Master’s Thesis
M.Sc. International Economics and Business
MONETARY POLICY TRANSMISSION:
THE EFFECTS OF LONG PERIODS OF LOW POLICY
RATES FOR HOUSING PRICES.
Abstract
Currently policy rates are low across advanced economies. At the same time, housing prices are rising. As highlighted in the literature, the Fed’s expansionary monetary policy is a cause of the credit and housing bubble that resulted in the global financial crisis. This paper examines this and explores the current monetary policy’s effect on housing prices in OECD-countries, using a vector-autoregressive model. The main findings are that policy rates and credit growth Granger cause housing prices. These results are robust across countries when accounting for differences in mortgage markets. Macroprudential policies reduce the effect of policy rates on housing prices.
Key words: policy interest rate, housing prices, monetary policy transmission JEL classification: E32, E52, E58, G21
Acknowledgements
1
1. Introduction
“Financial crises (…) are caused by excesses – frequently monetary excesses – which lead to a boom and an inevitable bust.”
– John B. Taylor (2009, p.1)
The global financial crisis highlighted the consequences asset bubbles can have for the real economy. It also disclosed the interconnectedness of the global financial system. Starting in 2007, it was the largest economic downturn in decades. Problems first arose in the sub-prime mortgage market in the United States and spread quickly to other parts of the financial system. Within months, other sectors in the economy were hit, also outside the U.S. (Agnello and Schuknecht, 2011).
Soon, a discussion about the cause of this misery started. Given the crisis’ origin, the loosening of credit standards in countries with higher levels of financial development were a first target. However, as Crotty (2009) states, this trigger is not the root cause of the crisis. The combination of financial innovation and deregulation since the 1980s allowed for financial booms that never ended well. Even worse, the bailouts during busts promoted the build-up of the next boom, allowing the financial sector to expand rapidly (Crotty, 2009).
Shiller (2012) provides another explanation. He argues that the sub-prime crisis – in essence the bursting of a real-estate bubble – was not ultimately caused by financial innovation. He continues by arguing that other frequently stated causes, such as regulatory inattention and central bank laxity were mere significant parts of the story, but not the central cause.
Recent literature proposes yet another cause of the housing boom: the low interest rates since the collapse of the dotcom bubble in 2000 (McDonald and Stokes, 2013b). The major advocate of this explanation is John B. Taylor. He argues that the low policy interest rate as set by the Federal Reserve in the period 2002-2006 has attributed to the house-price boom (Taylor, 2007; 2009). (See McDonald and Stokes (2013a) for a review of various advocates and opponents of this explanation.)
Taylor (2007, p. 2) states that “the federal funds rate was well below what experience during the previous two decades of good economic macroeconomic performance – the Great Moderation – would have predicted.” The Fed feared deflation, and hence deviated from the rule. However, given the actual inflation rate in the period 2002-2006, Taylor (2009) argues that the central bank should have increased the policy rate sooner. The actual interest rate was well below what the Taylor rule prescribed. Using a counterfactual model, he shows that the Fed’s loose monetary policy accelerated the build-up of the housing boom.
2 negative GDP growth) and be raised when inflation increases. To know precisely by how much the policy rate should change, one can use the formula as shown in equation 1.
! = 1 + 1.5' − 0.5 *+∗− +
+ - (1)
Here, ' is the inflation rate, +∗ is GDP with full employment and + actual GDP (Taylor, 2009;
McDonald and Stokes, 2013a). Taylor (2009) states that the Great Moderation (a period from the 1980s until the global financial crisis in which the US economy saw very low volatilities in GDP, inflation and interest rates) could be contributed for by the more systematic monetary policy approach. Even more striking, he shows, is that when policy rates deviated more from the by the Taylor rule described level, housing booms were larger.
Figure 1 reveals that the actual Federal funds rate is always lower than the Taylor rule advises in the period 2002-2006. This long period of lower interest rates allowed for cheap credit, given a lower cost of borrowing. For example, through the balance sheet channel, a looser monetary policy stance of the central bank lowers interest costs of borrowers, increasing their financial position and thus their creditworthiness, allowing for credit growth (Bernanke and Gertler, 1995).
Figure 1. The actual Federal funds rate against the policy rate as prescribed by the Taylor rule,
1993-2015q2. Source: The Economist (2015).
Taking a second look at the graph above, we see that since 2010, the actual policy rate in the U.S. is again below what the Taylor rule would prescribe. While the Fed when applying this rule directly would have already started raising its short-term interest rate in 2012, it actually started increasing the funds rate in December 2015 to 0.375%, and only recently speeded up these increases. Hence, there again is a long period in which actual rates are lower than the suitable level given the level of inflation and the output gap.
3 globe. In most advanced economies around the world, interest rates have been lower than 2% since early 2009 (see figure 2). Starting in 2017, interest rates have been increasing steadily in some countries such as the U.S. and the U.K., but it remained low in, among others, the Euro area and Japan.
Figure 2. Policy interest rates, various countries, daily data, 1/1/1999 – 19/11/2018. Data source:
Bank for International Settlements (2018a).
Following the analysis of Taylor (2007; 2009), this then implies that again, post-crisis, central banks in advanced economies accommodate the build-up of other credit booms which might translate into new housing booms. When interest rates were (historically) low for only a couple of successive years (2002-2006), it helped fuel one of the largest financial crises in decades. With the even lower interest rates now (2009-2017), sometimes even at the zero lower bound, coupled with the injection of billions into the economy as part of unconventional monetary policies, will the effect on housing and credit markets be the same, or even more intense now than a decade ago?
In this paper, the aim is to assess whether the low policy rates in this century are (i) effective in achieving the desired path and level of inflation and (ii) accelerate the build-up of credit and housing booms, i.e., contribute to asset price inflation. To find evidence for a new housing boom post-crisis, the sample will be split into two periods: 1999-2009 and 2009-2018. If we find that policy rates influence credit markets and housing prices and these results are similar for both periods, we are able to conclude that post-crisis, monetary policies in advanced countries contribute to the boom-bust structure of the credit and housing market. More formally, the research questions that will guide the remainder of this paper are as follows;
- What is the effect of long periods of expansionary monetary policy for the development of asset price
bubbles, such as housing price bubbles?
- Is this effect of similar magnitude, comparing pre-crisis years to post-crisis years?
4 the global financial crisis. The second question delves into possible differences between the periods 1999-2009 and 2009-2018. As we saw in figure 2, interest rates in the latter period are lower for a longer period than in the years before the 2008-crisis. What does this difference hold for the effect on both consumer price inflation and, more importantly, asset price inflation in the housing market?
To answer both questions, I will assess the relationship between the policy rate and housing prices for various advanced economies, all OECD countries. In order to evaluate this relationship, a separate set of regressions will be run with, instead of asset prices, consumer prices as dependent variable. These prices capture normal inflation, or CPI. Comparing these results with the results from the first set of regressions with housing prices provides us with an indication about how much asset prices are affected by interest rates, relative to consumer prices. Empirically, these questions are answered by running a VAR model and a Granger causality analysis. Extending the studies by Taylor (2007) and McDonald and Stokes (2013a, b), more variables are taken into account. As highlighted in the literature (see e.g., Alessi and Detken, 2017), a rising credit-to-GDP ratio is linked to financial crises.
