, monetary policy, bank lending channel, random walk, house price bubble, United States, advanced economies, emerging market economies

 

BIS Working Papers are written by members of the Monetary and Economic Department of the Bank for International Settlements, and from time to time by other economists, and are published by the Bank. The papers are on subjects of topical interest and are technical in character. The views expressed in them are those of their authors and not necessarily the views of the BIS.

This publication is available on the BIS website (www.bis.org).

 

© Bank for International Settlements 2017. All rights reserved. Brief excerpts may be reproduced or translated provided the source is stated.

ISSN 1020-0959 (print)

ISSN 1682-7678 (online)

 

Interest rates and house prices in the United States and around the world

Gregory D Sutton, Dubravko Mihaljek and Agne Subelyte 

 

Abstract

This paper estimates the response of house prices to changes in short- and long-term interest rates in 47 advanced and emerging market economies. We use data that statistical authorities selected as their best house price series, covering almost half a century of quarterly observations for the United States and over 1,000 annual observations for the rest of the sample. We find a surprisingly important role for short-term interest rates as a driver of house prices, especially outside the United States. Our interpretation is that this reflects the importance of the bank lending channel of monetary policy in house price fluctuations, especially in countries where securitisation of home mortgages is less prevalent. In addition, we document substantial inertia in house prices and find that changes in interest rates and other determinants affect house prices gradually rather than on impact. This suggests that modest cuts in policy rates are not likely to rapidly fuel house price increases. Finally, we find that US interest rates seem to affect house prices outside the United States.

 

Keywords: interest rates, house prices, monetary policy, bank lending channel, random walk, house price bubble, United States, advanced economies, emerging market economies

JEL classification: E39, E43, E58, G12, R31, R32.

October 2017


The views expressed in this paper are those of the authors and not necessarily those of the BIS. We are grateful to Michela Scatigna for help with the data and a description of the BIS data set on residential property prices. Dietrich Domanski, Mathias Drehmann, Ilhyock Shim, James Yetman, Fabrizio Zampolli and participants of seminars at the Central Bank of the Republic of Austria, BIS, Dubrovnik Economic Conference, and the Narodowy Bank Polski provided helpful comments.

 

1. Introduction

This paper estimates the response of house prices in 47 advanced and emerging market economies (EMEs) to changes in short- and long-term interest rates. Our study has four novel aspects. First, we analyse in some detail the impact of short-term interest rates on house prices. Second, we look at the responsiveness of house prices around the world to US interest rates. Third, we use a unique data set on house prices compiled by the BIS in cooperation with national statistical and monetary authorities. And fourth, our empirical framework tries to capture the important role of inertia in house prices.

Our focus on short-term interest rates is motivated by their link to monetary policy. As a house is a long-lived asset, the interest rate appropriate for relating the service flow from a house to its price is arguably a long-term rate. However, house prices also depend importantly on ease of access to credit, which is in turn significantly affected by the monetary policy stance. Bernanke and Blinder (1992), for instance, showed that changes in the US federal funds rate were associated with changes in lending by US banks, an effect that has become known as the bank lending channel of monetary policy. Short-term interest rates, which are more closely related to the stance of monetary policy, might therefore be just as important a "fundamental" for house prices as longer-term rates. Indeed, we find a surprisingly important role for short-term interest rates as drivers of house prices, especially outside the United States. Our interpretation is that this reflects an important role for the bank lending channel of monetary policy, especially in countries where securitisation of home mortgages is less prevalent.

The motivation for looking at the responsiveness of house prices around the world to not only domestic but also US interest rates is that the latter have become a key measure of the global cost of financing. We do find spillover effects from US interest rates, both short and long ones, on house prices outside the United States.

Our study draws on the BIS residential property price statistics and, in particular, the "preferred" house price series as identified by national statistical offices or central banks. We compiled over 1,000 annual observations on house prices for the non-US countries in our sample from these series and about a half century of quarterly house prices for the United States. We use these data to estimate the dynamic impact of changes in interest rates and other explanatory variables on real house prices around the world.

We find that real house prices display inertia when they adjust to new levels and that changes in short-term interest rates from up to five years in the past can have a significant impact on current changes in house prices. We interpret this as a consequence of the large transaction costs associated with trading residential real estate, which implies that the demand for housing adjusts slowly in response to changes in interest rates and other driving variables. Given this protracted response, modest cuts in policy rates are unlikely to rapidly fuel house price bubbles.

The finding of inertia in house price movements is not new: Case and Shiller (1989) documented it for several US cities as early as the 1980s. For the cities they studied, annual changes in real house prices tended to be followed by price changes in the same direction the following year. We show that this is also the case for real house prices at the country level in both advanced economies and EMEs.

In contrast, studies based on modern asset pricing theory assume that new information about economic fundamentals gets incorporated into house prices more or less instantaneously. Himmelberg et al (2005) provide an example of such an analysis and calculate that a 1 percentage point decline in real interest rates could raise house prices by as much as 19-33% in US cities with persistently high house price growth. Our empirical estimates indicate a much smaller response of real house prices to changes in interest rates. In the case of the United States, we find that a 1 percentage point decline in the US short rate leads to a 5% increase in house prices after three years. We find even smaller interest rate effects in other countries. Although the interest rate impacts we estimate are much smaller than the theoretical impacts reported by Himmelberg et al (2005), our estimates are between two and five times larger than those of Jarocinski and Smets (2008) and Crowe et al (2011). We believe that the difference between what asset pricing models predict and what we find can be explained by the large search and transaction costs associated with trading residential real estate, and shifting between owner-occupied and rental housing, that asset pricing theory ignores.

Concerning the relation of our estimates to previously published research, we believe that we document a larger impact of interest rates on house prices in part because we consider the effects of both nominal and real interest rates. Our results show a small but statistically significant impact of the level of a short-term real interest rate on real house price growth, even after allowing for a long distributed lag of changes in the short-term nominal rate.

The paper is organised as follows. Section 2 describes the data and the main stylised facts on house prices and interest rates in our sample. Section 3 presents the empirical model. Sections 4-6 look at the determinants of house prices in the United States, other advanced economies, and EMEs. Section 7 concludes.

 

2. Data and stylised facts

Data samples

Despite their importance, data on house prices are a relatively new statistical indicator. One reason is that house price data in many countries used to be compiled by real estate associations rather than statistical offices. That practice changed with the Great Financial Crisis. Another is that standardisation of house prices can be challenging even within a single country. Different house price indices cover different types of dwelling (new, existing, apartments, houses etc), forms of ownership (owner- occupied, built for wholesale) and geographical areas (capital city, regions).

The house price data used in this paper come from a newly compiled BIS data set on residential property prices (see Scatigna et al (2014)). The data set consists of several series for each country, differentiated by types of housing, geographical areas and time frequencies. The series are transmitted to the BIS by national statistical offices or central banks. These authorities also identified their preferred house price series, which facilitated the choice of data for cross-country comparisons.

For the majority of countries, we use a data set that contains the most homogeneous and internationally comparable house price series provided to the BIS, the so-called Selected Dataset. For 18 advanced economies we use an additional data set with longer historical data, backdated by the BIS in cooperation with national authorities.[1] We also construct longer series at an annual frequency for a sample of seven central and eastern European economies. Appendix 1 lists all the metadata.

The full data set covers 47 countries, of which 20 are advanced economies and 27 are EMEs.[2] We separately analyse house prices in the United States, other advanced economies and EMEs. Most regressions use data at an annual frequency. For the United States, we use a sample of quarterly house prices from 1970 to 2015.

The data set is unbalanced: the sample begins between 1950 (Norway) and 1988 (Portugal) for advanced economies, and between 1966 (South Africa) and 2009 (Romania) for EMEs. For 17 out of 20 advanced economies, the data go back at least to the late 1960s or early 1970s, giving us from 45 to 67 annual observations per country, and a total of 970 observations. Although long historical series are unavaila­ble for most EMEs (with the exception of Korea and South Africa), we gathered at least 15 years of data for two thirds (ie 17) of the countries. The full EME sample has 454 annual observations.

House prices in the long term Is housing a good long-term investment? Our data suggest that the answer is a qualified "yes": nominal house prices increased on average by close to 7% per annum and real prices by 2%% in the sample of 20 advanced economies for which there are 45 years of data on average (Table 1).[3] The corresponding figures for EMEs are 7% and 3% per annum over an average sample length of 15 years. The biggest single­year nominal increases in advanced economies ranged from 16% in Germany to over 50% in Italy (in 1974); the biggest single-year declines from -2% (also in Germany) to -19% (Ireland, in 2009) (Table 1). The EMEs have experienced even wider variations: nominal house prices increased up to 83% in a single year (Latvia, in 2002), and fell by 36-41% in a year (Latvia, the Czech Republic and Estonia, all in 2009).

