Monetary policy rules and financial stress : does financial instability matter for monetary

Monetary Policy Rules and Financial Stress: Does Financial Instability Matter for Monetary Policy? Jaromír Baxa, Roman Horváth, Borek Vašícek 11.01 De p artament d'Economia A p licada Facultat d'Economia i Empresa Aquest document pertany al Departament d'Economia Aplicada. Data de publicació : Departament d'Economia Aplicada Edifici B Campus de Bellaterra 08193 Bellaterra Telèfon: (93) 581 1680 Fax:(93) 581 2292 E-mail: [email protected] http://www.ecap.uab.es Gener 2011 1 Monetary Policy Rules and Financial Stress: Does Financial Instability Matter for Monetary Policy? Jaromír Baxa* Institute of Economic Studies, Charles University, Prague and Institute of Information Theory and Automation, Academy of Sciences of the Czech Republic Roman Horváth Czech National Bank and Institute of Economic Studies, Charles University, Prague Bořek Vašíček Universitat Autonoma de Barcelona Abstract We examine whether and how main central banks responded to episodes of financial stress over the last three decades. We employ a new methodology for monetary policy rules estimation, which allows for time-varying response coefficients as well as corrects for endogeneity. This flexible framework applied to the U.S., U.K., Australia, Canada and Sweden together with a new financial stress dataset developed by the International Monetary Fund allows not only testing whether the central banks responded to financial stress but also detects the periods and type of stress that were the most worrying for monetary authorities and to quantify the intensity of policy response. Our findings suggest that central banks often change policy rates: mainly decreasing it in the face of high financial stress. However, the size of a policy response varies substantially over time as well as across countries, with the 2008-2009 financial crisis being the period of the most severe and generalized response. With regards to the specific components of financial stress, most central banks seemed to respond to stock market stress and bank stress, while exchange rate stress is found to drive the reaction of central banks only in more open economies. JEL Classification: E43, E52, E58. Keywords: financial stress, Taylor rule, monetary policy, time-varying parameter model, endogenous regressors. * We thank Øyvind Eitrheim, Ekkehart Schlicht, Miloslav Vošvrda and the seminar participants at the 7 th Norges Annual Monetary Policy Conference, the Institute of Information Theory and Automation (Academy of Sciences of the Czech Republic), Universitat de Barcelona, Universitat de Girona, Universitat de les Illes Balears and Universidad Complutense de Madrid for helpful discussions. The views expressed in this paper are not necessarily those of the Czech National Bank. Emails: [email protected], [email protected], [email protected]. 2 1 Introduction The recent financial crisis has intensified the interest in exploring the interactions between monetary policy and financial stability. Official interest rates were driven sharply to historic lows and many unconventional measures were used to pump liquidity into the international financial system. Central banks pursued monetary policy under high economic uncertainty coupled with large financial shocks in many countries. The financial crisis also raised new challenges for central bank policies, in particular how to operationalize the issues related to financial stability for monetary policy decision-making (Goodhart, 2006, Borio and Drehmann, 2009). This paper seeks to analyze whether and how central banks reacted to the periods of financial instability, and in particular whether and how the interest-setting process evolved in response to financial instability over the last three decades. The monetary policy of central banks is likely to react to financial instability in a non-linear way (Goodhart et al., 2009). When a financial system is stable, the interest rate setting process largely reflects macroeconomic conditions and financial stability considerations enter the monetary policy discussions only to a limited degree. On the other hand, central banks may alter its monetary policy to reduce financial imbalances if these become severe. In this respect, Mishkin (2010) questions the traditional linear-quadratic framework 1 when financial markets are disrupted and puts forward the arguments for replacing it by nonlinear dynamics, describing the economy and the non-quadratic objective function resulting in the non-linear optimal policy. To deal with the complexity of monetary policy and financial stability nexus as well as to evaluate monetary policy in a systematic manner, this paper employs the recently developed time-varying parameter estimation of monetary policy rules appropriately accounting for endogeneity in policy rules. This flexible framework, together with a new comprehensive financial stress dataset developed by the International Monetary Fund, will allow not only testing whether the central banks responded to financial stress but also the quantification of the magnitude of this response and the detection of the periods and types of stress that were the most worrying for monetary authorities. Although theoretical studies disagree about the role of financial instability for central bank interest rate setting policy, our empirical estimates of time-varying monetary policy rules of the US Fed, the Bank of England (BoE), Reserve Bank of Australia (RBA), Bank of Canada (BoC) 1 Linear behavior of the economy and the quadratic objective function of monetary authority. 3 and Sveriges Riksbank (SR) shows that central banks often alter the course of its monetary policy in the face of high financial stress, mainly by decreasing their policy rates. However, the size of this response varies substantially over time as well as across countries. There is a certain across country and time heterogeneity as well when we look at central banks’ consideration of specific types of financial stress: Most of them seemed to respond to stock market stress and bank stress, and exchange rate stress drives central bank reactions only in more open economies The paper is organized as follows: Section 2 discusses related literature. Section 3 describes our data and empirical methodology. Section 4 presents our results. Section 5 concludes. An appendix with a detailed description of the methodology and additional results follows. 2 Related Literature First, this section gives a brief overview of the theory as well as empirical evidence on the relationship between monetary policy (rules) and financial instability. Second, it provides a short summary of various measures of financial stress. 