The Effects of Monetary Policy on the Dynamics of Housing Price Cycles in the United States, 1950-2016

Abstract

Housing cycles are a barometer of the macroeconomic situation in the United States, and their supply-side and demand-side dynamics have been discussed in detail in the existing literature. This paper proposes a long-run approach with an emphasis on the demand side and its policy determinants. It traces the empirical connections between the Effective Federal Funds Rate (EFFR), one of the main monetary policy tools used by the Federal Reserve, and the movements of housing prices in the United States from 1950 to 2016, as captured by the S&P Case-Shiller Home Price Index and influenced by other macroeconomic cyclical variables. The paper builds an empirical, long-run model which incorporates the propositions that the movements of the EFFR exhibit strong lagged correlation with housing market cycles in the United States. It also demonstrates that the influence of interest rate shocks on housing markets has been transmitted with a lag of between one and four years and that this lag has been changing as the underlying structure of the US financial system has altered. Finally, the paper proposes a combined approach of monetary policy intervention and macro-prudential regulation to moderate the duration and amplitude of housing cycles and weaken the interdependencies between them and the wider financial system.

Contents

Chapter                                                          Page

1. Introduction…………………………………….5
2. Theoretical frameworks………………………..7
3. Existing literature……………………………..10
4. Empirical discussion……………………………13

1. Empirical process and 1950-2016 VECM model……..13

2. Sub-period VECM models…………………………….18

3. Empirical conclusions…………………………………20

5. Policy implications……………………………23

1. Monetary policy and asset-market regulation………….23

2. The subprime crisis from a policy perspective…………23

3. Towards a combined monetary-regulatory approach….26

6. Conclusions……………………………………31
7. Bibliography………………………………….32
8. Appendix A – VECM specifications………….35
9. Appendix B – Data sources…………………..39

1. Introduction

US housing markets experienced a number of cyclical movements from 1950 to 2016. The last cycle ended in 2007 with the subprime mortgage crisis, causing the deepest recession since the 1930s and substantially affecting the global economy. Identifying the reasons behind US housing-price dynamics and developing empirical models to capture relationships between policy variables and outcomes is key to moderating the macroeconomic effects of housing cycles. Monetary policy is an important determinant of credit conditions underpinning mortgage lending, which supports demand for housing. Understanding the effects of monetary policy on the duration and amplitude of housing cycles, as well as the interaction between monetary policy and financial regulation, is crucial to avoiding a repetition of the subprime crisis.

Central banks hold the monopoly on ‘producing’ money and setting the basic interest rate in an economy, making their influence on credit and capital markets considerable. Asset markets, including housing markets, are significantly affected by credit conditions as many of their transactions are debt-financed. Housing markets are a class of asset markets with singular characteristics. Supply tends to be inelastic, as land supply is usually fixed and urban land needs to be redeveloped for housing. There exists a slow feedback mechanism between market conditions and supply due to the physical time required for construction and frictions in the user market, investment market, and developer market (these frictions are captured in detail in the four-quadrant model of real estate markets proposed by diPasquale & Wheaton, 1992, and by Keogh, 1994, for the UK). Housing markets have high importance in the US political economy, exemplified by the ideas of the home-owning democracy and the ‘ownership society’. Housing is subject to policy intervention and planning restrictions by local, state and federal government. There is high dependence on debt financing both on the supply side (for developers) and on the demand side (for homebuyers) because of the high land, building, and maintenance costs. Debt financing usually comes in the form of mortgage loans that employ the property as collateral. Mortgage financing conditions depend on the lender’s assessment of present and future economic conditions. Factors include: the current value of the property based on the situation in local and national housing markets; the current income status of the creditor (determined by their employment situation in the case of homebuyers and by the cash flow situation in the case of developers); the future value of the collateral; and the future income prospects of the creditor (likelihood of homeowner making loan payments on time for a homebuyer, or the ability of a developer to sell units under construction).

The purpose of this paper is to measure the long-run empirical connections between monetary policy actions taken by the Federal Reserve as represented by changes in the average quarterly Effective Federal Funds Rate and the cyclical movements of the composite S&P CoreLogic Case-Shiller Home Prices Index (an aggregate, indexed measure of the movement of US house prices, with the 1950 level taken to be 100). This effect will be estimated whilst considering other business-cycle variables that might influence the movements of the Case-Shiller Index, including Average Quarterly Unemployment levels in the US, year-on-year quarterly real GDP growth, the Average Quarterly Consumer Sentiment Index published by the University of Michigan, and the Total Dwellings and Residential Buildings Started for the United States Quarterly Index (RBI). The paper will also discuss the evolving understanding in the existing literature of the complex interactions between monetary policy and housing markets, and compare its findings on the magnitude and lag of monetary policy action on housing prices to the results of similar studies.