One striking feature of house price growth is its persistence. With the exception of Germany, Portugal and Switzerland, advanced economies have seen nominal house prices growing by an average of at least 6% per year for 40-50 years (Table 1, first and last columns). In the United States, for instance, this resulted in a 13-fold increase in nominal house prices over a period of 48 years; in Norway, in a 77-fold increase over 67 years.

House prices have also been persistent in real terms: in 13 out of 20 advanced economies, and 16 out of 27 EMEs, real house prices have been growing by more than 2% per year over average sample periods of, respectively, 50 and 19 years. Cumulative increase in real house prices over the sample period thus amounted up to 300-400% in Australia, Norway and the United Kingdom, and as much as 660% in Japan (Table 1, penultimate column). Estonia, Hong Kong SAR and Latvia have seen real house prices more than doubling over much shorter periods.

1  These data have already been used in a number of studies; see eg Drehmann et al (2012).

2 WP665 Interest rates and house prices in the United States and around the world

3 Nominal house prices are important because of their properties in the formation of expectations about future house prices. Even in equity markets there is evidence that investors suffer from inflation illusion: Modigliani and Cohn (1979) and Campbell and Vuolteenaho (2004) show, for instance, that investors apparently value equities using the nominal interest rate instead of the real interest rate.

Another way to appreciate the persistence of house prices is to contrast the length of their upswings and downswings. We define an upswing (downswing) as a period of nominal house price increases (decreases) sustained in an individual country for three years or more. Based on this definition, periods of nominal price upswing accounted for 80% of the advanced economy sample. The upswings lasted on average 12 years, with the longest one, in Australia, lasting half a century, from 1961 to 2010 (Table 2 and Appendix Table A1). By contrast, nominal price downswings accounted for only 10% of the advanced economy sample; they lasted on average five years, with the longest one, in Japan, lasting 18 years (from 1992 to 2009). In EMEs, nominal price upswings accounted for two thirds of the sample. They lasted on average six years, and the downswings four years (Table 2 and Appendix Table A2).

Periods of real price upswing have been shorter: they accounted for 54% of the sample in advanced economies and lasted on average six years (Table 2). Austria, Belgium, Denmark, France, Ireland, Japan, Spain, Sweden, Switzerland, the United States and the United Kingdom experienced uninterrupted real price upswings lasting from 10 to 22 years (Appendix Table A3). On the other hand, real price downswings accounted for 27% of the advanced economy sample, and lasted on average the same as upswings, six years (Table 2). Austria, Japan, Portugal and Switzerland experienced the longest downswings in real house prices, lasting from 11 to 18 years (Appendix Table A3). In EMEs, real house price upswings accounted for 45% and downswings for 28% of the sample (Table 2). They lasted on average six and five years, respectively. Brazil, Hong Kong SAR, Israel, Lithuania, Peru and South Africa experienced eight-year or longer upswings in real house prices; Croatia, Israel, Indonesia and Korea eight- year or longer downswings (Appendix Table A4).

The surge in house prices has been particularly pronounced since the turn of the millennium. Between 2000 and 2015, nominal house prices increased by 100% or more in half the economies in our sample (Graphs 1 and 2). Most experienced a housing boom before 2007 (red bar segments). But many have also seen very rapid house price growth since 2007 (blue bar segments). These included Australia, Austria, Canada, New Zealand, Norway, Sweden and Switzerland among advanced economies (Graph 1); and Brazil, Hong Kong SAR, Israel, Malaysia and Peru among EMEs (Graph 2). In real terms, 13 out of 20 advanced economies have seen real house prices rising by 50% or more in cumulative terms since 2000 (Appendix Graph A4). Even more

Data for individual countries also indicate that real house prices often move in the same direction as interest rates. This was the case for instance in many advanced economies during 2005-07 (Appendix Graph A1). But in other countries, eg Mexico, real house prices reacted little, if at all, to interest rate changes over the past decade (Appendix Graph A2). There are also many cases where real house prices simply kept on rising regardless of interest rate changes, eg in Australia, Austria, Canada, Hong Kong SAR, Malaysia, Norway, Peru, Sweden and Switzerland for the past 10 years or so (Appendix Graphs A1 and A2).

However, past changes in interest rates on average do affect real house price changes in the expected direction. For instance, changes in interest rates lagged by eight quarters in the United States, and by two years in non-US advanced economies and EMEs, are negatively correlated with current changes in real house prices; this negative relationship is statistically highly significant in all three samples (Graph 4).

The empirical relationship between changes in interest rates and real house prices might therefore not be as straightforward as implied by simple models. Changes in interest rates often seem to affect real house prices with long delays. This suggests that several lags of interest rate changes would have to be included in an empirical model of house price determination. And, in view of their extraordinary persistence, several lags of real house price changes would have to be included, too.

One way to illustrate the importance of inertia is to consider the likelihood of a fall in real house prices at a given point in time. Rosengren (2011) suggested, for instance, that an important myth contributing to the crisis was the belief that a diversified portfolio of US residential real estate had little risk of falling in value. As it turned out, US real house prices fell by as much as 31% over the course of 2007-09.
Graph 5 shows a histogram of three-year changes in real house prices for all countries except the United States: at 37%, the probability for real house prices to fall over a three-year period was quite high. The likelihood of a fall at least as large as that experienced in the United States was much lower: 2.3%. However, the histogram reports unconditional probabilities. Against the backdrop of a growing US house price bubble in 2000-06, and much new supply coming on stream due to elevated construction, the conditional probability of a marked house price decline was larger. How much larger is open to debate, but we can take 2.3% as a lower bound to the probability of the three-year decline experienced in the United States.

3. Empirical model

House price dynamics is usually analysed within the framework of either asset pricing or demand/supply models. The asset pricing model considers living in a house one owns and selling it and moving to rental housing as two alternative "assets" to be priced. The equilibrium occurs when the expected cost of owning a house equals that of renting it. While theoretically appealing, this approach is of limited use in cross­country empirical work: rental data are unavailable for most EMEs and unreliable for many advanced economies, as rental markets are often either over-regulated or very opaque. As a result, the available data on rents usually bear little resemblance to rents prevailing in the market.

Although we are unable to test the asset pricing model directly, we can shed some light on two of its key implications. First, the random walk hypothesis, which states that only new information about economic fundamentals (rents, interest rates) influences house prices, and implies that there is little or no inertia in house prices. And, second, that interest rates are likely to be a key determinant of house prices, given that the latter represent an expected present value of future rents.

The alternative specification is the standard demand/supply model, which looks at housing as a durable credit-financed consumption good. It typically analyses house price dynamics with the help of demand determinants such as household income and wealth, interest rates, labour market and demographic factors, and, on the supply side, the price of land, construction costs (wages of construction workers, material costs etc), and, ideally, supply restrictions such as zoning, building, environmental and other regulations. With the exception of supply side variables, which are difficult to obtain for most economies (especially the price of land and the extent and nature of regulations), our specification includes these basic house price determinants.

In particular, our dependent variable is the change in real (ie CPI-adjusted) (log) house prices, Ahp, where A is the first-difference operator. As explanatory variables we use real GDP growth, Ay (a proxy for real income growth); total employment growth, Al (a proxy for changes in labour market and demographic factors); and a number of interest rate variables, Ai. Growth of the domestic bank credit to the non- financial private sector was also initially included, but was eventually dropped because of its high correlation with real GDP growth.

We investigate the impact of both changes in and levels of interest rates; the former are specified in nominal terms, the latter in real, ie CPI-adjusted terms (ir). Theoretical models of the demand for housing would suggest a role for real interest rates in the demand for housing, but nominal interest rates are also important, as lending criteria (eg debt service or debt-to-income ratios) are typically specified in nominal terms. Nominal interest rates may also be important if a large number of potential home buyers suffer from money illusion. Changes in interest rates are expressed in percentage points, real interest rates in percent, and all other changes as differences in natural logarithms. A detailed description of explanatory variables is provided in Appendix 1. The baseline equation we estimate is thus:

where c is a constant and s an error term. This equation is estimated with lags of variable lengths, first on quarterly data for the United States and then on annual data for the two panels of advanced and emerging market economies.

Most studies rely on statistical tests to determine a single, "optimal" length of the time lag for individual variables. However, evidence from Graphs 3 and 4 and Appendix Graphs A1-A3 suggests that this approach might significantly underesti­mate the impact of changes in interest rates on a persistent variable such as changes in house prices. We therefore include many lags of explanatory variables and consider their cumulative effect on changes in house prices.

The use of contemporaneous values in the level of a real interest rate and the change in a nominal interest rate raise the possibility that our results are affected by simultaneous equations bias. The reason is that house prices embody information relevant for the stance of monetary policy, which is itself a key determinant of interest rates. Simple experiments consisting of replacing the current values of some of the interest rates in our regression model with their one-period lagged values suggest, however, that this bias is small, if it exists at all. We believe that this reflects inertia in house prices: it takes time for changes in economic fundamentals to be incorporated into prices of illiquid assets such as houses.

