Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Nouriel Roubini wikipedia , lookup
Pensions crisis wikipedia , lookup
Fear of floating wikipedia , lookup
Global financial system wikipedia , lookup
Fiscal multiplier wikipedia , lookup
Quantitative easing wikipedia , lookup
Money supply wikipedia , lookup
Austrian business cycle theory wikipedia , lookup
Inflation targeting wikipedia , lookup
Interest rate wikipedia , lookup
Business cycle wikipedia , lookup
targeting financial stability: macroprudential or monetary policy? David Aikman† Julia Giese‡ Sujit Kapadia§ ∗ Michael McLeay¶ October 2016 PLEASE DO NOT QUOTE OR DISTRIBUTE WITHOUT PERMISSION Abstract This paper examines monetary and macroprudential policy interaction in a calibrated two-period new-Keynesian model. The model incorporates the possibility of a credit boom leading to a future financial crisis, and a loss function reflecting financial stability considerations. Deploying the countercyclical capital buffer (CCyB) improves outcomes relative to when the interest rate is the only policy tool. The instruments are typically strategic substitutes, with monetary policy loosening when the CCyB tightens, though the CCyB should be activated later if monetary policy is constrained at the zero lower bound. We also examine when the instruments are strategic complements and assess how market-based finance and a risk-taking channel of monetary policy affect our results. JEL CLASSIFICATION: E52, E58, G01, G28. KEY WORDS: macroprudential policy; monetary policy; financial stability; countercyclical capital buffer; financial crises; credit. ∗ The views expressed in this paper are those of the authors and not necessarily those of the Bank of England. Thanks [to be added]. † Financial Stability Strategy and Risk, Bank of England ([email protected]). ‡ Monetary Analysis, Bank of England ([email protected]). § Research Hub, Bank of England ([email protected]). ¶ Monetary Analysis, Bank of England ([email protected]). 1. Introduction The global financial crisis highlighted deficiencies in macro-financial policy toolkits. Monetary policy focused on price stability; prudential regulation was mostly concerned with practices at individual firms. Some central banks had a mandate to consider the stability of the financial system as a whole, but few had tools to ensure the resilience of the system. The events of 2007-08 demonstrated starkly how price stability was no guarantee of financial stability. Some had argued before the crisis that monetary policy should be used to tackle financial imbalances (Borio and Lowe, 2002 [6]; White, 2006 [33]) while others argued that the costs outweighed the uncertain benefits (Bernanke and Gertler, 2001 [4]; Kohn, 2006 [25]; Posen, 2006 [28]). Since the crisis, macroprudential instruments have come to the fore as an important addition to the toolkit of many central banks or regulatory authorities. These are mainly prudential tools set with the stability of the whole financial sector in mind. Some tools seek to address cross-sectional risks while others are concerned with time-varying risks (Aikman, Haldane and Nelson, 2015 [1]). The countercyclical capital buffer, designed under Basel III, is an example of the latter. Time-varying macroprudential policies share similarities with monetary policy. Both look to identify cycles - the financial cycle in the case of macroprudential policies and the business cycle in the case of monetary policy - and seek to tighten the calibration of tools in the upswing. While macroprudential policy makers are primarily concerned with the resilience of the financial system, they should not achieve this at any cost and also need to take into account the effects of policies on growth and employment. Similarly, monetary policy seeks to keep inflation close to target, but not necessarily at the expense of output and employment. They also work through similar channels: monetary policy changes the price of credit through interest rates and has effects on asset prices, among others. And macroprudential policy may affect credit provision, either through directly affecting quantities or by changing banks’ funding costs and thereby lending spreads. 1 Policymakers hold different beliefs over the relative effectiveness of the tools in relation to financial stability. Some argue that only monetary policy is likely to be effective across the whole financial system because interest rates ’get in all the cracks’ (Stein, 2013) [31]. Others argue for a combination of policies: Shin (2015) [30] notes that ’both monetary policy and macroprudential policies have some effect in constraining credit growth and the two tend to be complements’ And yet others question whether monetary policy should have an explicit financial stability objective at all since small changes in monetary policy may not be effective in reducing the crisis probability and ensuring resilience, and monetary policy may not be powerful enough to clean up after a crisis (eg Kohn, 2015) [26]. According to some, if anything monetary policy should ’lean with the wind’ rather than against it (Svensson, 2015) [32]. In this paper, we explore the implications of giving the policymaker an additional tool to monetary policy, a countercyclical capital buffer (CCyB). Monetary policy can focus on what it is most effective at, ie keeping inflation stable and the output gap closed. And macroprudential policy can be used to reduce the probability of a crisis and thereby increasing resilience, allowing for a separation of objectives and staying true to the Tinbergen principle of one objective, one tool. But the use of the CCyB is not costless: it affects both aggregate demand and potential supply, triggering endogenous reactions of monetary policy. The balance of the two effects determines how optimal monetary policy should respond and allows us to study cases where the policies are strategic substitutes and strategic complements, ie they move in the opposite or same direction, respectively. We use a similar but extended framework to Ajello et al (2015) [2]. They study the effectiveness of monetary policy in a simple semi-structural new-Keynesian model with a twist: in addition to an IS and Phillips curve, they have a crisis probability and an equation governing credit growth. The higher credit growth, the higher the crisis probability, and the policymaker minimises a two-period loss function, where the second-period can either be a crisis or a non-crisis outcome. In our set up, the IS and Phillips curve also contain credit 2 spreads which in turn are affected by the setting of the CCyB and the crisis probability decreases the higher the CCyB. Moreover, the loss function contains an additional term that reflects the policymakers risk tolerance with regard to crises. While the private sector is aware that a financial crisis may occur in the second period, it does not take the effects of its own behaviour on the crisis probability into account in its decisions in the first period. In the paper, we explore different scenarios around the interaction of monetary and macroprudential policies. Following Ajello et al (2015), we use a 2 period version of the model in order to focus on the key trade-offs underlying the optimal policy decision.1 We might typically expect a higher CCyB to reduce demand and have limited effects on supply. In this case, monetary policy should loosen to offset the effects of the CCyB on output growth, and the policies are strategic substitutes. This also holds if monetary policy is constrained by the zero lower bound, although it may then be optimal to wait until financial stability risks are somewhat more elevated before activating the CCyB, because monetary policy cannot offset any negative aggregate demand effects. The policies are strategic complements in two cases: If the CCyB has larger negative effects on supply than on demand (or indeed the effect on demand is positive) or if the CCyB becomes less effective, for example if there are leakages of credit to unregulated sectors, it may be optimal to lean against the wind with monetary policy for financial stability purposes. This paper is organised as follows. Section 2 introduces the model, Section 3 discusses the baseline optimal policy results and Section 4 explores extensions. 2. Macroprudential policy in a New Keynesian model In this section, we introduce a parsimonious model for understanding the interaction of monetary and macroprudential policies. Our model develops Ajello et al (2015): a two1 Svensson (2015) [32] points out in the context of Ajello et al (2015)’s model that the two-period assumption is not innocuous, however. We plan to explore how optimal policy changes in an infitnite horizon version of the model in future work. 3 period New Keynesian model, which features the possibility for a crisis in period 2. A crisis is an event that causes a discontinuous drop in output, whose likelihood is positively related to the degree of leverage built up in the economy in period 1. 2.1. Benchmark model Aggregate demand and supply are determined by a variant of the canonical two-equation New Keynesian model, featuring a forward-looking IS curve and Phillips curve.2 ps ps y y1 = E1 y2 − σ(i1 − E1 π2 + ωs1 ) + ξ 1 ps π1 = E1 π2 + κy1 + νs1 + ξ 1π (1) (2) where y1 is the gap between output from its efficient level, π1 is the deviation of inflation from target, i1 is the deviation of the central bank nominal interest rate from its steady state level and s1 , as discussed below, is the deviation of the credit spread from its steady state y level. ξ 1 is a demand or consumption preference shock, while ξ 1π is a cost-push or mark-up shock. These equations contain two departures from the canoncial model. First, following Curdia and Woodford (2010) [14], we introduce a role for fluctuations in credit spreads from their steady state level, s1 , in driving macroeconomic equilibria. Credit spread shocks in this structure are isomorphic to ’cost push’ shocks in that they introduce a trade-off between stabilising real activity and inflation. Higher credit spreads push down on aggregate demand via the IS equation (1) by increasing the interest rates facing households and firms wishing to borrow for consumption and investment. The sensitivity of demand to credit spreads may differ from the sensitivity 2 See Woodford (2003) [34], Galı́ (2008) [17], for example. 