Monetary Policy and Shadow Banking

Transcription

1 University of Palermo; University of Coimbra and NIPE; University of Evry Val d Essone; University of Minho, NIPE and London School of Economics. Presentation at the Society for Economic Measurement (SEM) Annual Conference Paris, France July 22-24, 2015

2 Outline Overview 1 Overview

3 Overview Taylor (1993): the monetary authority adjusts the short-term interest rate in response to the current in ation rate and the output gap.

4 Overview Taylor (1993): the monetary authority adjusts the short-term interest rate in response to the current in ation rate and the output gap. Several studies developed di erent versions of the monetary policy rule. Lagged interest rate term due to inertia or interest rate smoothing behaviour (Woodford, 1999), data uncertainty (Orphanides, 2003) or misspeci cation (Rudebusch, 2002). Forward-looking nature of economic behaviour and the importance of in ation targeting (Clarida et al., 1998). Real-time data in the information set of the central bank (Orphanides, 2001).

5 The Taylor rule is still the workhorse of monetary policy. English et al. (2002): policy actions during the Greenspan period are described well by a Taylor rule with inertia. Blinder and Reis (2005): the Fed adjusted the federal funds rate in reaction to changes in the core in ation. Clarida et al. (2000): a forward-looking Taylor rule that links the short-term interest rate to expected future in ation and output gap during the Volcker and the Greenspan tenures. Mehra and Minton (2007): a forward-looking behaviour, a focus on core in ation and smoothed interest rates appear to be the general characterization of the policy rule adopted in the Greenspan Fed.

6 The nancial crisis of raised fundamental questions about the goals of monetary policy and the role that the central bank can have in weathering the various macroeconomic dynamics with which it is continuously confronted. Taylor (2009): excessively expansionary monetary policy in the pre-crisis period helps to explain the emergence and the severity of the nancial turmoil.

7 The nancial crisis of raised fundamental questions about the goals of monetary policy and the role that the central bank can have in weathering the various macroeconomic dynamics with which it is continuously confronted. Taylor (2009): excessively expansionary monetary policy in the pre-crisis period helps to explain the emergence and the severity of the nancial turmoil. Not surprisingly, a vast amount of work suggests that the original framework of Taylor (1993) should be extended to include other variables.

8 Monetary policy and stock prices: Cecchetti et al. (2000) and Chadha et al. (2004): strong response of central banks to variations in asset prices. Bernanke and Gertler (2001): once the predictive content of asset prices for the dynamics of in ation has been controlled for, central banks should avoid to react to asset prices. Monetary authorities should act only to the extent that movements in asset prices might in uence future in ation or damage real economic activity in the aftermath of a nancial bubble burst. Woodford (2003): monetary policy should be designed to achieve macroeconomic stability in the face of nominal frictions. Svensson (2010): monetary policy should focus on price and output stability.

9 Monetary policy and housing prices: Aoki et al. (2004): the importance of housing wealth in providing collateral services to household consumption. Chirinko et al. (2008): housing price shocks play a more relevant role in the design of monetary policy than stock price shocks. Iacoviello and Neri (2010): the sensitivity of residential investment and housing prices to changes in monetary policy measures. Castro and Sousa (2012): while wealth composition drives the formulation of monetary policy, the attempts of central banks to counteract undesirable uctuations in say, nancial wealth, can lead to disruptions in housing wealth.

10 Since the Great Recession, there has been an increasing interest by academics, central banks and policy makers in understanding how central banks actions can impact the dynamics of the balance sheet of nancial intermediaries (Adrian and Shin, 2008; Adrian and Shin, 2010a).

11 Since the Great Recession, there has been an increasing interest by academics, central banks and policy makers in understanding how central banks actions can impact the dynamics of the balance sheet of nancial intermediaries (Adrian and Shin, 2008; Adrian and Shin, 2010a). Some works have also focused on the risk-taking channel of the monetary policy transmission (Borio and Zhu, 2012).

12 Until the eighties, the dominant nancial intermediaries were traditional banks (Boyd and Gertler, 1993).

13 Until the eighties, the dominant nancial intermediaries were traditional banks (Boyd and Gertler, 1993). However, the role of market-based intermediaries increased exponentially over the last decades, and shadow banks, which lack the access to public sources of liquidity, became important providers of funding by translating long-term, opaque and risky assets into short-term liabilities (Pozsar et al., 2010).