2.1 Monetary policy (rules) and financial instability – some theories Financial frictions, such as an unequal access to credits or debt collateralization, were recognized to have important consequences for monetary policy transmission and Fisher (1933) has already presented the idea that adverse credit-market conditions can cause significant macroeconomic disequilibria. During the last two decades, the effects of monetary policy have been studied mainly within New Keynesian (NK) dynamic stochastic general equilibrium (DSGE) models, which assume the existence of nominal rigidities. The common approach to incorporate financial market friction within the DSGE framework is to introduce the financial accelerator mechanism (Bernanke et al., 1996, 1999), implying that endogenous developments in credit markets work to amplify and propagate shocks to the macroeconomy. Tovar (2009) emphasizes that the major weakness of the financial accelerator mechanism is that it only deals with one of many possible financial frictions. Goodhart et al. (2009) notes that many NK DSGE models lack the financial sector completely or modeled it in a rather embryonic way. Consequently, more recent contributions within this stream of literature examined other aspects of financial frictions such as the balance sheets in the banking sector (Choi and Cook, 2004), portfolio choice issue with complete (Engel or 4 Matsumoto, 2009) or incomplete markets (Devereux and Sutherland, 2007) or collateral constraints (Iacovello and Neri, 2010). 2 A few studies focus more specifically on the relationship between the monetary policy stance (or the monetary policy rule) and financial stability. However, they do not arrive at a unanimous view on whether a monetary policy rule should include some measure of financial stability. Brousseau and Detken (2001) present an NK model where a conflict arises between short-term price stability and financial stability due to a self-fulfilling belief linking the stability of inflation to the smoothness of the interest rate path and suggests that monetary policy should react to financial instability. Akram et al. (2007) investigate the macroeconomic implications of pursuing financial stability within a flexible inflation-targeting framework. Their model using policy rule, augmented with financial stability indicators, shows that the gains of such an augmented rule vis-à-vis the rule without financial stability indicators highly depends on the nature of the shocks. Akram and Eitrheim (2009) build on the previous framework, finding some evidence that the policy response to housing prices, equity prices or credit growth can cause high interest rate volatility and actually lower financial stability in terms of indicators that are sensitive to interest rates. Ceccheti and Li (2008) show in a static and dynamic setting that a potential conflict between monetary policy and financial supervision can be avoided if the interest rate rule takes (procyclical) capital adequacy requirements into account, in particular that the policy interest rates are lowered when financial stress is high. Bauducco et al. (2008) extends the current benchmark NK model to include financial systems and firms that require external financing. Their simulations show that if a central bank responds to financial instability by policy easing it achieves better inflation and output stabilization in the short term at the cost of greater inflation and output volatility in the long term and vice versa. For the US Fed Taylor (2008) proposes a modification of a standard Taylor rule to incorporate adjustments to credit spreads. Teranishi (2009) derives a Taylor rule augmented by the response to credit spreads as an optimal policy under heterogeneous loan interest rate contracts. He finds that the policy response to a credit spread can be both positive and negative depending on the financial structure. However, he also puts forward that when nominal policy rates are close to zero a commitment rather than discretional policy response is the key for reducing the credit spreads. Christiano et al. (2008) suggest augmenting the Taylor rule with aggregate private credit and find that such a policy would raise welfare by reducing the magnitude of the output fluctuations. Cúrdia and Woodford (2010) develop an NK DSGE model with credit frictions to evaluate the performance of alternative policy rules that are 2 The survey of this literature is provided by Tovar (2009). 5 augmented by a response to credit spreads and to aggregate the volume of private credit in the face of different shocks. They argue that the response to credit spreads can be welfare improving, but the optimal size of such a response is likely rather small. Like Teranishi (2009), they find little support for augmenting a Taylor rule by the credit volume given ― that the size and even the sign of the desired response is sensitive to the sources of shocks and their persistence ― which is information that is not always available during operational policy making. The related stream of literature focuses on a somewhat narrower issue of whether or not monetary policy should respond to asset prices. Bernanke and Gertler (1999, 2001) argue that the stabilization of inflation and output provides a substantial contribution to financial stability and there are little if any gains to responding to asset prices. Faia and Monacelli (2007) extend the model developed by Bernanke and Gertler (2001) by a robust welfare metric confirming that strict inflation stabilization offers the best solution. Cecchetti et al. (2000) takes the opposite stand arguing that developments in asset markets can have a significant impact on both inflation and real economic activity, and central banks might achieve better outcomes considering the asset prices provided they are able to detect their misalignments. Borio and Lowe (2002) support this view claiming that financial imbalances can build up even in a low inflation environment, which is normally favorable to financial stability. The side effect of low inflation is that excess demand pressures may first appear in credit aggregates and asset prices rather than consumer prices, which are normally considered by the policy makers. Gruen et al. (2005) argues that responding to an asset bubble is feasible only when the monetary authority is able to make a correct judgment about the process driving the bubble. Roubini (2006) and Posen (2006) provide the summary of this debate from a policy perspective. 2.2 Monetary policy (rules) and financial instability – empirical evidence The empirical evidence on central banks’ reaction to financial instability is rather scant. Following the ongoing debate about whether central banks should respond to asset price volatility (e.g. Bernanke and Gertler, 1999, 2001; Cecchetti et al., 2000; Bordo and Jeanne, 2002), some studies tested the response of monetary policy to different asset prices, most commonly to stock prices (Rigobon and Sack, 2003; Siklos and Bohl, 2008; Fuhrer and Tootel, 2008). They find some evidence that asset prices either entered the policy information set (because they contain information about future inflation) or that some central banks were directly trying to offset its 6 disequilibria. 