4 to the policy rate and ω 6= 1.3 Higher credit spreads also affect aggregate supply. They increase firms’ marginal costs and act as an endogenous cost-push shock, reducing the economy’s productive capacity.4 Second, we assume that the private sector’s expectations of period 2 outcomes, denoted ps E1 , are myopic and do not take into account the true model of how financial crises occur (Gennaioli, Schleifer and Vishny, 2015) [18]. In particular, risks are neglected such that expectations are taken to be a weighted average of crisis and non-crisis states, where the weights are exogenous and do not respond when financial system vulnerabilities build. ps E1 π2 = eπ2,c + (1 − e)π2,nc ps E1 y2 = ey2,c + (1 − e)y2,nc (3) (4) where e is an exogenous parameter reflecting the probability agents attach to a financial crisis occurring between period 1 and period 2, and the subscripts c and nc denote outcomes in the crisis and non-crisis states respectively. The financial side of our model consists of a law of motion for real credit growth, a macroprudential transmission mechanism, and an equation specifiying the probability of a financial crisis. B1 = φ0 + φi i1 + φs s1 + ξ 1B 3 There (5) are two opposing effects that could lead to this parameter being smaller or larger than 1. All else equal, the fact that only the subset of agents who are borrowers will be affected by changes in credit spreads would lead to a smaller effect than changes in risk-free rates, which affect everyone. But other factors could lead to a larger effect: credit constrained borrowers may have a higher sensitivity to changes in rates than savers. 4 This channel typically arises in a variety of ways in models of financial frictions. In models featuring only households, such as Curdia and Woodford (2010), it may come from the effect of higher credit spreads in distorting labour supply decisions. Or spreads could increase the cost of firms’ working capital, as in Carlstrom, Fuerst and Paustian (2010) [11]. In models with capital accumulation and binding credit constraints facing firms, such as Gertler and Karadi (2011) [19], higher credit spreads lead to capital shallowing, reducing labour productivity. 5 s1 = ψk1 γ1 = exp(h0 + h1 B1 + h2 k1 ) 1 + exp(h0 + h1 B1 + h2 k1 ) (6) (7) where B1 is defined as 3 year cumulative real credit growth, k1 is the setting of the CCyB in period 1, and γ1 is the probability of a financial crisis occurring between periods 1 and 2. Real credit growth depends (negatively - φs , φi < 0) on interest rates and credit spreads over the period. A constant term, φ0 , captures the steady state rate of credit growth, while ξ 1B is a shock to the quantity of credit, which may be correlated with the demand and y supply shocks ξ 1 and ξ 1π . The CCyB influences the likelihood of a financial crisis via two channels. First, all else equal, activating the CCyB directly reduces the probability of a bad outcome (h2 < 0 in (7)). This reflects the resilience benefits of higher bank capital, which increases the loss absorbing capacity of the banking system for a given distribution of shocks, and could also be viewed as affecting the cost of a crisis. This channel is consistent with the empirical evidence that banks that entered the crisis with more capital had higher survival probabilities (Berger and Bouwman, 2013) [3] and were better able to maintain their supply of loans during the crisis (Carlson, Shan and Warusawitharana, 2013) [10]. It is also consistent with theories that emphasise the role of bank capital in incentivising banks to monitor their loans over the cycle (Holmstrom and Tirole, 1997) [22]. We choose to model this channel as part of the crisis probability. An alternative would be to endogenise the cost of a crisis by modelling it separately from the probability of a crisis and making it dependent on the level of capital in the economy. The two options are qualitatively equivalent for our purposes. Second, activating the CCyB forces banks to rely on a more equity rich funding structure, which, in turn, pushes up their overall cost of capital. Banks, we assume, are able to pass on some of these increased costs, resulting in higher credit spreads for real economy borrowers and hence lower credit demand in the first period. We capture this relationship in equation 6 (6). The CCyB can influence the risk environment facing banks through this ’leaning’ channel: activating the CCyB tempers a built up credit growth and hence reduces the likelihood of a financial crisis. Note also that the CCyB will also have knock-on effects on both aggregate demand and supply via this channel, via the IS curve (1) and Phillips Curve (2). In the baseline version of our model, monetary policy can also be used to influence the likelihood of a financial crisis by leaning against the build-up risks associated with credit growth. Higher interest rates reduce credit growth by lowering credit demand and hence influence the risk environment facing the banking system. Credit supply, captured by (6), is independent of monetary policy in our baseline model. In Section 4, we consider relaxing that assumption. We also explore extensions of the model that emphasise further qualitative differences in the transmission mechanisms of the two tools. For instance, the scope of application of the CCyB is much narrower than monetary policy in that it is applied only to banks. 2.2. Loss function We focus in the main on the case of a single policymaker that sets monetary and macroprudential policy jointly to minimise an overall loss function. This allows us to analyse the normative question of how policy should be set to meet given objectives, free of institutional constraints. It closely approximates the institutional set-up of many central banks where monetary and financial policy committees have overlapping objectives, overlapping membership and common information sets. We assign our policymaker a simple loss function that captures the mandates of central banks with responsibility for monetary and financial stability policies. The loss function is an augmented version of that typically used in the monetary policy literature. Overall loss 7 is the sum of period 1 loss and discounted period 2 loss. L = L1 + βL2 (8) Period 1 loss features the central bank’s traditional monetary policy objectives, which typically involve the dual mandate of an inflation target plus a concern for output or employment volatility. In our formulation, the policymaker seeks to minimise the quadratic loss resulting from deviations of inflation and output from their respective targets. λ denotes the weight assigned to output deviations relative to those in inflation. L1 = 1 [(π12 + λy21 ) 2 (9) This is augmented with endogenous and exogenous financial stability concerns in period 2, where loss is given by: L2 = 1 2 2 [(1 − γ1 )(π2,nc + λy22,nc ) + γ1 (1 + ζ )(π2,c + λy22,c )] 2 (10) The first two terms capture outcomes if there is no crisis, and are similar to the period 1 objectives of the central bank. We capture the central bank’s financial stability concerns via the second two terms, which capture the output and inflation costs of a crisis event. Since these costs are larger than those in a non-crisis state, the policymaker has an endogenous motivation to minimise γ1 . The final terms are also increasing in ζ, an extra exogenous weight placed on avoiding the costs associated with a financial crisis. When ζ > 0, the policymaker attempts to minimise the probability of a financial crisis to a greater degree than quadratic expected losses alone would imply is optimal. In effect, the presence of the financial stability objective skews the central banks’ objectives towards crisis prevention. The greater is ζ, the more the central bank’s objectives are skewed in the direction of crisis avoidance relative to symmetrically stabilising inflation and output. This formulation of financial stability 8 objectives is novel and qualitatively distinct from simply increasing the weight attached to future output gaps . There are two compelling justifications for augmenting the standard loss function in this way. First, in the aftermath of the financial crisis many central banks are mandated to achieve financial stability goals by elected governments on behalf of the public. For example, the Bank of England’s Financial Policy Committee has a mandate to ’take action to remove or reduce systemic risks’ with a view to ’protect[ing] and enhance[ing] the resilience of the UKs financial system’. Such financial stability mandates in part reflect a desire by taxpayers to avoid bearing the bailout costs of future systemic crisis. This idea is explored in Peek, Rosengren and Tootell (2015) [27], who posit a quadratic financial stability term, arguing that too much financial stability is undesirable since there are costs in reducing credit availability to an unacceptably low level. We model the cost of achieving financial stability directly through its effect on potential output, which provides an endogenous motivation for avoiding ’too much’ financial stability. Financial stability mandates may also reflect a preference to avoid the distributional effects of financial crises, which may be difficult to offset with other policy tools. A second justification for ζ > 0 arises if there are uncertain welfare costs of crises. Robust control considerations would suggest avoiding the worst case outcome the output losses associated with a severe crisis in period 2 rather than targeting expected period 2 output. Crises may also generate additional costs that are not captured in our simple two-period framework. We show in the appendix, for instance, that if there are hysteresis effects associated with financial crises such that depressed output persists into future periods (Blanchard, Cerutti and Summers, 2015) [5], the policymaker optimally places a higher weight on avoiding crises in the first place. The policy problem facing the central bank therefore is to choose an interest rate and CCyB rate in period 1 to minimise the loss function (8) subject to the IS curve (1), the Phillips curve (2), the dynamics of credit (5) and the crisis probability function (7), taking 9 Table 1: Benchmark parameter values Parameter Description Standard Parameters β Discount Factor σ Interest-rate sensitivity of ouptut κ Slope of the Phillips Curve λ Weight on output stabilisation i∗ Long-run natural nominal rate of interest Effect of the CCyB ψ Effect of the CCyB on credit spreads ω Effect of spreads relative to policy rate on y ν Effect of spreads on the Phillips Curve Financial conditions equation parameters φ0 Average real credit growth φi Coefficient on interest rates φs Coefficient on spreads Crisis probability equation parameters h0 Constant h1 Coefficient on leverage vairable h2 Coefficient on k1 , (resilience effect of CCyB) e Private sector perception of crisis probability Period 2 parameters y2,c Deviation of output from efficient in crisis state Parameter Notes 0.99 0.57 1.03 0.05 3% Matches r*=1% Burgess et al (2013) Burgess et al (2013) Standard welfare-based Rachel and Smith (2015) 0.2 1 0.41 1pp equity = 20bps - MAG (2010) Cloyne et al (2015), updated Franklin, Rostom and Thwaites (2015) 0.21 -1.4 -6.1 Historical average Cloyne et al (2015), updated Cloyne et al (2015), updated -1.7 + 0.11h2 5.18 -27.8 0.0005 All estimated using dataset constructed in Bush et al (2015) Arbritarily small -0.032 3.2% lost output per year Brooke et al (2015) No effect Steady state Steady state Risk-neutral policy π2,c y2,nc π2,nc ζ Shocks y SD (ξ 1 ) Deviation of inflation from target in crisis state Deviation of output from efficient in non-crisis state Deviation of inflation from target in non-crisis state Extra weight on E(crisis cost) 0 0 0 0 Standard deviation of demand shocks 0.0125 SD (ξ 1π ) Standard deviation of cost-push shocks 0.0011 SD (ξ 1B ) Standard deviation of credit shocks 0.16 Similar to risk premium shock in Burgess et al (2013) Similar to mark-up shocks in Burgess et al (2013) Set to match historical data ps as given private sector expectations, E1 . The appendix contains a formal solution to the central bank’s problem. 