Monetary shocks, macroprudential shocks and financial stability

Economic Modelling 56 (2016) 11–24

Contents lists available at ScienceDirect

Economic Modelling journal homepage: www.elsevier.com/locate/ecmod

Monetary shocks, macroprudential shocks and financial stability夽 Matthew Greenwood-Nimmo a, * , Artur Tarassow b,1 a b

Department of Economics, University of Melbourne, Australia Faculty of Economics and Social Sciences, Department of Socioeconomics, University of Hamburg, Germany

A R T I C L E

I N F O

Article history: Received 19 November 2014 Received in revised form 29 February 2016 Accepted 3 March 2016 Available online xxxx JEL classification: C32 C51 C54

A B S T R A C T This paper examines the implications of monetary shocks and macroprudential shocks for aggregate financial fragility using a sign restricted VAR model estimated with US data spanning the period 1960Q1–2007Q4. Contractionary monetary shocks are found to exacerbate financial fragility, increasing both the credit to GDP ratio and the ‘financial ratio’, which is the ratio of firms’ debts to their internal funds. By contrast, when interest rates are fixed, credit-constraining macroprudential shocks may be able to reduce the credit to GDP ratio in the short run but are not able to reduce the financial ratio. However, when the interest rate is free to accommodate the macroprudential shock, both the credit to GDP ratio and the financial ratio decline, indicating a reduction of financial fragility and suggesting that there may be gains from a coordinated approach to macroeconomic management. © 2016 Elsevier B.V. All rights reserved.

Keywords: Financial stability Monetary policy Macroprudential policy Sign restrictions

“In these latter days, since the downfall, I know that there will be much talk of corruption and dishonesty. But I can testify that our trouble was not that. Rather, we were undone by our own extravagant folly, and our delusions of grandeur. The gods were waiting to destroy us, and first they infected us with a peculiar and virulent sort of madness.” [Anonymous (1933)] 1. Introduction Economic history has seen repeated booms and busts in the asset markets which seem neither predictable nor avoidable ex ante. A crude generalisation is that investors undertake progressively

夽 We are grateful for the constructive comments of the Editor and four anonymous referees and for the helpful discussion of Ulrich Fritsche, Ingrid Größl, Viet Hoang Nguyen, Yongcheol Shin and Tomasz Wozniak. ´ Our estimation routines employ Ambrogio Cesa-Bianchi’s VAR Toolbox for Matlab. Any errors or omissions are our own. * Corresponding author at: 3.12 Faculty of Business & Economics, University of Melbourne, 111 Barry Street, Carlton, VIC3053, Australia. Tel.: +61 3 8344 5354. E-mail addresses: [email protected] (M. Greenwood-Nimmo), [email protected] (A. Tarassow). 1

Tel.: +49 40 42838 8683.

http://dx.doi.org/10.1016/j.econmod.2016.03.003 0264-9993/© 2016 Elsevier B.V. All rights reserved.

more risky positions as rising speculative profits fuel an increasingly bullish outlook until confidence in the sustainability of asset prices eventually fails and the bubble collapses. Subsequently, many commentators are left wondering how so many investors, seasoned and novice alike, were swept up in an ex-post unsustainable clamour to realise speculative gains based largely on market euphoria. The historical inability of market participants to prevent the growth and subsequent collapse of bubbles has been well documented. This led to a lively debate in the years before the Global Financial Crisis (GFC) as to whether the central bank should — and indeed could — formulate monetary policy to intervene in financial markets (e.g. Cecchetti et al., 2000; Nickell, 2005; Posen, 2006; Roubini, 2006). The dominant view was that a sufficiently aggressive inflation-targeting policy could stabilise both output and inflation in the face of asset price volatility driven either by bubbles or by technology shocks or a combination of the two (Bernanke and Gertler, 2001). Consequently, it was generally believed that monetary policy should only respond to asset prices indirectly via their influence on the optimal inflation forecast. Rather than adjusting interest rates in the hope of preemptively deflating a nascent bubble, the central bank should act to ‘mitigate the fallout’ ex post in the event that the bubble were to burst (Greenspan, 2002). The primary responsibility of the central bank was therefore to maintain price stability which was, in turn, believed to beget financial stability (Schwartz, 1998).

12

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24

Consequently, financial shocks were not a primary concern of central banks and financial regulation mainly operated at the firm level rather than the systemic level. This consensus has largely dissolved following the GFC and the unprecedented macroeconomic policy response that it instigated. Constrained by the zero lower bound (ZLB), policymakers in many countries employed a mix of countercyclical fiscal policy and unconventional monetary policy. The Federal Reserve was quick to undertake large-scale asset purchases, manipulating its balance sheet as an implement of unconventional monetary policy (Jawadi et al., in press). Forward guidance also emerged as a prominent tool for guiding interest rate expectations in the anticipation of an eventual normalisation of interest rates. Interest rate normalisation will not, however, entail a simple return to prior policy arrangements. The policy framework which emerges must adapt to reflect Blanchard et al.’s (2010) observation that the maintenance of price stability is necessary but not sufficient to deliver macroeconomic stability. Indeed, Christiano et al. (2010) show that narrow inflation-targeting may actually exacerbate financial cycles. Their argument is predicated on the observation that asset market booms are not typically associated with high inflation as one would expect under the logic of Bernanke and Gertler (2001). Rather, over the last 200 years, asset booms in the US have been overwhelmingly associated with low inflation. In this environment, narrow inflation-targeting policies may deliver undesirably low interest rates, fueling the boom. It is well established that interest rate policy is non-neutral with respect to financial stability. For example, the credit channel literature stresses that transaction costs, informational asymmetries between borrowers and lenders and creditors’ risk aversion against insolvency may collectively generate financial frictions in imperfect capital markets (Bernanke and Gertler, 1995). An interest rate hike is likely to reduce loan supply and thereby initiate a flight-to-quality which will constrain the borrowing power of smaller firms (Gertler and Gilchrist, 1994; Kashyap and Stein, 1997). In addition, Christiano et al. (1996) show that a monetary contraction will typically reduce both aggregate demand and aggregate cash-inflows, thereby undermining the net worth of the representative borrower and increasing the probability of default, a combined effect that will generate an increased external financing premium (Bernanke et al., 1996). Consequently, both the cost of credit and the conditions governing its supply move in accordance with monetary policy, with the result that the contractionary influence of a rate hike will be concentrated disproportionately among smaller and more informationally opaque firms. Minsky’s (1982) financial instability hypothesis goes a step further, stressing that the effects of a monetary tightening are not felt only at the idiosyncratic level but also at the systemic level.2 Empirical evidence consistent with this view has been provided by Mallick and Sousa (2013), who document a strongly positive association between monetary tightenings and financial stress. Minsky holds that the link between monetary policy and financial fragility arises because as the central bank changes the interest rate in accordance with its policy objectives, it also changes the cash-commitments of leveraged firms in an imperfectly predictable manner. In an uncertain world, firms faced with long-lived and irreversible investment

2 The financial instability hypothesis offers a number of insights into the emergence of financial fragility, yet references to Minsky’s work are scarce in the current debate. This may, in part, reflect the absence of a canonical Minskyan model, a lacuna which has led to the establishment of several different interpretations of Minsky’s work. Our discussion is similar to that of Fazzari et al. (2008), who assert that a central bank may actually precipitate financial crises by pursuing active monetary policy. We are grateful to an anonymous referee for pointing out that alternative interpretations variously emphasise the role of commercial bank behaviour and of asset prices as drivers of cyclical behaviour (Ryoo, 2013; Skott, 1995).