The empirical analysis covers 30 OECD member countries. All of them are advanced economies and are selected based on availability of (reliable) data as well as their higher level of financial development. The latter matters because, for instance, their housing markets are more homogeneous than for a broader sample which also includes emerging economies. Nevertheless, there still exist various cross-country differences, such as different mortgage market institutions. These factors might influence the relationship between policy rates and asset prices and are therefore included in the analysis (see section 4; methodology).
The focus of this paper regarding asset prices is on the housing market, rather than the stock market for various reasons. Firstly, a limited scope allows for a more in-dept analysis. Secondly, as Bordo and Jeanne (2002) find, boom-bust episodes occur more frequently in housing markets than in stock markets. A third reason that supports this choice is the fact that the 2008 global financial crisis had its origin in the housing market (e.g., Taylor, 2007; McDonald and Stokes (2013a, b). Focusing on the housing market allows comparison of the results with these studies.
With this research, I contribute to the literature by bridging the gap between theories that explain how monetary policy affects economic growth and inflation (through various channels of monetary transmission) and theories that explain how monetary policy affects asset prices, with the potential of causing housing price bubbles. Furthermore, most research regarding the role of central bank policy in the development of boom and bust episodes that is conducted so far focuses on the U.S. only. This paper instead looks at other advanced economies as well; and reveals differences and similarities across these countries in the way monetary policy is transmitted.
5 functions indicate that the effect of policy rates on housing prices has become stronger since the global financial crisis. Combined with the observation that in the period 2009-2018, policy interest rates are lower for a longer period than in the decade preceding the crisis, this indicates that long periods of low interest rates are at least partly responsible for appreciating housing prices. This paper also finds Granger causality running from household credit to housing prices, revealing to close connection between credit and housing price bubbles.
In the next section, I provide a review of the relevant literature available in this research area. Section 3 provides an overview and exploration of the data used. In section 4, the empirical model that is employed in this paper is explained. Section 5 covers the empirical results, robustness tests and a discussion thereof. The last section concludes.
2. Literature Review
Monetary policy transmission
The mandates of most central banks in advanced countries state that their objective is to achieve price stability, and thus keep inflation low and stable. Various countries, including Japan and the U.K., have an inflation target of 2%, while the ECB aims to keep inflation below, but close to 2% over the medium term (Berk, 2018). To achieve this objective, central banks employ various instruments. The policy rate is often thought to be the most important one. By adjusting the policy rate, central banks can influence consumer price inflation.
There is a broad literature covering the various transmission channels of monetary policy. Besides the traditional channel, by which a lowering of the policy rate leads to a drop in the real rate causing an increase in investment, consumption and in turn output, the credit view proposes some alternative channels (Berk, 2018). As showed by Bernanke and Gertler (1995), this direct effect on interest rates is amplified by various endogenous channels which influence the external finance premium. For example, a lower short-term interest rate increases the creditworthiness of borrowers, allowing for credit growth. Creditworthiness increases due to (1) higher stock prices increasing the net worth of borrowers, (2) a lower interest rate raising cash flows, (3) unanticipated inflation lowering real debt and thereby increasing the borrowers’ net worth (Berk, 2018).
6 change. Indirectly, household wealth changes allow for transmission through balance sheet and credit channels (Mishkin, 2007).
Empirically, these theories hold. Regarding the relationship between monetary policy and asset prices, Rigobon and Sack (2004) show that an increase in the short-term interest rate negatively impacts stock prices, while it is positively related to market interest rates. Bernanke and Gertler (2000) state that asset price fluctuations can affect the real economy, especially given the boom-bust movements in asset markets. Furthermore, they argue that asset prices matter for firms’ creditworthiness, given the asymmetric information present in credit markets. Hence, asset price volatility can improve (or worsen) leverage. Moreover, they state that monetary policy in the form of inflation targeting “induces policymakers to automatically adjust interest rates in a stabilizing direction” (p. 14). When asset prices increase, interest rates should be raised.
Regardless of this relationship between asset prices and interest rates, Bernanke and Gertler (2000) argue that monetary policy should not be conducted based upon movements in asset prices. Gilchrist and Leahy (2002) agree and argue that, while there are some key arguments mentioned in the literature, they are not profound enough to argue that monetary policy should take into account asset price movements. This reflects the fact that before the crisis, given the widespread adoption of inflation targeting, the mainstream practice among central bankers was to take action when asset price bubbles burst, instead of trying to dampen the build-up in the boom-phase (Dell’Ariccia, Igan, Laeven and Tong, 2016).
After the crisis, the question arose whether this practice by central banks was still desirable. Bordo and Jeanne (2002) use a stylized model to find out whether central banks can be more successful in preventing asset market busts by dampening the booms, or whether cleaning up the mess afterwards is more appropriate. Their results yield that the optimal policy depends strongly on economic conditions and beliefs of various economic actors. Important factors are the chance a bust occurs and the costs of policy intervention. Given this complexity, as the fact that rules are generally blind to the fact that financial instability is endogenous to monetary policy, the Taylor rule is not likely to be an optimal policy rule.
Overall, there is a consensus about the role of asset prices in the determination of monetary policy. As Dell’Ariccia et al. (2016) finds, contractionary monetary policy does not result in fewer booms, the expected negative sign turns out insignificant. Furthermore, one of the dangers of tight monetary policy is an increase in capital inflows, making the monetary policy ineffective. Mishkin (2007) agrees by arguing that using monetary policy to slowdown the build-up of bubbles is tricky. It is difficult to assess real-time whether certain price developments are signs of a true bubble. Moreover, a higher rate can do more damage than good to the overall economy. He favors an approach whereby there is more prudential supervision.
7 2005). Claudio Borio (2014, p. 8) argues that monetary policy regimes that focus on low and stable inflation “can make it less likely that signs of unsustainable economic expansion show up first in rising inflation and more likely that they emerge first as unsustainable increases in credit and asset prices.” He refers to this as the ‘paradox of credibility’. Given this, the process of financial liberalization – which makes booms and busts more likely to drive economic fluctuations – and globalization, Borio (2014) argues that monetary policy can no longer focus only on controlling short-run inflation; it should also take into account financial cycles. He favors a policy that would tighten during a boom phase even if the inflation target is not yet reached, and that eases less sharply during a burst, making it more symmetrically across both phases. Given the major influence of monetary policy on boom-bust episodes, Christiano, Ilut, Motto and Rostagno (2007) also argue that monetary policy should ‘lean against the wind.’
The role of credit
There is a large literature describing the way that finance affects the economy. As noted by Schumpeter (1934), credit is a necessary condition for economic growth. However, more recent research discloses a downside of too much finance; a growing disparity between the growth rates of GDP and credit, causing over-indebtedness of the entire economy (see e.g., Bezemer and Hudson, 2016; Jordá, Schularick and Taylor, 2016; Arcand, Berkes and Panizza, 2015; Beck et al., 2012). Not only does too much credit relative to GDP harm the economy by extracting income (and thus wealth) from the real (producing) economy to repay debts and interests (rents) (Bezemer and Hudson, 2016). The shift in debt, whereby more recently homeowners have a higher share in total credit than non-financial firms, mostly in the form of mortgages, implies that a larger share of credit nowadays is used not to generate income, but solely to buy (existing) assets. This furthermore disentangles the historical one-to-one relationship between credit growth and income growth (Bezemer, Samarina & Zhang, 2017).