14 High leverage ratios, pro-cyclical balance sheets and strong reliance on short-term nancing are other relevant features of shadow banks (Adrian and Shin, 2009, 2010b; Du e, 2010), which became more visible during the 2000s (Kalemli-Ozcan et al., 2011).

15 High leverage ratios, pro-cyclical balance sheets and strong reliance on short-term nancing are other relevant features of shadow banks (Adrian and Shin, 2009, 2010b; Du e, 2010), which became more visible during the 2000s (Kalemli-Ozcan et al., 2011). Guimarães and Zinna (2013): the impact of the balance sheet of nancial intermediaries on bond risk premia. Three causes for their expansion: the "search-for yield" i.e. monetary policy; the savings glut due to global imbalances; and innovation and/or regulation (i.e. incentives). The asset growth of the shadow banking sector contributes to the counter-cyclical price of risk and compressed (increased) risk premia in the run up to (after) the crisis.

16 Only a few studies quantify the impact of monetary policy actions on the balance sheets of nancial intermediaries. Angeloni et al. (2015): monetary policy shocks have a signi cant e ect on di erent measures of bank risks. Den Haan and Sterk (2011): the e ect of monetary policy shocks on consumer credit and home mortgage credit prior to and during the Great Moderation. A tightening of monetary conditions leads to an increase in non-bank mortgages, especially, during the Great Moderation. Nelson et al. (2015): monetary policy surprises have a distinct impact on the balance sheet growth of commercial banks and shadow banks, even though their contribution to the asset growth of the nancial sector as a whole has been small. Unexpected monetary contractions reduce the asset growth of commercial banks and expand the asset growth of shadow banks, as securitization activity rises.

17 Yet, there is still an important gap in:

18 Yet, there is still an important gap in: 1 The literature concerning the response of central banks to developments in those balance sheets; and

19 Yet, there is still an important gap in: 1 The literature concerning the response of central banks to developments in those balance sheets; and 2 The opportunity to include them in the design of monetary policy measures.

20 Yet, there is still an important gap in: 1 The literature concerning the response of central banks to developments in those balance sheets; and 2 The opportunity to include them in the design of monetary policy measures. Should monetary policy be leaning against the wind of the growth in size of securities brokers and dealers and shadow banks?

21 We assess the response of the central bank to the growth rate of the size of non-bank nancial intermediaries.

22 We assess the response of the central bank to the growth rate of the size of non-bank nancial intermediaries. Using quarterly data for the United States over the period 1946:Q1-2014Q4 and both linear and nonlinear econometric frameworks, we nd that an increase in the asset growth of the securities brokers and dealers and the shadow banking sector leads to a rise in the federal funds rate, as it is perceived as posing nancial risks.

23 We assess the response of the central bank to the growth rate of the size of non-bank nancial intermediaries. Using quarterly data for the United States over the period 1946:Q1-2014Q4 and both linear and nonlinear econometric frameworks, we nd that an increase in the asset growth of the securities brokers and dealers and the shadow banking sector leads to a rise in the federal funds rate, as it is perceived as posing nancial risks. When controlling for endogeneity using IV-GMM estimators, our results con rm that in ation is the most important driver of monetary policy actions. However, the central bank is also concerned about the developments in the economic activity and the growth of the assets held by securities brokers and dealers and shadow banks.

24 The estimates of a Switching Transition Regression (STR) model show that two di erent regimes characterize the dynamics of the federal funds rate. Its relationship with the in ation rate, the output gap, the asset growth of securities brokers and dealers and the assets of the shadow banking sector enters di erently according to the regime under consideration, suggesting an active on/o link.

25 The estimates of a Switching Transition Regression (STR) model show that two di erent regimes characterize the dynamics of the federal funds rate. Its relationship with the in ation rate, the output gap, the asset growth of securities brokers and dealers and the assets of the shadow banking sector enters di erently according to the regime under consideration, suggesting an active on/o link. The results of a Markov-Switching Vector Autoregressive (MSVAR) model also indicate that the relationship between the federal funds rate and the asset growth rate of non-bank nancial intermediaries switches across two regimes. While the US monetary policy is generally neutral vis-à-vis changes in the size of non-bank nancial intermediaries, the (sharp) collapse in the growth rate of assets of securities brokers and dealers and shadow banks is associated with a decline of the federal funds rate in times of crises.

