1 Introduction

In the speech at the Annual Meeting of the American Economic Association (AEA) in January 2010, Ben S. Bernanke, who was the Chairman of Board of Governors of the Federal Reserve, pointed out the association between monetary policies and the recent housing crisis, noting that “the timing of the housing bubble does not rule out some contribution from monetary policy.” However, he concluded that “the direct linkages, at least, are weak.” In September 2010, in a statement before the Financial Crisis Inquiry Commission, he said, “Even if monetary policy was not a principal cause of the housing bubble, some have argued that the Fed could have stopped the bubble at an earlier stage by more-aggressive interest rate increase. For several reasons, this was not a practical policy option” (Bernanke 2010).

Motivated by the debate on the role of monetary policies in the recent housing crisis, this paper provides new insights into the timing and effectiveness of monetary policies for housing bubbles at disaggregate levels (Huang and Yeh 2015).Footnote 1 It characterizes the nonlinear interrelation between housing prices and the federal funds rate, which works as a proxy for the economic fundamental. The federal funds rate is regarded as the main nationwide shock to local housing markets and reflects the monetary policy criteria. A mounting body of empirical studies investigates whether local housing markets respond to an aggregate shock differently as attention grows regarding the possibly divergent sensitivities of disaggregate housing markets to a monetary policy shock.Footnote 2 Hence, this study utilizes the interest rate to associate the localized housing price dynamics with the nationwide economic fundamental.

In addition, because the interest rate has an impact on household wealth, this study links the empirical investigation into the interdependence between local housing markets and the interest rate with the theory of heterogeneous regional risk-sharing. Specifically, Lustig and Nieuwerburgh (2010) establish a dynamic general equilibrium model to show the divergent risk-sharing patterns across U.S. metropolitan areas. Based on this theory, tightly borrowing-constrained households change their consumption choices in response to income shocks more strongly than those owning more valuable housing collaterals do. High interest rates are likely to increase the sensitivity of consumption to the income change because they erode the values of housing collaterals. On the other hand, low interest rates encourage the demand of houses because they reduce the mortgage costs of house buyers and thus make households less collaterally constrained. Hence, a housing price climb is expected to be associated with a downward movement in interest rates.

The study contributes to the literature by modeling co-movements between housing and real sectors to shed insight on the underlying causes of housing crises. Noticeably, the threshold contemporaneous linkage between housing prices and the interest rate works as an instrument to capture metropolitan housing market dynamics. Hence, although lead–lag interrelations may exist, the modeled contemporaneous linkages enable us to investigate how the estimated low-growth regimes track the two housing bubble-like bust periods of the 1990s and 2006–2007 in the USA. In addition, the threshold model provides fresh housing bubble implications by showing whether housing prices and the economic fundamental are closely linked in the selected 25 metropolitan statistical areas (MSAs) housing markets. Although a vast literature exists on the linkage between housing price dynamics and economic fundamentals, this paper differs from previous studies along three dimensions. By modifying Perez-Alonso and Di-Sanzo (2011) framework, this study makes methodological contributions to the literature on nonlinear asset dynamics since it develops a new version of threshold framework which enriches our investigations into housing market dynamics. First, the model divides housing price growth into two components, a cycle (transitory) component and a trend (permanent) component, to provide more insights into housing market dynamics. Few existing studies adopt decomposition, and they ignore the distinction between transitory and permanent components.