As noted by Alessi and Detken (2017), there is broad evidence that a relation between credit growth and financial crises exists. Credit growth is therefore helpful in signaling the risks of a future banking or financial crisis, especially when looking at deviations of the credit-to-GDP ratio from the long-term trend (Aikman, Haldane and Nelson, 2015; Alessi and Detken, 2017).
Monetary policy and housing price inflation – analyzing the global financial crisis
Whether or not monetary policy should change its stance regarding the prevention of booms and busts, and thus include asset price developments (more) in their analyses that determine their monetary policy decisions might depend on the contribution central bank policy had to the development of global financial crisis.
8 empirical test of the effect of monetary policy on an index of housing prices to determine the extent to which monetary policy produced the housing price bubble.” (McDonald and Stokes (2013a, p. 1). Using the Federal funds rate and a housing price index for various U.S. metropolitan areas in a VAR model, they find that pre-crisis the policy rate Granger-causes the housing prices. They conclude that the low interest rate policy of the Federal Reserve in the period 2001-2004 was at least one of the causes of the housing bubble in the U.S.
Both papers of Taylor (2007) and McDonald and Stokes (2013a) however lack the inclusion of other variables that might be of influence. Their results could suffer from omitted variable bias (Miles, 2014). Given this critique, McDonald and Stokes (2013b) published a second paper in which they include the mortgage interest rate. Again, they find Granger causality running from the Federal funds rate to the housing index, while the mortgage rate is not significant in all model specifications.
Focusing on monetary transmission, Sandra Eickmeier and Boris Hofmann (2013) find that monetary policy shocks have a persistent effect on house prices. Moreover, using a factor-augmented vector auto-regressive (FAVAR) model as developed by Bernanke, Boivin and Eliasz (2005), they conclude that monetary policy contributed to the unsustainable developments in housing and credit markets before the crisis, concurring Taylor’s observations. In line with these results are the findings of Cerutti, Dagher and Dell’Ariccia (2017). Their paper shows that bank credit and house price booms are tightly linked. Also, the level of consumer price inflation during booms is often not that different compared to normal times. In other words, credit and asset-price imbalances can grow even if the inflation and output gap are low and stable.
Cerutti et al. (2017) find a statistically significant relationship between the movement of the Federal Funds Rate and the likelihood of a house price boom. i.e., a lowering of this policy rate may trigger the build-up of a housing boom. In accordance, Dell’Ariccia et al. (2016) find that expansionary monetary policies tend to promote credit booms. Just before the global financial crisis, there was an all-time high in the occurrence of credit booms worldwide. They argue that rising credit-to-GDP ratios are a powerful predictor of financial crises, as they indicate overleveraging of large economic sectors.
Disagreeing with these authors, a study by the Federal Reserve Board argues that monetary policy was not the primary reason for the housing boom. In Dokko et al. (2011), the authors note that the link between interest rates and housing markets is not strong enough to account for the enormous increase in house prices. They argue that the looser standards in credit markets have contributed more to the build-up of the crisis.
9 They further argue that the classic characteristics of financial bubbles also apply to the 2008 global financial crisis: initially, a new innovation yields above normal returns. However, over-optimism about future returns fuels over-investment, leading to rising asset prices. Combined with increased leverage and liquidity in the financial system, and the perception that ‘this time it’s different,’ an asset price bubble develops due to sustained excess investment; until the bubble bursts after an event that reassesses asset prices and expectations. In the first decade of this century, these financial innovations were nonprime mortgage-backed securities either packed into collateralized debt obligations or those insured by credit default swaps (Duca et al., 2010).
Looking at the development of housing prices over time and indicators of financial instability, Mishkin (2007) argues that there appears to be a correlation between declining housing prices and increased financial instability. However, given case studies of crises in the 1990s, he does not believe is relationship is causal. Nevertheless, given the rapid deregulation and financial innovation, he argues that at the time of writing the situation in the U.S. could be different.
Just before the global financial crisis, there was a peak in the number of housing booms around the globe (Cerutti et al., 2017). They argue that house price booms, especially those fueled by a credit boom, often end in busts that negatively impact the macro economy. Agnello and Schuknecht (2011) findings align with this statement. Their study shows that both domestic credit and interest rates are key in the developments of booms and ultimately busts. In particular, they find that for the boom to turn into a bust, the short-term interest rate has to increase significantly, and credit needs to decline. Based on their results, Agnello and Schuknecht (2011) argue that policies that slow down both money and credit growth make booms less likely to take place. A study by Christiano et al. (2007) finds that it is difficult to model boom-bust episodes. However, they occur naturally when the researchers introduce two frictions; a central bank that targets inflation and sticky wages. They show that inflation targeting can turn a small economic downturn into a major boom-bust episode.
Heterogeneity in national mortgage markets
Mishkin (2007) argues that the housing market, because of its role in the transmission of monetary policy, has a major effect on economic activity. The global financial crisis disclosed these consequences of fragilities of mortgage markets for financial stability (Stanga, Vlahu and de Haan, 2017). As noted by Calza, Monacelli and Stracca (2013), across advanced economies the structure of mortgage markets differs substantially. Some characteristics are the duration of mortgage contracts and the interest-rate structure. Their VAR-analysis reveals that monetary policy has a stronger effect on consumption and on housing prices in countries with more developed mortgage markets.
10 of mortgages is securitized (such as in the U.S.), lending standards declined more. Their results offer an alternative explanation for the credit boom of 2002-2006. Lower lending standards allowed for a boom in subprime mortgages. Combined with the low interest rates in this period, a credit boom emerged.
Post-crisis, regulation was strengthened among advanced countries. Stanga et al. (2017) examine the role of macroprudential policies in mortgage markets. Since 2009, these policies have become more popular. When, for example LTV ratios are more restrictive, this is associated with a reduction in mortgage defaults (Stanga et al., 2017). Moreover, Akinci and Olumstead-Rumsey (2018) and Carreras, Davis and Piggott (2018) both provide evidence that macroprudential tightening effectively reduces lending in mortgage markets, and thus dampens the booming episodes in credit and housing markets.
3. Data
The main variables
This paper makes use of several datasets from various sources. The core variables – policy rates, housing prices and consumer price inflation –, as well as two types of credit – are gathered from datasets developed and maintained by the Bank for International Settlements (BIS). An overview of all variables, their description and source is provided in appendix 1.
Firstly, the main explanatory variable, the central bank policy rate. The BIS (2018a) provides daily and monthly data, whereby the latter refers to the last available observation of the month. Given the availability of some of the other variables (see below), the final dataset used in this paper comprises of quarterly data. To obtain quarterly data for the policy rate, averages are calculated. For example, the average of end-of-the-month data for January, February and March equals the rate in the first quarter of each year. Data is available for a most of the advanced countries, and except for Japan and Korea, there are no gaps in the data for the period 1/1/1999 – 30/09/2018. This dataset is developed in cooperation with the central banks themselves, who informed the BIS on which interest rate to take as the policy rate (BIS, 2018b). The average policy rate in the chosen period is 2.25%, with a standard deviation of 2.12 (table 1).