26 Outline Overview 1 Overview

27 Linear frameworks DOLS We start by estimating an extended version of the Taylor rule using the DOLS regressor of Stock and Watson (1993): FFRATE t = c + β INFL INFL t + β OG OG t + +β SBRDL SBRDL t + ε SBRDL,t, (1) FFRATE t = c + β INFL INFL t + β OG OG t + +β SHADBNK SHADBNK t + ε SHADBNK,t, (2)

28 where ε SBRDL,t = k i= k + k i= k β INFL,i INFL t+i + k i= k β SBRDL,i SBRDL t+i + ε t, β OG,i OG t+i +

29 and ε SHADBNK,t = k i= k + k i= k β INFL,i INFL t+i + k i= k β OG,i OG t+i + β SHADBNK,i SHADBNK t+i + ε t,

30 FFRATE is the federal funds rate, INFL is the in ation rate, OG is the output gap, SBRDL is the growth rate of the assets of securities brokers and dealers, SHADBNK is the growth rate of the assets of the shadow banking sector, denotes the rst di erence operator, c is a constant, k is the number of leads and lags of the explanatory variables, ε t and is the error term. For the purpose of comparison, we also consider these modi ed versions of the Taylor rule allowing for inertia and with the various regressors entering in lag terms. Finally, we estimate the standard Taylor rule with and without inertia.

31 Linear frameworks IV-GMM To control for the presence of endogeneity in our model, we employ an e cient IV-GMM estimator, where the heteroscedasticity and autocorrelation are controlled for using the HAC procedure.

32 Linear frameworks IV-GMM To control for the presence of endogeneity in our model, we employ an e cient IV-GMM estimator, where the heteroscedasticity and autocorrelation are controlled for using the HAC procedure. As we are using time-series, autocorrelation should be a concern, hence the HAC procedure is suitable to deal with these issues.

33 Nonlinear frameworks STR In order to capture more interaction between the federal funds rate and its main determinants, as well as the growth rate of assets of securities brokers and dealers and the shadow banking sector, we extend equations (1) and (2) to a nonlinear framework.

34 Nonlinear frameworks STR In order to capture more interaction between the federal funds rate and its main determinants, as well as the growth rate of assets of securities brokers and dealers and the shadow banking sector, we extend equations (1) and (2) to a nonlinear framework. Using a particular class of threshold model, i.e. the STR models, we enable the relationship to be asymmetric and to exhibit time-variation and switching regime.

35 Nonlinear frameworks STR In order to capture more interaction between the federal funds rate and its main determinants, as well as the growth rate of assets of securities brokers and dealers and the shadow banking sector, we extend equations (1) and (2) to a nonlinear framework. Using a particular class of threshold model, i.e. the STR models, we enable the relationship to be asymmetric and to exhibit time-variation and switching regime. The STR models introduced by Granger and Teräsvirta (1993) are preferred to the TAR models of Tong and Lim (1980), as they capture not only extreme states, but also a continuum of intermediate regimes for which the transition is smooth.

36 A two-regime STR model for the dynamics of the federal funds rate can be written as follows: FFRATE t = ψ 0 z t + ω 0 z t G (η, c, s t ) + ε t, t = 1,..., T (3) where FFRATE t denotes the monetary policy instrument, z t = (1, INFL, OG t, SBRDL) 0 or z t = (1, INFL, OG t, SHADBNK ) 0 is a vector of explanatory variables.

37 A two-regime STR model for the dynamics of the federal funds rate can be written as follows: FFRATE t = ψ 0 z t + ω 0 z t G (η, c, s t ) + ε t, t = 1,..., T (3) where FFRATE t denotes the monetary policy instrument, z t = (1, INFL, OG t, SBRDL) 0 or z t = (1, INFL, OG t, SHADBNK ) 0 is a vector of explanatory variables. The vectors ψ and ω represent the parameter vectors in the linear and nonlinear parts of the model, respectively.

38 A two-regime STR model for the dynamics of the federal funds rate can be written as follows: FFRATE t = ψ 0 z t + ω 0 z t G (η, c, s t ) + ε t, t = 1,..., T (3) where FFRATE t denotes the monetary policy instrument, z t = (1, INFL, OG t, SBRDL) 0 or z t = (1, INFL, OG t, SHADBNK ) 0 is a vector of explanatory variables. The vectors ψ and ω represent the parameter vectors in the linear and nonlinear parts of the model, respectively. The error-term is assumed to be independent and identically distributed with zero mean and constant variance, ε t N 0, σ 2. G (η, c, s t ) is continuous and bounded between 0 and 1 in the transition variable s, which can be an element of z t or even a linear combination of elements of z t.