4. United States

How have changes in interest rates affected real house prices in the United States over the past half century? Table 3 reports OLS estimates of equation (1) relating real house price growth to short-term interest rates and the explanatory variables discussed above, from Q2 1970 to Q4 2015. Several results are worth highlighting.

House price changes are highly persistent

Changes in US house prices are highly persistent: four of the eight lags of quarterly real house price growth are statistically significant at traditional confidence levels. All lags except the fourth, sixth and eighth enter the regression with a positive sign. Thus, US house prices tend to move in the same direction for about a year after being hit by a disturbance, and then exhibit a modest reversal.

prices. In particular, this pattern of US house prices is unlikely to be a consequence of forecastable changes in the service yield to housing. Rather, it is more likely that it arises from the large search and transaction costs associated with trading residential property. Under this view, forecastable upward moves in house prices per­sist in equilibrium because the large search and transaction costs associated with house purchases make it difficult to develop profitable trading rules. It is also impos­sible to take advantage of forecastable declines in house prices, because there is no mechanism to short-sell homes. As we show below, correctly capturing inertia is im­portant to estimate accurately the long-run impact of a disturbance on house prices.

Income and employment growth seem less important Interestingly, real GDP growth (as a proxy for income) and total employment growth do not seem to affect house price growth strongly in this setup. Only two of the coefficients on GDP growth are statistically significant (for lags of one and five quarters, at the 10% level only), and only one on employment growth (for a lag of five quarters, at the 5% level). However, as with interest rates, cumulative effects of changes in real GDP and employment growth on house price growth are non- negligible and positive: about 0.15 percentage points for a 1 percentage point faster GDP growth sustained over six quarters, and 0.24 points for a 1 percentage point faster employment growth.

Interest rates matter for house prices

Another key finding is that of a small but protracted impact of changes in nominal short-term interest rates on house price growth. Only two of the 13 estimated coefficients on interest rate changes are statistically significant at traditional levels (lags 5 and 7). However, all but four are negative. This impact is difficult to show with the small sample sizes often used in the literature. Indeed, we show in the next section that changes in short-term interest from as long as five years in the past can have a statistically highly significant impact on current changes in house prices.

We also find a small but statistically significant impact from the real short-term interest rate on house price growth. The evolution of the US real short-term rate over the sample period is shown in Graph 6. There are both high- and low-frequency movements in this variable. In particular, there was a marked increase in the US real short-term rate from the mid-1970s to the mid-1980s, after which a long-term decline set in. In view of the importance of US financing conditions for global financial markets, this long-term trend has also been reflected in a trend decline in interest rates in other economies (shaded areas in Appendix Graphs A1-A3).

Given that monetary policy affects the level of the short-term nominal interest rate, and that policy rates may be set with a view to affecting house prices, the use of contemporaneous values of the level and change in short-term rates may introduce a simultaneous equation bias in our OLS estimates. When we re-estimate the regression replacing the contemporaneous value of the real short-term rate with its value lagged one quarter, the coefficient rises only slightly, from -0.0007 to -0.0006. We take this as evidence that potential simultaneous equations bias is most likely to be small.

How do US house prices respond to changes in long-term interest rates, which would at first glance seem to be more relevant for the path of house prices? Graph 7 plots two US long-term interest rates - the 10-year government bond yield and the 30-year mortgage rate - against the three-month nominal Libor used above. The 10- year bond yield and the 30-year mortgage rate co-move closely, so using one.

Estimates of the effects of long-term interest rates on house prices are provided in Appendix Table A5. They are broadly comparable to those in Table 3: real house prices are again persistent; GDP growth is somewhat more and employment growth somewhat less significant;   and most coefficients   on long-term interest rates are properly signed but fewer are significant than in the  case   of short-term   rates. As discussed below, this results in very similar cumulative effects on house prices of long­term and short-term interest rates.

Cumulative effects of interest rates are large

Our estimated model permits a straightforward calculation of the response of house prices to a 100 basis-point fall in the nominal short-term rate if two main conditions hold: (i) inflation expectations do not change in response to a 100 basis-point fall in the nominal short-term rate that takes place at time zero, so the real short-term rate also falls by 100 basis points at time zero; and (ii) the cut in rates remains in force for the next three years. Appendix 3 explains how we derive cumulative effects as reported in the graphs below.

Under these assumptions, our estimated model implies that a 100 basis-point fall in the nominal US short-term rate generates a rise in real house prices, relative to the baseline, of about 8 percentage points after three years.

There are reasons to suspect that this dynamic impact overstates the sensitivity of real house prices to short-term interest rates. The full sample includes the 2000­06 period, during which short-term rates were low and house prices surged. Yet the house price bubble of 2000-06 could have been the result of other factors that just happened to coincide with the period of low interest rates, such as the pronounced decline in lending standards associated with the US subprime mortgage boom. To eliminate this special effect, we consider a different time period: if a systematic reaction of house prices to interest rates was the main cause of the bubble, then we should be able to find a significant interest rate effect prior to 2000 as well.

Graph 8 plots the estimated interest rate effect on real US house prices, up to 12 quarters into the future, for a 100 basis-point fall in the nominal US short-term rate. The exercise uses the assumptions stated above, for the same regression model reported in Table 3, but estimated only on data up through 1999 Q4. Consistent with our conjecture, we find a smaller impact of changes in short-term interest rates on

US house prices: a 100 basis-point reduction in the short-term rate is now associated with a 5 percentage point rise in real house prices after three years relative to baseline (upper left-hand panel). Note that there is only a tiny response of house prices initially, but the effect steadily increases for the next 12 quarters.

Comparison with earlier studies

Persistence of house prices

Our finding of substantial inertia in real house prices suggests that Bayesian estima­tion methods that rely on a random walk prior - for instance Jarocinski and Smets (2008) - are likely to underestimate the effects of interest rate disturbances on house prices. The bottom right-hand panel of Graph 8 shows the estimated impact on house prices when we set all of the coefficients on lagged house price growth equal to zero. This is not a pure random walk model because it permits serially correlated changes in the other explanatory variables to generate fluctuations in real house prices that are partly predictable. It is nevertheless an approximate random walk model.

The results of the experiment show that imposing an approximate random walk model for real house prices leads to a substantial underestimation of the impact of changes in interest rates. This suggests that the random walk prior should not be employed in the estimation of price dynamics for illiquid assets like housing, for which it is not reasonable to expect new information about economic fundamentals to be rapidly incorporated into prices. Of course, the random walk prior makes good sense for assets traded in liquid markets with low transaction costs, where estimated forecastability is likely to be spurious.

Interest rate effects

Even the conservative estimates of dynamic interest rate effects shown in the top left- hand panel of Graph 8 are considerably larger than some previously reported ones. For instance, Sutton (2002), using a small VAR model, estimated that US real house prices increased by about half a percentage point, relative to baseline, four quarters after a surprise 100 basis-point fall in the real short-term rate. This is about a third of the size of the effect shown in the top left-hand panel of Graph 8. Jarocinski and Smets (2008), using a Bayesian VAR in levels on data through the first half of 2007, found that US real house prices fell by half a percentage point, relative to baseline, 10 quarters after a surprise 25 basis-point rise in the federal funds rate. Our conservative estimates would predict a bit more than a 1 percentage point fall in real house prices 10 quarters after the equivalent interest rate change, about twice the effect reported by Jarocinski and Smets (2008).

Do we find larger interest-rate effects because the studies mentioned above have estimated VAR models, while the present paper estimates a single regression equation that treats contemporaneous values of interest rates as exogenous variables? Our estimates suggest that the answer is no. The bottom left-hand panel of Graph 8 shows the impact of a 100 basis-point fall in the real short-term rate implied by our estimated model, setting all of the coefficients on changes in the nominal short-term rate equal to zero. This exercise predicts a rise in house prices of about half a percentage point after four quarters. This is the same impact Sutton (2002) found for a VAR model of real house price growth, estimated over roughly the same sample period, which includes the level of the US real short-term rate, but not changes in the nominal US short-term rate. This suggests that our larger estimated impact arises because the present study also includes a long distributed lag of changes in the nominal short-term rate. One explanation for its additional importance could be that a large part of the population does not distinguish between nominal and real interest rates, ie it suffers from money illusion.

The effects of reversing the experiment are shown in the top right-hand panel of Graph 8: it suggests that the level of the real short-term rates and the distributed lag of changes in the nominal short-term rate contribute roughly equally to our total estimated impact of a 5 percentage point gain after three years, following a 100 basis­point cut in short-term rates. We suspect that the long distributed lag of changes in the nominal short-term rate are partly driving this result. In particular, we found correctly signed interest rate coefficients for lags up to seven quarters. By comparison, Sutton (2002) used four, and Jarocinski and Smets (2008) five quarterly lags of interest rates. This implicitly sets the coefficients of longer interest rate lags to zero. Our results suggest that both studies should have employed more lags.