2.3. Parameter choices We calibrate the model using the parameters in Table 1. Our approach involved estimating equation (7) for the probability of a crisis and its determinants on a cross-country dataset. For our other parameters, governing the effect of our two policy tools on credit, aggregate demand and supply, we used empirical estimates from the literature for the UK, taken from a variety of econometric models and estimated DSGE models already in use for policy analysis at the Bank of England. We interpret the time periods in our model as 3 years to capture the notion that financial 10 system vulnerabilities build up in response to the prolonged credit booms that appear to be a feature of the data (Aikman, Haldane and Nelson, 2015 [1]; Borio, Drehmann and Tsatsaronis, 2012 [15]). And macroprudential policy tools such as the CCyB have longer implementation lags than monetary policy, such that a longer timeframe captures the interaction between the tools in a more meaningful way. For the CCyB, a change in the requirement in period 1 can be interepreted straightforwardly as a change to a new ratio, announced and implemented and reaching its full effectiveness all within the same three-year period. For monetary policy, there are range of different policy changes that could be implemented over a three-year horizon. The specific experiment we consider in our calibration exercise is a change in the policy rate in the first quarter of period 1, held constant for three years. In the various models we use, we implement this using a series of unanticipated monetary policy shocks. The model’s standard macroeconomic parameters, which determine the effect of monetary policy on demand and supply, are calibrated using the Bank of England’s main DSGE forecasting model: COMPASS, set out in Burgess et al, 2013 [9]. Rather than reading across directly from the Euler equation and Phillips Curves in that much larger model, we match σ and κ, respectively the interest rate sensitivity of output and the slope of the Phillips Curve, to the reduced form estimates from a series of monetary policy shocks in Burgess et al. For ease of comparison with other papers, all variables are meaured as average annual values. So σ = 0.57 implies that an increase in the annual policy rate of 1 percentage point, for a period of 3 years, reduces the level of GDP by 0.57%. While κ = 1.03 implies that a fall in GDP by 1% over the 3-year period reduces annual inflation by an average of 1.03pp. The parameters governing the probability of a financial crisis are estimated using the cross-country panel dataset in Bush et al (forthcoming). Figure 1 illustrates the relationship between credit growth, the CCyB and the likelihood of a crisis. Evaluated at a CCyB of zero and the sample average rate of credit growth of 21%, the estimated coefficients imply that an increase in real credit growth of 1pp per year to 24% would increase the 11 annual probability of a crisis by around 0.4 percentage points. In addition, the ratio of the coefficients h2 h1 implies that an increase in the CCyB of 1pp is sufficient to offset the impact on the crisis probability of just over 5% real credit growth over 3 years, or 1.8% per year. Figure 2 shows the (in-sample) performance of the estimated equation at predicting the probability of a crisis in the UK. The line peaks in 2008 with an estimated annual crisis probability of 12%. Turning to the transmission mechanism of the CCyB, we set φs , the sensitivity of credit growth to spreads, equal to -6.1, based on Cloyne et al [13]. This is over 4 times as large as the effect of interest rates on credit growth taken from the same paper. A 1 percentage point increase in the CCyB is set to increase spreads by 20 basis points. This number is at the top end of the estimates reported in the Macroeconomic Assessment Group (2010) [21] meta study, conducted by the Financial Stability Board and Basel Committee on Banking Supervision to quantify the potential impact of increasing bank capital requirements on lending and growth. Combined with the effect of spreads on credit growth, it implies that a 1pp increase in the CCyB reduces three year credit growth by 1.2 percentage points, similar to the median estimate reported in the Macroeconomic Assessment Group study (a reduction of 1.4% after 18 quarters). We set the sensitivity of output to spreads equal to that on interest rates, ω = 1, broadly consistent with Cloyne et al (2015). Together, these estimates imply that even ignoring its resilience benefits, the CCyB has an absolute advantage in dampening credit growth for a given cost to aggregate demand. We calibrate the parameter governing the impact of spreads on inflation, ν, jointly with φs and κ, using the estimates reported in Franklin, Rostom and Thwaites (2015) [16]. This paper finds that each percentage point fall in corporate lending in the recession reduced UK labour productivity by 0.25%. Combining this with the effect of the CCyB on lending described above gives its effect on potential supply. Our calibration implies a relatively large impact of the CCyB on potential supply . It may be an overestimate if one believes that some of fall in UK labour productivity in recent years has been driven by the scale and 12 nature of the financial crisis rather than being a more general phenomenon caused by lower corporate lending per se. In the event of a crisis, output contracts by an average of around 3% per year, in line with the post-resolution estimates from Brooke et al (2015) [7]. Inflation, however, remains at target in the crisis state, reflecting the experience internationally in the years following the 2008 crisis (Gilchrist et al, 2015) [20]. As we abstract from further shocks, the policymaker is able to achieve zero output and inflation gaps in period 2 if a crisis does not occur. Finally, the standard deviations of the demand and inflation shocks are calibrated close to the estimated standard deviations of similar shocks in Burgess et al (2013) [9] - a risk premium shock for demand, and a combination of two mark-up shocks for inflation. The standard deviation of the credit shock is set to loosely match the standard deviation of credit in the data. 3. Results In this section, we describe our results. We first examine the monetary policy-only case and show that welfare is higher with active use of the CCyB. We then discuss trade-offs for monetary and macroprudential policies and explore the interaction between monetary and macroprudential policy further under different assumptions. 3.1. Benchmark case: monetary policy-only economy We begin by examining the benchmark monetary policy-only economy, where the CCyB is switched off and spreads are exogenous. In this case, the first-order condition governing the central bank’s interest rate policy in period 1 can be shown to be (see appendix): σ(κπ1 + λy1 ) = where ∂γ1 ∂i1 ∂γ1 ∂L ∂i1 ∂γ1 (11) is the derivative of the crisis probability with respect to interest rates, and 13 ∂L ∂γ1 2 2 + λy2 ), the policymaker’s expected discounted = − β(π2,nc + λy22,nc ) + β(1 + ζ )(π2,c 2,c loss from a crisis, is a function of the severity of the crisis state and the skewness of the central bank’s objective function. This condition is an augmented version of the standard description of optimal monetary policy, λy1 = −κπ1 (Clarida, Gali, Gertler, 1999) [12]. Intuitively, it collapses to the standard condition when either the probability of a financial crisis γ1 is insensitive to credit growth, or when credit growth is insensitive to interest rates, φi = 0. Outside of these special cases, monetary policy must balance two trade-offs. First, the standard intra-temporal trade-off between stabilising inflation and output in period 1 in the presence of cost-push shocks. Second, the inter-temporal trade-off between stabilising output (and inflation) in period 1 and stabilising the probability of a financial crisis, which would depress output in period 2. Figure 3 presents the intertemporal trade off for monetary policy alone when faced by a credit shock of different sizes.5 Each point on the trade-off curves is the solution to the optimal policy problem for a different value of ζ. Points to the northeast of each line are strictly inferior, while points to the southwest are infeasable. The y-axis shows the period 1 loss from using monetary policy to counter credit growth if not required for monetary stability: raising interest rates results in an undershoot of the period 1 inflation and output targets for certain. The x-axis shows the financial stability benefit of doing so: lower credit growth reduces the crisis probability, so increases period 2 output in expectation. The stronger is credit growth, the further this intertemporal trade-off shifts out, and the more willing the central bank should be to raise interest rates. Period 1 welfare loss is scaled to be a proportion of the period 2 loss, conditional on a crisis occurring. This means that a policymaker with symmetric objectives, ζ = 0, has indifference curves with slope -1, through points (1,1), (2,2), etc. Skewing the policymaker’s objectives towards avoiding financial crises by increasing ζ has the effect of increasing the slope of the indifference curves, leading to tighter policy, lower inflation, output and crisis probability. But interestingly, this quantitative difference is the only effect of a positive ζ, the qualitative nature of the 5 We assume the shock is independent of any cost-push and demand shocks, which are set to zero. 14 intertemporal trade-off is unchanged to that under symmetric objectives. 