39 Granger and Teräsvirta (1993) and Teräsvirta (1994) suggest the (symmetric) Exponential function (ESTR) which identi es a central regime and two upper regimes and corresponds to h G (η, c, s t ) = 1 exp η (s t c) 2i, η > 0, (4) where η indicates the smoothness of the transition between regimes and c determines where the transition occurs.

40 Granger and Teräsvirta (1993) and Teräsvirta (1994) suggest the (symmetric) Exponential function (ESTR) which identi es a central regime and two upper regimes and corresponds to h G (η, c, s t ) = 1 exp η (s t c) 2i, η > 0, (4) where η indicates the smoothness of the transition between regimes and c determines where the transition occurs. The second function is Logistic (LSTR), which enables also to identify two di erent regimes (i.e. an inner regime and an upper regime): G (η, c, s t ) = f1 + exp [ η (s t c)]g 1, η > 0. (5)

41 The speci cation of STR model is carried out in three main steps. 1 Considering the linear benchmark model using information criteria and autocorrelation function. 2 The application of linearity tests (Luukkonen et al., 1988; Teräsvirta, 1994) is then required to test linearity against non-linearity. 3 Sequential Fisher tests (Teräsvirta, 1994) are applied to select the appropriate transition function. The estimation of STR model is done by Nonlinear Least Squares.

42 Nonlinear frameworks MSVAR An alternative approach to capture nonlinear aspects of monetary policy reaction function consists of estimating a Markov-Switching VAR models (MS-VAR). Multiple time series might be subject to regime shifts associated with events such as nancial crises (Jeanne and Masson, 2000; Hamilton, 2005) or abrupt changes in economic policy (Hamilton, 1988; Davig, 2004; Sims and Zha, 2006). Krolzig (1997) establishes taxonomy of models belonging to the MS-VAR class. They can be classi ed depending on which VAR parameters are allowed to shift across regimes.

43 The less restrictive MS-VAR speci cation is the one where the intercept (or mean), the autoregressive coe cients and the variance-covariance matrix are conditioned on the unobservable regime variable s t which indicates the regime prevailing at time t: 8 >< A 01 + p i=1 A i1y t i + Σ e t if s t = 1 Y t =., (6) >: A 0M + p i=1 A im Y t i + Σ 1 2 M e t if s t = M where Y t is the k-dimensional time-series, M stands for the number of regimes, p for the number of lags of the autoregressive terms to take into account and e t NID (0, I k ).

44 The unobservable realization of the regime s t 2 f1,..., Mg is governed by a discrete-state Markov stochastic process, which is de ned by transition probabilities: M p ij = Pr (s t+1 = j j s t = i), p ij = 1 8i, j 2 f1,..., Mg. j=1 (7) More precisely, it is assumed that s t follows an irreducible ergodic M state Markov process with the transition matrix 2 3 p 11 p 12 p 1M p 21 p 22 p 2M P = , (8) p M 1 p M 2 p MM where p im = 1 p i1 p i2... p i,m 1 for i = 1,..., M. Luca Agnello, Vitor Castro, Fredj Jawadi and Monetary Ricardo Policy M. and Shadow SousaBanking

45 In what concerns the estimation techniques, a Maximum likelihood (ML) estimation of the model is based on an implementation of the Expectation-Maximization (EM) algorithm proposed by Hamilton (1990) for this class of model. In our study, Y t is a (4 1) vector of endogenous variable including: the federal funds rate (FFRATE t ), output gap (OG t ), in ation rate (INFL t ) and the growth rate of assets held by securities brokers and dealers (SBRDL t ) or the shadow banking sector (SHADBNK t ).

46 We also assume that M = 2 (i.e. two regimes) and all the parameters of the model are allowed to switch across regimes. This latter assumption is particular important to account for the well-documented shifts in the conduct of US monetary policy during the great moderation (Benati and Surico, 2009). Finally, we remove the stringent assumption of homoscedasticity in the error structure.

47 Outline Overview 1 Overview

48 Overview Quarterly data for the United States over the period 1946:Q1-2014Q4. Policy instrument: the federal funds rate. Board of Governors of the Federal Reserve System.

49 In ation rate: computed using data for the implicit GDP de ator that is generated via the information about the nominal and the real GDP. Bureau of Economic Analysis (BEA). Output gap: computed using data for the real GDP. BEA. This is transformed into natural log terms and the Hodrick-Prescott lter is applied to generate the output gap in percentage of potential output.