decisions engage in forward planning based on optimal forecasts of future conditions which, owing to this very uncertainty, must be heavily conditioned on recent historic experience. A key decision facing firms is the choice of financing structure, with firms undertaking either hedge, speculative or Ponzi financing (Minsky, 1986). Following Sordi and Vercelli (2006) and Vercelli (2011), these financing structures can be defined with reference to the current and intertemporal financial ratios, kit and k∗it : h  

kit =

eit zit

and

k∗it =

n=0

(1 + q)−n e∗it+n

h  

n=0

∗ (1 + q)−n zit+n

 

where eit represents cash-outflows, zit denotes cash-inflows, an asterisk signifies an expected value, q is the discount rate and the subscripts i = 1, 2, . . . , N and t = 1, 2, . . . , T identify firms and time periods, respectively. For any horizon, h, the ith firm is hedge financing if kit < 1 for t = 0 and k∗it < 1 for 1 ≤ t ≤ h. It is engaged in speculative financing if kit > 1 for t = 0 and k∗it > 1 for t ∈ [1, . . . , s] provided that s < h is a relatively short horizon and k∗it < 1 for t ∈ [s + 1, . . . , h]. Finally, it is Ponzi financing if kit > 1 for t = 0 and k∗it > 1 for 1 ≤ t ≤ h − 1 under the expectation that k∗it << 1 in period t = h. Hedge financing is the most prudent strategy when faced with unanticipated shocks, while Ponzi financing involves a considerable risk of insolvency. An unforeseen interest rate hike is likely to raise cash-outflows (by raising the cost of debtservicing) while simultaneously reducing cash-inflows (by reducing aggregate activity), resulting in a rightward shift through the hedgespeculative-Ponzi spectrum and increasing financial fragility at the aggregate level. Acknowledging the financial stability implications of monetary policy, several commentators have broken with the prior consensus that asset prices should not enter the interest rate rule. For example, in light of their observation that a narrow inflation targeting policy is likely to deliver undesirably low interest rates during the growth phase of an asset market boom, Christiano et al. (2010) suggest that credit growth should be assigned an independent role in the interest rate rule beyond its influence on the inflation forecast. Similar reasoning underlies a fast-growing literature, both theoretical and empirical, which has sought to augment both monetary and fiscal policy reaction functions with a variety of asset price and wealth indicators (e.g. Agnello et al., 2012; Castro and Sousa, 2012; Mendicino and Punzi, 2014). In addition to reconsidering the use of existing policy tools, the literature is increasingly emphasising alternative instruments in light of the remarkable broadening of the policy mix brought about by the GFC. With the gradual withdrawal of quantitative easing, macroprudential policies aimed at limiting excessive credit growth and restraining asset price inflation are set to play a key role in mitigating the emergence of financial fragility in the future (Elliott et al., 2013). Macroprudential policies to curtail excessive credit creation and to maintain the creditworthiness of borrowers may be directed at either lenders, borrowers or both (Claessens et al., 2014). On the lenders’ side, countercyclical capital requirements of the type proposed in the Basel III Accord can curtail unsafe lending and protect the portfolios of financial institutions from large corrections in the value of collateral assets. Meanwhile, on the borrowers’ side, capping the loan-to-value and/or debt-to-income ratios can limit the potential for the emergence of Ponzi financing and strengthen borrowers’ incentives to manage funds responsibly by increasing their own stake in debt-funded projects, while also reducing bank losses in the event of default. Federal Reserve Chair Janet Yellen (2014) has indicated in a recent lecture at the International Monetary Fund that a judicious mix

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24

of monetary and macroprudential policies is likely to emerge as the preferred approach to macroeconomic management. The use of macroprudential policies to directly target credit markets provides a valuable complement to monetary policy and gives policymakers greater latitude to achieve multiple goals simultaneously. Based on a New Keynesian model with nominal rigidities and credit frictions, De Paoli and Paustian (2013) show that a combined approach to policymaking has the potential to be welfare-enhancing. However, developing an operational policy framework will be challenging. For instance, De Paoli and Paustian caution that difficult issues of policy coordination may arise due to the interaction of monetary and macroprudential policies, particularly where the monetary and macroprudential policy instruments work in a similar fashion. In this paper, we therefore study the effects of both monetary shocks and macroprudential shocks on aggregate financial fragility in the US over the period 1960Q1–2007Q4. To this end, we estimate a monetary VAR model augmented to include the prime lending rate, real credit and real internal funds of US non-financial corporate businesses and the real S&P 500 index as well as the key variables usually included in a monetary VAR — the federal funds rate, real output, the general price level and nonborrowed reserves. To identify monetary and macroprudential shocks, we adopt the pure sign restrictions approach of Uhlig (2005). Following a wellestablished literature, we identify a contractionary monetary shock as one which does not decrease the federal funds rate and which does not increase nonborrowed reserves, output or inflation. Meanwhile, in the absence of a strong precedent, we employ several identification strategies to define a ‘credit-constraining’ macroprudential shock, all of which stress that the shock does not increase credit and stock prices and does not reduce nonborrowed reserves. With the shocks identified, we then infer impulse response functions for the credit to GDP ratio and the corporate financial ratio (the ratio of corporate credit to internal funds, which is a proxy for Sordi and Vercelli’s (2006) current financial ratio) in order to evaluate the financial stability implications of both monetary and macroprudential policies. Our results indicate that a contractionary monetary shock will raise the credit to GDP ratio and drive a wedge between firms’ internal funds and the extent of their debt. Hence, as found by Mallick and Sousa (2013), a monetary tightening will exacerbate financial fragility. By contrast, a credit-constraining macroprudential shock in the absence of interest rate adjustments may be able to reduce the credit to GDP ratio in the short run but is unable to reduce the financial ratio. However, when the interest rate is free to accommodate the macroprudential shock, both the credit to GDP ratio and the financial ratio decline, indicating an unambiguous reduction of financial fragility. The interaction of monetary and macroprudential policy raises the possibility that a combined policy framework may deliver the best outcomes in terms of financial stability, although the attendant issues of policy coordination must be handled with care (De Paoli and Paustian, 2013). The paper proceeds in 4 sections. Section 2 introduces our empirical methodology, while our estimation results including several robustness tests are presented and discussed in detail in Section 3. Section 4 concludes. Details of the dataset and its construction may be found in the Data Appendix.