Second, the proposed model incorporates contemporaneous threshold interdependence between cyclical housing price return and the interest rate change to examine whether their associations exhibit two threshold-switching regimes and how the regime-switching patterns are connected with the housing boom–bust cycles at MSA levels. The threshold interrelations between housing prices and the macroeconomic fundamental facilitate the investigation into possible housing bubbles in the MSA-level housing markets. Figure 1 shows that the federal funds rate peaked in 2000Q4, declined sharply up to the late 2003, and remained at a low level until the end of 2004. It rose from less than 2% in 2004 to about 5.7% in 2007Q1 and then experienced a persistent downward movement from 2007Q2 to 2010Q4. Correspondingly, the real housing price in Los Angeles-Long Beach-Glendale experienced a sharp swing during 1998–2006, while Dallas-Plano-Irving displayed a quite stable movement during the whole sample. These results reflect the potentially divergent co-movements between housing prices and macroeconomic fundamentals across metropolitan housing markets. Also, Bernanke (2010) stated, “Although the house price bubble appears obvious in retrospect—all bubbles appear obvious in retrospect—in its earlier stages, economists differed considerably about whether the increase in house prices was sustainable; or, if it was a bubble, whether the bubble was national or confined to a few local markets.” Thus, the investigation into MSA-level housing markets sheds light on the question of whether the bubble was national or locally confined.

Fig. 1
figure 1

Federal Funds Rate and Real Housing Price Indexes: 1991–2010. Notes: The figure displays the dynamics of federal funds rate (right-scaled) and real housing price indexes of Dallas-Plano-Irving, TX (DA) and Los Angeles-Long Beach-Glendale, CA (LA). The housing price indexes are gathered from U.S. Department of Labor: Bureau of Labor Statistics, and they are deflated by CPI

Full size image

Finally and equally important, this study adopts moving-average thresholds that divide the sample period for the metropolitan housing markets into two regimes: high-growth (boom) and low-growth (bust) regimes. The proposed model shows that half of the selected 25 MSA-level housing markets experience significant threshold regime-switching patterns, and thus, the interrelation between housing prices and the federal funds rate contradicts common assumptionsFootnote 3 about estimated low-growth regimes. Specifically, movements in housing price are positively associated with interest rate changes as housing price growth shrinks during the periods of 1990s and 2006–2007 in many MSA housing markets. The empirical results provide new implications for the timing and effectiveness of monetary policies in forestalling MSA-level housing crises, a topic examined by few existing studies.

The remainder of this paper is organized as follows. Section 2 reviews the literature that motivates the study. Section 3 presents the data and the proposed model. Section 4 reports the main empirical findings, including switching-regime phenomena of MSA-level housing markets for the discussion about the timing of monetary policies, and the interrelations between cyclical housing price returns and the federal funds rate for the analysis of monetary policy effectiveness. Section 5 provides concluding remarks.

2 Literature review

The paper investigates the implications of housing bubbles at disaggregate levels, motivated by four strands of empirical studies. First, empirical studies report nonlinear interactions across asset markets and real activities. Particularly, Franses and Van-Dijk (2000) find that the TAR model is a representative framework used to investigate nonlinearities of time series. Chang et al. (2011) propose that equity real estate investment trusts and housing asset returns respond to the change in the federal funds rate under a Markov-switching autoregressive framework. Guo et al. (2011) analyze contagion co-movements among stock, housing, credit default, and energy markets for the recent crisis period under a Markov-switching vector autoregression framework. Tkacz and Wilkins (2008) find that a threshold model improves the abilities of housing and stock prices in predicting GDP growth and inflation, and Liu and Shu (2010) use a threshold methodology to discuss the relation between housing and stock markets in China. Motivated by these studies, this paper proposes a threshold unobserved components model that captures the associations between housing prices and the interest rate. In addition, asymmetric relations across macroeconomic aggregates such as the Taylor rule, which describes interactions among monetary policies, output, and inflation gaps, have attracted the attention of researchers recently.Footnote 4 These studies utilize threshold models in an attempt to capture the pattern that central banks’ responses and policy preferences depend on the levels of the inflation rate and the output. For instance, Aguiar and Martins (2005) show that asymmetry of central banks’ policy preferences exist in the Euro area. Bunzel and Enders (2010) point out that the Fed has asymmetric responses during 1965–2007. Cukierman and Muscatelli (2008) apply smooth transition regressions to study nonlinear features of central banks of the UK and USA. Fiodendji (2013) finds that the response of the Bank of Canada to high inflation is stronger than its response to low inflation. Sznajderska (2014) discusses the response function of the National Bank of Poland. In an international context, Teles and Zaidan (2010) examine the Taylor rule for 12 developing countries.