50 We follow Adrian and Shin (2010a, b), Adrian et al. (2010) and Berrospide and Edge (2010) and construct the growth rate of the assets of securities brokers and dealers and the growth rate of the assets of the shadow banking sector. from the Flow of Funds Accounts of the Board of Governors of the Federal Reserve System. Securities brokers and dealers are treated separately from shadow banks, which include three types of nancial intermediaries: 1) nance companies; 2) issuers of asset-backed securities (ABS); and 3) funding corporations.

51 1 Finance companies generate loans in a way that is similar to that of commercial banks. 1 Financial intermediaries of "information-problematic" borrowers and typically involved in larger risk-taking behaviour than commercial banks (Carey et al., 1998). 2 Issuers of asset-backed securities special purpose vehicles (SPVs) that hold pools of loans and use them as collateral in their issuances. 3 Funding corporations include subsidiaries of foreign bank and non-bank nancial rms that raise funds in the commercial paper market and transfer them to foreign parent companies abroad or their banking o ces in the United States.

52 Outline Overview 1 Overview

53 Linear frameworks DOLS Both the asset growth of the securities brokers and dealers and the asset growth of the shadow banking sector enter positively and signi cantly in the monetary policy reaction function. The central bank takes into account the size of these nancial intermediaries when setting up the interest rate: an increase in the asset growth of such nancial intermediaries is seen as posing nancial stability risks, thus, the federal funds rate is raised. Similar evidence is obtained when we allow for inertia in the conduct of monetary policy, as both the asset growth of the securities brokers and dealers and the asset growth of the shadow banking sector keep their positive and signi cant coe cients.

54 In what concerns the other regressors, we nd that: the coe cient associated with in ation is positive, signi cant and large in magnitude (monetary policy aggressively counteracts the in ationary pressures by raising the interest rate); the output gap also enters positively and signi cantly the monetary policy reaction function (a widening of the gap between actual output and potential output leads to an increase in the federal funds rate); and the coe cient associated with the lagged interest rate is also positive, relatively large in magnitude and statistically signi cant (inertia is a good characterization pattern of monetary policy conduct).

55 (9)

56 Some authors like Den Haan and Sterk (2011) and Nelson et al. (2015) emphasize that monetary policy actions also have an impact on the growth of non-bank mortgage lending, as well as on securitization activity. To explore this channel, we use the DOLS estimator to estimate the e ect of the federal funds rate on the asset growth of the securities brokers and dealers and the asset growth of the shadow banking sector.

57 In line with the work of Nelson et al. (2015), we nd that a rise in the federal funds rate is associated with an acceleration of the activity of non-bank intermediaries: the coe cient associated with the federal funds rate is positive and signi cant in both the equation for the asset growth of the securities brokers and dealers and the equation for the asset growth of the shadow banking sector; shadow banks tend to counteract the intensi cation of liquidity constraints imposed by tightening monetary policy actions with a potential rise in securitization activity.

58 In line with the work of Nelson et al. (2015), we nd that a rise in the federal funds rate is associated with an acceleration of the activity of non-bank intermediaries: the coe cient associated with the federal funds rate is positive and signi cant in both the equation for the asset growth of the securities brokers and dealers and the equation for the asset growth of the shadow banking sector; shadow banks tend to counteract the intensi cation of liquidity constraints imposed by tightening monetary policy actions with a potential rise in securitization activity. As for the other explanatory variables: an increase in in ation typically has a positive e ect on the activity of the nancial intermediaries under analysis; the coe cient associated with the output gap is also positive, although its statistical signi cance is weak; and the asset growth of the nancial intermediaries displays strong persistence. Luca Agnello, Vitor Castro, Fredj Jawadi and Monetary Ricardo Policy M. and Shadow SousaBanking

59

60 Linear frameworks IV-GMM Presence of heteroscedasticity in all speci cations and endogeneity. The HAC procedure is suitable to deal with serial autocorrelation. Four lags of INFL t and OG t are used as instruments for these variables; The second lag of the federal funds rate (FFRATE t ) is also used as an instrument for its lag, when it is included in the equation. Other variables have proved to be exogenous according to the C-statistic (di erence-in-sargan statistic). The Cragg-Donald statistic shows that the instruments are not weak.