2.       Theoretical frameworks

The effects of monetary policy on housing-price cycles are only one facet of the complex relationship between monetary conditions and financial stability. Standard economic theory suggests that monetary policy decisions affect housing prices through their impact on credit conditions in the macroeconomy, given that housing markets are highly sensitive to credit conditions. The impact of monetary policy on housing markets is transmitted through the influence of monetary policy on mortgage rates for house-buyers (affecting demand for housing) and on loan rates for house developers (influencing supply for housing).

In the United States, the Federal Reserve influences credit markets through its open-market operations, which keep the Effective Federal Funds Rate (EFFR) within the target limits set by the Federal Open Market Committee (FOMC). In theory, lowering the EFFR leads to cheaper credit, making housing more affordable. This in turn causes demand for housing to rise, which leads to higher housing prices in the short run, all other conditions being equal and supply being inelastic in the short run and adjusting slowly to higher demand in the long run. Alternatively, acting to increase EFFR leads to more expensive credit and makes homes less affordable, causing lower demand for housing and hence lower prices under the two conditions stated above. In the long run, lower rates should stimulate builders, thereby bringing more supply and reducing prices, and higher rates should do the opposite, taking out supply and raising prices. The relative strength of interest rates pulling demand and pushing supply depends on the relative elasticity of supply and demand with respect to changing interest rates.

It has long been argued in the theoretical literature that housing and land markets are primarily demand-driven due to supply constraints. The classical treatment of the subject (Ricardo, 1821), as well as modern models such as the bid-rent theory of Alonso (1964) give rise to surplus theory, which explains rents and prices as based on demand proxied by the income-generating abilities of the property. Following this theoretical argument, the general expectation should be that higher rates lead to lower house prices, and lower rates should lead to higher prices, an effect that is slightly dampened by the opposite, slower dynamics on the supply side. The latter takes much longer to adjust to changed credit conditions due to the specificities of housing markets discussed in Part I of this paper.

The theoretical approach of this paper to the US housing market follows the model developed by Arestis and Karakitsos (2007), which ascribes greatest importance to the mortgage rate and to real disposable income (in that order) as drivers of housing prices in the United States at the homogeneous, national level. The authors emphasise treating the US housing market as a homogeneous structure rather than as a compendium of regional markets. This model expresses demand for housing as a function of two short-run variables (real disposable income and the mortgage rate), and two long-run factors (the debt service burden and the net real estate of households). Supply for housing is a function of housing prices, the level of housing starts, and the level of real residential investment. In the long run, housing prices are a function of the real disposable income, the mortgage rate, the net real estate of households, the level of housing starts, and the level of real residential investment. The authors stipulate that the level of real disposable income, the debt service burden and the net real estate of households have a positive effect on housing prices, whilst the mortgage rate affects housing prices negatively. The mortgage rate is itself a function of the yield on 30-year Treasuries, which has a positive effect on the mortgage rate.

According to Kuttner (2012), there are two main theoretical views on how monetary policy should respond to fluctuating asset prices: the Bernanke-Gertler (1999) view that monetary policy should respond to the macroeconomic effects of asset-price movements rather than to asset prices themselves, and the view that monetary policy is itself a primary cause of asset-price cycles and should be calibrated to moderate them. The latter view is supported by Taylor (2007, 2009) and urges extreme caution in the design and implementation of monetary policy, especially as a countercyclical tool that could have unwanted effects on asset markets and the financial system, as some scholars believe happened in 2004-2008.

Building on the two theoretical approaches above, this paper would like to propose a long-run relationship between housing prices and monetary policy. This is expressed by the nationwide Case-Shiller index as a function of the average quarterly Effective Federal Funds Rate (targeted by the Federal Reserve), whilst taking into account a number of other explanatory variables such as the year-on-year quarterly GDP growth, average quarterly unemployment levels, the average quarterly Consumer Sentiment Index (as compiled by the University of Michigan), and the Total Quarterly Dwellings and Residential Building Starts Index for the United States. The long-run value of the Case-Shiller index will be modelled as:

1. Y=YR, GDP, UNEMP, CSENT, RESB        –      +            –               +             –

The empirical analysis will attempt to capture long-run relationships in a ‘theory-light’ way, as proposed by Sims (1980), which seeks to impose as few structural restrictions on the model as possible. Moreover, it will seek to bridge the gap between a policy variable such as the Effective Federal Funds Rate and a nationwide price index of a very complex market that is being treated homogeneously (following Arestis and Karakitsos, 2007). The inclusion of classic macroeconomic indicators as explanatory variables may or may not be helpful in explaining this long-run relationship. The empirical analysis stemming from this theoretical construction should help the paper avoid the “incredible identification restrictions” imposed by structural models, as described by Sims (1980).