Second, extant studies discuss whether low interest rates can be attributed to the remarkable housing boom in the 2000s. On the one hand, some empirical studies argue that interest rates in the recent housing cycle play a minor role (e.g., Campbell et al. 2009; Case and Shiller 2003; Dynan et al. 2006; Mayer 2003; Roy and Kemme 2012; Veld et al. 2011). Particularly, Campbell et al. (2009) argue that fluctuations in real risk-free interest rates fail to explain the housing price dynamics at MSA levels, and Dynan et al. (2006) suggest that the linkages between interest rates and housing markets dwindle recently compared to those in the previous periods. Roy and Kemme (2012) find that the recent housing bubble would have occurred even if relaxed monetary policies had not been adopted. On the other hand, some recent studies on housing markets claim that low interest rates act as the primary driving force behind the swing in housing prices. For example, Leamer (2007), Jarocinski and Smets (2008), and Taylor (2007) contend that relaxed monetary policies during 2003–2005 drove the recent housing price boom. Many academic studies, including Edelstein and Tsang (2007), Himmelberg et al. (2005), Jin and Zeng (2004), Lai and Van-Order (2010), McDonald and Stokes (2013), Shiller (2009), make similar arguments.

This study does not join the intense debate about influences of interest rates on housing price dynamics. Instead, it exposes the threshold regime-dependent associations between MSA-level housing markets and the federal funds rate, emphasizing that the estimated housing bust phases coincide with the bubble-like housing busts in 1990s and 2006–2007. Although they do not use any interest rate data, Huang and Yeh (2015) recently address the issue of monetary policy timings and effectiveness. They observe transitory shocks to asset markets, which are incorporated to track Markov-switching low-growth regimes of asset markets under an asymmetric unobserved components framework. Motivated by the previously discussed literature, this study delivers insights into the potentially different roles of monetary policies in metropolitan housing markets.

Third, a vast literature discusses local housing bubbles across MSA-level housing markets (e.g., Fratantoni and Schuh 2003; Ghent and Owyang 2010; Goodman 2005; Goodman and Thibodeau 2008; Hwang and Quigley 2006; Coleman et al. 2008; Leung and Teo 2011; McDuff 2011; Saiz 2010; among others). The study is particularly motivated by Goodman and Thibodeau (2008), who explore the presence of localized speculative housing bubbles in the USA. Goodman and Thibodeau (2008) estimate expected nominal appreciations in metropolitan house prices based on demand and supply fundamentals and compare their estimated growths with real growths computed by housing price data. Based on their comparisons, Goodman and Thibodeau (2008) examine whether possible housing bubbles exist in MSA-level housing markets during 2000–2005. As a result, they classify the analyzed metropolitan housing markets into two subgroups: 39 MSAs whose actual house prices are higher than expected and 45 MSAs whose prices are lower than expected. They propose that the 39 MSAs with higher actual house prices than expected are the housing markets that are subject to speculative housing bubbles. In addition, they find that most speculative activities of housing markets occur within 75 miles of the Atlantic coast or California Pacific coast. Motivated by these empirical findings, this paper investigates whether the proposed threshold framework provides consistent results with those of Goodman and Thibodeau (2008) and other related research.

Finally, the housing market literature supports the decomposition of housing price returns in this study. For instance, Cocco (2004) assumes a positive correlation between cyclical housing price movements and aggregate labor income shocks. Capozza et al. (2004) provide definitions for housing cycles and housing bubbles and differentiate price overshooting from cycles in housing markets by analyzing the parameters of correlation and reversion. Chen (2006), Lettau and Ludvigson (2004), and Sun et al. (2007) address how both transitory and permanent changes in housing wealth influence consumption. In addition, Fadiga and Wang (2009) point out that regional housing price dynamics display three common cyclical and two common trend components under a multivariate state-space framework. These studies suggest that the division of housing price dynamics into cyclical and trend components facilitates the investigation into housing markets. The previously discussed four strands of the literature conceptually and methodologically inspire this study to analyze the timings and effectiveness of monetary policies for 25 U.S. metropolitan housing markets.