11 2004. On average, API exceeds regular inflation. Compared to CPI, the volatility of housing prices is much larger.
The third variable of interest, already addressed above, is consumer price inflation, or CPI. The BIS (2018e) publishes monthly updated data about inflation for 60 countries, gathered from the national statistics offices. Similar to housing prices, inflation is measured as year-on-year changes, in percent, instead of an index. For the calculation of quarterly averages, the method as used for the policy rates is deployed. Data is available for all countries for the entire period. The mean equals 2.01. This indicates that, on average over time and across countries, a medium-term average inflation of around 2% is achieved with the implemented monetary policies.
Table 1. Descriptive statistics.
n Mean St. dev. Median Minimum Maximum
Policy rate 2570 2.253 2.119 2.000 -0.750 18.000 CPI 2605 2.099 2.084 1.934 -6.108 17.699 API (housing price) 2022 2.558 8.711 2.285 -44.860 59.580 Household Credit 1898 69.55 26.13 65.85 6.70 139.40 Bank Credit 1909 100.95 37.11 95.30 23.50 248.40
Source: Bank for International Settlements (2018a, c, e, f), own calculations.
Fourthly, credit statistics are used to allow for a test of monetary transmission channels. According to the credit view, lower policy rates increase the creditworthiness of borrowers for various reasons (higher stock prices or lower net debt increase financial wealth). In turn, this reduces the asymmetric information problem between borrowers and lenders (e.g., banks), allowing for credit growth (and in turn, investment in for example housing, boosting GDP) (Berk, 2018). Therefore, including credit data in the model allows us to see whether a lowering of policy rates indeed allows for a growth in credit provided by banks, and provided to households in later periods.
12 Household and bank credit mostly go in tandem across countries. Their overall correlation coefficient is 0.75. There are however some cross-country differences. In some countries in the sample, the two variables are closely related, and of similar magnitude (as percentage of GDP). Examples are Australia, Canada, the Netherlands, Portugal and the U.K. In the U.S., there is more household credit relative to bank credit. This might indicate that banks play a less important role in providing credit to the public relative to other financial firms. In France, Germany, Spain and Sweden the reverse is true, with bank credit exceeding household credit as percentage of GDP for all years in the period. Likely, this is the case because, besides households, small and middle-sized firms also rely strongly on banks for provision of credit. These observations can be linked to the division between market-based and bank-based financial systems.
Country sample
As stated earlier, this paper will extend the research conducted by McDonald and Stokes (2013a, b) in two ways. First of all, the time period is extended up to a decade after the global financial crisis. This allows for a comparison between the pre- and post-crisis developments of policy rates and the responses of asset price- and consumer price inflation, and credit. Secondly, instead of focusing solely on the U.S., other countries are investigated as well. Their selection started with the list of advanced economies as categorized by the International Monetary Fund (IMF, 2018). In the World Economic Outlook (2018, p. 132), they classify 39 advanced economies; the United States, 19 Euro area countries, Japan, the United Kingdom, Canada and 16 other advanced countries.
From this list, some countries are removed due to a lack of data; for example, Taiwan and Hong Kong. Singapore is excluded because their main monetary policy instrument is the exchange rate, which is managed against a basket of currencies of major trading partners, benefitting the small open country (MAS, 2014). Eventually, the sample consists of 30 countries, listed in appendix 2. All countries are OECD-members.
Most countries in the sample are part of the Euro area, and therefore have a common policy rate determined by the European Central Bank (ECB). This yields that for all these countries, data on the policy rate is only available from 1999q1 and beyond. This marks the starting point of the period investigated in this paper. A benefit of including Euro area countries in the analysis is that boom-bust episodes in housing markets seem to happen relatively more in smaller countries than in larger economies (Bordo and Jeanne, 2002). Furthermore, it will be interesting to find out whether the common short-term interest rate transmits similar across the Eurozone or whether there are cross-country differences.
A first exploration of the data: trends
13 the housing price timeseries have the greatest number of peaks. In most of the countries, we clearly see the building up, and bursting of the housing bubble in 2007-2008. Most housing price timeseries started at a high in 1999 and maintained positive levels until late 2007. Then growth rates of housing prices suddenly dropped and soon turned negative. They stayed below zero for a couple of successive quarters. While in Australia, Canada, Sweden and the U.K. positive growth figures were already seen late 2009, the trouble in the housing market continued for several years in for example France, the Netherlands, Spain and the U.S.
The second remark to be made about the housing prices growth rates – and the main reason this paper concerns the interaction between policy rates and housing prices in the most recent decade – is the high growth of housing prices in the most recent years. In for instance France and the U.K. this growth is quite modest. However, in countries such as Canada, Germany and Sweden, decade record highs have been broken since 2016. In Spain and the Netherlands, after years of negative growth rates, there is now a steep growth in housing prices visible. One that coincides with policy rates at the zero lower bound.
Looking at the CPI, we see that on average, two dips are present. One around 2009, when most countries were in recession and one around 2015-2016. For this latter dip various central banks implemented quantitative easing programs to boost economic growth and prevent the economy from going into deflation. These programs might have had an effect because since 2017, CPI slowly rises in all countries shown in the figure except for Australia. This is also the case for most other countries in the sample.
14 enterprise credit benefits economic growth, credit provided to households (e.g., mortgages) does not. The trend of rising household credit as percentage of GDP is therefore worrisome. It could put pressure on future economic growth and stability.
Figure 3. The policy rate, CPI, housing prices (all percentages, or rates, left axis) and household credit
15 Figure 3 (continued).
4. Methodology
The statistical methodsMost of the papers reviewed above dealing with monetary policy and credit or housing bubbles use VAR models for their empirical analysis. As Dokko et al. (2011) state, the theoretical link between monetary policy and housing and credit bubbles is weak. Furthermore, the relationships between monetary policy, asset prices and financial stability are complex, due to their non-linearity and the occurrence of extreme events (Bordo and Jeanne, 2002). Therefore, including many of the variables that central banks also use to determine their policies will better capture the true relationship. Using a VAR model is thought to be most useful in these cases (see e.g., Bernanke et al., 2005; Dokko et al., 2011).
As defined by Bernanke and Gertler (1995, p. 30), a vector autoregression (VAR) “is a system of ordinary least-squares regressions, in which each of a set of variables is regressed on lagged values of both itself and the other variables in the set.” Using this estimation technique allows one to disclose the dynamic relationships between the various variables.
For the empirics of this paper, I follow McDonald and Stokes (2013a), using both a Granger causality analysis and a vector auto-regression model. While their paper is focused on the U.S. only, with various metropolitan areas separately analyzed, this research paper focuses on various advanced (OECD) countries. One important concern to notice here is that while McDonald and Stokes (2013a) concentrate on one country with one type of mortgage market institutions and regulations, Calza et al. (2013) show that across industrialized countries, the characteristics of these markets differ. This in turn affects the transmission of (similar, or comparable) monetary policy in each country. Therefore, I augment the U.S. study by McDonald and Stokes (2013a) to include some measures that capture these cross-country differences. Among them are data regarding economic growth (e.g., GDP growth and unemployment) as well as some mortgage market features.