2. Empirical methodology 2.1. VAR specification Our aim is to build a parsimonious macroeconometric model with which to evaluate the effect of monetary and macroprudential shocks on financial fragility in the aggregate. To this end, we estimate

13

a pth order VAR of the form:

Y t = b 0 + bd t +

p 

bi Y t−i + 4t

(1)

i=1

where Yt is a vector of endogenous variables observed over periods t = 1, 2, . . . , T, b0 is a vector of intercepts, bd is a vector of deterministic trend terms, the bi  s are matrices of autoregressive parameters and 4t is a zero-mean error process with positive definite covariance matrix S. As we will be working with quarterly data, we follow the widespread rule-of-thumb and adopt a lag length of p = 4. The vector Yt contains  the following eight variables: the effective fedf eral funds rate it ; the log of real credit to non-financial corporate   businesses (ct ); the prime lending rate ilt ; the log of real internal funds of non-financial corporate businesses ( ft ); the log of real nonborrowed reserves (rt ); the log of real GDP ( yt ); the log of the GDP deflator (pt ); and the log of the real S&P 500 index (qt ). A detailed summary of our data sources and of the transformations applied to each series may be found in the Data Appendix. It is common practice in the literature to include the federal funds rate, output, the price index and nonborrowed reserves within a monetary VAR model. The additional variables that we include support our sign identification strategy for the macroprudential shock (as discussed below) and also allow us to study several key indicators of aggregate financial fragility. First, following Uhlig (2005), we compute the rate of  inflation (pt ) and thereafter the real prime lending rate ilt − pt+1 to measure the effective stance of monetary policy. Second, we define the credit to GDP ratio (ct − yt ) to measure the degree of credit extension in the economy. High and increasing values of the credit to GDP ratio are often interpreted in the literature as signs of a credit boom (Mallick and Sousa, 2013). Third, to measure the ability of the representative firm to meet its debt-servicing obligations, we define the aggregate financial ratio, (ct − ft ), which measures the extent of firms’ borrowing relative to their internal funds. Under the assumption that credit and internal funds can be used to proxy for the cash-outflows and cash-inflows of firms, respectively, an increase (decrease) in the financial ratio will be associated with an increase (decrease) in Sordi and Vercelli’s (2006) current financial ratio for the representative firm. Therefore, an increase (decrease) in the financial ratio signals increasing (decreasing) financial fragility in the aggregate. As we will show below, it is possible to infer impulse responses for the inflation rate, the real prime lending rate, the credit to GDP ratio and the financial ratio even though none of these terms enter our VAR model directly. 2.2. Sign restricted VAR As is common in the VAR literature, we are not directly interested in the eight reduced form innovations in 4t . By contrast, we are interested in just two mutually independent fundamental shocks — a monetary shock and a macroprudential shock. Traditional approaches to the identification of fundamental shocks include the Wold-causal identification routine of Sims (1986), the use of shortrun restrictions as in Blanchard and Watson (1986) and the use of long-run restrictions following Blanchard and Quah (1989). These traditional approaches invoke strong assumptions, including recursivity and variable exclusion (zero) restrictions. More recently, the sign restrictions identification approach associated with Uhlig (2005) has gained traction in the literature. Sign restrictions are generally considered to be weaker than traditional identifying restrictions because they are based on inequality constraints as opposed to point restrictions and they yield results which are invariant to the ordering of the variables in the system. Furthermore, sign restrictions allow for the agnostic identification of shocks in a manner which imposes minimal structure on the impulse responses of the variables

14

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24

in the system to the identified shocks. Consequently, we employ sign restrictions in this paper. Consider the following relationship between the one-step ahead prediction error in Eq. (1), 4t , and the set of fundamental shocks, vt :

responses for the variables in Yt . First, following Uhlig (2005), the hstep ahead impulse response for the inflation rate, Rp, a (h), may be computed as follows: Rp, a (h) = Rp, a (h) − Rp, a (h − 4)

4t = Avt

where A is an identifying matrix. Since our VAR model contains 8 variables, 4t is an 8 × 1 vector. Furthermore, we assume that vt is also an 8 × 1 vector with E (vt vt ) = I 8 and that A is an 8 × 8 matrix. Our interest is limited to the identification of just two fundamental innovations. Hence, we need only identify two of the column vectors within A, which we shall denote amon and amac and which correspond to the monetary shock and the macroprudential shock, respectively. These two column vectors — known as impulse vectors — define the impact effect of the respective shock on each of the 8 variables in the VAR system. Following Uhlig (2005), it is common to use the Cholesky decomposition to define the matrix A such that AA = S, although this choice is not central to the resulting identification. Having defined A, Uhlig shows that if there exists a conformable vector of unit length, m, then the corresponding impulse vector is a = Am. Consequently, given the impulse vector a, the corresponding impulse response at the h-period horizon, Ra (h), is simply a linear combination of the impulse responses obtained under the Cholesky decomposition of S:

Ra (h) =

(4)

(2)

8 

mi Ri (h)

(3)

i=1

where mi is the ith element of m and Ri (h) is a vector containing the h-period ahead impulse responses with respect to the i-th shock in the Cholesky decomposition of S. Following Mountford and Uhlig (2009), the generalisation to two or more impulse vectors is straightforward. Uhlig (2005) proposes two Bayesian routines to implement the sign restrictions, both of which account for the sampling uncertainty of the least squares estimates of the reduced form coefficient matrices (the B s) and the reduced form covariance matrix (S) as well as the non-exact identification of the impulse vectors. Under the pure sign restrictions approach, all impulse vectors that satisfy the sign restrictions are given equal weight, while under the penalty function approach the best impulse vector is chosen by minimising a criterion function. While the penalty function approach may yield narrower intervals around the impulse response functions, we employ the pure sign restrictions approach here as we wish to remain agnostic concerning the set of impulse responses arising from our sign restrictions. In other words, we do not wish to impose any additional assumptions — whether implicit or explicit — beyond those which are transparently stated in the setting of our sign restrictions. Estimation proceeds by first drawing from a Normal-Wishart posterior for (B, S) and computing the Cholesky factor A. For each of these draws, one draws repeatedly from a uniform distribution over the unit sphere for m and each time computes the impulse vector a. The resulting impulse responses are tested against the proposed sign restrictions — if the sign restrictions are not violated then the draw is retained, otherwise it is discarded. The process is repeated until a sufficiently large number of draws have been retained. Analysis and inference can then proceed on the basis of the set of retained draws. As mentioned above, we are particularly interested  in evaluating the response of the real interest rate ilt − pt+1 , the credit to GDP ratio (ct − yt ) and the financial ratio (ct − ft ) to both monetary and macroprudential shocks. Even though none of these quantities directly enter our specification, it is possible to infer impulse responses in each case as linear combinations of the impulse

with Rp, a (h) = 0 for h < 0. Using Rp, a (h), we may then compute the h-step ahead impulse response for the real interest rate as follows: Rir , a (h) = Ril , a (h) − Rp, a (h + 1)

(5)

where Rp, a (h + 1) represents the rational expectations forecast of inflation in the next quarter. Next, we approximate the impulse response for the change in the credit to GDP ratio as: R(c−y), a (h) =

1 + Rc, a (h) −1 1 + Ry, a (h)

(6)

while the impulse response for the change in the financial ratio is approximated by: R(c−f ), a (h) =

1 + Rc, a (h) −1 1 + Rf , a (h)

(7)