3 Data and model

3.1 Data

This study uses purchase-only house price indexes (estimated using sales price data) of the largest 25 MSAs in the USA, which are gathered from Federal Housing Finance Agency. The housing price data, which are seasonally adjusted and deflated by the core consumer price index, span from 1991Q1 to 2010Q4. The consumer price index for all urban consumers includes all items less food and energy and is obtained from the U.S. Department of Labor’s Bureau of Labor Statistics. The real housing price return of each metropolitan housing market is computed as the first difference in logs of the real housing price index. The effective federal funds rate, which represents the monetary policies and the economic fundamental, is obtained from the Board of Governors of the Federal Reserve System and is seasonally adjusted by the U.S. Census Bureau’s X12 seasonal adjustment.

3.2 Threshold unobserved components model

This study is methodologically motivated by Hansen (1997) who proposed the threshold autoregressive (TAR) model, and Perez-Alonso and Di Sanzo (2011) who proposed the decomposition framework. Perez-Alonso and Di Sanzo (2011) extend the methodology of Jaeger and Parkinson (1994) to develop a model that describes the threshold impact of cyclical unemployment on the employment rate. They analyze unemployment persistence by decomposing the unemployment rate into two components: a cycle and a trend (natural employment rate) to consider threshold interactions between the cycle and trend of the unemployment rate and investigate whether the short-term adjustment in labor markets has an impact on the long-term employment in Italy, France, and the USA.

Based on the TAR framework in Hansen (1997), this study further modifies the Perez-Alonso and Di Sanzo (2011) framework to observe the threshold interrelation between cyclical housing price returns and the federal funds rate for MSA-level housing markets. The proposed model is

$$ P_{t} = P_{t}^{T} + P_{t}^{C} $$
(1)
$$ P_{t}^{T} = P_{t - 1}^{T} + \theta P_{t - 1}^{C} + \varepsilon_{t}^{T} $$
(2)
$$ P_{t}^{C} = \rho_{1} P_{t - 1}^{C} + \rho_{2} P_{t - 2}^{C} + \varepsilon_{t}^{C} $$
(3)
$$ R_{t} = \beta R_{t - 1} + \delta_{1} P_{t}^{C} I\left( {q_{t - d} \ge \gamma } \right) + \delta_{2} P_{t}^{C} I\left( {q_{t - d} < \gamma } \right) + \varepsilon_{t}^{R} . $$
(4)

\( P_{t} \) refers to the housing price return, which is decomposed into a nonstationary trend component, \( P_{t}^{T} , \) and a stationary cyclical component, \( P_{t}^{C} \). Equation (2) shows the cycle–trend interaction, which is measured by the coefficient θ. Equation (3) shows the cycle of the housing price return, \( P_{t}^{C} \), which is assumed to follow a stationary second-order autoregressive process.Footnote 5 Equation (4) links the cyclical housing price return \( (P_{t}^{C} ) \) with the change in the federal funds rate (Rt) in the threshold framework. This equation uses a 1-period-lag interest rate change as the explanatory variable in the regression to take into account the first-order autoregressive process of the dynamics of the federal funds rate. Thus, the model examines whether MSA-level housing price cycles deviate from the economic fundamental. The coefficients δ1 and δ2 stand for the linkages in the housing boom and housing bust phases, respectively.