3.2. Two instruments: adding macroprudential policy With one instrument, balancing monetary and financial stability objectives requires higher interest rates than if monetary policy was focused exclusively on inflation and output stabilisation. The stronger is credit growth, the greater the conflict between monetary and financial stability objectives and the larger the relative losses under the inactive macroprudential regime. By contrast, under an active macroprudential regime, the CCyB can be directed at stabilising the crisis probability by matching the resilience of the banking system to the risk environment it faces, while interest rates can focus on inflation and output stabilisation. This, of course, is Tinbergen’s famous result: that to achieve n independent targets there must be at least n effective instruments. If using the CCyB were costless, there would be perfect separation of instruments. The green circle in figure 4 shows that in this case, (and with unbounded instruments), then the first-best outcome of no crises could be achieved. This could be achieved with ν = 0 by setting capital requirements to an (infinitely) high level. Even if ω > 0, the negative effects of high capital ratios on demand could be perfectly offset by monetary policy, absent the zero lower bound on interest rates. Away from that special case, while losses are unambiguously lower under the active macroprudential regime, there is no free lunch. This is shown in the blue lines in figure 4. Although much improved relative to monetary policy only, a trade-off continues to exist between monetary and financial stability objectives. This is because actively using the macroprudential tool makes intermediation more costly, affecting the supply side of the economy. This has the side-effect of worsening the traditional monetary policy trade-off between output and inflation. To examine these trade-offs in more detail, we first define the natural rates of output 15 and interest as the levels of each consistent with inflation being at target.6 Setting all expectations terms and the cost-push shock to zero in (2), setting y1 = y1n , and substituting in (6) gives the the natural level of output, as a deviation from its steady state. y1n ≡ − νψ k1 κ (12) Substituting (12) for y1 and (6) for s1 into (1), and again setting expectations to 0 gives the natural rate of interest: r1n ≡ ( νψ y − ωψ)k1 + σ−1 ξ 1 κσ (13) Evidently, the natural rate can rise or fall when the CCyB increases. If the supply effects of the CCyB are large, the first term will dominate, and the natural rate will be increasing in the CCyB. Similarly, if increasing the CCyB has only a small negative effect on demand, then the natural rate will increase. Given that the CCyB introduces a monetary policy trade-off, tracking the natural rate is not optimal policy, however. But it provides a useful benchmark to compare optimal policy to below. For optimal policy, the policymaker sets period 1 interest rates and the CCyB to minimise the Loss Function (8) subject to the constraints imposed by the private-sector equilibrium conditions. As we show in the appendix, with two instruments, this gives one first order condition for each policy instrument, each trading off output, inflation and the crisis probability. But with two condtions and these three target variables, we show that we can also derive conditions giving independent trade-offs between any two of target variables. Any two of these trade-offs give the optimality conditions that completely characterise optimal policy. For ease of comparability with the monetary policy only case, we show the optimality condition governing the intra-temporal monetary trade-off between output and inflation in 6 In the standard New Keynesian model, the natural rates of output and interest would equate to the levels of each that would prevail were prices and wages perfectly flexible. 16 period 1. We then discuss the second inter-temporal monetary-financial stability trade-off between period 1 output and the probability of a crisis in period 2. Intra-temporal trade-off The optimal split between output and inflation in our model has a lower weight on output than the policymaker’s actual weight λ. λy1 + κπ1 = 0 where λ ≡ h1 ψ(φs −φi ω )+h2 λ ν h1 ψ(φs −φi (ω − κσ )+h2 (14) < λ. The intuition underlying λ can be seen by defining χ ≡ ∂γ1 ∂k ( ∂γ1 1 ∂i1 ∂y1 ∂i1 ∂y1 ∂k1 − 1) as the comparative advantage of macroprudential policy at affecting the crisis probability, versus monetary policy at affecting the IS curve/aggregate demand. This allows us to express λ as λ = ∂y χ ∂k 1 1 ∂yn 1 + χ ∂y1 ∂k1 ∂k1 λ. When the instruments become orthogonal, either because interest rates do not affect the ∂y 1 1 crisis probability ( ∂γ ∂i → 0), or the CCyB does not affect aggregate demand ( ∂k → 0), then 1 1 the comparative advantage tends to infinity (χ → ∞). In that case, or if the cost of using ∂yn the CCyB tends to 0 ( ∂k1 → 0), then (λ → λ) and the optimal monetary policy prescription 1 is the same as in the textbook model. When either the tools have an infinite comparative advantage, or when macroprudential policy is costless, there is no advantage to monetary policy adjusting to try to affect the crisis probability. But when using the CCyB is costly, then at the margin there will be some benefit in using interest rates to take on some of the responsibility. Adding the CCyB as a second instrument makes the optimal inflation-output choice independent of the expected crisis loss, unlike (11), the monetary policy only case, where both output and inflation are lower if the crisis probability is higher. The financial stability motive does still affect the equilibrium inflation and output outcomes in two distinct ways, 17 however. First, is that with the CCyB as a an active policy instrument, it imparts an endogenous cost-push type effect, so that even in the absence of any exogenous cost-push shocks, it will be impossible to close both output and inflation gaps at once. Second, is that the effective weight on output stabilisation in (14) is lower than the weight in the loss function (8). This is because the CCyB imparts positive cost-push effects that push up on inflation and down on output, so placing a lower weight on output means setting a higher interest rate for a given level of the CCyB. This effect is present because the CCyB is costly. Similarly to the monetary policy only case, it is always optimal at the margin for monetary policy to be somewhat tighter than with no financial stability concerns, in order for the CCyB to be somewhat lower. The difference is that with two instruments, tighter monetary policy always means bringing inflation closer to target Inter-temporal trade-off The policymaker’s optimality condition trading off period 1 potential output with the probability of a financial crisis in period 2 is given by: λy1 (− νψ ∂γ ∂L ∂γ νψ ) = ( 1 + 1( − ωψ))(− ) κ ∂k1 ∂i1 κσ ∂γ1 (15) The intution behind the equation can be seen using the natural rate definitions above: λy1 ∂y1n ∂γ ∂γ ∂r n ∂L = ( 1 + 1 1 )(− ) ∂k1 ∂k1 ∂i1 ∂k1 ∂γ1 (16) The optimality condition sets the policymaker’s marginal cost of increasing the CCyB/credit spreads in terms of lower potential output today, on the LHS, equal to the marginal gain in terms of a lower crisis probability, multiplied by the perceived cost of a crisis, ∂L ∂γ1 . The first term in brackets is the lower risk of a systemic crisis with a higher CCyB, while the second term is the additional effect from changing interest rates in response to any macroprudential policy induced change in the natural rate of interest. If the demand effect of macroprudential policy dominates, this second term will be negative, so the gain 18 from increasing spreads will be somewhat offset by looser monetary policy, for unchanged inflation. Policy settings Figure 5 shows, under our benchmark calibration, the optimal policy settings required to implement the two conditions above. The charts show the policy functions for the CCyB and the interest rate in response to a credit shock of varying size. As the size of the credit shock increases, the policymaker chooses to increase the CCyB, to lean against credit growth and increase resilience. At the same time, policy reduces the interest rates in order to offset some of the reduction in demand brought about by the CCyB. To explore the mechanism in detail, the corresponding equilibrium outturns for four other variables are shown in Figure 6. Absent policy, the rate of credit growth would increase one for one with the size of the credit shock. With policy, a higher CCyB and a lower interest rate both affect this increase, with the CCyB leading to higher spreads, leaning against the credit shock, while the lower interest rate exacerbates it. In our calibrated model, the first effect dominates, such that a small amount of the credit shock is offset in the top-left panel. Qualitatively, optimal policy is not particularly dependent on the exact parameterisation of the credit growth equation (5), however. Even if we were to set φs = 0, so that only monetary policy is able to lean against credit growth, policy should still cut interest rates. The reason is that it is the crisis probability, rather than credit growth itself, that has a direct effect on the policymaker’s loss. And the CCyB still has a comparative advantage in affecting the crisis probability due to its resilience benefits, even if it is unable to lean against the credit cycle. Thus in the top-right panel, a higher CCyB is able to mitigate the effect of the credit shock on the crisis probability, despite most of it feeding through into higher credit growth. Instead of leaning against the credit cycle, the optimal role of monetary policy is largely to best offset the effects of the CCyB on output and inflation. First, the demand effects of the 19 CCyB are larger than the supply effects, so with a higher CCyB the natural rate of interest falls - to which monetary policy responds by cutting rates. In addtion, the policymaker places a positive weight on output deviations, so optimally chooses to cut interest rates even further than the fall in the natural rate, in order to split the cost-push effects of the CCyB between output and inflation according to (14). As a result, equilibrium output falls by less than y1n , while inflation rises. 3.3. The zero-lower bound on interest rates Until now, we have discussed optimal policy under the assumption that both instruments are set optimally in an unconstrained fashion. But in practice, monetary policy may face a zero lower bound (ZLB) constraint. Figure 7 shows how this would affect the trade-off facing the policymaker. The blue lines show the policymaker’s menu of options in response to a credit shock with both instruments freely available. The red lines show the equivalent policy frontier when monetary policy is assumed to be constrained, all else equal. With monetary policy unable to offset any of the demand effects of a higher CCyB, using it becomes far more costly. As a result, the trade off facing the policymaker rotates out from the point on the horizontal axis where the CCyB is set to zero. For our baseline calibration, the trade-off remains less steep than when monetary policy is the only instrument available, however. This is due to the comparative advantage of the CCyB in dealing with credit shocks that affect the crisis probability - for a given reduction in the crisis probability, the demand cost is much smaller if that is achieved using the CCyB than via monetary policy. Given the steeper trade-off, for given policymaker preferences, it is optimal to use the CCyB less when monetary policy is constrained. Figure 8 shows how the CCyB policy function changes at the zero lower bound. As the credit shock grows, the optimal response of the CCyB, while still positive, is far more muted. The larger the CCyB response would be if monetary policy were unconstrained, the harder the constraint bites on macroprudential policy, and the more that financial stability concerns have to be balanced with monetary 20 stability. 