In each case, the inferred impulse responses are computed draw-bydraw and so the construction of intervals around them proceeds as usual. 2.3. Identifying restrictions A large body of research has attempted to identify monetary shocks using sign restrictions but macroprudential shocks have yet to receive much attention in this literature. Table 1 reports the set of sign restrictions that we use to establish a baseline. Each of these restrictions is imposed both on impact and in the subsequent two quarters. This is consistent with the established precedent arising from the literature. In his original paper, Uhlig (2005) imposes sign restrictions over 6 months (2 quarters) and many subsequent applications have followed suit, including Mallick and Sousa (2013). In keeping with the monetary VAR literature, we identify a contractionary monetary shock as one which does not decrease the federal funds rate and which does not increase nonborrowed reserves or inflation. This is an uncontroversial identification strategy which is similar in spirit to that of Uhlig (2005) but with the additional restriction that output does not increase following a contractionary monetary shock, in line with a large theoretical and applied literature. We employ a similar approach to identify a credit-constraining macroprudential shock. The macroprudential shock that we have in mind can be thought of as a policy-initiated adverse credit supply shock. Specifically, it is a shock which should deflate a creditfueled bubble by constraining growth in credit and asset prices while simultaneously encouraging banks to increase their reserve holdings. Hence, to establish a baseline, we impose the restrictions that a credit-constraining macroprudential shock does not increase credit or stock prices and does not reduce nonborrowed reserves. In the absence of a strong precedent for the sign identification of a macroprudential shock, we experiment with several alternative and more stringent identification schemes in Section 3.4. The nature of these Table 1 Identifying sign restrictions. f

it Monetary shock Macroprudential shock

ct

+ −

ilt

ft

rt

yt

pt

− +

−

−

qt −

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24

restrictions is guided by the literature which has studied macroprudential policies using dynamic stochastic general equilibrium (DSGE) models (e.g Akram, 2014; Alpanda et al., 2014; Bailliu et al., 2015; Brzoza-Brzezina et al., 2014, 2013, 2015). Based on a careful reading of this literature — and, in particular, on consideration of the impulse responses derived from these DSGE models — we consider combinations of restrictions in which: (i) output is constrained not to rise following a macroprudential shock; and (ii) the federal funds rate is constrained not to rise and the prime lending not to fall after a macroprudential shock. Our results are robust to these changes. It is important to note that, by design, our identification scheme imposes no direct restrictions on the inferred impulse responses of the credit to GDP ratio and the financial ratio, which are to be our key measures of financial fragility. Furthermore, we impose only minimal identifying restrictions on their components.3 In all cases, the restrictions that we do impose are necessary for the proper identification of the shocks (for example, the restriction that credit should not increase following a credit-constraining macroprudential shock is essential to our definition of the shock). Overall, therefore, our approach is as close as possible to Uhlig’s (2005) notion of agnostic identification and we take all reasonable steps to ensure that our results are not merely a reflection of the structure that we impose on the model. 3. Estimation results Our sample covers the period 1960Q1–2007Q4, ending before the switch to unconventional monetary policy in the US. Due to the absence of monthly data for several series — notably GDP and the GDP deflator — we use quarterly data to estimate our model. Details of the data sources including series codes and any required transformations are collected in the Data Appendix. Our choice to end the sample in 2007Q4 is made in light of both theoretical and practical concerns. Monetary policy since the GFC has been characterised by a ZLB environment in which a combination of quantitative easing and forward guidance has emerged as the preferred policy. Hofmann and Bogdanova (2012) stress that no systematic relationship can be discerned between the federal funds rate and the majority of standard macroeconomic indicators at this time. Consequently, estimation over the ZLB period is likely to hinder our efforts to identify a contractionary monetary shock. This becomes a concrete issue when one studies the data pertaining to the unprecedented accumulation of bank reserves since the GFC. While the value of nonborrowed reserves averaged just US$29.2 bn between 1960Q1 and 2007Q4, it increased by a factor of almost 60 to reach a value of US$1725 bn by 2013Q1. Consequently, setting theoretical concerns aside, it would be infeasible to estimate our model beyond 2007Q4 due to the presence of a profound structural break late in the sample. 3.1. A contractionary monetary shock Fig. 1 reports the impulse responses of each variable in the system with respect to a contractionary monetary shock. The impulse responses are computed using 10,000 retained draws and are shown alongside their respective 68% intervals over a horizon of 20 quarters. In addition, the final row of the figure reports corresponding graphs for the inferred impulse responses of the inflation rate, the real interest rate, the credit to GDP ratio and the financial ratio.

3 For example, our baseline restrictions state that real credit does not rise following a credit-constraining macroprudential shock but this does not restrict the sign of the response of the credit to GDP ratio. In fact, even if both real credit and real GDP are jointly restricted not to rise following a credit-constraining macroprudential shock (as in two of the robustness exercises that we shall consider in Section 3.4), this still does not impose any restriction on the sign of the response of the credit to GDP ratio.

15

In light of our identification scheme, it follows that the shock is associated with an immediate increase in the federal funds rate and decreases in nonborrowed reserves, output and the price level. In practice, the funds rate response dies away relatively rapidly and subsequently reverses sign. Uhlig (2005) observes the same phenomenon and attributes it to the Federal Reserve’s attempts to correct the interest rate error implied by the initial monetary shock. An equivalent sign-reversal can be seen in the responses of nonborrowed reserves and output, both of which dip in the short-run due to the contractionary monetary shock but subsequently recover as the Federal Reserve undertakes corrective action. Meanwhile, the price level falls in a sustained manner before stabilising at a new lower level after approximately 15 quarters. Consequently, the rate of inflation also displays a reversal, falling significantly for 15 quarters before returning to its previous level. Interestingly, our results indicate no significant response of the stock market to the monetary shock at any horizon, although the median impulse response is mildly negative in the immediate wake of the shock. However, given that financial markets typically react very rapidly to news, it is possible that our quarterly sampling frequency masks a more pronounced short-run response of the stock market to the monetary shock.4 Moving onto the key credit variables, the response of the prime lending rate tracks that of the funds rate relatively closely as one would expect. Consequently, the inferred impulse response for the real prime lending rate shows a marked increase in the short run before falling after approximately 5 quarters, in line with the Federal Reserve’s correction. It is not surprising that the emergence of lower interest rates after 5 quarters acts to stimulate borrowing after 5 quarters. What is more notable is that we observe an expansion of real credit in the period immediately following the shock despite higher short-run borrowing costs at this time. This accords with Christiano et al.’s (1996) observation that firms are unable to quickly adjust their expenditure following a contractionary monetary shock. Firms may be obliged to borrow more in the immediate wake of a monetary contraction for at least two reasons. First, firms may borrow to meet their debt servicing obligations, which are likely to rise with higher interest rates. Second, firms may borrow to address a shortfall in their income resulting from the contractionary influence of the shock, a shortfall which is at least partially reflected in the decline that can be seen in the impulse response for firms’ real internal funds. The combination of an expansion of real credit with a contraction of real GDP leads to an unambiguous increase in the credit to GDP ratio. A similar result has recently been documented by Mallick and Sousa (2013). The expansion of real credit is sufficiently marked that the credit to GDP ratio remains elevated even after real GDP growth is resumed as a result of the Federal Reserve’s policy correction. Furthermore, the credit expansion is accompanied by a strong and sustained increase in the financial ratio, indicating a reduction in the average firm’s ability to service its debt and the prevelance of increasingly fragile financing arrangements (Sordi and Vercelli, 2006).5 The combination of rapid credit growth and an increase in

4 To further explore the response of the stock market to a contractionary monetary shock, we re-estimated the model using the valuation ratio instead of the real S&P500 index. It is reasonable to expect that the valuation ratio may react to a monetary shock in a more prolonged manner than real stock prices. The results support this view, indicating a two-thirds probability that the valuation ratio will have adjusted within the range 0% to −1.5% after 20 quarters have passed. Full results of this exercise are available on request. 5 Recall the definitions of the current and intertemporal financial ratios offered by Sordi and Vercelli (2006). The combination of increasing cash-outflows (proxied by credit growth) and falling cash-inflows (proxied by falling internal funds) in both the short- and the long-run will cause both kit and k∗it to increase for the representative firm. Consequently, at the aggregate level, a monetary contraction will be reflected in a general shift through the hedge-speculative-Ponzi spectrum.