3 All these papers estimate time-invariant policy rules, which means that they test a permanent response to these variables. However, it seems more plausible that if central banks respond to asset prices, they do it only when their misalignments are substantial, in other words their response is asymmetric. There are two additional controversies related to the effects of asset prices on monetary policy decisions: (i) The first concerns the measure, in particular whether the stock market index that is typically employed is sufficiently representative or whether some other assets, in particular the housing prices, should be considered as well, and (ii) the second issue is related to the (even ex-post) identification of the asset price misalignment. Finally, it is likely that the perception of misalignments is influenced by general economic conditions and that a possible response could evolve over time. Detken and Smets (2004) summarize some stylized facts on macroeconomic and monetary policy developments during asset price booms. Overall they find that monetary policy was significantly looser during the high-cost booms that were marked by the investment and real estate prices crash in the post-boom periods. A few empirical studies measure monetary policy response using broader measures of financial imbalances. Borio and Lowe (2004) estimate the response of four central banks (Reserve Bank of Australia, Bundesbank, Bank of Japan and the US Fed) to imbalances proxied by the ratio of private sector credit to GDP, inflation-adjusted equity prices and their composite. They find either negative or ambiguous evidence for all countries except for the USA confirming that the Fed responded to financial imbalances in an asymmetric and reactive way, i.e. that the federal fund rate was disproportionally lowered in the face of imbalance unwinding but it was not tightened beyond normal as imbalances built up. Ceccheti and Li (2008) estimate a Taylor rule augmented by a measure of banking stress, in particular a deviation of leverage ratios (total loans to the sum of equity and subordinated debt; total assets to the sum of bank capital and reserves) from its Hodrick-Prescott trend. They find some evidence that the Fed adjusted the interest rate in order to counteract the procyclical impact of a bank’s capital requirements, while the Bundesbank and the Bank of Japan did not. Bulíř and Čihák (2008) estimate the monetary policy response to seven alternative measures of financial sector vulnerability (crisis probability, time to crisis, distance to default or credit default swap spreads) in a panel of 28 countries. Their empirical framework is different in the sense that the monetary policy stance is proxied along the short-term interest rate by measures of domestic liquidity and external shocks are controlled for. In the panel setting, they find a statistically significant negative response to many variables representing vulnerability (policy easing) but surprisingly not in country-level regressions. Belke 3 A similar but somewhat less polemic debate applies to the role of the exchange rate, especially for small open economies (Taylor, 2001). 7 and Klose (2010) investigate the factors behind the interest rate decision of the ECB and the Fed during the current crisis. They conclude that the estimated policy rule was significantly altered only for the Fed and they put forward that the ECB gave greater weight on inflation stabilization at the cost of some output loss. 2.3 Measures of financial stress The incidence and determinants of different types of crises have been typically traced in the literature by a means of narrative evidence (expert judgment). This was sometimes complemented with selected indicators (the exchange rate devaluation, the state of foreign reserves) that point to historical regularities (e.g. Eichengreen and Bordo, 2002; Kaminsky and Reinhart, 1999; Reinhart and Rogoff, 2009; Laeven and Valencia, 2008). The empirical studies (e.g. Goldstein et al., 2000) used binary variables that were constructed based on these narratives. Consequently, some contributions strived to provide more data-driven measures of financial stress. Most of the existing stress indices are based on high-frequency data but they differ in the selected variables (bank capitalization, credit ratings, credit growth, interest rate spreads or volatility of different asset classes), country coverage and the aggregation method. An important advantage of continuous stress indicators is that it may reveal periods of small-scale stress that did not result in full-blown crisis and were neglected in studies based on binary crisis variables. The Bank Credit Analyst (BCA) reports a monthly financial stress index (FSI) for the USA that is based on the performance of banking shares as compared to whole stock market, credit spreads and the slope of the yield curve, and the new issues of stocks, bonds and consumer confidence. JP Morgan calculates a Liquidity, Credit and Volatility Index (LCVI) based on seven variables: the US Treasury curve error (standard deviation of the spread between on-the-run and off-the-run US Treasury bills and bonds along the entire maturity curve), the 10-year US swap spread, US high-yield spreads, JP Morgan’s Emerging Markets Bond Index, foreign exchange volatility (weighted average of 12-month implied volatilities of several currencies), the Chicago Board of Exchange equity volatility index VIX, and the JP Morgan Global Risk Appetite Index. Illing and Liu (2006) develop a comprehensive FSI for Canada. Their underlying data covers equity, bond and foreign exchange markets as well as the banking sector. They use a standard measure and refined measure of each stress component, where the former refers to the variables and their transformations that are commonly found in the literature, while the latter incorporates 14 but it is rather a factor such as the lagged interest rate, i.e. it may explain why the actual interest rate t r deviates from the target. Moreover, placing it in the regression on the same level as a lagged interest rate, we can directly test whether this variable representing ad-hoc policy decisions decreases the interest rate inertia ρ as suggested by Mishkin (2009). The common logic also suggest that the coefficients ρ and δ shall move in the opposite direction because the central bank either smoothes the interest rate changes or adjusts the rates in the face of financial stress. In the latter case, the response is likely to be quick and substantial. We set i equal to 2, j equal to 0 and k equal to -1. 