4. When are the policies complements or substitutes? In this section, we focus specifically on the interaction between monetary and macroprudential policies in our model. We explore when the two policies are complements and move in the same direction in response to a shock, or substitutes, and move in opposite directions. In our benchmark model, the policymaker may choose to move interest rates and the CCyB in the same direction for two distinct reasons. First, is that although we have assumed so far that our credit and macro shocks are uncorrelated, in practice the business cycle and credit cycle are highly intertwined. We first explore how even if the policymaker largely uses the interest rate to offset macro shocks and the CCyB to offset credit shocks, the correlation between the two types of shocks may drive a correlation in the instruments. Second, is that even if shocks are completely independent, the strategic substitutability or complementarity (Bulow, Geanakoplos and Klemperer, 1985 [8]) between the two instruments means that one instrument will be changed purely in response to the change in the other. We next explore how our varying the calibration of our model to capture different beliefs about the structure of the economy can switch the instruments from being strategic substitutes to complements. This alters the nature of interaction between the instruments - such that the policymaker tightens and loosens both together even when faced with independent shocks. Away from our benchmark model, we consider other views about policy interaction. This section finally explores how some small changes to our model could give rise to different types of interaction often cited by policymakers, within our parsimonious framework. 4.1. Responses to different shocks The optimal response of period 1 policy to a one standard deviation innovation in each of the three different shocks is shown in Figure 9. The first 6 bars show the response of each 21 instrument to each shock under the assumption that they are completely independent. The final 2 bars show the response to a joint shock to credit and aggregate demand, under the other extreme assumption that these are perfectly correlated. In response to an isolated positive credit shock, as discussed in the results section, it is optimal for the CCyB to tighten. It does so partly to dampen credit growth, but even more so to enhance resilience and reduce the probability that high credit growth will lead to a systemic financial crisis in period 2. Although the setting of interest rates optimally takes into account benefit from tightening them in reducing the crisis probability, the main role for monetary policy is to offset the negative effects of a higher CCyB on aggregate demand. So monetary policy is loosened slightly. In response to a positive demand shock, in contrast, the impact of the shock can largely by offset by a tightening in interest rates, leaving period 1 output little changed. The optimal response of the CCyB is to loosen slightly, since with higher interest rates (and lower credit growth), the crisis risk becomes lower, so the marginal benefit of using the CCyB falls. Finally, following a positive cost-push shock, the optimal response is again for the interest rate to tighten and the CCyB to loosen. But in this case, the monetary policy tightening is smaller, partly due to the need to balance the effects shock between higher inflation and lower output. The CCyB loosens, and by more, since with a positive cost-push shock already pushing up on inflation and down on output, the marginal cost of using the CCyB and reducing potential supply becomes larger. The final two bars show the case of perfectly positively correlated credit and demand shocks. This could be thought of as the typical policy setting when the credit and business cycles are closely aligned. Or equivalently, it could represent optimal policy when credit shocks had large endogenous effects on aggregate demand, which we abstract from in our model. With correlated positive shocks, both instruments should be tightened. The direct tightening of monetary policy in response to a demand shock dominates the indirect effect coming from strategic substitutability between the instruments, which would tend towards 22 Table 2: Optimal policy under different sets of parameters as crisis cost/financial stability weight rises Policy interaction ∆k1 ∆i1 Parameter case Intuition Strategic complements +ive +ive νψ κ2 κ+λ > σωψ (Supply effect of CCyB)*(policymaker weight on inflation) > demand effect of CCyB νψ κ2 κ+λ < σωψ, Strategic substitutes +ive -ive > ∂γ1 ∂k1 ∂γ1 ∂i1 Relative effect of the CCyB/interest rates < (σωψ − νψ κ2 κ+λ )σ−1 on crisis probability < relative effect of the CCyB/interest rates on demand and supply (σωψ − νψ κ2 κ+λ )σ−1 Strategic substitutes and -ive instrument switches +ive Demand effect of CCyB is bigger than (weighted) supply effect, but the CCyB is still relatively more effective at reducing crisis probability ∂γ1 ∂k1 ∂γ1 ∂i1 loosening. Similarly, the direct tightening required in the CCyB in response to the positive credit shock is larger than the indirect loosening needed in response to the rise in interest rates. 4.2. Response to a credit shock under different calibrations Even in response to a credit shock that leaves aggregate demand conditions unchanged, changing our assumptions with regard to a key parameter in the model changes the way monetary and macroprudential policies interact. In this sub-section, we examine three mutually exclusive cases, summarised in Table 2. As in our benchmark case, policies may be strategic substitutes - a tightening in one policy instrument reduces the marginal benefit to tightening the other. Or they may be strategic complements - a tightening in one policy instrument increases the marginal benefit to tightening the other. And if they are strategic substitutes, there may be circumstances when the objectives of each instrument switch, such that the CCyB should target inflation and output, and interest rates should aim to reduce the probability of financial crises. Table 2 shows that the interaction between the policies depends cruicially on two features: whether the (weighted) supply cost of macroprudential policy is larger than its 23 demand cost; and which policy has a comparative advantage at acheiving which goal. On the first, when macroprudential policy tightens, it reduces both actual and potential output, relative to their effecient level. The policymaker’s desire to smooth volatility in actual output means they care directly about the reduction in output. But they also care about any change in actual output relative to potential output, via the inflation term. If the fall in potential output is large enough, such that the output gap increases, and if the policymaker’s weight on inflation is large enough, then the policymaker will respond to the macroprudential policy tightening by tightening monetary policy. This will also lower the crisis probability somewhat, reducing the required macroprudential tightening. Note that the result relies on the relative supply and demand effects of the CCyB. If an increase were to increase aggregate demand, rather than reduce it as in our benchmark case, it would also be optimal to tighten monetary policy even if the supply effect of the CCyB were to be small. As the demand effect gets more negative relative to the supply effect, or the relative weight on inflation falls, the policymaker switches to wishing to reduce interest rates in response to increases in the CCyB, as in the benchmark case. The policymaker sets jointly optimal policy, so has no in-built preference to use each instrument for a specific goal. The decision as to which policy to tighten, when there is an increased desire to target the financial stability goal, depends entirely on which policy has a comparative advantage. The definition of comparative advantage is slightly more general than the χ term defined earlier. Macroprudential policy is tightened to reduce the crisis probability if and only if it is relatively more effective than monetary policy at doing so, relative to their effects at traditional monetary policy goals. Effectiveness at monetary policy goals is given by the weighted average of the marginal effect of the policy on output, with the policymaker’s output weight, and the output gap, with the policymaker’s inflation weight. Figure 10 shows how the monetary policy response varies when parameters are set such that the policies are strategic complements. The blue lines show the standard case 24 when policies are strategic substitutes. But in the green lines, the CCyB is assumed to boost aggregate demand, and optimal policy involves monetary policy tightening to offset this. In the red line, the CCyB negatively impacts supply by 5 times as much as in the benchmark case. This has two effects, first it makes using macroprudential policy more costly, so less active as a result. But as with positive demand effects, it also makes the policymaker choose to tighten monetary policy at the same time as tightening the CCyB. The other extreme case is when the comparative advantage of the instruments switches. The parameters in this theoretical example are unlikely in practice, and the mechanism partly depends in there being no lower bound for the CCyB. It could occur if the (negative) demand effect of macroprudential policy (ω) became very large. As the demand effect grows, monetary and macroprudential policy operate on more and more similar margins - their comparative advantage falls. This means that the policies generally have to both move further in opposite directions to achieve a similar outcome for output, inflation and the crisis probability. At the point where the comparative advantage switches, it quickly becomes optimal for the instruments to switch roles, and monetary policy tightens to reduce the crisis probability, while macroprudential policy loosens to offset some of the spillovers to output and inflation. 4.3. Introducing a market-based finance sector In this subsection, we return to the calibration of our benchmark model but extend it to allow for a market-based finance sector that is not subject to macroprudential policy alongside a banking sector to which macroprudential policies apply. The intuition is that as the banking system is subject to tighter and costly macroprudential policy, there might be leakage to the non-banking system, ie non-banks step in and extend credit more cheaply. This implies that there are limits to the effectiveness of the CCyB in reducing the crisis probability as it cannot lower the probability of a crisis emanating from the market-based finance system. 