The error terms in Eqs. (2)–(4) (i.e., \( \varepsilon_{t}^{T} , \varepsilon_{t}^{C} , \varepsilon_{t}^{R} \), respectively) are assumed to be uncorrelated and normally distributed with corresponding variances (i.e., \( \sigma_{t}^{T} \), \( \sigma_{t}^{C} ,\sigma_{t}^{R} \), respectively). I(.) is the usual indicator function with a threshold indicator, \( q_{t - d} \),which splits the whole sample period into favorable (high-growth; \( q_{t - d} \ge \gamma \)) and unfavorable (low-growth; \( q_{t - d} < \gamma \)) regimes based on the estimated threshold value γ. The threshold indicator is the moving-average housing price growth (i.e., \( q_{t - 2} = \frac{{y_{i,t} + y_{i,t - 1} }}{2} \); \( q_{t - 3} = \frac{{y_{i,t} + y_{i,t - 1} + y_{i,t - 2} }}{3} \)). The moving-average period d \( \in \) {2,3} is estimated along with the threshold value γ by the maximum likelihood method of the Kalman filter. A maximization log-likelihood grid search over each two-dimensional space (γ, d) is solved, and γ is restricted to within the range [30% quantile of \( q_{t - d} \), 70% quantile of \( q_{t - d} \)]. According to Perez-Alonso and Di Sanzo (2011), bootstrap methods to calculate p values can be used to test nonlinearities for metropolitan housing markets.

Noticeably, Perez-Alonso and Di Sanzo (2011) choose the long difference of the cyclical unemployment rate between two or three lags as the threshold variable. However, this study finds that the long-differenced threshold fails to capture metropolitan housing boom–bust cycles in reality although it works well in the real activity as analyzed by Perez-Alonso and Di Sanzo (2011). Thus, this study establishes moving-average thresholds to classify housing boom–bust regimes, which are reported in Sect. 4.3 of monetary policy timing analyses.

This study utilizes the threshold contemporaneous linkages between housing prices and the federal funds rate to work as the instrument for co-movements between housing markets and the fundamental (through the two estimated coefficients, δ1 and δ2). The exact lead–lag housing price–interest rate connections require further research and beyond the scope of this paper.

4 Empirical results

This section analyzes the empirical results of the proposed threshold unobserved components model. Table 1 shows the descriptive statistics of housing price returns for the 25 largest MSA-level housing markets. The MSAs whose distributions are peaked (leptokurtic) relative to the normal (Kurtosis is larger than 3) are highlighted, and those with nonnormal dynamics are indicated by the bold Jarque–Bera statistics. The table also highlights standard deviations that are larger than 2.5, means and medians that are greater than 0.25 and minimums (maximums) that are smaller (larger) than − 5 (5).

Table 1 Descriptive statistics of MSA housing price returns
Full size table

4.1 Cycles versus trends in housing price returns

Table 2 shows that only 7 (Chicago-Joliet-Naperville; Edison-New Brunswick, NJ; Houston-Sugar Land-Baytown; Nassau-Suffolk, NY; Philadelphia; Riverside-San Bernardino-Ontario, CA; San Diego-Carlsbad-San Marcos, CA) out of the 25 MSA housing markets display significant interrelations between cycles and trends in housing price returns (θ). Particularly, the MSAs in New Jersey (θ = − 0.03) and California (θ = − 0.23 and − 0.17 for both Riverside and San Diego) exhibit significantly negative relations between these two components. Empirical studies widely regard the local housing markets in the cities of these two states as subject to housing bubbles (e.g., Davis and Heathcote 2007; Follain and Giertz 2011; Holly et al. 2010; McCarthy and Peach 2004; Rapach and Strauss 2009; Schiller 2006; among others).