The paper will empirically test three main hypotheses. The first is that the monetary policy of the Federal Reserve has been strongly correlated with the cyclical movements of housing prices on the national aggregate level in the US, as captured by the S&P Case-Shiller CoreLogic Home Prices Index in the period 1950-2016. Secondly, interest rates have a cyclical pattern of acting on housing markets, with their influence peaking a certain time after they are changed and then receding. Finally, the response lag at which house prices respond most vigorously to interest rate shocks has been changing over the decades between 1950 and 2016 and could be inferred from the underlying data.

The use of the selected explanatory variables is designed to capture the supply- and the demand-side dynamics of housing markets, as interest rates are likely to affect both sides. The effects of changes in the EFFR on the supply side are captured by the movements of the Residential Buildings Starts index with a suitable lag that allows for developers to adjust to new credit conditions. GDP growth, the Consumer Sentiment Index, and unemployment levels reflect macroeconomic conditions and proxy demand for housing in the economy.

The reason for selecting the period of 1950 to 2016 is that this period is the longest that is useful for informing current monetary policy debates. This is the period in which the Federal Reserve has existed with its present legal powers and independence. The Federal Reserve was founded as a decentralised central bank in 1913 in response to several financial panics in the early twentieth century. In the first half of the century, the Federal Reserve faced three major monetary abnormalities – the First World War, the Great Depression, and the Second World War – and was forced to respond to those emergencies with measures that were not independent of fiscal policy. In 1951, the Treasury Accord was signed, which restored the independence of the Federal Reserve by eliminating its obligation to monetise the debt of the Treasury at a fixed rate. This became essential to the independence of central banking in the US and to how monetary policy is pursued by the Federal Reserve today.

3. Existing literature

The cyclical behaviour and interdependence of monetary policy and housing markets have been researched and documented extensively.  Claessens et al (2011) show that housing markets exhibit strong pro-cyclicality, suggesting that they peak at points of the business cycle when rates are low and GDP is growing fast, in their analysis of 21 advanced economies over the period 1960-2007. Moreover, financial and housing cycles are highly synchronised across countries, as evidenced by Hirata et al (2013). Ahearne et al (2005), in a study of 18 major industrial countries between 1970 and 2005, discover that housing booms tend to be preceded by periods of monetary easing, with a lag of between one and three years. They note that central bankers have generally been unwilling to intervene to stem rises in asset prices and have followed the Bernanke-Gertler (1999) consensus regarding the relationship between monetary policy and asset markets.  However, their empirical analysis does not purport to trace a causal connection between housing-price movements and interest rates and is merely suggestive of a more complex relationship. Campbell et al (2009) sought to decompose house price movements for 23 metropolitan areas in the United States into contributing factors pertaining to real interest rates, rents, and risk premia. The authors found that risk premia are the main contributor to housing-price fluctuations in their model, and changing interest rates has negligible effects on housing-price dynamics. Dokko et al (2009) provided empirical backing for Taylor’s (2007) assertion that over-expansionary monetary policy caused the subprime housing boom by using VAR forecasting to predict hypothetical house prices under different monetary policy regimes. They found little evidence that overshooting the ‘Taylor rule’ (the point beyond which further monetary easing will cause the creation of a bubble) could satisfactorily explain the subprime boom and advocated a Bernanke-Gertler approach to monetary policy coupled with macro-prudential regulation of mortgage lending. Reinhardt and Reinhardt (2011), analysing historical housing prices and real interest rates, conclude that a ‘moderately’ different monetary policy strategy could not have prevented the housing bubble, and advocate the view that monetary policy has no long-term effects on real interest rates or housing prices, placing the emphasis on global capital flows as the main driver behind credit conditions and housing prices. Kuttner (2012) summarises the empirical findings of four studies on the dynamic effects of a 25-basis-point monetary policy shock on housing prices as displayed in Table (1):