25 Rather than equation (7), let the probability of a crisis be equal to: γ1 = bγ1BL + (1 − b)γ1SL (17) where γ1BL is the probability of a crisis arising in the banking sector and γ1SL is the probability of a crisis arisisng in the shadow banking sector. The parameter b is the share of lending done by banks in steady state. Lending growth for each sector is determined by an analogous equation to (5): BL1 = φ0B + φi i1 + φsB s1 + ξ 1B (18) SL1 = φ0S + φi i1 + φsS s1 + ξ 1S (19) We assume that φsB < 0 - higher capital requirements (and bank lending spreads) reduce the amount of lending done by banks. But we also assume there is some leakage of macroprudential policy to the unregulated, shadow banking sector, such that φsS > 0. Both equations have the same parameter governing the effect of interest rates on credit growth, however, capturing the Stein (2013) argument that monetary policy gets in all the cracks. For simplicity, we assume that both shocks are perfectly correlated and ξ 1B = ξ 1S We calibrate φsS = −3φsB , based on Aiyar, Calomiris and Wieladek (2014), and set the shadow banking sector equal to 20% of the overall stock of lending. If macroprudential policy reduces bank credit growth by 1pp, shadow bank credit growth increases by 3pp. But given the smaller stock of shadow bank lending, together, these parameters imply that around three-quarters of the reduction in regulated bank lending leaks into the shadow banking sector. Figure 11 shows how the optimal policy response to a credit shock changes when the shadow banking channel is switched on, for different policymaker preferences for avoiding crises. The channel leads the policymaker to attenuate their policy response. This is because 26 the marginal benefit of tighter macroprudential policy will always be smaller, now that it boosts shadow banking credit (all else equal). But less tight macroprudential policy also means that monetary policy can set interest rates higher, both because there is less macroprudential policy induced cost-push shock to effect, and also because with a less effective macroprudential tool, there is more need to lean against the wind. 4.4. Introducing a strong risk-taking channel of monetary policy Similar conclusions can be reached when we allow for non-linear interaction between the two policies in their effect on credit. In particular, we add such an interaction term to equation (5). This captures the argument of Shin (2015) [30] that monetary policy and macroprudential policies are complements, with higher levels of one increasing the marginal effectiveness of the other. Intuitively, this could be interpreted as a ’risk-taking channel’ of monetary policy, under which combinations of low interest rates and high capital requirements encourage banks to lend to riskier borrowers. Under this interpretation, B1 would proxy for the growth rate of credit risk, which may increase even if the growth rate of credit quantities remains unchanged.7 Let real credit growth by given by: B1 = φ0 + φi i1 + φs s1 + φi,s i1 s1 + ξ 1B , (20) where φi,s < 0. Then lower interest rates make the CCyB less effective at reducing lending growth. This is illustrated in Figure 12 - although the interaction coefficient has to be large to make policies strategic complements and therefore give monetary policy a large role in leaning 7 An alternative way of introducing a risk-taking channel of monetary policy to the model would be to allow credit spreads to be negatively correlated with the level of interest rates. Given the endogenous cost-push effect of credit spreads in our Phillips Curve, this would be equivalent to introducing a negative version of a cost channel of monetary policy (Ravenna and Walsh, 2006). [29] This would make using macroprudential policy more costly, since monetary policy would be unable to costlessly offset its effects. But it would not qualitatively change the results presented in Section 3. 27 against the wind. 5. Conclusion Our simple framework allows us to explore various assumptions around how monetary and macroprudential policies affect the economy and interact with each other. We show that depending on the assumptions made, the two policies may be strategic substitutes or complements, ie they move in the opposite or the same direction. In particular, while our benchmark case suggests that they would typically be substitutes, they may be complements when an increase in capital requirements has particularly large negative supply effects or when such an increase affects aggregate demand positively. And under certain conditions, monetary policy can be seen as a last line of defence, for example when the effectiveness of capital in reducing the probability of crises decreases with its level. But under most assumptions the implications for monetary policy are relatively small. The findings do, however, speak to understanding the impact of changes in capital requirements better, in order to determine the optimal policy mix. References [1] David Aikman, Andrew Haldane, and Benjamin Nelson. Curbing the credit cycle. The Economic Journal, 2014. [2] Andrea Ajello, Thomas Laubach, David López-Salido, and Taisuke Nakata. Financial stability and optimal interest-rate policy. 2015. [3] Allen Berger and Christa Bouwman. How does capital affect bank performance during financial crises? Journal of Financial Economics, 109(1):146–176, 2013. [4] Ben S Bernanke and Mark Gertler. Should central banks respond to movements in asset prices? The American Economic Review, 91(2):253–257, 2001. 28 [5] Olivier Blanchard, Eugenio Cerutti, and Lawrence Summers. Inflation and activity–two explorations and their monetary policy implications. Technical report, National Bureau of Economic Research, 2015. [6] Claudio EV Borio and Philip William Lowe. Asset prices, financial and monetary stability: exploring the nexus. 2002. [7] Martin Brooke, Oliver Bush, Robert Edwards, Jas Ellis, Bill Francis, Rashmi Harimohan, Katharine Neiss, and Caspar Siegert. Measuring the macroeconomic costs and benefits of higher UK bank capital requirements. Financial Stability Paper, 35, 2015. [8] Jeremy I Bulow, John D Geanakoplos, and Paul D Klemperer. Multimarket oligopoly: Strategic substitutes and complements. Journal of Political economy, 93(3):488–511, 1985. [9] Stephen Burgess, Emilio Fernandez-Corugedo, Charlotta Groth, Richard Harrison, Francesca Monti, Konstantinos Theodoridis, and Matt Waldron. The Bank of England’s forecasting platform: COMPASS, MAPS, EASE and the suite of models. 2013. [10] Mark Carlson, Hui Shan, and Missaka Warusawitharana. Capital ratios and bank lending: A matched bank approach. Journal of Financial Intermediation, 22(4):663–687, 2013. [11] Charles Carlstrom, Timothy Fuerst, and Matthias Paustian. Optimal monetary policy in a model with agency costs. Journal of Money, Credit and Banking, 42(s1):37–70, 2010. [12] Richard Clarida, Jordi Gali, and Mark Gertler. The science of monetary policy: A New Keynesian perspective. Journal of Economic Literature, 37:1661–1707, 1999. [13] James Cloyne, Ryland Thomas, Alex Tuckett, and Samuel Wills. An empirical sectoral model of unconventional monetary policy: The impact of QE. The Manchester School, 83(S1):51–82, 2015. 29 [14] Vasco Curdia and Michael Woodford. Credit spreads and monetary policy. Journal of Money, Credit and Banking, 42(s1):3–35, 2010. [15] Mathias Drehmann, Claudio Borio, and Kostas Tsatsaronis. Characterising the financial cycle: don’t lose sight of the medium term! 2012. [16] Jeremy Franklin, May Rostom, and Gregory Thwaites. The banks that said no: banking relationships, credit supply and productivity in the UK. 2015. [17] Jordi Galı́. Monetary policy, inflation, and the business cycle: an introduction to the New Keynesian framework. Princeton University Press, 2008. [18] Nicola Gennaioli, Andrei Shleifer, and Robert Vishny. Neglected risks: The psychology of financial crises. Technical report, National Bureau of Economic Research, 2015. [19] Mark Gertler and Peter Karadi. A model of unconventional monetary policy. Journal of Monetary Economics, 58(1):17–34, 2011. [20] Simon Gilchrist, Raphael Schoenle, Jae Sim, and Egon Zakrajsek. Inflation dynamics during the financial crisis. 2015. [21] Macroeconomic Assessment Group. Assessing the macroeconomic impact of the transition to stronger capital and liquidity requirements, 2010. [22] Bengt Holmstrom and Jean Tirole. Financial intermediation, loanable funds, and the real sector. the Quarterly Journal of economics, pages 663–691, 1997. [23] Sujit Kapadia. Optimal monetary policy under hysteresis. Department of Economics, University of Oxford, 2005. [24] Mervyn King. Twenty years of inflation targeting, 2012. [25] Donald Kohn. Monetary policy and asset prices. Technical report, 2006. 30 [26] Donald Kohn. Implementing macroprudential and monetary policies: The case for two committees, 2015. [27] Joe Peek, Eric Rosengreen, and Geoffrey Tootell. Should US monetary policy have a ternary mandate? The Economic Journal, 2014. [28] Adam S Posen. Why central banks should not burst bubbles. International Finance, 9(1):109–124, 2006. [29] Federico Ravenna and Carl Walsh. Optimal monetary policy with the cost channel. Journal of Monetary Economics, 53(2):199–216, March 2006. [30] Hyun Song Shin. Macroprudential tools, their limits and their connection with monetary policy, 2015. [31] Jeremy Stein. Overheating in credit markets: Origins, measurement, and policy responses, 2013. [32] Lars Svensson. Discussion of Ajello, Laubach, López-Salido, and Nakata, ’Financial stability and optimal interest-rate policy’, 2015. [33] William R White. Is price stability enough? 2006. [34] Michael Woodford. Interest and prices: foundations of a theory of monetary policy. 2003. Appendix 1 - Solving the model Period 1 private sector expectations of period 2 The private sector has all of the same information as the policymaker conditional on the state, but does not know the probability of a crisis occurring between periods 1 and 2. Instead, agents assume there is an exogenous probability e of a crisis occurring. 