One additional reason why the estimated dynamic effects of interest rates on house prices shown in Graph 8 are larger than those reported in the previous studies we cite is that our calculations assume that interest rates remain 100 basis points lower than they were at the beginning of period zero for the next three years. In particular, we do not allow interest rates to exhibit any mean reversion over the course of the experiment. In contrast, impulse response functions obtained from VAR models permit interest rate shocks to dissipate steadily over time.

Another finding that differs from some previous studies is that changes in interest rates influence real house prices only gradually over time rather than mostly on impact. For instance, Del Negro and Otrok (2007) estimated impulse responses from monetary policy shocks that display a relatively large initial impact on house prices, which then declines over time. One reason is that monetary policy shocks are identified in part by the requirement that they have the largest effect on house prices. This identification assumption leads to large contemporaneous effects. Iacoviello and Neri (2010), who used a DSGE model, also reported impulse responses showing a large initial effect of a monetary policy shock on US house prices that subsequently dissipated. In contrast, Iacoviello (2005) reported impulse responses that showed real US house prices falling for about eight quarters following an unexpected rise in the nominal federal funds rate. His VAR model was identified by way of a Choleski decomposition of the reduced form error terms.

Short-term vs long-term interest rate effects

The left-hand panel of Graph 9 shows the estimated effects on US real house prices of a 100 basis-point fall in nominal and real US long-term interest rates. The estimation period is the same as in Graph 8. The impact of a change in long-term rates is stronger initially, but then rises less quickly than that for short-term rates. After 12 quarters, the effects are basically the same.

Why would short-term rates be as important to house prices as long-term rates? As noted above, one possibility is that short-term rates better capture the additional boost to house prices resulting from the bank lending channel of monetary policy. Kishan and Opiela (2000) found that small US banks reduced their real estate lending in response to a tightening of monetary policy. They argued that small banks - especially undercapitalised ones - found it more difficult to replace demand deposits with other funding sources and were hence more likely to reduce lending when financing conditions tightened. While they did not provide separate results for lending for residential and commercial real estate, it seems plausible that the cuts in lending induced by rising funding costs affected housing demand and eventually led to some drop in house prices.

Arguably, the bank lending channel is more important when home mortgages are not securitised. When they are, funding for additional home loans can be obtained by selling, for the purpose of securitisation, loans the bank has already made. This

suggests that the relative importance of short- vs long-term rates in determining house prices would be greater in those countries where securitisation of home loans is less prevalent than in the United States. This is exactly what we find in other advanced economies and EMEs (see Sections 5 and 6).

A related issue is whether changes in lending standards play a role in driving house prices, presumably by being part of the bank lending channel. Data on lending standards for US residential real estate are available from 1990 (Graph 9, right-hand panel). The lending standards series is dominated by two large spikes associated with recessions in 1990-91 and 2008-09. This suggests that a large component of the lending standards series is endogenous: weak economic conditions and probably also falling prices of residential real estate lead to a tightening of standards, which are then relaxed as economic conditions return to more normal levels. Also, there does not appear to be a strong correlation between the lending standards series and the US real short-term rate, our proxy for the stance of US monetary policy (purple line in Graph 9, right-hand panel).

5. Other advanced economies

We next study the impact of interest rates on real house prices in advanced economies other than the United States. For a panel of 20 countries, we have a total of 718 annual observations. We use panel least squares estimates with country and time fixed effects. Table 4 reports two sets of estimates: one with changes in domestic short-term interest rates and the level of the US real rate (Panel A), and another with only domestic interest rates, both changes and the level (Panel B). Table A6 in the Appendix provides the corresponding estimates with long-term interest rates.

House prices are persistent, fundamentals important

Panel estimates for non-US advanced economies provide additional evidence that house price moves in one year tend to be followed by moves in the same direction the following year (Table 4). The positive coefficients on lagged real house price growth are both large and statistically highly significant. The coefficient turns negative at lag two, suggesting some reversal in house price growth over the medium term, as we also found in US quarterly data.

GDP growth affects house price growth in non-US advanced economies almost one-for-one: in Panel A, the coefficient on contemporaneous GDP growth is equal to 0.95 and is statistically highly significant. Output growth lagged two years has a weaker but also significant impact on house price growth. In addition, contempora­neous growth in total employment (estimated coefficient of 0.65) strongly affects the growth in house prices. In Panel B, the effects of GDP growth on house prices are even stronger; those of employment are about the same.

Domestic vs US interest rates

Estimates for non-US advanced economies provide further evidence for a small but persistent effect of interest rate changes on house price growth: in Panel A, five of the six coefficients have the expected negative sign; three are statistically significant at the 5% level or higher (lags one, three and five); and the one on current interest rate change is significant at the 10% level. The finding that interest rate changes from up to five years in the past can significantly affect current house prices underlines the importance of considering a long distributed lag for interest rate changes.

The estimated coefficient on the contemporaneous value of the US real short­term interest rate is small (-0.0034) but statistically highly significant, suggesting that shifts in the stance of US monetary policy might also affect house prices in other advanced economies. For instance, the real US short-term rate fell rapidly when monetary policy was eased to offset the effects of recessions in 1989, 2000 and 2007. This could have put upward pressure on house prices outside the United States if overseas banks funded themselves in part in the US interbank market.

The upper left-hand panel of Graph 10 plots the estimated impact on house prices of a global, perfectly correlated fall in short-term interest rates across countries. This includes a 100 basis-point fall in the nominal short-term interest rate in every country in the sample, plus the equivalent fall in the US short-term real interest rate. Of course, as short-term interest rate changes are not normally so highly correlated, our estimates can be taken as an upper bound to the impact of changes in US interest rates on house prices in other advanced economies. The estimates suggest that a 100 basis-point fall in global short-term rates is associated with a rise in advanced economy real house prices by 3% percentage points, relative to baseline, after three years. The effect continues to grow in years four and five, reaching almost 5 percentage points after five years.

domestic counterpart. The coefficient on the country-specific real short-term rate is slightly smaller in absolute value than the one on the US real short-term rate in Panel A, but remains statistically significant.

Interestingly, the response of real house prices to a purely domestic fall in short­term rates (Graph 10, upper right-hand panel) is smaller than the one accompanying an easing of US monetary policy (upper left-hand panel). Five years after a 100 basis­point fall in domestic short-term rates, house prices are less than 3 percentage points higher than they would have been otherwise, all else equal. This is 2 percentage points less than when domestic and US interest rates fall together.

Another interesting finding is that house prices in non-US advanced economies seem to be much less sensitive to changes in domestic short-term interest rates than they are in the United States: following a 100 basis-point cut in interest rates, they rise by slightly less than 2 percentage points after three years, compared with 5 points in the US case.

Long-term vs short-term interest rates

The lower panels of Graph 10 show the impulse responses we obtain when we replace the estimated coefficients on short-term rates from Table 4 with their long-term counterparts from Appendix Table A6.

The lower left-hand panel shows the estimates for the US real long-term rate: we obtain a larger impact of 6% after five years. A more striking change is shown in the lower right-hand panel: when we include only domestic long-term rates, the response after five years is only about a third of that with domestic short-term rates. Thus, house prices in non-US advanced economies seem to be more sensitive to changes in short-term than in long-term interest rates. This is consistent with the view that the bank lending channel of monetary policy is relatively more important in these countries, partly because of the less widespread use of mortgage securitisation than in the United States, and partly because of the prevalence of variable-rate mortgages.

Comparison with earlier studies

Our estimates suggest that house price growth in other advanced economies is less sensitive to changes in short-term interest rates than it is in the United States, where the equivalent exercise led to a 5 percentage point increase in house prices after three years (Graph 8, upper left-hand panel). Nevertheless, the effect is not negligible. Moreover, our estimate of this effect is five times larger than that found by Crowe et al (2011), who examined the impact of policy rates on house prices in 22 countries within the context of a VAR model. They reported that an unexpected 100 basis-point rise in the policy rate was associated with only a 1 percentage point fall in house prices, relative to baseline, after five years. This small effect is even more surprising given that their VAR model included real GDP and permitted the hike in interest rates to slow down GDP growth. This should have led to a larger impact of rate hikes on house prices than in our analysis, which treats GDP growth as exogenous. We suspect that our finding reflects the inclusion of a long distributed lag of changes in domestic interest rates, and the level of the US real interest rate.

Our estimates are also relevant for discussions about how much US short-term rates would need to have been raised to have prevented, or at least delayed, the onset of the 2008-09 crisis (see Appendix 4). Jorda et al (2015) argued that US short-term rates would need to have been raised by 8 percentage points to prevent the most recent boom-bust housing cycle in the United States. They reached this conclusion mainly by studying the response of house prices to interest rates in countries other than the United States, and assuming that the relationship they found was applicable to the United States. However, our finding that real house prices respond to interest rates more strongly in the United States than they do in other advanced economies suggests that US short-term rates might need to have been raised by less than the claimed amount.