10 Consequently, the disturbance term t ε is a combination of forecast errors and is thus orthogonal to all information available at time t ( t Ω ). The empirical studies on monetary policy rules have moved from using time-invariant estimates (Clarida et al., 1998) through sub-sample analysis (Taylor, 1999, Clarida et al., 2000) towards more complex methods that allow an assessment of the evolution in the conduct of monetary policy. There are two alternative methods to modeling structural changes in monetary policy rules that occur on an unknown date: (i) regime switching models, in particular the state-dependent Markov switching models (Valente, 2003; Assenmacher-Wesche, 2006; Sims and Zha 2006) and (ii) statespace models, where the changes are characterized by smooth transitions rather than abrupt switches (Boivin, 2006; Elkhoury, 2006; Kim and Nelson, 2006; Trecrocci and Vasalli, 2009). As argued in Baxa et al. (2010), we consider the second approach as preferable for the estimation of policy rules given that it is more flexible and allows the incorporation of a simple correction of endogeneity (Kim, 2006; Kim and Nelson, 2006), which is a major issue in forward-looking policy rules estimated from ex-post data. 11 The state-space approach or time-varying coefficient model seems also suitable when one wants to evaluate the effect of factors such as financial stress that can, for a limited length of time, alter (rather than permanently change) the monetary policy conduct. 10 Although the targeting horizon of central banks is usually longer (4-8 quarters), we prefer to proxy inflation expectations by inflation in t+2 for the following reasons: First, the endogeneity correction requires a strong correlation between the endogenous regressor and its instruments. Second, the prediction error logically increases in longer horizons. In case of the output gap, we instead assume a backward-looking reaction. The reason is that in the absence of real time data we have to rely on the output gap construction of statistical methods. It is arguable that besides the prediction error there is also a construction error that both might be magnified if an unobserved forecast is substituted by the output gap estimate for future periods. At last, we assume that central bankers’ response (if any) to financial stress is rather immediate (see Mishkin, 2009). Therefore, we use one lag of the FSI and its subcomponents in the benchmark case. However, as a robustness check we allow for different lags and leads, allowing the central bankers’ response to be preemptive rather than reactive. 11 The time-varying parameter model with the specific treatment of endogeneity is still relevant when real-time data are used (Orphanides, 2001). When the real-time forecast is not derived under the assumption that nominal interest rates will remain constant within the forecasting horizon (Boivin, 2006) or in the case of measurement error and heteroscedasticity (Kim et al., 2006). 15 The state-space models are commonly estimated by means of a maximum likelihood estimator via the Kalman filter or smoother. Unfortunately, this approach has several limitations that can turn problematic in applied work. First, the results are somewhat sensitive to the initial values of the parameters, which are usually unknown, especially in the case of variables whose impact on the dependent variable is not permanent and whose size is unknown, which is the case of financial stress and its effect on interest rates. Second, the log likelihood function is highly nonlinear and in some cases, optimization algorithms fail to minimize the negative of the log likelihood. In particular, it can either fail to calculate the Hessian matrix throughout the iterations process or, when the likelihood function is approximated to facilitate computations, covariance matrix of observation vector can get singular for provided starting values. The alternative is a moment-based estimator proposed by Schlicht (1981, 2005) and Schlicht and Ludsteck (2006), which is employed in our paper and briefly described below. This framework is flexible enough so as to incorporate the endogeneity correction proposed by Kim (2006). Kim (2006) shows that the conventional time-varying parameter model delivers inconsistent estimates when explanatory variables are correlated with the disturbance term and proposes an estimator of the time-varying coefficient model with endogenous regressors. The endogeneity may arise not only in forward-looking policy rules based on ex-post data (Kim and Nelson, 2006, Baxa et al., 2010), but also in the case of variables that have a two-sided relation with monetary policy. Financial stress unquestionably enters this category. Following Kim (2006) we rewrite Eq. 3 as follows: ( ) 1 1 t t t t t i t t j t t t t k t r y r x ρ α β π γ ρ δ ε + + − +   = − + + + + +   (4) 1 1, t t t α α ϑ − = + , ( ) 1 2 1, ~ . . . 0, t ii d N ϑ ϑ σ (5) 1 2, t t t β β ϑ − = + , ( ) 2 2 2, ~ . . . 0, t ii d N ϑ ϑ σ (6) 1 3, t t t γ γ ϑ − = + , ( ) 3 2 3, ~ . . . 0, t ii d N ϑ ϑ σ (7) 1 4, t t t δ δ ϑ − = + , ( ) 4 2 4, ~ . . . 0, t ii d N ϑ ϑ σ (8) 1 5, t t t ρ ρ ϑ − = + , ( ) 5 2 5, ~ . . . 0, t ii d N ϑ ϑ σ (9) ' t i t m t Z ϕ π ξ σ ϕ + − = + , ( ) ~ . . . 0,1 t ii d N ϕ (10) ' t j t m t y Z ν ψ σ ν + − = + , ( ) ~ . . . 0,1 t ii d N ν (11) ' t k t m t x Z ι ο σ ι + − = + , ( ) ~ . . . 0,1 t ii d N ι (12) 16 The measurement Eq. (4) of the state-space representation is the monetary policy rule. The transitions in Eqs. (5)-(9) describes the time-varying coefficients as a random walk process without drift. 12 Eqs. (10)-(12) track the relationship between the potentially endogenous regressors ( it+ π , t j y + and t k x + ) and their instruments, t Z . We use the following instruments: 1−t π , 12 t π − ( 4 t π − for CAN and SWE), 1 t y − , 2 t y − , 1−t r , and when k≥ 0 also 1 t x − and 2 t x − ). Following Kim (2006), we assume that the parameters in Eqs. (10)-(12) are time-invariant. The correlation between the standardized residuals t ϕ , t ν and t ζ and the error term t ε is εϕ κ , , εν κ , and , ζ ε κ respectively (note that ϕ σ , ν σ , and ζ σ are the standard errors of t ϕ , t ν and t ζ , respectively). The consistent estimates of the coefficients in Eq. (4) are obtained in two steps. In the first step, we estimate Eqs. (10)-(12) and save the standardized residuals t ϕ , t ν and t ζ . In the second step, we estimate Eq. (13) below along with Eqs. (5)-(9). Note that Eq. (13) now includes bias correction terms, i.e. (standardized) residuals from Eqs. (10)-(12), to address the aforementioned endogeneity of the regressors. Consequently, the estimated parameters in Eq. (13) are consistent, as t ε is uncorrelated with the regressors. ( ) [ ] 2 1 1 1 1 t t t t t t t t t t t , ε ε,t t ν,ε ε t ,ε ε t t r π y r x κ σ κ σ ν κ σ ζ ϕ ρ α β γ ρ δ ζ ϕ ζ + − − − = − + + + + + + + + , ( ) 2 2 2 2 , , , , ~ 0,(1 ) t v t N ε ϕ ε ι ε ε ζ κ κ κ σ − − − (13) As we noted before, instead of the standard framework for second-step estimation, the maximum likelihood estimator via the Kalman filter (Kim, 2006), we use an alternative estimation framework, the “varying coefficients” (VC) method (Schlicht, 1981; Schlicht, 2005; Schlicht and Ludsteck, 2006). This method is a generalization of the ordinary least squares approach that, instead of minimizing the sum of the squares of residuals 2 1 T t= ε ∑ , uses the minimization of the weighted sum of squares: 2 2 2 2 1 1 2 2 n n 1 1 1 1 T T T T t= t= t= t= +θ+θ+θ ζ ϑ ϑ ϑ …+ ∑ ∑ ∑ ∑ (14) where the weights i θ are the inverse variance ratios of the regression residuals t ζ and the shocks in time-varying coefficients t ϑ , that is 2 2 / i i θ=σ σ . This approach balances the fit of the model and the parameter stability. Additionally, the time averages of the regression coefficients, estimated by a weighted least squares estimator, are identical to their GLS estimates of the 12 Note that while a typical time-invariant regression assumes that 1 t t a a − = , in this case it is assumed that [ ] 1 t t E a a − = . 