31 So the private sector’s period 1 expectations are equal to: ps E1 π2 = eπ2,c + (1 − e)π2,nc = eπ2,c (21) ps E1 y2 = ey2,c + (1 − e)y2,nc = ey2,c (22) At time t = 1, the values these expectations are known by the policymaker, and crucially, independent of period 1 policy. So the period 1 policymaker’s problem reduces to a static problem. Period 1 The period 1 policymaker chooses period 1 settings of the CCyB and interest rates to minimise the Loss function: L= 1 2 (π + λy21 ) + βγ1 (1 + ζ ) E1 [ L2,c ] + β(1 − γ1 ) E1 [ L2,nc ] 2 1 (23) Subject to the Phillips Curve: ps π1 = κy1 + E1 π2 + νs1 + ξ 1π (24) and the IS curve ps ps y y1 = E1 y2 − σ(i1 − E1 π2 + ωs1 ) + ξ 1 (25) Since L2,nc = 0, the policymaker’s problem reduces to: mini1 ,k1 L = 1 2 (π + λy21 ) + βγ1 (1 + ζ ) E1 [ L2,c ] 2 1 (26) We then substitute (24) in place of inflation in (26), and (25) in place of output. And ps ps noting that E1 [ L2,c ], E1 π2 , and E1 y2 are exogenous parameters from the perspective of 32 the period 1 policymaker, differentiating with respect to i1 and k1 gives the two first order conditions which hold at the minimised value of L: ∂L ∂γ = −σ(κπ1 + λy1 ) + 1 β(1 + ζ ) E1 [ L2,c ] = 0 ∂i1 ∂i1 (27) ∂γ ∂L = −σωψ(κπ1 + λy1 ) + νψπ1 + 1 β(1 + ζ ) E1 [ L2,c ] = 0 ∂k1 ∂k1 (28) The crisis probability is given by: γ1 = exp(h0 + h1 B1 + h2 k1 ) 1 + exp(h0 + h1 B1 + h2 k1 ) (29) We can substitute into this the equation determining credit growth, as well as the Phillips Curve, IS curve and the macroprudential policy effect, to give an equation in terms of spreads and interest rates. Differentiating this gives the two marginal crisis risk equations: ∂γ1 exp(h0 + h1 B1 + h2 k1 ) = h (φ ) ∂i1 (1 + exp(h0 + h1 B1 + h2 k1 ))2 1 i (30) exp(h0 + h1 B1 + h2 k1 ) ∂γ1 = (h (ψφs ) + h2 ) ∂k1 (1 + exp(h0 + h1 B1 + h2 k1 ))2 1 (31) Which means the relative effect of each policy on the crisis risk is a constant parameter, which we define as A: A≡ ∂γ1 ∂k1 ∂γ1 ∂i1 = h1 ψφs + h2 h1 φi (32) Multiplying through (27) by − A, and adding to (28), gives: ( Aσ − σωψ)(κπ1 + λy1 ) + νψπ1 = 0 33 (33) This can be be expressed as λy1 + κπ1 = 0 (34) where λ := χσωψ νψ κ + χσωψ λ (35) and χ is a parameter defined as the comparative advantage of macroprudential policy at affecting the crisis probability, versus monetary policy at affecting the IS curve/aggregate demand. And νψ κ is the marginal effect of macroprudential policy on the natural rate of output. χ≡ ∂γ1 ∂k ( ∂γ1 1 ∂i1 ∂y1 ∂i1 ∂y1 ∂k1 − 1) (36) Equation (33) can also be used to back out the optimal relationship between the two policy instruments, interest rates and spreads. This can then be substituted into one of the two first order conditions, to give an equation determining the optimality condition trading off changes in the crisis probability with output/inflation losses in period 1: λy1 (− where ∂L ∂γ1 ∂γ ∂γ νψ ∂L νψ ) = ( 1 + 1( − ωψ))(− ) κ ∂k1 ∂i1 κσ ∂γ1 (37) is the policymaker’s expected discounted cost of a financial crisis, taking into account the extra weight ζ that they place on the expected cost of a crisis. ∂L = β((1 + ζ ) E1 [ L2,c ] − E1 [ L2,nc ]) ∂γ1 (38) The intution behind equation (37) can be seen by defining the natural rates of interest 34 and output as those that would obtain with flexible prices (but with credit frictions): λy1 ∂y1n ∂γ ∂γ ∂r n ∂L = ( 1 + 1 1 )(− ) ∂k1 ∂k1 ∂i1 ∂k1 ∂γ1 (39) At the optimal setting of policy, the marginal loss from changing output is equated to that from changing inflation. So the condition can equivalently be expressed as an optimality condition trading off inflation today with the cost of a financial crisis tomorrow (with output today unchanged.) This gives: ∂π ∂γ ∂γ π1 1 = ( 1 − 1 ∂k1 ∂k1 ∂i1 ∂y1 ∂k1 ∂y1 ∂i1 )(− ∂L ) ∂γ1 (40) The LHS is the marginal cost of increasing inflation through the cost-push effect of higher credit spreads. This is set equal to the marginal gain from a lower crisis probability, made up of two terms. The first term on the RHS is the marginal gain from a lower crisis probability of higher spreads. The second is the marginal increase in the crisis probability from offsetting the demand effects of higher spreads using interest rates. Appendix 2 - Uncoordinated policies Until now, we have considered monetary and macroprudential policies being set jointly, capturing cases when these are set by the same policymaker, or at least in a co-ordinated fashion. But in some locations the institutional set-up is such that two policymakers with different objectives have responsibility for setting the policies. In this section we examine the extent to which this leads to suboptimal outcomes. Outcomes may of course depend on how objectives are split. We try to capture how this has often been done in practice, by assigning our monetary policymaker a standard monetary policy loss function that ignores tail outcomes such as financial crises: LM = 1 2 (π + λi y21 ) 2 1 35 (41) For the macroprudential policymaker, one option would be to assign what is left from our joint loss function - its effect on the crisis probability. But this would be somewhat one sided. In practice, macroprudential seeks to balance its benefits with the costs of excessive regulation on intermediation and ultimately on the economy’s supply potential. We therefore also assign the macroprudential policymaker a desire to care about the natural rate of output, defined by (12). This gives the following macroprudential policy loss function (setting E1 [ L2,nc ] = 0): LF = 1 λ (yn )2 + βγ1 (1 + ζ ) E1 [ L2,c ] 2 k 1 (42) Nash policies In this subsection we examine how policy differs in the uncoordinated case, under the assumption that the two policymakers take each other’s policy as given in a Nash equilibrium, with both acting as Stackelberg leaders with the respect to the private sector. Minimising the monetary policymaker’s assigned section of the loss function: mini1 L M = 1 2 (π1 + λi y21 ) 2 (43) gives the familiar monetary policy first order condition under discretion (Clarida, Gali, Gertler, 1999) [12] σ (κπ1 + λi y1 ) = 0 (44) compared to the jointly optimal case (also given earlier by (11)): σ(κπ1 + λy1 ) = ∂γ1 ∂L ∂i1 ∂γ1 (45) Assuming (λi = λ), meaning that the monetary policymaker is assigned the same weight on output losses as the joint policymaker, then (44) implies that the under Nash, conditional on 36 a given level of the CCyB k1 , monetary policy is set looser than is optimal. This is because the monetary policymaker fails to take into account that raising interest rates also creates a positive externality for financial stability. The rise in interest rates also dampens credit growth and reduces the probability of a crisis in the subsequent period. Delegation means that the Nash monetary policymaker suboptimally fails to lean against the credit boom. The macroprudential policymaker minimises their loss function: mink1 L F = 1 λ (yn )2 + βγ1 (1 + ζ ) E1 [ L2,c ] 2 k 1 (46) to give the first order condition: ∂L F νψλk n ∂γ1 ∂L F y1 + =− =0 ∂k1 κ ∂k1 ∂γ1 (47) or rearranging, λk y1n (− ∂γ νψ ∂L ) = 1 (− F ) κ ∂k1 ∂γ1 (48) The equation sets the marginal loss from a higher CCyB lowering the natural rate of output equal to the marginal benefit from reducing the crisis probability. This compares to the jointly optimal intertemporal condition (given earlier as (15)): λy1 (− νψ ∂γ ∂L ∂γ νψ ) = ( 1 + 1( − ωψ))(− F ) κ ∂k1 ∂i1 κσ ∂γ1 (49) Again assuming that (λk = λ), there are two differences to the macroprudential policymaker’s decision under Nash. First, on the left hand side of (48), the Nash policymaker is assigned to care about the natural rate of output, rather than the actual level of output. This leads to a higher marginal cost of using the CCyB, since the Nash policymaker disregards that the cost-push effect of using the CCyB can be optimally split between two goals (output and inflation), rather than borne solely by one (potential output). Given the quadratic nature of losses, this leads them to overstate the marginal cost of the CCyB. On the right hand side of (48), however, the Nash macroprudential policymaker is 37 missing the term that depends on the marginal effect of changing the interest rate on the crisis probability. This second missing effect derives from the fact that under Nash, the macroprudential policymaker ignores the externality that the CCyB exerts on demand and inflation via the output gap. If, as under our benchmark calibration, the demand effect of the CCyB is larger than the supply effect, this is a negative externality. The CCyB creates a negative output gap which is costly over and above its effect on potential output. Even though interest rates can be cut to offset the externality - and are, in the Nash case - such a monetary loosening increases the crisis probability. The Nash macroprudential policymaker fails to take account of this, since the output gap does not enter their loss function. Consequently, they overstate the marginal benefits of using the CCyB. If the supply effect of the CCyB were larger, the opposite would be true, and the Nash policymaker would understate the policy’s marginal benefits, relative to the jointly optimal case. Depending on the relative sizes of the two effects, the CCyB may be set suboptimally high or low by the uncoordinated macroprudential policymaker. Under our benchmark calibration delegation means that the CCyB is set too low for a given monetary policy setting. And in response, interest rates loosen by less under Nash policies than if they were responding to the higher, jointly optimal setting for the CCyB. Overall, the Nash settings of policy implies too tight a setting for macroprudential policy and too loose a setting for monetary policy. But for practical policymaking purposes, there may be other gains from delegating responsibilities. A key question is therefore whether these suboptimal settings are quantiatively large. To get a sense of this, Figure 13 plots the policy functions for interest rates and the CCyB as credit growth varies in response to a credit shock, assumed for simplicity to be uncorrelated with the other shocks. Relative to the co-ordinated case, delegation has only a small effect on policy settings. For our benchmark calibration, while there it is not strictly true that macroprudential policy and monetary policy should only focus on their respective goals, doing so is nonethelesss a fairly close approximation to jointly optimal 38 policy. If there are any other material gains from splitting objectives not captured in our simple framework - such as improved accountability, or greater specialised expertise on committees, for example - then these could easily outweigh any losses. Stackelberg policies An alternative way of delegating responsibilities would be to assign one of the policymakers to act first within period 1, as a Stackelberg leader, with the other policymaker as a Stackelberg follower.8 Given the longer timeframe that macroprudential policy operates over, it seems natural to assign the macroprudential policymaker as the first mover, the Stackelberg leader. As the Stackelberg follower, the monetary policymaker’s decision is identical to under Nash equilibirum policy - macroprudential policy is predetermined, so taken as given. Condition (44) again determines interest rates for a given setting of the CCyB, ignoring the externality from monetary policy to financial stability. By acting first, the macroprudential policymaker is able to take advantage of this, however, and take into account the effect that changes in the CCyB have on interest rates and the knock-on effect to financial stability. The size of this effect can be seen by substituting the IS curve (1), Phillips Curve (2) and spreads equation (6) into (44). Rearranging gives the policy function for interest rates, conditional on k1 : i1 = σ (κ 2 κ 1 ps ps ps ( E1 π2 + νψk1 + ξ 1π ) + E1 y2 + E1 π2 − ωψk1 σ + λi ) (50) Differentiating this with respect to k1 gives the effect of the CCyB on interest rates: di1 κ 2 νψ = 2 − ωψ dk1 κ + λi σκ (51) The Stackelberg macroprudential policymaker takes this effect into account when setting 8 In both case the private sector is assumed to be a Stackelberg follower to both policymakers, however. 