16 prices. Given the consequences of the global financial crisis, various countries have implemented more restrictive regulations for the financial sector. To test for the robustness of the results and evaluate the effectiveness of these macroprudential policies, this paper will take into account some of them to assess their effectiveness in securing financial stability.
The vector autoregressive model
To investigate the relationship between the policy rate (./0,2) and the two types of inflation – consumer price inflation (3.40,2) and asset price inflation (5.40,2) –, this paper deploys a similar
method as John F. McDonald and Houston H. Stokes (2013a), which is based on Granger (1969). Focusing on consumer price inflation for the moment (equation 2), ./2 will Granger
cause 3.40,2 if model 3.40,2 = 6 + 7 89:93.4 0,2 ; 9<= + 7 >9:9./ 0,2 ; 9<= + ?2 (2)
has a significantly lower sum of squared errors than model 3.40,2 = 6 + 7 89:93.4
0,2 ;
9<=
+ ?2 (3)
In other words, the model in equation 3 restricts >9 = 0, for A = 1, … , C. : is the lag operator
defined as :9./
0,2 ≡ ./0,2E9. Furthermore, F denotes country, G denotes quarter (i.e., the
time), and 6 is the constant. To ensure there is no autocorrelation in the error term ?2, the
number of lags, C, should be set large enough.
Similarly, equation 4 shows the identical approach for asset price inflation, measured by the housing price index. Here, ./2 will Granger cause 5.40,2 if the model in equation 4 has a significantly lower sum of squared errors than a model where >9 = 0, for A = 1, … , C.
5.40,2 = 6 + 7 89:95.4 0,2 ; 9<= + 7 >9:9./ 0,2 ; 9<= + ?2 (4)
17 Building on equations 2 and 4, equations 5 and 6 respectively show the term H0,2, which
represents other variables of importance. In principle, H0,2 = [J0,2, K0,2], whereby J0,2
represents endogenous variables and K0,2 represents exogenous variables.
3.40,2 = 6 + 7 89:93.4 0,2 ; 9<= + 7 >9:9./ 0,2 ; 9<= + 7 M9:9H 0,2 ; 9<= + ?2 (5) 5.40,2 = 6 + 7 89:95.4 0,2 ; 9<= + 7 >9:9./ 0,2 ; 9<= + 7 N9:9H 0,2 ; 9<= + ?2 (6) Hypotheses
The current monetary policy goal of central banks in advanced countries usually mentions ‘to maintain price stability’ by ‘keeping inflation low and stable’. This is usually defined as an inflation rate of around 2% per year. Central banks raise their main policy rate when the real inflation rate reaches or exceeds its target rate. Since the global financial crisis, the reverse happened in most advanced economies, because inflation rates were low. Therefore, it is expected that there is a negative relationship between the policy rate and CPI. In terms of the coefficients in equation 2, >9 is expected to be negative. Because of the interdependence
between the two variables, causality likely moves in both directions.
Regarding the main relationship of interest in this paper (PR and API), it is expected that a lower policy rate will lead to higher housing prices. This hypothesis is based on the results of articles assessing the causes of the housing boom leading up to the 2008 housing market bust. Given the higher volatility of API relative to CPI, and the literature concerning housing booms, it is predicted the impact of policy rates on housing prices is of higher magnitude relative to the impact of policy rates on consumer price inflation.
One of the endogenous Z variables is the credit-to-GDP ratio. As discussed in the data section, two different types of credit are assessed; household credit and bank credit. A higher level of credit allows for economic growth. However, if demand exceeds supply, more credit will lead to price increases. Therefore, it is likely a higher credit-to-GDP ratio allows for rising housing price inflation. Regarding housing prices, more credit relative to GDP allows for higher bids on the housing market and allows more households to build new houses. As highlighted in the literature (e.g., Cerutti et al., 2017), credit bubbles are associated with housing price bubbles. It is worth noting that policy rates are likely to influence credit growth, given that most lending rates are based on the policy rate to which a risk premium is added. As we saw for the majority of other countries in the sample, pre- and post-crisis low policy rates are associated with rising credit-to-GDP ratios.
18 fall and confidence levels drop across sectors, GDP growth usually declines (Cerutti et al., 2017). Therefore, in the model specification where H0,2 represents GDP growth, M9 and N9 are both
expected to show positive coefficients.
The opposite signs are expected for unemployment ratios. If a higher percentage of workers is unemployed, demand falls, causing dropping prices (holding supply constant). For housing prices, given that credit is provided based upon the creditworthiness of the borrower; lower unemployment increases the wealth of households, and thus allows for more mortgages and thus supports housing price growth.
Another variable included in the extended model is the GDP gap. It captures the economic performance relative to the potential economic performance. A larger output gap implies that the economy is not growing, nor performing, as well as it could be. It is likely this negatively impacts consumer price as well as housing price inflation.
Going beyond the economic variables, housing and mortgage markets usually have different characteristics across countries (see e.g., Cerutti et al., 2017; Calza et al., 2013). Using data collected by Cerutti et al. (2017), four features of mortgage markets are taken into account in the analysis. Firstly, the maximum loan-to-value (LTV) ratio differs across countries. I expect a higher maximum LTV ratio to drive up housing prices, as households will generally borrow more in order to win a bid for a house. The same holds for a longer term to maturity. If borrowers are able to spread out their repayments over a longer period, they are likely to be able to borrow more. There is a positive relationship between the maximum LTV ratio and the term to maturity (see appendix 2).
Thirdly, another feature of mortgage markets is the tax deductibility of interest payments. As Cerutti et al. (2017) notes, tax deduction is possible in two thirds of countries in the sample (see appendix 2). I expect that this variable is positively related to asset price inflation, given that it allows households to borrow more, and thus positively contributes to credit growth. A final feature of mortgage market included in the Z-variables is the type of interest rate: variable, fixed, or a mix. As noted by Stanga et al. (2017), variable interest rates, with more volatile interest payments lead to higher default rates. The less volatile interest payments of fixed rate mortgages are likely to increase lending if interest rates are low at the time the loan in initiated. In periods with low interest rates, fixed mortgage rates are thus expected to boost credit growth. However, unlike variable rates, fixed rates are stickier, they tend to move more slowly with short-term rates than variable rates due to the (long) duration of most mortgages. It could therefore also be that variable rates boost credit and housing prices growth, rather than fixed mortgage interest rates.
5. Empirical Results
19 models. We start by analyzing the relationship between the policy rate (PR) and housing prices (API) for the pre- and post-crisis period (model 1). These periods are defined as 1999q1 – 2009q4 and 2009q4 – 2018q3. The dataset is divided in these two broad periods to allow for enough data points when using multiple lags. The year 2009 is included in both periods because as the financial crisis was spreading globally the timing central banks lowering their policy rates to address the crisis varied.