Table 2 MSAs with significant linkages between housing prices and the federal funds rate
Full size table

4.2 Threshold linkages between housing prices and the federal funds rate

Table 2 reports the MSA-level housing markets that have significant coefficients of interrelations between cyclical housing price returns and the federal funds rate (i.e., δ1 and δ2) in at least one of the threshold regimes.Footnote 6 However, only seven MSAs (Atlanta-Sandy-Springs-Marietta; Cleveland-Elyria-Mentor, OH; Houston-Sugar Land-Baytown; Phoenix-Mesa-Glendale, AZ; Riverside-San-Bernardino-Ontario; San Diego-Carlsbad-San Marcos; Washington, DC-Arlington-Alexandria) display significant housing price–interest rate linkages in both the housing boom and bust phases. Noticeably, only Atlanta and Cleveland MSAs display negative interdependence in the favorable regime, whereas the Houston, Phoenix, Riverside, San Diego, and Washington, DC MSA-level housing markets have counterintuitively positive linkages in both regimes. Interestingly, most metropolitan housing markets (except for Houston) are regarded as the cities subject to housing bubbles in the literature as introduced in Sect. 2.

In particular, 14 out of the 25 metropolitan housing markets have counterintuitively positive linkages between housing price growths and the interest rate in at least one of boom–bust regimes, with the significant coefficients ranging from 0.82 to 15.89. Eighteen out of the selected MSAs display insignificant housing price–interest rate interactions in at least one of boom–bust regimes. In general, the linkages are weaker in booms than busts, implying the weak explanatory power of the interest rate in stabilizing the swings of metropolitan housing prices as the recent bubble-like housing boom occurred.

4.3 Threshold regime-switching phenomena: timings of monetary policies

This section addresses the switching housing growth regimes during 1991–2010 and examines how the threshold regime-switching phenomena of the MSA-level housing markets are associated with interest rate dynamics. Table 3 reports the estimated threshold values and moving-average periods (i.e., the parameter d). During 1991–2010, the downward movements of the federal funds rate occur in the three periods: 1991–1993Q2, 2000Q4–2003Q3, and 2007Q1 to the present. As Case and Shiller (2003) suggest, the US local housing markets display bubble-like boom–bust cycles around 1990 during which the peaks of housing prices are reached and followed by sharp declines in many metropolitan areas. Although the presence of a housing bubble is a lasting debate, this study observes how the threshold-switching regimes are synchronous with the bubble-like housing busts in the 1990s and 2006–2007.

Table 3 Threshold values and estimated lags
Full size table

Figure 2 shows the threshold regime-switching phenomena for the MSA housing markets that have significant co-movements between housing prices and the interest rate in one or both threshold regimes. The periods in which housing returns are lower than the estimated threshold values are specified as housing bust regimes. Although only 7 out of 25 MSAs have significant regime-switching patterns of housing price growths for both regimes, 15 MSA-level housing markets have significant bust regimes. Specifically, the estimated low-growth regimes coincide with the first bubble-like housing bust, spanning from 1991 to 1995–1997, depending on the metropolitan housing markets. The housing bust phases lagged the first downward movement of the federal funds rate (1991–1993Q2) by different time periods across the MSA-level housing markets.

Fig. 2
figure 2

MSA-level housing price returns for MSAs with significant low-growth regimes. Notes: The figure shows the MSA-level housing returns for the MSAs with significant bust regimes. The horizontal lines indicate the values of threshold indicators (\( q_{t - d} , \) moving-average housing price returns). Thus, the bust regimes are the periods in which housing price returns are lower than the estimated threshold values. The shaded periods refer to the NBER-dated recessions

Full size image

More important, the metropolitan housing markets switch from high-growth to low-growth phases in 2006–2007, which markedly correspond to the third downward movement in the federal funds rate. Riverside, San Diego, and Washington, DC enjoy short threshold high-growth phases in 2009. Overall, the empirical results succeed in tracking timings of monetary policies in stabilizing housing market dynamics at MSA levels during the recent housing boom–bust cycles.