16 M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24 Fig. 1. Impulse response functions following a contractionary monetary shock. Notes: This figure shows the impulse responses to a contractionary monetary shock alongside their respective 68% intervals. The figure is generated from 10,000 draws that satisfy the sign restrictions detailed in Table 1 both on impact and in the following two quarters. The horizontal axis shows the time horizon in quarters.

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24

the financial ratio is indicative of a credit bubble. Consequently, our results indicate that a contractionary monetary shock is likely to erode financial stability. 3.2. A credit-constraining macroprudential shock Fig. 2 replicates Fig. 1 in the case of a credit-constraining macroprudential shock. Recall that, under our baseline identification scheme, we define the macroprudential shock as one which does not lead to real credit or asset price growth and which does not reduce nonborrowed reserves. By construction, the macroprudential shock causes a contraction of real credit and real stock prices in the short run. The stock market response dies away after approximately 4 quarters while the response of credit is sustained for 10 quarters. Similarly, nonborrowed reserves increase on impact by construction and then remain elevated for 16 quarters following the shock. This pattern is consistent with a policy intervention which requires lenders to amass additional reserves and curb their credit creation. The link between reduced lending and reduced asset prices may arise through both direct and indirect channels. A direct effect would naturally arise if credit funds speculation, in which case a reduction in the availability of credit will reduce speculative activity. An indirect effect may arise if investors perceive the shock as a signal that markets are overheated and adjust their positions accordingly. The macroprudential shock is mildly contractionary in the shortrun and it exerts a disinflationary influence after a lag of approximately 5 quarters. These two observations are important because they collectively help to explain the behaviour of the federal funds rate, which declines in response to the shock. This suggests that a monetary accommodation occurs alongside the macroprudential shock. This effect does not arise mechanically due to our identifying sign restrictions — recall that the only restrictions that we impose to identify the macroprudential shock are on credit, the stock market and nonborrowed reserves. Rather, as described by Alpanda et al. (2014) in the context of a macrofinancial DSGE model, the Federal Reserve lowers the funds rate in accordance with its reaction function in order to counteract the emergence of disinflationary and contractionary pressures. The reduction in the funds rate is rapidly reflected in both the nominal and the real prime lending rate, indicating a reduction in the cost of debt-servicing. Furthermore, the shock does not exert a strong influence over internal funds, causing only a mild and shortlived increase in the medium term. This is a key result because, unlike in the case of a monetary contraction, it indicates that the nonborrowed funds available to firms are not decreasing in response to an otherwise contractionary shock. Consequently, the financial ratio falls as a result of the credit-constraining macroprudential shock, indicating an improvement in firms’ aggregate financial health (Sordi and Vercelli, 2006). Furthermore, the shock achieves a reduction in the credit to GDP ratio. Collectively, these results suggest that credit-constraining macroprudential policy in conjunction with expansionary monetary policy may be effective in calming credit booms and in reducing financial fragility. 3.3. Macroprudential policy at the zero lower bound Several recent papers have observed that macroprudential policy is typically implemented in conjunction with monetary policy (e.g. Akinci and Olmstead-Rumsey, 2015). As shown by De Paoli and Paustian (2013), the interactions between monetary policy and macroprudential policy may be complex. The results presented in Fig. 2 provide an interesting example. On one hand, a monetary accommodation in response to a credit-constraining macroprudential shock may be undesirable as it dampens the effect of the macroprudential shock. On the other hand, the combination of a credit-constraining macroprudential shock and an interest rate

17

accommodation may prove a suitable policy formulation if one wishes to reduce credit growth without adversely affecting the cost of servicing debt. A natural question arising from Fig. 2 is what effect a creditconstraining macroprudential shock would have in the absence of any interest rate adjustment. This is a particularly compelling question in the context of policymaking in proximity to the ZLB, where there is no possibility of any reduction in the interest rate. To study this case, Fig. 3 presents the results of a counterfactual experiment in which we shut down the impulse responses of both the federal funds rate and the prime lending rate so that the nominal cost of credit remains constant in the wake of the shock.6 In the absence of interest rate adjustment, the macroprudential shock is considerably more contractionary and is also more disinflationary at longer horizons. Since the nominal prime lending rate remains unchanged, this disinflation translates into a higher real rate of interest on loans after approximately 10 quarters. All else equal, a higher real interest rate may be expected to depress real activity, real credit and real asset prices. However, despite brief contractions to each of these variables in the short-run, we find that output growth resumes after the first year alongside renewed growth in real credit and real asset prices. Consequently, after an initial fall, the credit to GDP ratio increases markedly, indicating a credit boom. Furthermore, our results indicate a relatively high probability of a sustained negative response of real internal funds, which contributes to an increase in the financial ratio at longer horizons. To interpret these results, it is useful to envisage the scenario in which, throughout our sample period, a macroprudential authority sets macroprudential policy in accordance with a reaction function, much like the Federal Reserve in the case of monetary policy. Consequently, just as Uhlig (2005) notes that a monetary shock can be viewed as a monetary policy error which the Federal Reserve will strive to correct, so a credit-constraining macroprudential shock can be viewed as a policy error that the macroprudential authority will attempt to correct. Recall Uhlig’s observation that, after a brief lag, the Federal Reserve will engage in accommodatory monetary policy in order to correct a contractionary monetary shock and that this is reflected in the changing sign of the funds rate impulse response. The same logic can be applied here. A credit-constraining macroprudential shock is associated with a reduction in credit and an increase in reserves. Viewing the shock as an unanticipated policy error, the macroprudential authority will act to correct it, which will be reflected in a change toward credit expansion and a disaccumulation of reserves. The overall picture emerging from Fig. 3 supports the notion that macroprudential policy can be used in isolation to influence macroeconomic conditions, exerting a non-negligible influence over credit growth and broader measures of economic activity. In particular, our results suggest that a credit-constraining macroprudential shock may be able to reduce the credit to GDP ratio in the short run, thereby cooling a credit boom. It is to be expected that macroprudential policies should be able to influence credit aggregates as macroprudential policy can directly affect banks’ balance sheets (Claessens et al., 2014). However, the macroprudential shock does not appear to be effective in reducing the financial ratio. This implies that a macroprudential intervention may fail to reduce the debt burden of the representative firm if firms elect to run down their internal funds to offset the reduced availability of credit. Overall, therefore, a comparison of Figs. 2 and 3 suggests that a coordinated approach to monetary and macroprudential policymaking may yield superior

6 Of course, one must remain cognisant of the Lucas critique when conducting such an exercise. Nonetheless, the results provide a useful framework within which to think about the issues involved in the use of macroprudential policy in a ZLB environment.

18 M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24 Fig. 2. Impulse response functions following a credit-constraining macroprudential shock. Notes: This figure shows the impulse responses to a credit-constraining macroprudential shock alongside their respective 68% intervals. The figure is generated from 10,000 draws that satisfy the sign restrictions detailed in Table 1 both on impact and in the following two quarters. The horizontal axis shows the time horizon in quarters.

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24 Fig. 3. Impulse response functions following a credit-constraining macroprudential shock conditional on no interest rate adjustment. Notes: This figure shows the impulse responses to a credit-constraining macroprudential shock alongside their respective 68% intervals in the case where neither the federal funds rate nor the prime lending rate can adjust in response to the shock. The figure is generated from 10,000 draws that satisfy the sign restrictions detailed in Table 1 both on impact and in the following two quarters. The horizontal axis shows the time horizon in quarters.