For comparison with the housing prices, we also analyze the relationship between the PR and the CPI (model 2). Then, more variables are added. Given the influence of credit channels in monetary transmission, household credit and bank credit are added to the model. To capture economic growth, we take into account the GDP growth rate, the unemployment ratio and the GDP gap. Furthermore, to account for liquidity preference, the yield spread is included in one model. Finally, two models capture the mortgage market features as well as macroprudential indicators.
Table 2. VAR models.
Model Core variables Endogenous variables Exogenous variables
1 PR, API 2 PR, CPI
3 PR, API, CPI Household credit, Bank credit 4 PR, API, CPI Household credit, GDP growth,
Unemployment, GDP gap 5 PR, API, CPI Household credit, GDP growth,
Yield spread
6 PR, API, CPI Household credit LTV max, Maturity, Tax deductibility, Interest type 7 PR, API, CPI Household credit Capital buffer, LTV limit,
Capital requirements
Before we estimate the models, it is important to test the variables for stationarity. Stationary VARs allow us to interpret the Granger causality tests as well as the impulse-response functions. Using Im-Pesaran-Shin (2003) panel unit root tests (see appendix 3), suited for unbalanced data, I find that the main variables (PR, API and CPI) are stable, also when a linear time trend is included. Credit data needs modification; taking the first difference makes both household and bank credit data stationary timeseries. All other variables are stationary as well, except for two of the macroprudential policy variables which have limited availability throughout the time period.
20
The core model
To answer the first research question, the empirical analysis starts by looking at the relationship between policy rates and housing prices. This subsection reports multiple Granger causation tests of the model presented in equations (2) and (4). If one variable is found to Granger cause the other, the direction of this influence, as well as the influence over time is analyzed using the provided impulse-response functions.
Table 3 reports the chi-squared statistics and p-values of the panel VAR Granger causality Wald tests for model 1. The null hypothesis states that the excluded variable (the row variable) does not Granger-cause the equation variable (the column variable). The alternative hypothesis states that there is Granger causation in this direction. The pre-crisis (1999q1-2009q4) and post-crisis period (2009q1-2018q3) are analyzed separately, as well as the overall relationship in the past two decades (i.e., the whole period). For the entire period, bidirectional Granger causality is found between the PR and API. In the post-crisis years, there is unidirectional Granger causality from the policy rate to housing prices, as one can reject the null hypothesis with a 5% significance level. For the period 1999-2009, there is Granger causality in the other direction. All periods for model 1 are stable.
Table 3. Granger causality tests for model 1.
1999q1 – 2018q3 1999q1 – 2009q4 2009q1 – 2018q3
PR API PR API PR API
PR 14.029 0.007** 6.150 0.188 13.132 0.011* API 14.611 0.006** 14.976 0.005** 3.119 0.538
Notes: Ho: row variable does not Granger-cause column variable; Ha: row variable Granger-causes column variable. ** p<0.01, * p<0.05. df =4. The first line reports the chi-squared statistic while the second line reports the p-value. All VAR models satisfy the eigenvalue stability condition.
To provide background to the findings of causality between interest rates and asset prices, the same set of VAR models are estimated using the policy rate and CPI. The results are shown in table 4. At 1% significance level, PR Granger causes CPI in the pre-crisis period and at 5% significance level, CPI Granger causes the policy rate. However, the VAR model for this period is not stable. At least one eigenvalue lies outside the unit circle. For the full period and for 2009-2018q3, bidirectional Granger causality is found, at 1% significance level. These periods are both stable, and therefore allow for statistical inference.
21 model. However, more recently and for the entire period, this initial empirical result reveals that the central banks’ monetary policy has a strong influence on price developments in the housing markets of advanced countries.
Table 4. Granger causality tests for model 2.
1999q1 – 2018q3 1999q1 – 2009q4 2009q1 – 2018q3
PR CPI PR CPI PR CPI
PR 28.383 0.000** 15.704 0.003** 19.286 0.001** CPI 13.821 0.008** 12.016 0.017* 38.175 0.000**
Notes: Ho: row variable does not Granger-cause column variable; Ha: row variable Granger-causes column variable. ** p<0.01, * p<0.05. df =4. The first line reports the chi-squared statistic while the second line reports the p-value. The VAR models satisfy the eigenvalue stability condition, except for the pre-crisis period.
Given this evidence in favor of the existence of Granger causality between the three core variables, it is important to take a look at their dynamics. In figure 4, impulse-response functions are provided for model 1. Figure 5 shows similar IRFs for model 2. These IRFs are created using Monte-Carlo confidence intervals, at 95%. They show the impact of one variable on the other for the first eight quarters. Overall, the dynamics are similar in both periods. However, magnitudes differ.
Looking at the effect of a policy shock – an increase in the policy interest rate – on housing prices, we see that pre-crisis it negatively affects housing prices. After four quarters, the decline of the API is roughly 3 times the change in the interest rate. Eight quarters after the policy shock housing prices decline by 7 times the magnitude of the policy shock. Post-crisis, the effect has the same direction. However, the decline in housing prices is stronger; the response function is steeper. Five quarters after the lowering of the policy rate, the API decreases with a factor 6 relative to the magnitude of the change of the interest rate.
These results are in line with those presented by McDonald and Stokes (2013a), who also find a negative relationship for the U.S. This relationship is in accordance with monetary theory that states that a decrease of the interest rate has a positive effect on housing prices (see e.g., Eickmeier and Hofmann, 2013). Combined with the results of the Granger causality tests, it can be concluded from the IRFs that the effect of policy rates on housing prices has become stronger since the global financial crisis. This indicates that long periods of low interest rates are at least partly responsible for increasing housing prices.
Figure 4. Impulse-response functions for model 1. Notes: left: period 1999q1-2009q4; right: period 2009q1-2018q3. Impulse: response. Monte-Carlo 95%
CI reported using Gaussian approximation with 100 iterations.
Figure 5. Impulse-response functions for model 2. Notes: left: period 1999q1-2009q4; right: period 2009q1-2018q3. Impulse: response. Monte-Carlo 95%
CI reported using Gaussian approximation with 100 iterations.
23 post-crisis impulse-response function contradicts this theory. A higher policy rate increases inflation with roughly the same magnitude. The accuracy of this result is not very good, given the spread of the 95% confidence interval.
Overall, these first results are in line with the hypotheses. A lowering of the policy rate raises housing prices. However, it was not expected there is bidirectional Granger causality between these two variables. Although the effect, being significant, is very small. Looking at policy rates and CPI, bidirectional causality was hypothesized. The effect of CPI on PR is however marginal, while policy rates in turn affect inflation in the opposite direction as expected since 2009. To get a better image of these relationships, we extend the VAR model.
Extending the core model
Credit bubbles have often been linked to housing bubbles (see e.g., Cerutti, Dagher and Dell’Ariccia, 2017). Therefore, we extend the core model and estimate the VAR models as presented in equations (5) and (6). In the first specification (model 3), we add household credit (dCreH) and bank credit (dCreB) data to the model. Results are shown in table 5 for both periods. Most previous results still hold. Pre-crisis, the policy rate Granger causes all other variables at either 1% or 5% significance level. Including credit data in the model thus changed the relationship between PR and API in the period 1999-2009, which is now in line with the findings of Taylor (2007) and McDonald and Stokes (2013a, b). Post-crisis, the policy rate Granger causes API with 99.979 significance and CPI with a significance of 99.996.