4.4 Counterintuitive interrelations between metropolitan housing prices and the federal funds rate: implications for monetary policy effectiveness

Section 4.3 shows that the estimated housing bust regimes closely match historical timings of monetary policies. The section further examines monetary policy effectiveness. Although only 7 of the 25 MSA-level housing markets display significant coefficients in both boom and bust regimes, up to 14 MSAs show significant counterintuitive (positive) interrelations between cyclical housing prices and the federal funds rate in the low-growth regimes. Noticeably, only Dallas-Plano-Irving and Nassau-Suffolk have negative interactions in their unfavorable regimes, whereas all other cities display positive linkages as their housing price growths curtail.

The contemporaneous linkages reflect divergent vulnerabilities to housing bubbles across the MSA-level housing markets analyzed. Interestingly, most MSA-level housing markets with significantly positive linkages in their low-growth regimes are areas that Goodman and Thibodeau (2008) suggest are subject to housing speculative bubbles, whereas many of the MSAs without significant threshold regime-switching patterns in this study are cities whose actual house prices are lower than fundamentally estimated in Goodman and Thibodeau (2008). For instance, Los Angeles, Miami, Florida, Minneapolis, Philadelphia, Phoenix, Riverside, San Diego, and Washington experience excessive housing price appreciation, and the results of this study show significantly positive contemporaneous interrelations between housing prices and the interest rate. Otherwise, the MSAs whose estimated associations are not significant in their bust regimes, such as St. Louis, Missouri, Seattle, Washington, Denver, Colorado, and Chicago, experience milder housing price increases than expected during 2000–2005 and do not have speculative housing bubbles as suggested by Goodman and Thibodeau (2008). Thus, the findings shed light on Bernanke’s question of whether the bubble occurred nationally or was confined to specific local markets. These results show that the housing bubble did not spread across all markets but rather was confined to a few local markets.

In short, the results suggest low effectiveness of monetary policies: the Fed may have great difficulty in protecting against two bubble-like housing busts in many MSA-level housing markets because the negative linkages between housing price dynamics and the federal funds rate are broken in these cities’ bust phases. In this sense, the results echo Bernanke’s argument in terms of the weak direct linkages between monetary policies and the recent housing bubble. Particularly, he emphasized in the conclusion of his 2010 AEA meeting speech: “Although the most rapid price increases occurred when short-term interest rates were at their lowest levels, the magnitude of house price gains seems too large to be readily explainable by the stance of monetary policy alone.” The study demonstrates that threshold interrelations are informative in capturing bubble-like housing busts in the USA, and the results are consistent with those of the previous literature on metropolitan housing markets.

5 Conclusions

This paper provides fresh implications for monetary policy timings and effectiveness for housing bubbles by investigating the contemporaneous interrelations between MSA-level housing prices and the federal funds rate during 1991–2010. The proposed threshold unobserved components model is capable of tracking the timings of monetary policies: roughly half of the largest U.S. metropolitan housing markets experience significant housing bust regimes that coincide with the U.S. bubble-like housing busts in 1990s and 2006–2007 and the downward movements in the federal funds rate. Nevertheless, the findings suggest the low effectiveness of monetary policies in forestalling the recent housing crisis and echo the remarks of Bernanke and the arguments in the existing literature in terms of the weak direct linkages between monetary policies and housing bubbles. The counterintuitively positive linkages between housing price and the federal funds rate are indicative of possible housing bubbles in some metropolitan housing markets, particularly for Phoenix, Riverside, San Diego, and Washington, DC.

While this study facilitates our investigations into different monetary policy effectiveness across metropolitan housing markets, several important questions remain open. For example, along the lines of Moench and Ng (2011) and Del Negro and Otrok (2007), it would be of interest for future research to investigate divergent housing market dynamics and vulnerabilities to housing bubbles under dynamic factor frameworks. Also, using other variables, such as the GDP gap or per capita income, to proxy for economic fundamentals would enrich our investigation into housing bubble vulnerabilities for disaggregate housing markets. Overall, this study reveals the challenges of monetary policy-making during crisis periods and invites further research on the monetary policy effectiveness and housing bubble implications in asset markets.