61 Monetary authority reacts signi cantly to in ation and, to a lesser degree, the output gap. When in ation increases, the monetary authority reacts quite strongly by raising the federal funds rate Its reaction to an improvement in the economic conditions has not proved to be so strong or so highly signi cant.

62 Monetary authority reacts signi cantly to in ation and, to a lesser degree, the output gap. When in ation increases, the monetary authority reacts quite strongly by raising the federal funds rate Its reaction to an improvement in the economic conditions has not proved to be so strong or so highly signi cant. Even when endogeneity is controlled for, there is a robust indication that monetary authorities respond to the asset growth of securities brokers and dealers and the asset growth of shadow banks.

63 Monetary authority reacts signi cantly to in ation and, to a lesser degree, the output gap. When in ation increases, the monetary authority reacts quite strongly by raising the federal funds rate Its reaction to an improvement in the economic conditions has not proved to be so strong or so highly signi cant. Even when endogeneity is controlled for, there is a robust indication that monetary authorities respond to the asset growth of securities brokers and dealers and the asset growth of shadow banks. A dynamic model indicates that there is a signi cant degree of persistence, but the main conclusions remain unchanged.

64

65 To account for the fact that monetary authorities may react to in ation and output gap with some lag, we test these extended monetary rules using the rst lag of all variables. monetary authorities react not only to lagged in ation and output gap, but also to the lagged e ects of the asset growth of securities brokers and dealers and the asset growth of shadow banks.

66 To account for the fact that monetary authorities may react to in ation and output gap with some lag, we test these extended monetary rules using the rst lag of all variables. monetary authorities react not only to lagged in ation and output gap, but also to the lagged e ects of the asset growth of securities brokers and dealers and the asset growth of shadow banks. Usual ndings with a Taylor rule: monetary authorities react strongly to in ation and, to a lesser degree, the output gap, which is clearer when the persistence in the process is controlled for.

67 To account for the fact that monetary authorities may react to in ation and output gap with some lag, we test these extended monetary rules using the rst lag of all variables. monetary authorities react not only to lagged in ation and output gap, but also to the lagged e ects of the asset growth of securities brokers and dealers and the asset growth of shadow banks. Usual ndings with a Taylor rule: monetary authorities react strongly to in ation and, to a lesser degree, the output gap, which is clearer when the persistence in the process is controlled for. These simple Taylor rules should be extended with information from securities brokers and dealers and shadow banks. the central bank indeed incorporates it in the decisions about the interest rate.

68 Next, we explore a bit further the reaction of monetary authorities, allowing for di erent estimators to control for endogeneity and a di erent set of instruments. Our benchmark model considers all regressors lagged one period and uses as instruments lags 2 to 4 of the federal funds rate, in ation, output gap and nancial variables. Endogeneity remains an issue, especially when the two-step GMM HAC procedure is employed. Monetary authorities react strongly to in ation (and, to a lesser degree, the output gap) and the growth in brokers and dealers assets and shadow banking assets.

69

70 We also test for the possibility of reverse e ects on the nancial variables. We replace the dependent variable by either the asset growth rate of securities brokers and dealers or the asset growth rate of the shadow banking sector. No evidence of endogeneity is found in this case, which means that we do not need to incur in this complexity to estimate this di erent equation. Despite the persistence of each nancial variable, only the output gap has a signi cant (negative) impact on the behaviour of the asset growth of securities brokers and dealers and shadow banks. Over time, a better economic environment tends to undermine such growth.

71

72 Nonlinear frameworks STR In line with Granger and Teräsvirta (1993) and Teräsvirta (1994), we rst specify the linear model and apply linearity tests. We start by assessing the null hypothesis of linearity against its alternative of a STR type of nonlinearity using Luukkonen et al. (1988) linearity tests. We nd that the most appropriate linear model should include one lag, which enables to capture inertia in the monetary policy reaction function.

73 The linearity is rejected for several transition variables, but the strongest rejection takes place in the case of the output gap (at the 1% signi cance level), suggesting it as a potential transition variable.

74 Next, we check the suitable transition function, by testing linearity against nonlinearity of exponential or logistic type and choosing the one that displays the strongest evidence of linearity rejection (Teräsvirta, 1994). The logistic function is preferred to the exponential function for both models, so we focus on the LSTR speci cation.

75

76 Regarding the model with brokers and dealers: In the rst regime, the central bank increases its policy rate after a rise in in ation and the output gap, as well as when it observes an expansion of the size of nancial intermediaries. In the second regime, there is also a response to the asset growth of securities brokers and dealers, but the adjustment of the federal funds rate is faster.