19

20

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24

outcomes in terms of economic and financial stability (De Paoli and Paustian, 2013).

Table 2 Alternative identifying sign restrictions for the macroprudential shock. f

3.4. Alternative identification schemes for the macroprudential shock As noted above, while the literature provides a strong precedent for the identification of monetary shocks via sign restrictions, the same is not true of macroprudential shocks. To test the sensitivity of our results to the chosen identification scheme, we therefore consider three alternative sets of identifying sign restrictions which are derived from careful consideration of the DSGE literature. Table 2 presents the restrictions used to identify the credit-constraining macroprudential shock in each case alongside those used to obtain our baseline results for ease of comparison. In each of cases I–III, we impose additional binding restrictions relative to our baseline case in order to achieve a more stringent identification scheme. In general, the computational time required in estimation will increase as additional binding sign restrictions are implemented. Therefore, our robustness exercises are each based on 1000 retained draws instead of the 10,000 retained draws used to generate our baseline results. The first case that we consider is motivated by several recent papers in the DSGE literature which record an immediate contraction of output in response to a credit-constraining macroprudential shock (e.g. Brzoza-Brzezina et al., 2014, 2015). To this end, in case I we impose the additional restriction that a macroprudential shock should not prove expansionary with respect to output either on impact or in the subsequent two quarters. This contrasts with our baseline results presented in Fig. 2, where a credit-constraining macroprudential shock is contractionary only after a lag, not on impact. Next, we focus on the dynamics of the two interest rate series in our model. By definition, the type of macroprudential shock that we consider leads to an immediate reduction in real credit. As discussed above, such a reduction in lending activity is likely to prove contractionary and, consequently, it may lead the central bank to reduce its nominal policy rate (as seen in Fig. 2). However, several recent studies have argued that lenders are likely to raise the interest rates that they demand on new loans in response to credit-constraining macroprudential policies (e.g. Akram, 2014; Alpanda et al., 2014; Bailliu et al., 2015). This may occur if a macroprudential shock increases the funding costs faced by lenders who then pass some or all of these costs on to borrowers. Similarly, the lending rate may rise if a macroprudential shock leads to a reduction in borrowers’ net worth and a commensurate increase in their default probability. To reflect the differential behaviour of the base rate and the rate of interest on loans, in cases II and III (which are identical aside from the treatment of output) we stipulate that a credit-constraining macroprudential shock must not raise the federal funds rate nor lower the nominal lending rate on impact.7 In practice, it is infeasible to impose these two restrictions jointly over any horizon longer than the quarter in which the shock occurs. This is to be expected in light of the data — during our sample, there have been just thirteen quarters in which the federal funds rate and the prime lending rate have moved in opposite directions and only one occasion where this has occurred in two consecutive quarters (in 1981Q2–Q3, during the Volcker disinflation). Consequently, our imposition of restrictions on the federal funds rate and the prime lending rate only on impact is consistent with the time series properties of the interest rate data. Furthermore, note that it is not necessary for the prime lending rate and the federal funds rate to move in opposite directions over prolonged periods

7 A desirable feature in this case is that the combination of reduced credit extension coupled with a higher price of credit is indicative of a credit supply shock (i.e. a leftward move of the credit supply curve in the context of a standard supply-anddemand framework). Such an identification strategy was recently applied by Busch et al. (2010) studying loan supply shocks in Germany and Hristov et al. (2012) analysing loan supply shocks in the Euro area using panel data.

Baseline Case I Case II Case III

it

ct

 

− − − −

ilt

⊕ ⊕

ft

rt + + + +

yt − −

pt

qt − − − −

Notes: As in Table 1, sign restrictions denoted by ‘+’ and ‘−’ apply on impact and during the following two quarters. By contrast, ‘⊕’ (‘’) implies that the relevant variable does not decrease (increase) on impact only. The restrictions used to identify the monetary shock are the same as in Table 1 in all cases and so they are not shown here to conserve space.

in order to generate a considerable degree of variation in the prime rate–funds rate spread. Indeed, over our sample, the spread averages 1.92% with a sample standard deviation of 1.05%. Fig. 4 summarises the key results in each case. Given that our identification of the monetary shock is unchanged, the figure only reports impulse responses with respect to the macroprudential shock. Furthermore, to conserve space, the figure focuses on the inferred impulse response functions for the real interest rate, the credit to GDP ratio and the financial ratio — a full set of impulse responses is available on request. In addition, to demonstrate the effect of the retrictions applied to the federal funds rate and the prime lending rate in cases II and III, the figure reports implied impulse response functions for the prime rate–funds rate spread, which are computed as R(il −if ), a (h) = Ril , a (h) − Rif , a (h). Our principal findings remain intact under the alternative identification schemes. Firstly, in all cases, a credit-constraining macroprudential shock is met by a fall in the real interest rate on loans in the medium-term. In case I, the real interest response is very similar to the baseline case. Meanwhile, under cases II and III, the reduction in the real interest rate is slightly more subdued and it arises with a lag due to the imposition of a binding positive sign restriction on the prime lending rate on impact. The impact of the interest rate restrictions in cases II and III on the prime rate–funds rate spread is marked, with the spread increasing by approximately 20 basis points on impact and remaining elevated for more than a year thereafter. By contrast, in case I where no restrictions are imposed on the interest rates, although the median impulse response suggests an increase in the prime–funds rate spread, the effect is weaker and the width of the 68% interval indicates a high degree of uncertainty. Secondly, as in the baseline case, a credit-constraining macroprudential shock leads to a reduction in the credit to GDP ratio after roughly one year in all cases. The effect is slightly more gradual under case I than in the baseline case because the macroprudential shock now exerts an immediate contractionary effect on GDP by definition which, ceteris paribus, would raise the credit to GDP ratio. Meanwhile, the credit to GDP ratio falls somewhat more deeply in cases II and III as a result of a higher real interest rate on loans. Finally, the financial ratio decreases after roughly one year in all cases, showing little sensitivity to the choice of identification scheme. Overall, therefore, we conclude that our results do not depend critically on the sign restrictions that we employ to identify the macroprudential shock.8 As a final exercise, Fig. 5 reports historical decompositions with respect to our two identified shocks. To conserve space, the analysis focuses on the three variables which underpin our key measures

8 In addition to the robustness tests reported here, we also conducted an additional robustness exercise based on the Minskyan literature. Much of this literature stresses that it is chiefly nominal cash-outflows and inflows that are of practical relevance to firms, not their real values (e.g. Caskey and Fazzari, 1987). To test whether our results are robust to the use of nominal data, we re-estimated our model entirely using nominal GDP, nominal credit, nominal internal funds, nominal nonborrowed reserves and the nominal stock price in place of their real counterparts. For this test, we retained our baseline sign restrictions. Our central findings remain intact using either real or nominal magnitudes. Results of this exercise are available on request.

(b) Case II: The prime rate does not fall and the funds rate does not rise

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24

(a) Case I: Output does not rise

(c) Case III: The prime rate does not fall; the funds rate and output do not rise Fig. 4. Alternative sign restrictions for the identification of the credit-constraining macroprudential shock. Notes: This figure reports selected impulse responses with respect to a credit-constraining macroprudential shock identified under the alternative sign restrictions detailed in Table 2. 68% intervals are reported in each panel. The results are based on 1000 retained draws. The horizontal axis shows the time horizon in quarters.