The influence of policy rates on credit is not visible in the post-crisis period. This could be due to the fact that since the crisis, banks have adopted stronger criteria for borrowers. Moreover, banks themselves now face stronger regulations. The impact of for instance tougher capital requirements is analyzed later in the paper.
Table 5. Granger causality tests for model 3.
1999q1 – 2009q4 2009q1 – 2018q3
PR API CPI dCreH dCreB PR API CPI dCreH dCreB
PR 13.827 0.008** 10.287 0.036* 16.218 0.003** 12.776 0.012* 11.553 0.021* 15.497 0.004** 1.704 0.790 6.588 0.159 API 20.160 0.000** 2.036 0.729 4.221 0.377 3.731 0.444 10.556 0.032* 12.140 0.016* 3.479 0.481 3.624 0.459 CPI 19.279 0.001** 5.025 0.285 10.341 0.035* 6.988 0.137 8.784 0.067 4.426 0.351 0.368 0.985 4.636 0.327 dCreH 8.123 0.087 6.287 0.179 3.422 0.490 0.007** 14.078 9.024 0.061 15.357 0.004** 0.288 0.991 32.791 0.000** dCreB 3.233 0.520 10.328 0.035* 5.422 0.247 8.326 0.080 5.875 0.209 5.876 0.209 3.865 0.425 5.206 0.267
24 Another difference between the first and second period is the relation between credit and housing prices. Prior to the global financial crisis, bank credit significantly Granger causes the API. After the crisis, this is no longer true. Instead, household credit Granger causes housing prices with a significance of 99.996. dCreH also Granger causes dCreB in both periods. This causality is unidirectional and could indicate that across countries most bank credit is directed towards households. The finding that bank credit no longer plays a significant role is likely a consequence of the already mentioned tougher regulations, stemming from the Lucas critique. Borrowers might shift to lenders other than banks when policy makers increase bank lending regulations (Alessi and Detken, 2017). However, it might also indicate that post-crisis banks have lend relative more credit to parts of the private, non-financial sector, other than households’ mortgages.
These test results for household credit are supported by the credit view. If households increase their credit demand, banks extend more loans. In turn these loans (mortgages) are used to invest in housing, which boosts GDP. Looking at the impulse-response functions for both periods (figure 6 and 7), the effect of household credit on housing prices is positive only for the first six quarters, thereafter it decreases to below zero. The pre-crisis causation effect of bank credit on housing prices is negative. This contradicts the hypothesis. An increase in bank credit results in a small but significant decline of housing prices that increases over time. A possible reason for this finding is the fact that while housing prices were already falling in some countries since 2006 (for example, France and the U.S), the credit bubble was still growing.
Figure 6. Impulse-response functions for model 3, period 1999q1-2009q4. Notes: impulse: response.
Monte-Carlo 95% CI reported using Gaussian approximation with 100 iterations.
25 Figure 7. Impulse-response functions for model 3, period 2009q1-2018q3. Notes: Impulse: response.
Monte-Carlo 95% CI reported using Gaussian approximation with 100 iterations.
A policy rate shock has similar effects on the API and CPI as in model 1 and 2. An increase of the interest rate results in a substantial drop of housing prices. Compared to the impact in model 1, the pre-crisis impulse-response function is steeper while post-crisis it is initially less steep. After four periods, housing prices decline by 3 times the magnitude of the policy shock in 1999-2009, and by 5 times in 2009-2018q3. Taking into account credit growth therefore implies that the impact of policy rates on API is more stable over time. As expected, credit growth itself is negatively impacted by a change of the policy rate. A policy shock that lowers the interest rate leads to a small decline in loans extended by banks and provided to households. This effect is, as stated before, only significant before 2009.
Having established a role for credit in the determination of housing prices, it is important to check whether changes in credit (as well as changes in the policy rates and two types of inflation) are influenced by other real economic variables. Therefore, model 4 includes GDP growth, the unemployment ratio and the GDP gap into the VAR. To save space, the Granger results for each period are provided in Appendix 4, tables A4.1-3. The impulse-response functions for the pre- and post-crisis period are shown in figures 8 and 9.
Extending the model has increased the significance of the Granger causality running from the policy rate to both the API and CPI, for both periods. This confirms our earlier results. Moreover, the policy rate seems to influence household credit over the entire time frame (table A4.1). A decrease of the interest rate results in a higher amount of credit being extended to households in the first two quarters after the policy shock. This can be explained using
26 household liquidity effects. The lower rate generally increases stock prices, which increases consumers’ financial wealth. In turn, this boosts their creditworthiness and allows households to borrow more (Berk, 2018). The credit view also explains that household credit Granger causes housing prices, at 5% significance level in both periods.
The impulse-response functions show that household credit has a similar effect on housing prices both before (figure 8) and after the crisis (figure 9). Initially, up to four quarters after an increase in credit extended, the effect on housing prices is positive. In later quarters, this effect is reduced. However, given the width of the confidence interval, it is unclear in which direction housing prices change as a result of a change in the credit-to-GDP ratio.
The relation between policy rates and housing prices is more stable across the models. While both periods’ impulse-response functions show that an increase in the interest rate lowers housing prices, the post-crisis effect is initially of higher magnitude. Four quarters after the policy shock, housing prices decline by five times the change in the interest rate. Before the crisis, this was only two times the change in the interest rate.
Turning to the role of the newly added variables in the model, we see that GDP growth Granger causes the unemployment rate and the GDP gap in the pre-crisis period. After 2009, only the causality running towards the GDP gap is still significant. Furthermore, from 2009 onwards, GDP growth explains changes in inflation and household credit. This effect is relatively small, but positive (figure 9).
In the latter period, unemployment Granger causes the API with a significance of 99.984. The impulse-response function reveals that an increase in the percentage of workers unemployed lowers housing prices. Rising unemployment changes expectations about economic growth, which could lead to lower investment in housing. Although not visible in the individual periods, there is bidirectional Granger causality between the unemployment rate and housing prices in the overall sample period. This result supports the findings by Cerutti et al. (2017) and Agnello and Schuknecht (2011) that housing price busts have crucial (negative) consequences for the macroeconomy. In 2008, the housing bust led to recessions in various countries, increasing unemployment ratios.
A final observation is that while prior to the crisis, causality running from policy rates to GDP growth is highly significant, this is not the case post-crisis. There are multiple explanations for this result. Firstly, only a few of the countries in the sample legally require their central bank to aid economic growth and thereby promote maximum employment. Examples are the U.S. and Australia. Secondly, since the global financial crisis, the recessions in e.g., most European countries more pronounced and continued for more time than expected.
Figure 8. Impulse-response functions for model 4, period 1999q1-2009q4. Notes: impulse: response. Monte-Carlo 95% CI reported using Gaussian
approximation with 100 iterations. Impulse functions of unemployment omitted to increase visibility of other impulse-response functions.
Figure 9. Impulse-response functions for model 4, period 2009q1-2018q3. Notes: impulse: response. Monte-Carlo 95% CI reported using Gaussian
approximation with 100 iterations.