Houses vs equities: differential impact of US interest rates

To highlight the differential impact of changes in US short-term interest rates on illiquid and liquid assets, in Table 5 we replaced changes in house prices with those in equity prices in non-US advanced economies.

In sharp contrast to what we found with house price growth, the coefficient on the level of the US real short-term rate is positive and statistically highly significant. The positive coefficient is to be expected for liquid assets such as equities: favourable global economic developments that coincide with a tightening of the US policy rate are rapidly incorporated into stock prices. When we lag the real US short-term rate by one year, the coefficient falls from 0.014 to -0.002 and completely loses statistical significance.


The estimates with equity price growth also reveal what at first seems to be a surprising result: the coefficient on GDP growth lagged one year is negative and statistically highly significant. This could be caused by measurement error in GDP that is positively and contemporaneously correlated with equity prices. High measurement error in period t-1 pushes up GDP growth from t-2 to t-1, as well as equity prices in period t-1. Since the high measurement error is likely to be unwound in period t, this tends to reduce equity price growth from t-1 to period t, all else equal.

We sometimes also get a negative coefficient on GDP growth lagged one period in regressions with real house price growth. But consistent with our view that information about fundamentals is less rapidly incorporated into house prices, the coefficient on GDP growth lagged one year is statistically insignificant in those regressions (one exception is Table 6).

6. Emerging market economies

How do interest rates affect house prices in EMEs? As with advanced economies, we first look at the effects of short-term interest rates and then compare these results with what we obtain with long-term rates.

In some ways, the results for EMEs reported in Table 6 are similar to what we found for non-US advanced economies. For instance, there is again strong evidence against the random walk model for real house prices: annual house price growth lagged one period is statistically highly significant. As the coefficient on this variable is positive, house price moves in one year tend to be followed by moves in the same direction the following year. And the coefficient on the second lag of real house price growth is negative, suggesting - as in advanced economies - a reversal of real house price growth two years after a disturbance. We also find a positive and statistically significant coefficient on contemporaneous GDP growth, but not on contemporane­ous employment growth.

Changes in short-term rates affect real house prices in EMEs also with long lags, and the coefficient on the level of the US real rate is negative, as expected, and significant at the 5% level. The predicted impact for a 100 basis-point fall in US and domestic short-term rates is quite large: real house prices are 5% percentage points higher over the course of five years (Graph 11, upper left-hand panel). This is slightly higher than in non-US advanced economies (Graph 10, upper left-hand panel).

As in other advanced economies, the response of real house prices in EMEs to a purely domestic fall in short-term rates (Panel B in Table 6 and Graph 11, upper right- hand panel) is smaller than the one accompanying the fall in US rates (Panel A and Graph 11, upper left-hand panel). Five years after a 100 basis-point fall in domestic short-term interest rates, house prices in EMEs would be about 2% percentage points higher than otherwise, all else equal. This is less than half the impact of the fall in the US real interest rate. It is also half a percentage point less than the equivalent impact in other advanced economies.

The lower left-hand panel of Graph 11 shows the dynamic impact we get when we replace short-term with long-term rates, including the real US long-term rate (see Appendix Table A7 for regression estimates). We get much smaller effects in this case. The lower right-hand panel shows the results when only domestic long-term rates are included in the regression. In this case, the impact turns slightly negative from the fourth year on. This suggests that short-term rates are a more important determinant of house prices in EMEs than long-term rates. Again, this is consistent with our view that the bank lending channel of monetary policy is relatively more important in countries for which securitisation of home mortgages is less prevalent. Interestingly, spillover effects from the US to EME housing markets seem strongest with US short­term rates. This is consistent with the view that shifts in US monetary policy drive capital flows to EMEs, which eventually seem to filter through to housing finance growth in EMEs.

7. Conclusion

Most empirical studies assume that short-term interest rates do not influence house price growth other than through the domestic cost of borrowing, ie by their influence on long-term interest rates. The findings in this paper suggest that this view might be mistaken: changes in short-term interest rates seem to have a strong and persistent impact on house price growth. Moreover, global, ie US short-term interest rates - not just domestic ones - seem to matter, both in advanced economies and EMEs. We interpret the relative importance of short-term interest rates in driving house prices as indicating an important role for the bank lending channel of monetary policy in determining housing financing conditions, especially outside the United States, where securitisation of home mortgages is less prevalent.

The larger effect of interest rates on house prices we find reflects in part the use in our regressions of a long distributed lag of interest rate changes. For the United States, our estimates for the period from 1970 to the end of 1999 suggest that a 100 basis-point fall in the nominal short-term rate, accompanied by an equivalent fall in the real short-term rate, generated a 5 percentage point rise in real house prices, relative to baseline, after three years. We find an even larger effect if we include the data through end-2015. For other advanced economies and EMEs, we estimate that a 100 basis-point fall in domestic short-term interest rates, combined with an equivalent fall in the US real rate, generates an increase in house prices of up to 3% percentage points, relative to baseline, after three years.

Another reason we find larger interest rate effects is by allowing for inertia in house price movements. We find strong evidence against the random walk hypothesis: real house prices around the world tend to move in the same direction for about a year after being hit by a disturbance, then exhibit a modest reversal. We think that this inertia in house prices reflects the large search and transaction costs associated with trading residential real estate and shifting between owner-occupied and rental housing. These costs are ignored in the user cost model, which predicts a fairly high interest rate sensitivity for house prices.

Our findings also suggest a potentially important role for monetary policy in countering financial instability. While higher short-term interest rates alone cannot significantly dampen the demand for housing, slower house price growth can give supervisors more time to implement measures to strengthen the financial system. At the same time, the finding that house prices adjust to interest rate changes gradually over time suggests that modest cuts in policy rates are not likely to rapidly fuel house price bubbles.

Our paper leaves a number of unanswered questions for future research. First, why are house prices in the United States more responsive to changes in interest rates than is the case for other countries? One possibility is that house price sensitivity to interest rates is inversely related to the size of the transaction costs for trading real estate, which are relatively small in the United States. While we think this is a reasonable conjecture, we offer no direct evidence that it is true. Second, in the United States, we find a roughly equal contribution to house price fluctuations from real and nominal interest rates. While this could arise from a large number of potential home buyers suffering from money illusion, we cannot offer direct evidence that this is the case. Third, the relatively large response of house prices in EMEs to US short-term interest rates suggests that US rates may have a larger global impact than previously recognised. This could arise, for example, if house prices in EMEs were especially sensitive to global liquidity conditions.

The bank lending channel of monetary policy gives us confidence that our use of the US short-term interest rate to capture global liquidity is reasonable; however, we recognise that there are other proxies. Cerrutti et al (2014) suggest, along with other measures, the term spread as an indicator of the availability of global liquidity. A low term spread implies reduced profit opportunities from borrowing short and lending long domestically. This can encourage banks to increase their risk-taking, extending loans to those who were previously denied credit, including borrowers abroad. Another candidate is repo borrowing: Adrian et al (2015), for instance, argue that the extent of repo borrowing by the US Federal Reserve's primary dealers is a measure of their risk appetite. Bruno and Shin (2015) present evidence consistent with the view that greater risk appetite is associated with greater cross-border bank lending. And the thesis that greater cross-border bank lending would affect house prices around the world is supported by the findings of Cesa-Bianchi et al (2015).

Future research could also investigate which measures of global liquidity are most correlated with house prices in advanced economies and EMEs and how sensitive these findings are to the inclusion of the 2000-06 period in the analysis. This is for the same reason that our preferred estimates for the impact of interest rates on US house prices rely on data only through the end of 1999: the 2000-06 period may be unusual. Shiller (2008) argues that the period was associated with a feedback loop between the demand for housing and expected capital gains on housing assets: price increases fuelled demand by raising expected capital gains, and the greater demand fed back on prices. He also suggests that "new era" stories that supported boom thinking, in the United States and elsewhere in the world, played an important role in this feedback loop. Ample global liquidity arguably helped this process along, but the effects of global liquidity on house prices might be smaller in the absence of a strong feedback loop from capital gains on housing assets to demand.

 

APPENDIX

1. Data description and sources House prices Advanced economies

Australia: from 2003 Q3 onward: all dwellings (8 cities), pure price, NSA. From 1986 Q3-2003 Q2, all detached houses (8 cities), pure price, NSA; from 1970 Q1-1986 Q2: Median dwelling prices, capital cities. BIS code: Q:VWBA:AU:01.

Austria: Price per square meter of usable space. Based on the EUR/m2 purchase price for new and used apartments and single-family houses in Vienna. Data inputs were generated by AIB (Austria Immobilienborse, a platform of 17 real estate agencies). BIS code: Q:VSJA:AT:44.