21

22

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24

(a) Baseline

(b) Case I: Output does not rise

(c) Case II: The prime rate does not fall and the funds rate does not rise

(d) Case III: The prime rate does not fall; the funds rate and output do not rise Fig. 5. Historical decompositions for key variables, comparing identification schemes. Notes: The figure reports historical decompositions for real credit, real internal funds and real GDP, having controlled for their respective deterministic components. Results are presented for each of the identification schemes detailed in Table 2. The historical contribution of the monetary shock is shown in black, the contribution of the macroprudential shock is shown in dark gray and the remaining variation is shown in light gray.

of financial conditions: real credit, real internal funds and real GDP.9 Furthermore, to provide a sense of the relative importance of our identified shocks, we also report the remaining variation in each variable which is not due to either of our identified shocks.

9 These three variables are the inputs required to construct the credit to GDP ratio and the financial ratio. A complete set of historical decompositions is available on request.

Given that we only identify two shocks, the variation due to unidentified shocks is often the largest component of the historical decompositions, as one may expect. However, regardless of which identification scheme is considered, the monetary and macroprudential shocks explain a non-negligible proportion of the variation, particularly in the case of real credit and real GDP. Under each identification scheme, the monetary shock plays a considerably more important role than the macroprudential shock and displays clear cyclicality. Furthermore, the role of the monetary shock appears

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24

to increase around the time of the Volcker disinflation, perhaps reflecting the transition to a new monetary policy regime at this time. The time variation in the role of the macroprudential shock is interesting. It is important to acknowledge that macroprudential policy has not been used systematically over our sample period. However, as noted above, the credit-constraining macroprudential shock that we consider can be thought of as a deliberate policyinitiated adverse credit supply shock. Consequently, it is reasonable to expect that our identification scheme should capture major credit supply shocks that have occurred throughout our sample period. This is precisely what we see in Fig. 5, where our identification scheme captures several apparent credit supply shocks, including notable adverse shocks in recent years during the recessions of the early 1990s and 2000s and at the onset of the subprime mortgage crisis.

results of which may provide valuable insights into the cyclical limits of macroprudential policy. Appendix A. Data Appendix The series used in estimation were sourced from the Federal Reserve Economic Data Service (FRED), the Flow of Funds Accounts Release Z1 (FoF) and Robert Shiller’s website. The variables are defined as follows: f

it

ct

ilt 4. Concluding remarks This paper studies the effect of monetary and macroprudential shocks on financial fragility in the US over the period 1960Q1– 2007Q4. Financial fragility is measured using both the credit to GDP ratio and the corporate financial ratio, which we define as the ratio of corporate credit to internal funds. The former is a measure of credit extension while the latter captures the extent of the debt burden facing firms. Our analysis is based on an extended monetary VAR model where both a contractionary monetary shock and a creditconstraining macroprudential shock are identified using the pure sign restrictions approach developed by Uhlig (2005). This has the benefit of identifying the shocks of interest while imposing minimal structure on the response of either the credit to GDP ratio or the financial ratio. Our results indicate that a contractionary monetary shock is likely to exacerbate financial fragility, increasing both the credit to GDP ratio and the financial ratio. A similar result has been documented by Mallick and Sousa (2013). By contrast, a credit-constraining macroprudential shock in the absence of any interest rate adjustment may reduce the credit to GDP ratio in the short run but is unlikely to similarly reduce the financial ratio. Consequently, our results indicate that macroprudential policy in isolation is likely to have ambiguous effects on financial fragility. However, as noted by Akinci and Olmstead-Rumsey (2015), macroprudential policy is seldom used in isolation. When interest rates are free to adjust in response to the macroprudential shock, both the credit to GDP ratio and the financial ratio decline substantially, indicating an unambiguous reduction of financial fragility. This finding suggests that a combined monetary and macroprudential approach to the pursuit of financial stability may be desirable, in line with the view espoused by Yellen (2014). However, as shown by De Paoli and Paustian (2013), the attendant issues of policy coordination must be handled very carefully. Our paper provides a foundation for continuing research and we conclude by mentioning two promising avenues. First, by broadening the specification of the model, it may be possible to separately identify several distinct types of macroprudential shock, such as loan-tovalue restrictions on the borrowers’ side and capital requirements on the lenders’ side. Although this would involve the estimation of a considerably larger model and the identification of the relevant shocks would be highly challenging, it would represent a valuable addition to the burgeoning literature on the efficacy of different macroprudential policy tools (e.g. Claessens et al., 2014). Second, as noted by Cerutti et al. (2015), macroprudential policy is typically employed to dampen the upturn of financial cycles rather than to soften their downturn and it may be considerably more effective at the former than the latter. In principle, one could accommodate such asymmetry via the use of a threshold VAR model, for example, the

23

ft

rt

yt pt qt

is the effective federal funds rate (FRED: FEDFUNDS, NSA), which is converted from monthly to quarterly frequency by taking the period average. is the log of nonfinancial corporate business loans (FRED: NCBLILQ027S, NSA). The series is deflated using the GDP deflator before being logged. is the bank prime loans rate (FRED: MPRIME, NSA), which is converted from monthly to quarterly frequency by taking the period average. is the log of the book value of U.S. internal funds of the nonfinancial corporate sector (FoF: FA106000135.Q, SA). The series is deflated by the GDP deflator before being logged. is the log of nonborrowed reserves of depository institutions (FRED: BOGNONBR, SA), converted from monthly to quarterly frequency. The series is deflated by the GDP deflator before being logged. is the log of real GDP (FRED: GDPC96, SA). is the log of the implicit price deflator based at 2009=100 (FRED: GDPDEF, SA). is the log of the nominal S&P500 composite price index downloaded from Robert Shiller’s website. The series is deflated by the GDP deflator before being logged.

(N)SA denotes that a series is (not) seasonally adjusted. We do not seasonally adjust the interest rate or stock market data. With the exception of nonfinancial corporate business loans, the remaining series are downloaded in seasonally adjusted form. Seasonally adjusted data for the level of nonfinancial corporate business loans is not available from FRED. However, the correlation between the unadjusted series and an adjusted version computed using Census X12 is 0.99994 over our sample period. Consequently, we work with the unadjusted data in this case.

References Agnello, L., Castro, V., Sousa, R.M., 2012. How does fiscal policy react to wealth composition and asset prices?. J. Macroecon. 34, 874–890. Akinci, O., Olmstead-Rumsey, J., 2015, May. How effective are macroprudential policies? An empirical investigation. Discussion Paper 1136, Federal Reserve Board of Governors, Washington D.C. Akram, Q.F., 2014. Macro effects of capital requirements and macroprudential policy. Econ. Model. 42, 77–93. Alpanda, S., Cateau, G., Meh, C., 2014. A policy model to analyze macroprudential regulations and monetary policy. BIS Working Paper 461, Bank for International Settlements, Basel. Anonymous, 1933. The royal road to bankruptcy. Atlantic Mag. 49–54. CLI(I). Bailliu, J., Meh, C., Zhang, Y., 2015. Macroprudential rules and monetary policy when financial frictions matter. Econ. Model. 50, 148–161. Bernanke, B., Gertler, M., 1995. Inside the black box: the credit channel of monetary policy transmission. J. Econ. Perspect. 9, 27–48. Bernanke, B., Gertler, M., 2001. Should central banks respond to movements in asset prices?. Am. Econ. Rev. 91, 253–257. Bernanke, B., Gertler, M., Gilchrist, S., 1996. The financial accelerator and the flight to quality. Rev. Econ. Stat. 78, 1–15. Blanchard, O., Dell’Ariccia, G., Mauro, P., 2010. Rethinking macroeconomic policy. J. Money, Credit, Bank. 42, 199–215. Blanchard, O., Quah, D., 1989. The dynamic effects of aggregate demand and supply disturbances. Am. Econ. Rev. 79, 655–673. Blanchard, O., Watson, M., 1986. Are all business cycles alike?. In: Gordon, R. (Ed.), The American Business Cycle. University of Chicago Press, Chicago, pp. 123–160.