17 corresponding regression with fixed coefficients, that is 1 1 ˆ ˆ T t GLS t= a = a T ∑ . 13 The method is useful in our case because: • (i) it does not require knowledge of initial values even for non-stationary variables prior to the estimation procedure. Instead, both the variance ratios and the coefficients are estimated simultaneously, • (ii) the property of the estimator, that the time averages of estimated time-varying coefficients are equal to its time-invariant counterparts, permits the easy interpretation of the results in relation to time-invariant results, • (iii) it coincides with the MLE estimator via the Kalman filter if the time series are sufficiently long and if the variance ratios are properly estimated. 14 However, this method suffers certain limitations of its own. In particular: (a) it requires that the time-varying coefficients are described as random walks and (b) the shocks in time-varying coefficients t ϑ are minimized (see Eq. (14)). While this does not represent a major problem for the estimation of the coefficients of common variables such as inflation, where the monetary policy response is permanent, it can lead to a loss of some information about ad-hoc response factors in monetary policy-making that are considered by central bankers only infrequently but once they are in place the policy response can be substantial. A financial stress indicator t k x + seems to be this kind of factor. The way to deal with this problem is the estimation-independent calibration of the variance ratios in Eq. (14) such that the estimated coefficient is consistent with economic logic, i.e. it is mostly insignificant and it can turn significant (with no prior restriction on its sign) during the periods of financial stress, i.e. when the financial stress indicator is different from zero. Therefore, we first estimate Eq. (13) using the VC method first and study whether the resulting coefficients at the FSI correspond to economic intuition, and especially whether the coefficient is not constant or slowly moving (a so-called pile-up problem, see Stock and Watson, 1998). When this problem occurs, we compare the results with models where k belongs to (-2, -1, 0, 1, 2) and calibrates the variance ratios in Eq. (13) by the variance ratios 13 See Schlicht and Ludsteck (2006) and Baxa et al. (2010) for more details. 14 The Kalman filter as implemented in common econometric packages typically uses the diffusion of priors for its initiation but it still produces a lot of corner solutions and often it does not achieve convergence. Schlicht and Ludsteck (2006) compare the performance of the moment estimator and the Kalman smoother in terms of the mean squared error on simulated data and they conclude that the moment estimator outperforms the Kalman filter on small samples with a size of up to 100 observations. For comparison, we estimated Eq. (13) using the conventional Kalman filter in the GROCER software using the function tvp (Dubois-Michaux, 2009). We parameterized the model by initial conditions taken from the OLS estimates of the parameters on the full sample and the initial forecast error covariance matrix set to 0. The matrix of the residuals of time-varying coefficients is assumed to be diagonal like in the VC method. The results were very similar to those obtained from the VC method when the estimated variances were the same in both methods. 18 estimated for the model with the largest variances in the FSI. This step was necessary for Australia and Sweden. The Taylor rule coefficients were compared with the initial estimates and they were consistent in both cases. 15 The results of our empirical analysis should reveal whether central banks adjusted its interest rate policy in the face of financial stress. However, the time-varying framework also allows inferring whether any response to financial stress led to the temporal dismissal of other targets, in particular the inflation rate. Therefore, we are mainly interested in the evolution of the financial stress coefficient t δ . We expect it to be mostly insignificant or zero given that episodes of financial stress are rather infrequent and even if they occur the monetary authorities may not always respond to them. Moreover, the size of the estimated coefficient does not have any obvious interpretation since the FSI is a composite indicator normalized to have a zero mean. Consequently, we define the stress effect as a product of the estimated coefficient t δ and the value of IMF financial stress index t k x + . The interpretations of the stress effect is straightforward: it shows the magnitude of interest rate reactions to financial stress in percentage points or, in other words, the deviation from the target interest rate as implied by the macroeconomic variables due to the response to financial stress. 4 Results This section summarizes our results on the effect of financial stress on interest rate setting. First, the results on the effect of the overall measure of financial stress on interest rate setting are presented. Second, the effect of specific components of financial stress on monetary policy is examined. Third, we briefly comment on the monetary policy rule estimates that served as the input for the assessment of financial stress effects. Finally, we perform a series of robustness checks. Figure 2 presents our results on the effect of financial stress on interest rate setting in all five countries (labeled as financial stress effect hereinafter). Although there is some heterogeneity across countries, some global trends on the effect of financial stress are apparent. While in good times, such as in the second half of the 1990s, financial stress has virtually no effect on interest rate 15 Stock and Watson (1998) propose a medium unbiased estimator for variance in the time-varying parameter model but its application is straightforward only in the case of one time-varying coefficient and, more importantly, it requires the variables being stationary. 