Belgium: from 2005 Q1 onward: all dwellings, pure price, NSA; from 1973 Q1-2004 Q4: existing dwellings, price per dwelling, NSA; from 1970 Q1-1972 Q4: Index of small- and medium-sized dwellings. BIS code: Q:VWBA:BE:01. Canada: from 1980 Q1 onwards: National residential average price, NSA; from 1970 Q1-1979 Q4: Average price of existing homes. BIS code: Q:VWBA:CA:01.

Denmark: from 2002 Q1 onwards: all types of dwelling in the country as a whole; from 1970 Q1 to2002 Q3: single­family houses, pure price, NSA. BIS code: Q:VWBA:DK:01.

Finland: from 2010 Q1 onwards: all dwellings, pure prices, NSA; 2005 Q1-2009 Q4: existing dwellings, per m2; 1983 Q1-2004 Q4: existing flats and terraced houses, total, per m2, NSA; 1970 Q1-1982 Q4: existing flats. BIS code: Q:VWBA:FI:01.

France: from 2006 Q2 onwards: all dwellings, pure price, Q-All, NSA; 1996 Q1-2006 Q1: existing dwellings, pure price, Q-All, NSA; 1970 Q1-1995 Q4: Produits dérivés, un sous-jacent immobilier, Ministère de l'Equipement, February 1999. BIS code: Q:VWBA:FR:01.

Germany: from 2003 Q1 onwards: all owner occupied dwellings, pure price, NSA; 1975 Q1-2002 Q4: New dwellings in West Germany; 1972 Q1-1974 Q4: Average of sales prices of owner-occupied flats, per m2, in Frankfurt, West Berlin, Hamburg and Munich. 1970 Q1-1971 Q4: Construction costs. BIS code: Q:VWBA:DE:01.

Ireland: from 2005 Q1 onwards: all dwellings, pure price, NSA; 1970 Q1-2004 Q4: price index, new houses. BIS code: Q:VWBA:IE:01.

Italy: from 1990 Q1 onwards: all dwellings, pure price, NSA; 1971 Q1-1989 Q4: Bank of Italy historical residential property price index. BIS code: Q:VWBA:IT:01.

Japan: from 2008 Q2 onwards: all dwellings, pure price, NSA; 1970 Q1-2008 Q1: Land prices, residential, urban areas, per m2, NSA. BIS code: Q:VWBA:JP:01.

The Netherlands: from 2005 Q1 onwards: all dwellings, pure price; 1995 Q1-2004 Q4: all existing dwellings, pure price, NSA; from 1976 Q1-1995 Q4: existing dwellings; 1970 Q1-1975 Q4: sales of houses and flats brokered by real estate agents. BIS code: Q:VWBA:NL:01.

New Zealand: from 1979 Q4 onwards: all dwellings, per dwelling, NSA; 1970 Q1-1979 Q3: Quarterly house price index main urban areas; quotable value New Zealand Limited. BIS code: Q:VWBA:NZ:01.

Norway: from 1992 Q1 onwards: all dwellings, pure price, NSA; 1970 Q1-1991 Q4: House prices, from O Eitrheim and S Erlandsen, House price indices for Norway, 1819-2003, pp 349-76, 2004. BIS code: Q:VWBA:NO:01.

Portugal: from 2008 Q1 onwards: all dwellings, pure price (INE), NSA; 1988 Q1-2007 Q4: all dwellings (Confidencial Imobiliario). BIS code: Q:VWBA:PT:01.

Spain: from 2007 Q1 onwards: all dwellings, pure price, NSA; 1987 Q1-2006 Q4: all dwellings, per m2, NSA; 1975 Q1-1986 Q4: House prices in the capital city area; 1971 Q1-1974 Q4: OECD historical statistics. BIS code: Q:VWBA:ES:01.

Sweden: from 2005 Q1 onwards: all types of dwelling in the country as a whole; 1986 Q1-2004 Q4: all owner- occupied houses, per dwelling, NSA; 1970 Q1-1985 Q4: index of owner-occupied one- and two-dwelling buildings. BIS code: Q:VWBA:SE:01.

Switzerland: from 1970 Q1 onwards: unweighted average of owner occupied flats and houses in the country as a whole. BIS code: Q:VWBA:CH:01.

United Kingdom: from 1970 Q1 onwards: all dwellings (ONS), per dwelling. BIS code: Q:VWBA:GB:01.

United States: from 1975 Q4 onwards: existing dwellings, per dwelling, SA; 1970 Q1-1975 Q3: Average sale price of existing single-family homes. BIS code: Q:VWBA:US:01.

Emerging market economies

Bulgaria: from 2009 Q1 onwards: all flats, NSA; 1993 Q1-2008 Q4: existing flats in big cities. BIS code: Q:VWBA:BG:01.

Brazil: all types of new and existing dwelling in the metropolitan area. BIS code: Q:VSFA:BR:R1.

China: new residential buildings, 70 city average, CEIC, National Bureau of Statistics.

Croatia: all types of new and existing dwelling in the whole country, BIS code: Q:VSFA:HR:R1.

Czech Republic: from 2008 Q1 onward: all owner occupied dwellings, pure price BIS code: Q:VSKA:CZ:04; 1999 Q1- 2004 Q1: existing flats, price per square meter (BIS code: Q:VSJA:CZ:90); from 2004 Q1-2007 Q4: existing flats, price per square meter, data compiled from tax returns (BIS code: Q:VSKA:CZ:04).

Estonia: from 2003 Q3: all flats in Tallinn, average price per square meter BIS code: Q:VSJA:EE:24. From 1997 Q1- 2003 Q2: 2-room apartments in Tallinn, average price per square meter, national statistical office.

Hong Kong SAR: all types of new and existing dwelling in the country, price per square meter, weighted average of five component indices categorised by saleable areas, BIS code: Q:VSJA:HK:00.

Hungary: FHB-House-Price-Index, http://www.fhbindex.com.

Indonesia: new houses in big cities, price per dwelling, BIS code: Q:VSLA:ID:86.

Israel: owner occupied dwellings, pure price, BIS code: Q:VSJA:IL:00.

Korea: all dwellings, pure price, BIS code: Q:VSJA:KR:00.

Latvia: from 2006 Q1: new dwellings, pure price, BIS code: Q:VSLA:LV:00; from 2000 Q1-2005 Q4: flats, urban areas, price per dwelling, BIS code: Q:VSJA:LV:91.

Lithuania: all dwellings, price per square meter, BIS code: Q:VSJA:LT:00.

Morocco: existing dwellings, price per square meter, BIS code: Q:VSKA:MA:00.

Macedonia: all flats in Skopje, pure price, BIS code: Q:VSJA:MK:44.

Mexico: all dwellings, pure price, BIS code: Q:VSJA:MX:00.

Malaysia: all dwellings, price per square meter, BIS code: Q:VSJA:MY:00.

Peru: new and existing flats in high-end neighbours in Lima, BIS code: Q:VSJA:PE:44.

Philippines: flats and commercial properties in Makati (in metropolitan Manila), price per square meter, BIS code: Q:VSJA:PH:44.

Poland: from 2006-Q3: new flats in Warsaw, price per square meter, BIS code: Q:VSLA:PL:44. From 2002 Q4-2006 Q2: new flats in Warsaw, price per square meter, old definition, BIS code: Q:VSLA:PL:94.

Romania: all dwellings, pure price, BIS code: Q:VSJA:RO:00.

Russia: existing dwellings, price per square meter, BIS code: Q:VSKA:RU:10.

Singapore: all dwellings, price per square meter, BIS code: Q:VSJA:SG:10.

Slovakia: from 2005 Q1 all dwellings, price per square meter, BIS code: Q:VSJA:SK:00. 2002-2005, same series but annual data. BIS code: A:VSJA:SK:00.

Slovenia: from 2007 Q1, new dwellings, pure price, BIS code: Q:VSLA:SI:00; 2003 Q1-2006 Q4, existing dwellings, price per square meter, BIS code: Q:VSKA:SI:91.

South Africa: all middle segment dwellings, price per dwelling, BIS code: Q:VSJB:ZA:05.

Thailand: new and existing detached houses in Bangkok, BIS code: Q:VSJB:TH:32.

Inflation rates

All countries (except Australia and New Zealand): Consumer price index, BIS code: M:VEBA:cc:LZ (cc = country code), monthly.

Australia and New Zealand: Consumer price index, BIS code: Q:VEBA:cc:LZ (cc = country code), quarterly.

Real GDP

All countries (except Ireland and Morocco): gross domestic product, at market prices, volume. Source: national data via the BIS databank.

Ireland: from 1997, national data via the BIS databank. Before 1997, gross domestic product, at 2013 chained prices, Datastream code: IRGDPCON, annual.

Morocco: gross domestic product, at constant 1998 prices, Datastream code: MCGDP...C, quarterly.

All real GDP data were seasonally adjusted.

Employment Advanced economies

Austria: Total employment, thousands of persons, Datastream code: OEEMPTOTP, quarterly.

Australia: Employed population (aged 15 and over), thousands of persons, Datastream code: AUQLFT12P, quarterly. Belgium: National employment, thousands of persons, Datastream code: BGEMPALLO, quarterly.