Table 1

Analysing the existing literature on the subject should also consider the ideological pendulum swings of twentieth-century economic thought, which is key to properly understanding adopted policy tools and the zeitgeist of economic literature on the topic. The periodical division this paper proposes consists of three main periods, which delineate the shifting paradigm on the role of housing markets in the US macroeconomy and their responsiveness to monetary policy interventions. The first period is the classic, Keynesian understanding of the subject, characteristic of the literature in the years 1950-1970.  Examples include Klaman (1956) and Naylor (1967). The second period is the era of the ‘Great Moderation’ of the years 1980-2007, heavily influenced by the decline of orthodox Keynesian thinking and the rise of monetarism and deregulation as topics of intense research interest.  The ‘manifesto’ of this new wave of economic thinking is Friedman and Schwartz’s A Monetary History of the United States (1963), which treats financial malfunctions, most importantly the Great Depression, as failures of monetary policy rather than as market imperfections to be corrected through regulation.  This school of economic thought and its relationship to the problem of monetary policy’s effects on housing markets are discussed at length by Goodhart and Hoffman (2000), Bernanke and Gertler (1999), and Cecchetti et al (2000). The transition between this period in the development of economic thought on the subject and the emerging post-crisis consensus (2008-2016) began after the bursting of the subprime bubble in 2006. Some early attempts at explanation and reconciliation between the theory and practice of monetary policy conduct and asset market instability were proposed by Mishkin (2007) and Taylor (2007) at the Jackson Hole Economic Policy Symposium in 2007, as well as by Del Negro and Otrok (2007).  As the financial crisis ran its course and the Federal Reserve reset its cycle of monetary tightening to support the flailing American economy, a new synthesis of the monetarist deregulatory understanding and the older, regulation-minded, Keynesian thinking developed. Some representatives of this emerging approach are Glaeser, Gottlieb, and Gyourko (2010), Allen and Rogoff (2011), and Williams (2015). This paper adheres to the consensual approach proposed by these authors, and whilst espousing the importance of monetary policy in regulating cyclical movements in housing markets, it also recommends the active implementation of regulatory mechanisms to achieve a more ‘peaceful’ coexistence of monetary policy interventions and stability in housing markets.

4. Empirical discussion

1. Empirical process and 1950-2016 VECM model

The modelling process commenced with a linear regression equation of only rates and housing prices of the type:

2. Yt= α+ βRt-n+ ϵ,

where Y_t is the value of the Case-Shiller in quarter t, R is the interest rate in period t-n, and

ϵis an error term. Performing an ordinary-least-squares regression on this equation produced a low Adjusted-R² of only 0.152, which suggested that the inclusion of the other business-cycle variables might be able to improve the model’s fit. Including the variables (average quarterly unemployment, quarterly GDP growth, consumer sentiment, and the residential building starts index) in the regression equation as follows:

(3)

Yt= α+ βRt-n+ γGDPt+ δUNEMPt+εCSENTt+θRESBt-4+ ϵ

raised Adjusted-R² to 0.31, but unemployment and consumer sentiment were insignificant. Ordinary-least-squares analysis suggested that interest rates and GDP growth are negatively correlated with the Case-Shiller index, while the Residential Building Starts index is positively correlated with the Case-Shiller index when lagged four quarters. This matched the average reported time to complete a house in the US, estimated at one year (Shiller, 2008). Interest rates are more strongly correlated with the C-S index when lagged, suggesting the peak response of house prices to interest rates shocks occurs with a delay. Removing the insignificant explanatory variables and reducing the regression equation to the form

(4)

Yt= α+ βRt-n+ γGDPt+θRESBt-4+ ϵ

produced results (by the ordinary-least-squares method) which suggested that all three remaining variables were significant at the 99% confidence level, with Adjusted-R² of 0.318.

All variables in the model (the housing-price index, interest rates, GDP growth, residential building starts index) are time series and are likely to exhibit non-stationarity, which would bias ordinary-least-squares model estimates as it makes the method prone to reporting spurious regression (Granger and Newbold, 1974). Indeed, the reported Adjusted-R² of 0.318 seems to be unrealistically large for such a complex, long-run, multivariate econometric relationship. This proposition was confirmed by the Augmented Dickey-Fuller test, which confirmed the existence of a unit root for three of the time series (the EFFR, the Case-Shiller index, and the Residential Building Starts index) and demonstrated that these three time series are non-stationary in its levels I(0), but are stationary in their differences I(1), at the 95% confidence interval, with four lagged differences to account for any serial correlation of the error term in the Dickey-Fuller regression. The year-on-year Quarterly GDP Growth series is I(0) stationary under the same conditions.