19 setting or it is slightly positive, 16 the reaction of monetary authorities to financial stress was highly negative during the 2008-2009 global financial crisis. While the previous evidence on the effect of financial stress on monetary policy is somewhat limited, our results broadly confirm the timeinvariant findings of Cecchetti and Li (2008), which show that the US Fed adjusted the interest rates to the procyclical impact of bank capital requirements in 1989-2000. Similarly, Belke and Klose (2010) estimate the Taylor rule on two sub-samples (before and during the 2008-2009 global financial crisis) and find that the Fed reacted systematically not only to inflation and output gap, but to asset prices, credit and money as well. 16 Note that the positive effect of financial stress on interest rate setting is to some extent a consequence of scaling the financial stress indicator; its zero value corresponds to a long run average stress. Hence, we do not place much attention onto the positive values of stress unless caused by the temporarily positive and significant regression coefficient associated with the FSI. 20 Figure 2 – The Effect of Financial Stress on Interest Rate Setting USA -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 UK -2.5 -2 -1.5 -1 -0.5 0 0.5 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Sweden -2 -1 0 1 2 3 4 5 1981 1983 1986 1989 1992 1995 1998 2001 2004 2007 Canada -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 1981 1983 1986 1989 1992 1995 1998 2001 2004 2007 Australia -0,8 -0,6 -0,4 -0,2 0 0,2 0,4 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Notes: The figure depicts the evolution of the financial stress effect. The stress effect (y-axis) is defined as the product of the estimated coefficient on the financial stress indicator in monetary policy rule and the value of the IMF financial stress indicator. The stress effect shows the magnitude of the interest rate reaction to financial stress in percentage points. The size of financial stress effects on interest rate setting during the recent financial crisis is somewhat heterogeneous with the strongest reaction found for the UK The results suggest that all central banks, except the Bank of England, kept its policy rates by about 50-100 basis points 21 lower, as compared to the counterfactual of no reaction to financial stress. The size of this effect for the UK is assessed to be about three times stronger (i.e. 250 basis points). This implies that about 50 percent of the overall policy rate decreased during the recent financial crisis was motivated by the financial stability concerns in the UK (10-30 percent in the remaining sample countries), while the remaining half falls on unfavorable developments in domestic economic activity. This finding is interesting when confronted with the BoE’s very low consideration of expected inflation over the last decade (found Baxa et al., 2010, using the time-varying model and in Taylor and Davradakis, 2006, in the context of the threshold model) that further decreased during the current crisis similarly as for the USA but less so for the other central banks. It is also evident that the magnitude of the response is unusual for all five central banks. Yet, the results for Australia, Canada and Sweden show that a similar magnitude of the response to financial stress recorded during the recent financial crisis was already seen in previous periods of high financial stress. Given that the 2008-2009 global crisis occurred right at the end of our sample (there is a peak in the stress indicator of 5 standard deviations that has not returned to normal values yet), we have performed an additional check to avoid possible end-point bias. In particular, we have run our estimation excluding the observation from the period of 2008-2009 crisis: These results (are available upon request) confirmed the robustness of the reported findings. With regards to the effect of the current crisis, the largest uncertainty is associated with the results for Canada, for which the shortest data sample was available, and it ends in the fourth quarter of 2008. When the possibility of a preemptive reaction of the central bank to financial stress is considered, the effect of financial stress in the current crisis is somewhere at 1-2 percent (see Appendix 3). These additional results suggest the response of the Bank of Canada is rather underestimated. The question of which components of financial stress influence interest rate setting is addressed in Figure 3. Some heterogeneity across countries is again apparent; although it seems that bank stress and stock market stress dominated the central bankers in less open economies. On the other hand, exchange rate stress matters in more open economies such as Canada and Sweden. More specifically, the US Fed seemed to be worried about financial instability, especially during the 1980s. We can see that the main concern in the early 1980s was banking stress, which is arguably related to the Savings and Loans crisis. Another concern was that of stock market stress 22 in particular during the stock market crash of 1987 when interest rates were lower by 30 b.p. with respect to the benchmark case. The Bank of England was, in general, much more perceptive to financial stress. We find its response mainly to stock market stress again notably in 1987. Interestingly, we find little response to exchange rate stress, not even during the 1992 ERM crisis. Nevertheless, it has to be emphasized that the interest rate reaction to speculative attack was subdued in comparison to, for example, the Riksbank (Buiter et al., 1998). The coefficient at the FSI remains constant until the devaluation of the pound sterling in September 1992. Since then the effect of financial stress on interest rate setting approaches zero from originally negative values. Besides this, the response of the Bank of England to inflation has decreased. From this perspective, it seems the pound sterling’s withdrawal from the ERM allowed for both a more rule-based and less restrictive monetary policy. The reaction of the Riksbank to the ERM crisis was different. First, after a series of speculative attacks on the Swedish krona in mid-September 1992, the Riksbank still tried to keep the current fixed exchange rate in place and the marginal interest rate jumped up 500 percent in order to offset the outflow of liquidity and other speculative attacks (see the large positive stress effect on the interest rate in 1992 in Figure 2). However, not even such an increase was sufficient and the fixed exchange rate had to be abandoned later in November. 17 The Reserve Bank of Australia significantly loosened its policy during the 1980s, which can be attributed to the stress in the banking sector with an exception of the reaction to the stock market crash in 1987 (see Figure 3). The exchange rate as well as bank stress seems to matter for interest rate considerations at the Bank of Canada. Interestingly, the results suggest that the Bank of Canada often responded to higher exchange rate stress by monetary tightening. A possible explanation for this finding could be that the Canadian central bank tightened the policy when the currency stabilized at the level that the monetary authority considered to be undervalued. 