Canada: Total employment, thousands of persons, Datastream code: CNEMPTOTO, monthly.

Denmark: Employment, domestic concept, thousands of persons, Datastream code: DKOAYOOPP, annual.

Finland: Total employment (15-74 years), thousands of persons, Datastream code: FNEMPTOTP, monthly.

France: Total employment, thousands of persons, Datastream code: FREMPTOTO, quarterly.

Germany: Employed population (aged 15 and over), Datastream code: BDQLFT12O, quarterly.

Ireland: Employed population (aged 15 and over), Datastream code: IRYLFT12O, annual.

Italy: Employment, domestic concept, thousands of persons, Datastream code: ITOAYOOPP, annual.

Japan: Total employment, thousands of persons, Datastream code: JPEMPTOTO, monthly.

The Netherlands: Employment, domestic concept, thousands of persons, Datastream code: NLOAYOOPP, annual. New Zealand: Civilian employment, domestic, thousands of persons, Datastream code: NZANECD., annual. Norway: Total employment, thousands of persons, Datastream code: NWEMPTOTP, quarterly.

Portugal: Total employees, thousands of persons, Datastream code: PTYLF007O, annual.

Spain: Total employment, thousands of persons, Datastream code: ESEMPTOTP, quarterly.

Sweden: Total employees, thousands of persons, Datastream code: SDQLF007O, quarterly.

Switzerland: Employees, thousands of persons, Datastream code: SWEMPTOTO, quarterly.

United Kingdom: Total workforce jobs, thousands of persons, Datastream code: UKEMPTOTO, quarterly.

United States: Total civilian employment, thousands of persons, Datastream code: USEMPTOTO, monthly.

Emerging market economies

Bulgaria: Total employment, thousands of persons, Datastream code: BLEMTOT.P, monthly.

Brazil: from 2002 Q1: employment (PME Survey), historical metropolitan areas, millions of persons, Datastream code: BREMPTOTP, monthly. Before 2002 Q1: total employment, thousands of persons, Datastream code: BREMPTO6P, monthly.

China: Total employment, tens of thousands of persons, Datastream code: CHEMPTOT, annual.

Croatia: Total employment, persons in paid employment in legal entities, thousands of persons, Datastream code: CTEMPALLP, monthly.

Czech Republic: Total employment, domestic concept, thousands of persons, Datastream code: CZYETP01O, annual.

Estonia: Total employment, thousands of persons, Datastream code: EOEMPTOTP, quarterly.

Hong Kong SAR: Total employment, thousands of persons, Datastream code: HKEMPTOTP, monthly.

Hungary: Employment, domestic concept, thousands of persons, Datastream code: HNYETP01O, annual. Indonesia: Total employment, actual, Datastream code: IDEMPALL, annual.

Israel: Total employment, thousands of persons, Datastream code: ISEMPTOTO, quarterly.

Korea: Total employment, thousands of persons, Datastream code: KOEMPTOTP, monthly.

Latvia: Total employment, Labour Force Survey, thousands of persons, Datastream code: LVEMPTOTP, quarterly. Lithuania: Total employment, thousands of persons, Datastream code: LNEMPTOTP, quarterly.

Morocco: Employment, thousands of persons, Datastream code: MCI67EP, quarterly.

Macedonia: Total employment, actual, Datastream code: MKEMPTOTP, quarterly.

Mexico: Total employment, actual, Datastream code: MXEMPTOTP, quarterly.

Malaysia: Employment, thousands of persons, Datastream code: MYEMPTOT, annual.

Peru: Employment, firms with more than 100 employees, 2010 October = 100, Datastream code: PEEMPALLH, monthly.

Philippines: Total employment, thousands of persons, Datastream code: PHEMPTOTP, quarterly.

Poland: Total employment, thousands of persons, Datastream code: POEMPTOTP, quarterly.

Romania: Registered employment, thousands of persons, Datastream code: RMEMPTOTP, monthly.

Russia: Total employment, millions of persons, Datastream code: RSEMPTOTP, monthly.

Singapore: Total employment, thousands of persons, Datastream code: SPEMPTOT, annual.

Slovakia: Total employment, thousands of persons, Datastream code: SXEMPTOTP, quarterly.

Slovenia: Registered employment, actual, Datastream code: SJEMPTOTP, monthly.

South Africa: Employees, actual, Datastream code: SAEMPTOTP, quarterly.

Thailand: Total employment, thousands of persons, Datastream: THEMPTOTP, monthly.

Short-term interest rates Advanced economies

Austria: from 1999 Q1: 3-month Euribor rate, Datastream code: EIBOR3M, daily. Before 1999 Q1: money market rate, IMF IFS line 60b, monthly averages.

Australia: from 1976 Q2: 90-day bank accepted bills rate, BIS code: D:HEJA:AU:32, daily. Before 1976 Q2: money market rate, IMF IFS line 60b, monthly averages.

Belgium: from 1999 Q1: 3-month Euribor rate, Datastream code: EIBOR3M, daily. Before 1999 Q1: money market rate, IMF IFS line 60b, monthly averages.

Canada: 3-month prime corporate paper rate, BIS code: M:HEHA:CA:02, monthly averages.

Denmark: from 1988 Q3: 3-month interbank rate, Datastream code: CIBOR3M, daily. Before 1988 Q3: money market rate, IMF IFS line 60b, monthly averages.

Finland: from 1999 Q1: 3-month Euribor rate, Datastream code: EIBOR3M, daily. Before 1999 Q1: money market rate, IMF IFS line 60b, monthly averages.

France: from 1999 Q1: 3-month Euribor rate, Datastream code: EIBOR3M, daily. Before 1999 Q1: money market rate, IMF IFS line 60b, monthly averages.

Germany: from 1999 Q1: 3-month Euribor rate, Datastream code: EIBOR3M, daily. Before 1999 Q1: money market rate, IMF IFS line 60b, monthly averages.

Ireland: from 1999 Q1: 3-month Euribor rate, Datastream code: EIBOR3M, daily. Before 1999 Q1: 3-month fixed interbank deposit rate, BIS code: M:HEEA:IE:92, monthly averages.

Italy: from 1999 Q1: 3-month Euribor rate, Datastream code: EIBOR3M, daily. Before 1999 Q1: 3-month interbank deposit rate, Datastream code: ITINTER3, monthly averages.

Japan: from 1989 Q4: 3-month Japanese yen Libor rate, Bloomberg code: JY0003M Index, daily. Before 1989 Q4: call money rate, collateralized and unconditional, overnight, BIS code: M:HEBA:JP:92, monthly averages.

The Netherlands: from 1999 Q1: 3-month Euribor rate, Datastream code: EIBOR3M, daily. Before 1999 Q1: interest rate on 3-month cash loans to local authorities, BIS code: M:HEEA:NL:94, monthly averages.

New Zealand: 90-day bank bills rate, BIS code: D:HEHA:NZ:32, daily.

Norway: from 1986 Q1: 3-month NIBOR rate, Datastream code: NWIBK3M, daily. From 1978 Q3 to 1985 Q4: 3- month NIBOR rate, BIS code: M:HEEA:NO:02, monthly averages. Before 1978 Q3: money market rate, IMF IFS line 60b, monthly averages.

Portugal: from 1999 Q1: 3-month Euribor rate, Datastream code: EIBOR3M, daily. Before 1999 Q1: 3-month interbank deposits rate, BIS code: M:HEEA:PT:93, end-of-month.

Spain: from 1999 Q1: 3-month Euribor rate, Datastream code: EIBOR3M, daily. Before 1999 Q1: 3-month interbank rate, Datastream code: ESINTER3, end-of-month.

Sweden: from 1987 Q1: 3-month Stibor rate, BIS code: D:HEEA:SE:02, daily. From 1969 Q1 to 1986 Q4: marginal rate, BIS code: M:HBHA:SE:91, end-of-month.

Switzerland: Tom/next deposit rate, BIS code: M:HEBA:CH:02, monthly averages.

United Kingdom: from 1987 Q1: 3-month British pound sterling Libor rate, Bloomberg code: BP0003M Index, daily. From 1975 Q1 to 1986 Q4, 3-month interbank rate, Datastream code: LDNIB3M, daily. Before 1975 Q1: 3-month interbank sterling deposit rate, BIS code: M:HEEA:GB:01, end-of-month.

United States: 3-month US dollar Libor rate, Datastream code: USINTER3, monthly averages.

Emerging market economies

Bulgaria: from 2003 Q1: 3-month interbank rate, Datastream code: BLIBK3M, daily. From 1997 Q3 to 2002 Q4: interest rate on newly accepted deposits on the interbank money market, over 1 month, BIS code: M:HEEA:BG:75, monthly.