77 Regarding the model with brokers and dealers: In the rst regime, the central bank increases its policy rate after a rise in in ation and the output gap, as well as when it observes an expansion of the size of nancial intermediaries. In the second regime, there is also a response to the asset growth of securities brokers and dealers, but the adjustment of the federal funds rate is faster. As for the LSTR model including the shadow banking sector: In regime 1, the reaction of monetary policy towards the change in the in ation rate is still positive and signi cant, but the output gap no longer plays a role. An increase in the asset growth of shadow banks enters with a positive and signi cant coe cient. In regime 2, when the output gap exceeds the threshold level, there is a switch in the monetary policy rule, as an increase of the in ation rate implies a cut in the federal funds rate.

78 Regarding the estimates of nonlinear parameters, both the threshold and the transition speed estimates are statistically signi cant. The values of the transition speed show that the transition is sharper when considering the asset growth of the shadow banking sector. Moreover, while the model with securities brokers and dealers implies that the switch between regimes occurs in 1997:Q4, the model with shadow banks suggests that it occurs later (i.e. 2007:Q1).

79 Quite similar for both models.

80 Two distinct regimes that characterize the federal funds rate.

81 Nonlinear frameworks MSVAR We estimate two MS-VAR models where the growth rates of nancial assets of securities brokers and dealers and shadow banks are alternatively treated as endogenous variables. Each equation can be interpreted as an augmented US monetary policy reaction which explains whether and how the balance sheet dynamics of non-bank nancial intermediaries imply a response by the central bank.

82 Figure 3 depicts the smoothed probabilities of regime one as (endogenously) identi ed from the two MS-VAR models. The red bars denote the US recession periods as dated by the National Bureau of Economic Research (NBER).

83 The dating of regime 1 is broadly the same for both the two model speci cations. It encompasses the four major nancial crises over the last 40 years in the US: 1 the commercial bank capital squeeze in ; 2 the savings and loan crisis (S&L) in ; 3 the burst of the dotcom bubble in ; and 4 the nancial crisis in In most of the times, and, notably, since the eighties, regime 1 and US recessionary episodes tend to overlap.

84

85 As expected, the asset growth of securities brokers and dealers and shadow banks share the same pattern, with the former being more volatile than the latter. At a rst glance, both series tend to decline during regime 1 and to moderately increase during regime 2 and notably in the run-up to nancial crises. With regard to the dynamics of federal funds rate, if we exclude the period of the seventies, we observe that the most signi cant cuts in the federal funds rate occurred during regime 1, while monetary tightening tends to occur during regime 2.

86 Overall, the US monetary policy behaves in a counter-cyclical way. The signs and the statistical signi cance of the coe cients associated with in ation and output gap suggest that, during both regimes, the Federal Reserve takes action by dampening in ationary pressures or boosting growth when the economy is lagging.

87 Overall, the US monetary policy behaves in a counter-cyclical way. The signs and the statistical signi cance of the coe cients associated with in ation and output gap suggest that, during both regimes, the Federal Reserve takes action by dampening in ationary pressures or boosting growth when the economy is lagging. Moreover, the estimates across regimes indicate that the federal funds rate adjusts to uctuations in the asset growth rate of non-bank nancial intermediaries in a non-linear fashion: In regime 2, there is no statistical evidence of a causal relationship going from asset growth to interest rates (i.e. US monetary policy is neutral vis-à-vis changes in asset values) In regime 1, the (sharp) collapse of asset growth rates a ecting brokers and dealers and shadow banks is associated with a decline in the federal funds rate. Luca Agnello, Vitor Castro, Fredj Jawadi and Monetary Ricardo Policy M. and Shadow SousaBanking

88 Even though, our estimates indicate that part of such monetary policy corrections are consistent with the policy strategy that prescribes an accommodative stance during bad times (i.e. negative output gap and low in ation), the statistical signi cance of the asset growth variables and the positive sign associated with their coe cients also suggests that the Fed further focuses on the dynamics of these nancial intermediaries as they gauge nancial risks. Thus, in line with the so-called "leaning against the wind" policy view, nancial stability objectives appear to lead to the adoption of a more aggressive posture by the central bank: in response to negative shocks, as re ected in the sharp deterioration of nancial intermediaries balance sheets, interest rate cuts are deeper than otherwise.