29
Robustness checks
Accounting for liquidity preference: the yield spread
In post-Keynesian economic literature, it is argued that short-term interest rates are set by central banks, while liquidity preference determines the mark-up of long-term rates (Dow and Dow, 1989). If the real sector increases its liquidity preference, for instance because of increased financial uncertainty, it will demand more liquid assets (Lavoie, 2014). In theory, this will cause a decline in investment in illiquid assets, such as housing. Furthermore, because of liquidity preference, it could be that there is no relation between interest rates and inflation. Regardless of the price of credit (the interest rate), if households for some reason have a higher liquidity preference, they will demand less credit, and spend less in the real or financial economy which puts downward pressure on inflation.
To test for these theories, a proxy for liquidity preference of the real economy is introduced: the yield spread. This variable is added to a VAR model including the core variables, household credit and GDP growth (model 5 in table 2). In this paper the yield spread is defined as the difference between the short-term (policy) rate and the long-term rate of 10-year government bonds. Although this long-term interest rate might not fully capture liquidity preferences of the real economy, they do proxy for country-cross differences in general confidence of investors. Granger causality test results per period are again available in appendix 4, table A4.4 and A4.5. Impulse-response functions are given in figures 10 and 11.
The relation between the PR and CPI as estimated in the above VAR models remains intact. With a p-value of 0.010 in the pre-crisis period and 0.002 post-crisis, the policy rate Granger causes CPI. In turn, inflation only significantly Granger causes the policy rate prior to the financial crisis. A similar result as for model 4.
The Granger causality running from the policy rate to housing prices is no longer statistically significant between 1999 and 2009, but it remains significant at 5% between 2009 and 2018. The same holds for household credit Granger causing housing prices. The direction of all the effects, as provided in the impulse-response functions, are of equal sign and magnitude as in models 3 and 4. Combined, these results disclose that accounting for the yield spread between long- and short-term interest rates shows that the findings for the post-crisis period are robust. However, the causation running from interest rates and credit towards housing prices is less robust during 1999-2009.
30 Figure 10. Impulse-response functions for model 5, period 1999q1-2009q4. Notes: impulse: response.
Monte-Carlo 95% CI reported using Gaussian approximation with 100 iterations.
Figure 11. Impulse-response functions for model 5, period 2009q1-2018q3. Notes: impulse: response.
Monte-Carlo 95% CI reported using Gaussian approximation with 100 iterations.
31 Finally, the relationship between PR and the yield spread is not stable over time. The former Granger causes the latter in both periods with a significance of 100 pre-crisis and 99.910 crisis. Pre-crisis a more contractionary monetary policy lowers the yield spread but post-crisis such a policy change leads to a higher yield spread. This could indicate that liquidity preference is higher today than it was before the financial crisis. A plausible reason for this argument is that since the crisis, the public became aware of risks they did not know or take into account previously. Lenders or investors might require higher interest rates now, even if policy rates are lower now than in the pre-crisis period. Overall, the previous estimated results still hold. There is a role for the yield spread in influencing the policy rate and two types of inflation, but these relationships are not stable over time.
Accounting for cross-country differences: mortgage market features
So far, the analysis has not taken into account the effect of cross-country differences in mortgage market features. However, as stated before these features might influence the behavior of both borrowers and extenders of credit, and therefore impact the development of the housing price index (see e.g., Cerutti et al., 2017; Calza et al., 2013). To test the behavior of the core variables and household credit when restricting the sample to capture certain characteristics of national mortgage markets, we estimate model 6. These characteristics, as collected by Cerutti et al. (2017), are (1) the maximum observed loan-to-value ratio, (2) the term to maturity of a mortgage, (3) tax deductibility of interest payments, and (4) the interest type of the mortgage rate, being either variable, fixed or a mix of both. The data per country is provided in appendix 2. Granger causality tests for model 6 are given in appendix 4, tables A4.6-9.
To test for the effect of LTV-ratios, the sample is split in half; countries with a maximum LTV-ratio of below, or above 91% – the average ratio. Table A4.6 shows that in countries with a higher maximum LTV-ratio, in both periods the policy rate significantly Granger causes housing prices, while for the other group this is not significant. In 1999-2009, the causation from the interest rate to household credit is also significant. This supports the view that in countries with higher LTV-ratios, loose monetary policy can lead to credit- and housing price bubbles. Moreover, the result is in line with the hypothesis.
A similar test is conducted for the maturity of mortgage loans. Here, the sample is divided in countries with a maturity below or above 24 years (the median is 25). The Granger causality tests show that in countries with shorter maturities, credit significantly Granger causes the API (table A4.7). This is surprising and not expected, given that on average higher maturities are associated with higher LTV-ratios (see figure 4). The results regarding the effect of policy rates on housing prices are inconclusive, given that regardless of the term of maturity, Granger causality from the former to the latter variable is only significant in the first period.
32 causality remains significant at 5%. These results indicate that regardless of the tax deductibility, in the period 1999-2009 the causation running from the policy rate to housing prices is robust. After the crisis, the lack of tax deduction of interest payments reduces the impact of low policy rates on housing prices.
Finally, the interest type of mortgages matters. Results are provided in table A4.9 for the period 1999-2018q3. If mortgage interest rates are variable, which is the case of half of the countries in the sample, the policy rate significantly Granger causes housing prices. For a mix of variable and fixed interest rates (in six countries), the p-value is 0.082, weakly significant at 10%-level. In the third case, fixed interest rates, there is no significant causal relationship between PR and API.
Similar results are found for the Granger causality running from household credit to housing prices. For variable and mixed interest rates, this causality is significant (respective p-values are 0.000 and 0.014). With fixed rates, there is no significant causality. Therefore, it seems that fixed interest rates are less associated with credit and housing bubbles than variable rates. This finding confirms the idea that variable rates, which follow short-term rates more closely than fixed rates, tend to be closer related to credit and housing booms than fixed rates. In summary, I find that high LTV ratios, tax deductibility and variable mortgage rates are strong predictors of the likelihood of a credit and housing boom happening in a given country. Accounting for cross-country differences: macroprudential policies
Since the global financial crisis, there has been increased interest in the usage and effectiveness of macroprudential policies. In advanced economies, they are often targeted at the housing sector (Akinci and Olmstead-Rumsey, 2018), and could restrain the build-up of credit bubbles (Aikman et al., 2015). They aim at containing systemic risks in the financial system, by (1) increasing the resilience of both individual firms and the system as a whole to shocks and (2) limiting the build-up of vulnerabilities over time (Cerutti, Correa, Fiorentino and Segalla, 2017). Using the database developed as part of the International Research Banking Network we assess the effectiveness of some of the macroprudential policies in containing credit and housing bubbles. The three policies, in order of discussion, are (1) real estate credit capital buffers, (2) loan-to-value limits and (3) capital requirements. Capital buffers require banks to finance a larger part of their credit exposures with capital. The LTV-limits restrict the amount households can borrow to the value of their house. Capital requirements refer to the implementation of the Basel agreements (Cerutti, Correa, Fiorentiono and Segalla, 2017).