Brazil: 3-month Brazilian real retail certificate of deposit rate, Bloomberg code: BCCDBCE CMPL Currency, daily. China: from 2002 Q1: 3-month interbank rate, Datastream code: CHIB3MO, daily. From 1996 Q1 to 2001 Q4: 90- days national interbank offered rate, CEIC code: 7059101 (CMOAF), monthly.

Croatia: from 2006 Q2: 3-month Zibor rate, Datastream code: CTZIB3M, daily. From 1997 Q1 to 2006 Q1: money market rate, IMF IFS line 60b, monthly averages.

Czech Republic: 3-month Prague interbank offered rate (PRIBOR), Datastream code: CZINTER3, monthly.

Estonia: 3-month Euribor rate, Datastream code: EIBOR3M, daily.

Hong Kong SAR: 3-month interbank rate, Datastream code: HKIBF3M, daily.

Hungary: 3-month interbank rate, Datastream code: HNIBK3M, daily.

Indonesia: 3-month interbank rate, Bloomberg: JIIN3M Index, daily.

Israel: 3-month Tel Aviv interbank rate, Datastream code: ILIBK3M, daily.

Korea: from 1992 Ql: 91-day negotiable security deposit rate, Datastream code: KOCD91D, daily. From 1991 Q1 to 1991 Q4: З-month certificate of deposit rate, BIS code: M:HELA:KR:32, monthly averages. Before 1991 Ql: money market rate, IMF IFS line 60b, monthly averages.

Latvia: З-month Euribor rate, Datastream code: EIBOR3M, daily.

Lithuania: З-month Euribor rate, Datastream code: EIBOR3M, daily.

Morocco: Money market rate, IMF IFS line 60b, monthly averages.

Macedonia: 1-month interbank rate, BIS code: M:HEEA:MK:32, monthly averages.

Mexico: 91-day Mexico CETES rate, Datastream code: MXCTC91, daily.

Malaysia: З-month interbank rate, Datastream code: MYIBK3M, daily.

Peru: Interbank rate, IMF IFS line 60b, monthly averages.

Philippines: 91-day Treasury bill rate, BIS code: M:HEPA:PH:72, monthly averages.

Poland: З-month deposit rate, Datastream code: POIBK3M, daily.

Romania: З-month interbank bid rate, BIS code: D:HEEA:RO:36, daily.

Russia: З1 to 90-day interbank rate, Datastream code: RSIBK90, daily.

Singapore: from 1999 Q3: З-month Singapore dollar Sibor rate, Bloomberg code: SIBF3M Index, daily. Before 1999 Q3: З-month interbank rate, Datastream code: SGIBK3M, daily.

Slovakia: З-month Euribor rate, Datastream code: EIBOR3M, daily.

Slovenia: З-month Euribor rate, Datastream code: EIBOR3M, daily.

South Africa: from 19S1 Ql: 91-day T reasury bill rate, Datastream code: SATBL3M, daily. Before 19S1 Ql: 3-month Treasury bill rate, Datastream code: SAGBILL3, end-of-month.

Thailand: З-month interbank rate, Datastream code: THBRF3M, daily.

4. Short-term interest rates and the US house price bubble

One important reason to focus on the impact of short-term interest rates on house prices is that a number of studies, beginning with Taylor (2008), argue that easy US monetary policy was an important determinant of the 2000-06 US house price bubble, whose bursting contributed to the Great Financial Crisis of 2007-09. It has been suggested that tighter monetary policy in the years leading up to the financial crisis could have reduced US house price growth, perhaps even enough to have averted the worst aspects of the financial crisis. A question that naturally arises is how much higher short-term interest rates would need to have been to have reduced cumulative house price growth substantially.

US real house prices rose strongly over the 2003-05 period, increasing by a total of 35%. The target federal funds rate also rose over this period, from 1%% to 4%%; however, the period did not witness a steady rise in the federal funds target rate, as it actually fell by 25 basis points in 2003. Our conservative model - ie our model estimated with data up through end-1999 (Graph 8) - suggests that the 2003 cut in the federal funds rate target added a bit over 1 percentage point to US real house price growth by the end of 2005 (Graph A4). Over the course of 2004 and 2005, the Fed raised its target for the federal funds rate, in a series of actions, by a total of 325 basis points. However, this did not have a large impact on US real house price growth up through end-2005 because most of the increases came toward the end of the period. Graph A4 indicates that the cumulative effect of changes in the federal funds target rate over the 2003-05 period was to reduce real US real house price growth by about 2 percentage points over this period. In contrast, a 400 basis-point increase in the federal funds rate target at the end of 2002 would have slowed real house price growth by about 20 percentage points through end-2005 (Graph 8).

Higher short-term interest rates by themselves probably would not have significantly dampened the demand for housing and the increase in construction activity that followed the easing of credit constraints to subprime borrowers. However, lower house price growth over 2003-05 would arguably have given financial sector supervisors more time to implement measures such as tighter lending standards to strengthen the US financial system. Thus, it seems that not too large an increase in US short-term rates could have affected the course of events over 2007-09. In fact, even a hike smaller than 400 basis points may have worked, because our model treats GDP growth - which has an additional impact on house prices - as exogenous, ie unaffected by increases in short-term interest rates.

With hindsight, one could argue that the Fed should have tightened monetary policy at the end of 2002. This argument would be stronger if it had already been clear at that time that the United States was in the middle of a national house price bubble threatening financial stability. Based on data available up through 2002, Case and Shiller (2003) found that house prices in seven US states, including California and some other large states, looked increasingly bubble-like. However, the main evidence that led these researchers to conclude that there was indeed a significant bubble in US house prices was survey results for four cities, two of which are in California, showing that homebuyers had very high expectations of future house price growth in 2002. These survey responses were only available in 2003.

References

Adrian, T, E Etula and H Shin (2015): "Risk appetite and exchange rates", Federal Reserve Bank of New York Staff Reports, December.

Bernanke, B and A Blinder (1992): "The federal funds rate and the channels of monetary transmission", American Economic Review, vol 82, pp 901-21.

Bruno, V and H Shin (2015): "Capital flows and the risk-taking channel of monetary policy", Journal of Monetary Economics, vol 71, pp 119-32.

Campbell, J and T Vuolteenaho (2004): "Inflation illusion and stock prices", American Economic Review, vol 94, pp 19-23.

Case, K and R Shiller (1989): "The efficiency of the market for single-family homes", American Economic Review, vol 79, pp 125-37.

Case, K and R Shiller (2003): "Is there a bubble in the housing market?", Brookings Papers on Economic Activity, no 2, pp 299-362.

Cerutti, E, S Claessens and L Ratnovski (2014): "Global liquidity and drivers of cross-border bank flows", IMF Working Papers, no 14/69.

Cesa-Bianchi, A, L Cespedes and A Rebucci (2015): "Global liquidity, house prices, and the macroeconomy: evidence from advanced and emerging economies", Journal of Money, Credit and Banking, vol 47, pp 301-35.

Crowe, C, G Dell'Ariccia, D Igan and P Rabanal (2011): "How to deal with real estate booms: lessons from country experiences", IMF Working Papers, no 11/91.

Del Negro, M and C Otrok (2007): "99 Luftballons: monetary policy and the house price boom across US states", Journal of Monetary Economics, vol 54, pp 1962-85.

Drehmann, M, C Borio and K Tsatsaronis (2012): "Characterising the financial cycle: don't lose sight of the medium term!", BIS Working Papers, no 380, June.

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 (2005): "House prices, borrowing constraints, and monetary policy in the business cycle", American Economic Review, vol 95, pp 739-64.

Iacoviello, M and S Neri (2010): "Housing market spillovers: evidence from an estimated DSGE model", American Economic Journal: Macroeconomics, vol 2, pp 125-64.

Jarodnski, M and F Smets (2008): "House prices and the stance of monetary policy", Federal Reserve Bank of St. Louis Review, July/August.

Jorda, O, M Schularick and A Taylor (2015): "Interest rates and house prices: pill or poison", Federal Reserve Bank of San Francisco, Economic Letter, 3 August.

Kishan, R and T Opiela (2000): "Bank size, bank capital, and the bank lending channel", Journal of Money, Credit and Banking, vol 32, pp 121-41.

Mihaljek, D and A Subelyte (2014): "Do we understand what drives house prices?", in E Nowotny, D Ritzberger-Grunwald and P Backe (eds), Financial cycles and the real economy: lessons for the CESEE Countries, Palgrave, pp 147-70.

Modigliani, F and R Cohn (1979): "Inflation, rational valuation and the market", Financial Analysts Journal, vol 35, pp 24-44.

Rosengren, E (2011): "The role of 'financial myths' in financial crises", speech given at the Boston University conference on "The state of financial reform", Boston, MA, 28 February.

Scatigna, M, R Szemere and K Tsatsaronis (2014): "Residential property price statistics across the globe", BIS Quarterly Review, September.

Shiller, R (2008): The subprime solution, Princeton University Press.

Sutton, G (2002): "Explaining changes in house prices", BIS Quarterly Review, September.