89

90 Given that all the variables entering our MS-VAR models are treated as endogenous, a reversal causal relationship running from changes in the federal funds rate to the balance sheet growth of nancial intermediaries cannot be ruled out ex-ante. To shed some light on this potential linkage, we now focus on the equations that explain the dynamics of asset growth of securities brokers and dealers and shadow banks. We nd that a two-way (reversal causality) relationship exists, but is mainly con ned to a particular regime, i.e. regime 1 (when we consider the shadow banking sector) and regime 2 (when we account for securities brokers and dealers).

91 We show that, in both cases, the interest rate policy has a positive impact on the dynamics of the asset growth. In regime 2, monetary contractions tend to expand the size of the shadow banking sector instead of reducing it. In regime 1, monetary easing reduces the activity of securities brokers and dealers.

92 We show that, in both cases, the interest rate policy has a positive impact on the dynamics of the asset growth. In regime 2, monetary contractions tend to expand the size of the shadow banking sector instead of reducing it. In regime 1, monetary easing reduces the activity of securities brokers and dealers. Our empirical ndings are in line with the work of Haan and Sterk (2011). Following a monetary policy contraction, non-bank mortgages increase, as opposed to standard bank mortgages which display a signi cant reduction. Nelson et al. (2015) also document this result as a waterbed e ect, whereby commercial banks circumvent tighter funding liquidity constraints following a contractionary monetary policy shock by increasing their securitization activity.

93

94 Outline Overview 1 Overview

95 Overview In this paper, we investigate the response of the central bank to the growth rate of the size of non-bank nancial intermediaries.

96 Overview In this paper, we investigate the response of the central bank to the growth rate of the size of non-bank nancial intermediaries. Using quarterly data for the United States over the period 1946:Q1-2014Q4 and both linear and nonlinear econometric frameworks, we nd that an increase in the asset growth of the securities brokers and dealers and the shadow banking sector leads to a rise in the federal funds rate, as it is perceived as posing nancial risks.

97 When controlling for endogeneity, using IV-GMM estimators, our results con rm that in ation is the most important target for the monetary authority, but it is also concerned over the developments in the economic activity and nancial variables, such as the asset growth of securities brokers and dealers and the shadow banks.

98 When controlling for endogeneity, using IV-GMM estimators, our results con rm that in ation is the most important target for the monetary authority, but it is also concerned over the developments in the economic activity and nancial variables, such as the asset growth of securities brokers and dealers and the shadow banks. In particular, the central bank reacts by raising the interest rates to increases in those assets.

99 When controlling for endogeneity, using IV-GMM estimators, our results con rm that in ation is the most important target for the monetary authority, but it is also concerned over the developments in the economic activity and nancial variables, such as the asset growth of securities brokers and dealers and the shadow banks. In particular, the central bank reacts by raising the interest rates to increases in those assets. Hence, we argue that the monetary rule should be extended with information about the size of the balance sheet of securities brokers and dealers and shadow banks.

100 From the STR modeling, we nd a threshold relationship between the federal funds rate and the asset growth of shadow banks, suggesting a monetary policy reaction function that changes by regime with regard to variations in the in ation rate, the output gap and the asset growth of non-bank nancial intermediaries.

101 From the STR modeling, we nd a threshold relationship between the federal funds rate and the asset growth of shadow banks, suggesting a monetary policy reaction function that changes by regime with regard to variations in the in ation rate, the output gap and the asset growth of non-bank nancial intermediaries. In particular, when the output gap is low, the Fed adjusts its policy rate vis-à-vis changes in the asset growth of securities brokers and dealers; when the output gap is high, the monetary authority is more concerned about the dynamics of in ation.

102 The results of the MS-VAR model indicate that, in bad times, the Fed focuses on the dynamics of the size of the shadow banking sector, as it gauges nancial risks.

103 The results of the MS-VAR model indicate that, in bad times, the Fed focuses on the dynamics of the size of the shadow banking sector, as it gauges nancial risks. In line with the so-called "leaning against the wind" policy view, the mandate of nancial stability increases the aggressiveness of monetary policy.

104 The results of the MS-VAR model indicate that, in bad times, the Fed focuses on the dynamics of the size of the shadow banking sector, as it gauges nancial risks. In line with the so-called "leaning against the wind" policy view, the mandate of nancial stability increases the aggressiveness of monetary policy. Thus, in response to a sharp deterioration of the balance sheet of non-bank nancial intermediaries, interest rate cuts are deeper than otherwise.

105 END