17 For Sweden, we add a dummy variable for the third quarter of 1992 (ERM crisis) to Eq. 13. At this time the Swedish central bank forced upward short-term interest rates in an effort to keep the krona within the ERM. From the perspective of our model, it was a case of a strong positive reaction to the actual stress that lasted only one period. When this dummy variable was not included, the model with a lagged value of the FSI was unable to show any link between stress and the interest and estimates of other coefficients were inconsistent with economic intuition. 23 We would like to highlight a comparison of Figures 2 and 3. First, it should be noted that the positive response to one stress subcomponent may cancel out against a negative response to another one, making the response to the overall stress negligible (as in the case of Canada). Second, the stress effects related to individual subcomponents need not necessarily sum up to the stress effect related to the entire FSI. All in all, the results suggest that the central bank tends to react to financial stress and different components of financial stress matter in different time periods. The effect of financial stress on interest rate setting is found to be virtually zero in good times and economically sizable during the period of high financial stress. 30 Borio, C. and P. Disyatat (2009): “Unconventional Monetary Policies: An Appraisal”, BIS Working Papers, No. 292. Borio, C. and M. Drehmann (2009): “ Towards an Operational Framework for Financial Stability: ‘Fuzzy’ Measurement and Its Consequences”, BIS Working Papers, No. 284. Brousseau, V. and C. Detken (2001): “Monetary Policy and Fears of Financial Instability.” ECB Working Paper, No. 89. Buiter, W.H., Corsetti, G.M. and P.A. Pesenti (1998): “Interpreting the ERM Crisis: CountrySpecific and Systemic Issues”, Princeton Studies in International Economics, 84, Princeton University. Bulíř, A. and M. Čihák (2008): “Central Bankers’ Dilemma When Banks Are Vulnerable: To Tighten or Not To Tighten?” IMF mimeo. Cardarelli, R., Elekdag, S. and S. Lall (2009): “Financial Stress, Downturns, and Recoveries”, IMF Working Paper, No. 09/100. Carlson, M.A., King, T.B. and K.F. Lewis (2009): “Distress in the financial sector and economic activity”, Finance and Economics Discussion Series 2009-01, Board of Governors of the Federal Reserve System. Cecchetti, S., Genberg, H., Lipsky, J. and S. Wadhwani (2000): “Asset prices and central bank policy”, Geneva Reports on the World Economy, No. 2. Cecchetti, S. and L. Li (2008): “Do Capital Adequacy Requirements Matter for Monetary Policy?”, Economic Inquiry, Vol. 46, No. 4, 643–659. Choi, W: and D Cook (2004): “Liability Dollarization and the Bank Balance Sheet Channel”, Journal of International Economics, 64 (2), 247–275. Christiano, L., Ilut, C., Motto, R. and Rostagno, M. (2008): Monetary Policy and Stock Market Boom-Bust Cycles, ECB Working Paper No. 955. Clarida, R., Galí, J., and M. Gertler (1998): “Monetary Policy Rules in Practice: Some International Evidence”, European Economic Review, No. 42, 1033–1067. Clarida, R., Galí, J., and M. Gertler (1998): “Monetary Policy Rules and Macroeconomic Stability: Evidence and Some Theory”, The Quarterly Journal of Economics, No. 115, 147–180. Cúrdia, V. and M. Woodford (2010): “Credit Spreads and Monetary Policy”, Journal of Money, Credit and Banking, 42, 3-35. . Devereux, M. and A. Sutherland (2007): “Country Portfolios in Open Economy Macro Models”, Journal of the European Economic Association, 5 (2-3), 491–499. Detken, C. and F. Smets (2004): “Asset Price Booms and Monetary Policy”, ECB Working Paper, No. 364. 31 Dubois É. and Michaux E. (2009): “Grocer 1.4: an econometric toolbox for Scilab”, available athttp://dubois.ensae.net/grocer.html. Eichengreen, B. and M. D. Bordo (2002): “Crises Now and Then: What Lessons from the Last Era of Financial Globalization”, NBER Working Paper, No. 8716. Engel, C. and A. Matsumoto (2009): “The International Diversification Puzzle When Prices are Sticky: It’s Really about Exchange-Rate Hedging not Equity Portfolios”, American Economic Journal: Macroeconomics, 1 (2), 155–188. Faia, E. and T. Monacelli (2007): “Optimal Monetary Policy Rules, Asset Prices and Credit Frictions”, Journal of Economic Dynamics and Control, Vol. 31, No. 10, 3228–3254. Fisher, I. (1933): “Debt-deflation Theory of Great Depressions”, Econometrica, Vol. 1, No.4, 337–357. Fuhrer, J. and G. Tootell (2008): “Eyes on the prize: How did the Fed respond to the stock market?”, Journal of Monetary Economics, Vol.55(4), 796–805. Goldstein, M., Kaminsky, G. and C. Reinhart (2000): “Assessing Financial Vulnerability: An Early Warning System for Emerging Markets”, Washington, DC: Institute for International Economics. Goodhart, C., (2006): “A framework for assessing financial stability?”, Journal of Banking and Finance, vol 30 (12), 3415–3422. Goodhart, C., Osorio, C., and D. Tsomocos (2009): “An Analysis of Monetary Policy and Financial Stability: A New Paradigm”, CESifo working paper, No. 2885. Gruen, D., Plumb, M. and A. Stone (2005): “How Should Monetary Policy Respond to AssetPrice Bubbles?”, International Journal of Central Banking, December, 1–31. Hakkiko, C.S. and W.R. Keeton (2009): “Financial Stress: What Is It, How is It to be Measured, and Why Does It Matter?”, Economic Review (2.Q 2009), Federal Reserve Bank of Kansas City, 1–50. Iacovello, M. and S. Neri (2010): “Housing Market Spillovers: Evidence from an Estimated DSGE Model”, American Economic Journal: Macroeconomics, 2(2), 125–164. Illing, M. and Y, Liu, (2006): “Measuring financial stress in a developed country: An application to Canada”, Journal of Financial Stability, 2(3), 243–265. Kaminsky, G. and C. Reinhart (1999): “The Twin Crises: The Causes of Banking and Balance-ofPayments Problems”, American Economic Review, Vol. 89, No. 3, 473–500. Kim, Ch.-J. (2006): “Time-Varying Parameter Models with Endogenous Regressors”, Economics Letters, 91, 21–26. Kim, Ch-J. and Ch. R. Nelson (2006): “Estimation of a Forward-Looking Monetary Policy Rule: A Time-Varying Parameter Model Using Ex-post Data”, Journal of Monetary Economics, No. 53, 1949–1966. 32 Kim, Ch.