24

M. Greenwood-Nimmo, A. Tarassow / Economic Modelling 56 (2016) 11–24

Brzoza-Brzezina, M., Gelain, P., Kolasa, M., 2014. Monetary and macroprudential policy with multi-period loans. Working Paper 16/2014. Norges Bank, Oslo. Brzoza-Brzezina, M., Kolasa, M., Makarski, K., 2013. The anatomy of standard DSGE models with financial frictions. J. Econ. Dyn. Control. 37, 32–51. Brzoza-Brzezina, M., Kolasa, M., Makarski, K., 2015. Macroprudential policy and imbalances in the euro area. J. Int. Money Financ. 51, 137–154. Busch, U., Scharnagl, M., Scheithauer, J., 2010. Loan supply in Germany during the financial crisis. Deutsche Bundesbank Discussion Paper Series 1: Economic Studies 2010.05 Caskey, J., Fazzari, S., 1987. Aggregate demand contractions with nominal debt commitments: is wage flexibility stabilizing?. Econ. Inq. 25, 583–597. Castro, V., Sousa, R., 2012. How do central banks react to wealth composition and asset prices?. Econ. Model. 29, 641–653. Cecchetti, S.G., Genberg, H., Lipsky, J., Wadhwani, S., 2000. Asset prices and central bank policy. Geneva Reports on the World Economy. International Center for Monetary and Banking Studies, London. Cerutti, E., Claessens, S., Laeven, L., 2015, March. The use and effectiveness of macroprudential policies: new evidence. Working Paper 15/61. International Monetary Fund, Washington D.C. Christiano, L.J., Eichenbaum, M., Evans, C., 1996. The effects of monetary policy shocks: evidence from the flow of funds. Rev. Econ. Stat. 78, 16–34. Christiano, L.J., Ilut, C., Motto, R., Rostagno, M., 2010. Monetary policy and stock market booms. Proceedings of the Federal Reserve Bank of Kansas City Economic Policy Symposium, Jackson Hole. pp. 85–145. Claessens, S., Ghosh, S.R., Mihet, R., 2014, August. Macro-prudential policies to mitigate financial system vulnerabilities. Working Paper 14/155. International Monetary Fund, Washington D.C. De Paoli, B., Paustian, M., 2013. Coordinating monetary and macroprudential policies. Staff Report 653. Federal Reserve Bank of New York, New York. Elliott, D.J., Feldberg, G., Lehnert, A., 2013. The history of cyclical macroprudential policy in the United States. Finance and Economics Discussion Series 2013-29, Board of Governors of the Federal Reserve System, Washington D.C. Fazzari, S., Ferri, P., Greenberg, E., 2008. Cash flow, investment, and Keynes–Minsky Cycles. J. Econ. Behav. Organ. 65, 555–572. Gertler, M., Gilchrist, S., 1994. Monetary policy, business cycles, and the behavior of small manufacturing firms. Q. J. Econ. 109, 309–340. Greenspan, A., 2002, August 30. Economic volatility. Speech at the Federal Reserve Bank of Kansas City Economic Symposium, Jackson Hole WY. Hofmann, B., Bogdanova, B., 2012. Taylor rules and monetary policy: a global “great deviation”?. BIS Q. Rev. Q3, 37–49. Hristov, N., Hlsewig, O., Wollmershaeuser, T., 2012. Loan supply shocks during the financial crisis: evidence for the euro area. J. Int. Money Financ. 31 (3), 569–592. Jawadi, F., Sousa, R.M., Traverso, R., 2015. On the macroeconomic and wealth effects of unconventional monetary policy. Macroecon. Dyn. in press. Kashyap, A.K., Stein, J.C., 1997. The role of banks in monetary policy: a survey with implications for the European monetary union. Econ. Perspect. 2–18. (Sep). Mallick, S.K., Sousa, R.M., 2013. The real effects of financial stress in the eurozone. Int. Rev. Financ. Anal. 30, 1–17. Mendicino, C., Punzi, M.T., 2014. House prices, capital inflows and macroprudential policy. J. Bank. Financ. 49, 337–355. Minsky, H.P., 1982. Can “It” Happen Again? Essays on Instability and Finance. M.E. Sharpe, New York. Minsky, H.P., 1986. Stabilizing an Unstable Economy. Yale University Press, New Haven CT. Mountford, A., Uhlig, H., 2009. What are the effects of fiscal policy shocks?. J. Appl. Econ. 24, 960–992. Nickell, S., 2005. Practical Issues in UK Monetary Policy, 2000-2005. Speech delivered at the British Academy Keynes Lecture, London. September 20, 2005

Posen, A.S., 2006. Why central banks should not burst bubbles. Int. Financ. 9 (1), 109–124. Roubini, N., 2006. Why central banks should burst bubbles. Int. Financ. 9, 87–107. Ryoo, S., 2013. The paradox of debt and Minsky’s financial instability Hypothesis. Metroeconomica 64, 1–24. Schwartz, A.J., 1998. Why financial stability depends on price stability. In: Wood, G. (Ed.), Money, Prices and the Real Economy. Edward Elgar, Cheltenham, pp. 34–41. Sims, C., 1986. Are forecasting models usable for policy analysis?. Minneapolis Fed. Reserve Bank Q. Rev. 2–16. Winter. Skott, P., 1995. Financial innovation, deregulation and Minsky cycles. In: Epstein, G., Gintis, H. (Eds.), Macroeconomic Policy after the Conservative Era. Cambridge University Press, pp. 255–273. Sordi, S., Vercelli, A., 2006. Financial fragility and economic fluctuations. J. Econ. Behav. Organ. 61, 543–561. Uhlig, H., 2005. What are the effects of monetary policy on output? Results from an agnostic identification procedure. J. Monet. Econ. 52, 381–419. Vercelli, A., 2011. A perspective on Minsky moments: revisiting the core of the financial instability hypothesis. Rev. Polit. Econ. 23, 49–67. Yellen, J.L., 2014, July. Monetary Policy and Financial Stability. Michel Camdessus Central Banking Lecture at the International Monetary Fund, Washington D.C.

Dr. Matthew Greenwood-Nimmo is a Senior Lecturer in Economics at the University of Melbourne. Matthew completed Undergraduate and Postgraduate degrees in economics at the University of Leeds, graduating with his PhD in 2009. Matthew’s research focuses on macroeconomic modelling with particular interests in the analysis of multivariate systems and regime-switching processes.

Dr. Artur Tarassow is a post-doctorate fellow at the University of Hamburg. He graduated with his PhD in 2015 at the Universitiy of Hamburg after completing his Postgraduate degree in economics at the University of Leeds. Artur’s research focuses on applied macroeconometrics covering monetary policy aspects and firm financing issues.