In the case of non-stationary time series, applying an unrestricted vector autoregression (VAR) model (Sims, 1980) is not advisable as it is likely to bias the estimated regression coefficients by spuriously rendering them higher (Brooks, 2014 & Gujarati, 2015). In the proposed model, performing a restricted VAR with first-differenced time series is also not advisable, as this is likely to disregard crucial information about the long-term relationships between the variables (Gujarati, 2015). The Johansen test (Johansen, 1988) suggests that long-run cointegrating relationships exist between the four variables of the model.

Therefore, the empirical process proceeded with a vector autoregression model with error correction (made possible by the cointegration relationships demonstrated by the Johansen test), producing a vector error correction model (VECM). The overall VECM model for the period 1950-2016 consists of the following four autoregressive equations (one for each of the four variables of the model) with 10 lags each, whose details are presented in the table below:

Figure 1

The VEC autoregressive equation for the housing-price index contains 38 parameters and has R² of 0.7912. The overall model’s Akaike information criterion is 7.45. Additionally, four other VECM models were used for the periods 1950-1970 (10 lags), 1970-1990 (10 lags), 1990-2006 (10 lags), and 2006-2016 (four lags due to the shorter duration of the period), as the Johansen test showed proof for cointegration between the series for all four periods, and the Augmented Dickey-Fuller test showed that all time series were non-stationary in their levels but stationary in their differences within the sub-periods. The orthogonalised impulse-response function of an interest-rate shock on housing prices (which takes account of the fact that the error terms of each of the four autoregressive equations are correlated) for the overall period between 1950 and 2016 is represented in Figure (2).

Figure 2

The impulse-response function demonstrates that on average for the nearly 60-year period, a unit shock (100bp) in the Effective Federal Funds Rate (an interest rate shock) will propagate through the economy to cause a decrease in the value of Case-Shiller index, peaking at around 14 quarters after the shock (at approximately -0.9%) and then receding. This result is in line with the estimates of Kuttner (2012), whose impulse-response function analysis obtained through a structural VAR model with error correction suggests a peak in the response of housing prices (-0.35%) to an interest-rate shock of 10 basis points at 12 quarters.  The VECM model appears to operate as expected with regards to the other explanatory variables, yielding the following impulse-response function for the effects of an interest rate shock on quarterly GDP growth, shown on Figure (3).

Figure 3

The VECM model also yields a theoretically-expected response of the residential building starts index to an interest-rate shock, as captured by the impulse-response function graph on Figure (4):

Figure 4

The above response function suggests an empirical conclusion in line with the discussion of Arestis and Karakitsos (2004) that residential building (but not supply of ready housing) responds quickly to housing-price shocks. In the medium- to long-run, as the market gets saturated with new buildings, increased housing prices could exert a negative pressure on residential building activity, as shown in Figure (5), which captures the response of the Residential Building Starts Index to a housing-price shock:

Figure 5

2. Sub-period VECM models

The empirical analysis proceeded with breaking down the initial period (1950-2016) into four sub-periods (1950-1970, 1970-1990, 1990-2006, and 2006-2016), to capture the changing dynamics of the monetary policy transmission mechanism throughout the main period, as reflected by the changing lag in the peak response of housing prices to an interest-rate shock. The following four graphs [Figures (6), (7), (8), and (9)] show the impulse-response functions for each of the above periods and demonstrate that the lag after which housing prices reach peak response to an interest-rate shock has been changing over the four sub-periods:

Figure 6

Figure 7

Figure 8

Figure 9

The three VECM models for 1950-70, 1970-90, 1990-2006 and the impulse-response functions of housing prices to an interest rate shock shown above [Figures (6), (7), (8), and (9)], suggest
that the time lag after which housing prices respond most strongly to interest-rate shocks
has been decreasing since 1950: from 23 quarters for 1950-1970, to 15 quarters for 1970-
1990, to 13 quarters for 1990-2006. Possible reasons for this could be the
increased levels of mortgage debt in the economy, the higher velocity of money since 1950, and ‘easier’ mortgages due to government policy stimulating home ownership, including the purchase of mortgage securities on the secondary mortgage market by government-backed enterprises. However, the tendency for the lag to decrease breaks down for the period 2006-2016, where the VECM analysis suggests the peak response to an interest rate shock arrives at around 17 quarters after the shock.

3. Empirical conclusions

The following table summarises the results obtained with the four models and the overall model:

Table 2: Effects of a 100bp increase in the EFFR on the Case-Shiller Index