-J., Kishor, K. and Ch.R. Nelson (2006): “A Time-Varying Parameter Model for a Forward-Looking Monetary Policy Rule based on Real-Time Data”, mimeo. Laeven, L. and F. Valencia (2008): Systemic Banking Crises: A New Database. IMF Working Paper, No. 08/224. Melvin, M. and M. P. Taylor (2009): “The crisis in the foreign exchange market”, Journal of International Money and Finance, 28 (8), 1317–1330. Merton, R. (1974): “On the Pricing of Corporate Debt: The Risk Structure of Interest Rates”, Journal of Finance, 29(2), 449–470. Mishkin, F. (2009): “Is Monetary Policy Effective during Financial Crises?”, American Economic Review Papers & Proceedings, 99(2), 573–577. Mishkin, F. (2010): “Monetary Policy Flexibility, Risk Management and Financial Disruptions”, Journal of Asian Economics, forthcoming. Kiyotaki, N. and J. Moore (1997): “Credit Cycles. The Journal of Political Economy”, Vol. 105, 211-248. Orphanides, A. (2001): “Monetary Policy Rules Based on Real-Time Data”, American Economic Review, 91, 964–985. Posen, A. (2006): “Why Central Banks Should Not Burst Bubbles”, International Finance, 9 (1), 109–124. Reinhart, C. and K. Rogoff (2009): “This Time is Different: A Panoramic View of Eight Centuries of Financial Crises” . Princeton University Press: Princeton. Reis, R. (2010): “Interpreting the Unconventional U.S. Monetary Policy of 2007-09”, NBER Working Paper, No. 15662. Rigobon, R. and B. Sack (2003): “Measuring the Reaction of Monetary Policy to the Stock Market”, The Quarterly Journal of Economics, Vol. 118, No. 2, 639–669. Roubini, N. (2006): “Why Central Banks Should Burst Bubbles”, International Finance, 9 (1), 87– 107. Rudebusch, G. (2006): “Monetary Policy Inertia: Fact or Fiction?”, International Journal of Central Banking, Vol. 2, No. 4, 85–136. Siklos, P. and M. Bohl (2008): “Asset Prices as Indicators of Euro Area Monetary Policy: An Empirical Assessment of Their Role in a Taylor Rule”, Open Economies Review, Vol. 20, No. 1, 39–59. Sims, Ch. and T. Zha (2006): “Were there Regime Switches in US Monetary Policy”, American Economic Review, Vol. 96, No. 1, 54–81. 33 Schlicht, E. (1981): “A Seasonal Adjustment Principle and a Seasonal Adjustment Method Derived From this Principle”, Journal of the American Statistical Association, 76(374), 374–378. Schlicht, E. (2005): “Estimating the Smoothing Parameter in the So-called Hodrick-Prescott Filter”, Journal of the Japan Statistical Society, 35(1), 99–119. Schlicht, E. and J. Lundsteck (2006): “Variance Estimation in a Random Coefficients Model, Department of Economics”, IZA Discussion Paper, No. 2031. Stock, J.H. and M.W. Watson (1998): “Median Unbiased Estimation of Coefficient Variance in a Time-Varying Parameter Model”. Journal of the American Statistical Association, Vol. 93, No. 441 (Mar., 1998), pp. 349–358. Taylor, J.B. (1993): “Discretion versus Policy Rules in Practice”, Carnegie-Rochester Conference Series on Public Policy, 39, 195–214. Taylor, J.B. (Ed.), 1999. Monetary Policy Rules. The University of Chicago Press: Chicago. Taylor, J.B. (2001): “The Role of Exchange Rate in Monetary-Policy Rules”, American Economic Review, Vol.91, No. 2, 263–267. Taylor, J.B. (2008): “Monetary Policy and the State of the Economy”, testimony before the Committee on Financial Services, U.S. House of Representatives, February 26, 2008. Taylor, J.B. and J. Williams (2009): “A Black Swan in the Money Market”, American Economic Journal: Macroeconomics, Vol. 1(1), 58–83. Taylor, M.P. and E. Davradakis (2006): “Interest Rate Setting and Inflation Targeting: Evidence of a Nonlinear Taylor Rule for the United Kingdom”, Studies in Nonlinear Dynamics & Econometrics, 10(4), Article 1. Teranishi, Y. (2009): “Credit Spread and Monetary Policy”, IMES Discussion Paper Series 2009E-14, Bank of Japan. Terrones, M. and Ch. Otrok (2004): “The Global House Price Boom”, In: IMF: World Economic Outlook, Chapter 3, April. Washington: International Monetary Fund. Tovar, C. E. (2009): “DSGE Models and Central Banks”. Economics: The Open-Access, OpenAssessment E-Journal, 3, 200916. Valente, G. (2003): “Monetary Policy Rules and Regime Shift”, Applied Financial Economics, 13, 525–535. Zhang, Ch., Osborn, D.R. and H Dong (2008): “The New Keynesian Phillips Curve: From Sticky Inflation to Sticky Prices”, Journal of Money, Credit and Banking, 40, 667–699. 34 Appendix 1 Figure A.1.1 – Time-Varying Monetary Policy Rules: U.S. Response on inflation -3 -2 -1 0 1 2 3 4 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Response on output gap -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Interest rate smoothing 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Response on financial stress -0.12 -0.1 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Note: The estimated coefficients of time-varying monetary policy rule are depicted with a 95% confidence interval. 35 Figure A.1.2 – Time-Varying Monetary Policy Rules: U.K. Response on inflation -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Response on output gap -0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Interest rate smoothing -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Response on financial stress -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Note: The estimated coefficients of time-varying monetary policy rule are depicted with a 95% confidence interval. 36 Figure A.1.3 – Time-Varying Monetary Policy Rules: Sweden Response on inflation Response on output gap Interest rate smoothing Response on financial stress Note: The estimated coefficients of time-varying monetary policy rule are depicted with a 95% confidence interval. 0 0.5 1 1.5 2 2.5 3 3.5 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 37 Figure A.1.4 – Time-Varying Monetary Policy Rules: Australia Response on inflation 0 0.5 1 1.5 2 2.5 3 3.5 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Response on output gap Interest rate smoothing -0.2 0 0.2 0.4 0.6 0.8 1 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Response on financial stress Note: The estimated coefficients of time-varying monetary policy rule are depicted with a 95% confidence interval. -0.2 0 0.2 0.4 0.6 0.8 1 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 -0.12 -0.1 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 38 Figure A.1.5 – Time-Varying Monetary Policy Rules: Canada Response on inflation 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Response on output gap 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Interest rate smoothing -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Response on financial stress -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Note: The estimated coefficients of time-varying monetary policy rule are depicted with a 95% confidence interval. 39 Appendix 2 The Results with the Interbank Rate as the Dependent Variable in the Policy Rule Figure A2.1 – The Effect of Financial Stress on Interest Rate Setting USA -2.5 -2 -1.5 -1 -0.5 0 0.5 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 UK -2.5 -2 -1.5 -1 -0.5 0 0.5 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Sweden -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 1981 1983 1986 1989 1992 1995 1998 2001 2004 2007 Canada -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 1981 1983 1986 1989 1992 1995 1998 2001 2004 2007 Australia -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 1981 1983 1984 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Notes: The figure depicts the evolution of the financial stress effect. The stress effect (y-axis) is defined as the product of the estimated coefficient on the financial stress indicator in monetary policy rule and the value of the IMF financial stress indicator. The stress effect shows the magnitude of the interest rate reaction to financial stress in percentage points.