39 the level of the CCyB. With i1 a function of k1 , rather than taken as given, the (rearranged) first order condtion for the problem mink1 L F = 1 λ (yn )2 + βγ1 (1 + ζ ) E1 [ L2,c ] 2 k 1 (52) becomes λk y1n (− ∂γ ∂γ di ∂L νψ ) = ( 1 + 1 1 )(− F ) κ ∂k1 ∂i1 dk1 ∂γ1 (53) Substituting (51) into this gives the macroprudential policymaker’s optimality condition: νψ λk y1n (− ) κ ∂γ1 ∂γ1 ∂L κ 2 νψ =( − ωψ))(− F ) + ( 2 ∂k1 ∂i1 κ + λi σκ ∂γ1 (54) As with the Nash policymaker, the left hand side of (54) shows that the Stackelberg policymaker overstates the marginal cost of using the CCyB. But differently, the second term on the right hand side is unambiguously smaller than the equivalent term in the jointly optimal policy condition - the Stackelberg policymaker always understates the marginal benefit. Intuitively, this is because the macroprudential policymaker is forced to not only internalise all of the the externality the CCyB exerts by creating an output gap, but also an additional effect, coming from the monetary policymaker choosing to run a positive output gap in response to the cost-push effects of the CCyB. Put differently, interest rates are cut to fully offset effects of the CCyB on demand, but only raised enough to partly offset the effects of the CCyB on supply. This asymmetry means the Stackelberg macroprudential policymaker is aware that any tightening will be offset by loose monetary policy more than is jointly optimal. As a result of higher marginal costs and lower marginal benefits, the CCyB is set unambiguously lower than under jointly optimal policy. For a given level of the CCyB, interest rates are set lower, although if the two instruments are substitutes, they may not be unconditionally lower. Figure 13 also plots the policy functions under Stackelberg competition. Again, quantitatively, the differences between the three sets of policies are 40 very small. Appendix 3 - Long-run potential supply effects - motivating different weights in the loss function In our benchmark model macroprudential policy and credit spreads have a negative impact on the economy’s supply potential. This is essentially a static effect, however. Potential output is lower when macroprudential policy is tight, but the effect reverses the moment policy is loosened. And the event of a financial crisis has no impact on the economy’s supply potential. In practice, one possible cost of macroprudential policy is that it may have persistent effects on the supply capacity of the economy. Similarly, to the extent running negative output gaps induces hysteresis, tight monetary policy may have a similar effect. As might the financial crisis itself. In principle one could model each of these effects by allowing policy or financial crises to have effects on level of potential output consistent with stable prices. But as we show below, doing so is isomorphic to endogenising the weights on different parameters in the loss function. Specifically: If macroprudential policy affects period 2 potential output, the policymaker should include minimising deviations in the macroprudential tool in the loss function, where the weight on the CCyB depends positively on the potential output effect If the period 1 output gap affects period 2 potential output, the policymaker should place a larger weight (λ) on period 1 output in their loss function the larger is the potential output effect If a financial crisis reduces period 3 potential output, the policymaker should place a larger weight (ζ) on the expected cost of a financial crisis in the loss function, where the weight depends positively 41 on the potential output effect We show this in the next three subsections. For simplicity and without loss of generality, we assume no shocks materialise and that this is known by everyone. So the only uncertainty is over whether a financial crisis occurs or not. We maintain the assumption that the crisis is a one off event, which either occurs between periods 1 and 2, or not at all. Macroprudential policy affects period 2 potential output The simplest way of introducing such effects would be to make period 2 potential output n ), a linear function of period 1 credit spreads.9 in the absence of a crisis, (y2,nc n y2,nc = τs1 = τψk1 (55) In period 2, in the event of no financial crisis, the policymaker would then minimise the following welfare loss function: L2,nc = 1 2 (π + λy22,nc ) 2 2,nc (56) subject to the Phillips Curve: (in the absence of shocks and since period 2 is the terminal period, future expectations play no role) n π2,nc = κ (y2,nc − y2,nc ) (57) giving the first order condition: κπ2,nc + λy2,nc = 0 9 One (58) could also assume that period 2 potential output is affected if there is a crisis too, although it seems reasonable to think of crises as being large negative demand shocks that policy is unable to fully offset, such that a small fall in potential output would not make much difference to the crisis loss, and may actually reduce it, be preventing inflation falling as far below target. 42 which can be substituted back into (57) for π2,nc to give (κ 2 + λ)y2 = κ 2 y2n (59) and minimised welfare loss of L2,nc = 1 λκ 2 τ 2 ψ2 2 1 λκ 2 n 2 ( y ) = k 2 κ2 + λ 2 2 κ2 + λ 1 (60) Therefore the expected discounted welfare loss in the event of no crisis can be given as: 1 λk k21 2 (61) λκ 2 τ 2 ψ2 κ2 + λ (62) βL2,nc = where λk ≡ This implies that the period 1 policymaker’s loss function, L= 1 2 (π + λy21 ) + βγ1 (1 + ζ ) E1 [ L2,c ] + β(1 − γ1 ) E1 [ L2,nc ] 2 1 (63) can then be written L= 1 2 (π1 + λy21 + (1 − γ1 )λk k21 ) + βγ1 (1 + ζ ) E1 [ L2,c ] 2 (64) The weight that the policymaker places on minimising deviations in the CCyB from zero, (1 − γ1 )λk , is positively related to the size of the hysteresis effect, τ. It also depends endogenously on the probability of a crisis. As the probability of a crisis increases, the policymaker adjusts optimal policy less due to potential output effects, because the effects only matter when a crisis does not occur. 43 Hysteresis effects - output gaps affect period 2 potential output As in Kapadia (2005) [23], we could also let period 2 potential output be affected by period 1 output, irrespective of whether period 1 ouptut was affected by monetary or macroprudential policy. This is a true hysteresis effect, rather than a long-run supply effect of macroprudential policy. Again, we assume that the effect only affects potential when a crisis does not occur. n y2,nc = f y1 (65) As in Kapadia (2005), we impose an upper bound on hysteresis by having the period 2 policymaker choose i2,nc or equivalently y2,nc to minimise the loss function: L2,nc = 1 n 2 (λ(y2,nc − y2,nc )2 + π2,nc + λh (yn − y2,nc )2 ) 2 (66) subject to the Phillips Curve: n π2,nc = κ (y2,nc − y2,nc ) (67) n Gives optimal period 2 output as a weighted average of y2,nc and its maximum value: y2,nc = λ + κ2 λh n y + yn 2,nc λ + κ 2 + λh λ + κ 2 + λh (68) Substituting this back into the Phillips Curve (67) gives optimal period 2 inflation of π2,nc = κλh n (yn − y2,nc ) λ + κ 2 + λh (69) So minimised welfare loss is equal to: L2,nc = 1 (λ + κ 2 )λh n n ( (y − y2,nc )2 ) 2 2 λ + κ + λh 44 (70) or equivalently L2,nc = 1 (λ + κ 2 )λh 2 yn 2 ( )y ) + t.i.p. f ( 1 − 2 λ + κ 2 + λh f y1 1 (71) where t.i.p stands for the term in yn that is independent of policy. This implies the period 1 loss function can be rewritten as: L= 1 2 (π + λy21 ) + βγ1 (1 + ζ ) E1 [ L2,c ] + t.i.p. 2 1 (72) where (λ + κ 2 )λh 2 yn λ ≡ λ + β(1 − γ1 ) ) f ( 1 − f y1 λ + κ 2 + λh (73) All of the optimality condtions from Appendix 1 still hold, but now the weight on period 1 output in the loss function is an endogenous function of the hysteresis effects, and of the crisis probability. As long as the limit on the hysteresis channel is large enough n that y2,nc < yn , then the weight placed on output deviations lambda will be strictly greater than the policymaker’s period preferences, reflecting the additional hysteresis benefit the following period. A low crisis probability also means a greater weight is placed on period 1 output than in the case with no hysteresis, since there is a high probability that the policymaker will get to benefit from higher period 2 potential output. Financial crises have persistent effects on the economy’s supply capacity A final type of hysteresis effect that could affect optimal policy is if financial crises have persistent effects on potential supply, but smaller reductions in output do not. We can model this simply by assuming a third period, where potential output is unchanged from steady state if there was not a crisis in period 2, but falls if there was. The parameter p 45 represents the size of this hysteresis channel. y3∗ = 0 if no crisis in period 2 − p if crisis in period 2 (74) In the event of no crisis between periods 1 and 2, the policymaker will be able to achieve 0 welfare loss in period 3. If there is a crisis however, the policymaker will select i3 , or equivalently y3 , to minimise the loss function: L3,c = 1 2 (π + λy23,c ) 2 3,c (75) subject to ∗ π3,c = κ (y3,c − y3,c ) (76) Doing so gives the following standard first order condition, specifying the optimal output inflation trade-off: ∂L3,c = κπ3,c + λy3,c = 0 ∂y3,c (77) ∗ and rearranging gives an equation for y Substituting in (76) for π3,c , (74) for y3,c 3,c as a function of the hysteresis effect, p. y3,c = − κ2 p κ2 + λ (78) κλ p +λ (79) And substituting into (76) for y3,c gives π3,c = κ2 46 In the event of a crisis, period 3 loss is therefore equal to: L3,c = 1 2 κ2 λ [π3,c + λy23,c ] = p2 2 2(κ 2 + λ ) (80) We assume that there is only one possible crisis event - between periods 1 and 2. So period 2 discretionary policy has no effect on period 3 outcomes. The period 1 loss function is therefore: L= 1 2 (π + λy21 ) + (1 − γ1 )( βE1 [ L2,nc ] + β2 E1 [ L3,nc ]) + γ1 (1 + ζ )( βE1 [ L2,c ] + β2 E1 [ L3,c ]) 2 1 (81) Using the fact that expected welfare loss in periods 3 is 0 if there is no crisis as well as equation (80), this can be rewritten as L= 1 2 β2 κ 2 λ 2 (π1 + λy21 ) + (1 − γ1 ) βE1 [ L2,nc ] + γ1 (1 + ζ )( βE1 [ L2,c ] + p ) 2 2(κ 2 + λ ) (82) or equivalently as L= 1 2 (π1 + λy21 ) + (1 − γ1 ) βE1 [ L2,nc ] + γ1 β(1 + ζ ) E1 [ L2,c ] 2 (83) where ζ ≡ ζ + (1 + ζ ) κ2 λ βp2 2(κ 2 + λ) E1 [ L2,c ] (84) ζ is therefore equal to the additional weight placed on avoiding crises, plus an additional term that is an increasing function of the size of those hysteresis effects. This set-up is equivalent to a model with no perisistent hysteresis effects but a higher value of ζ. Indeed, such persistent effects could be a justification for the ζ term itself. If ζ = 0, so that the 47 policymaker only cares about mean expected losses, then (84) becomes ζ≡ κ2 λ βp2 2(κ 2 + λ) E1 [ L2,c ] (85) implying that any additional weight placed on avoiding crises is solely due to their persistent effects on potential output. Figures Annual probability of a crisis (%) 7 CCyB of 0% CCyB of 1% CCyB of 2.5% Steady state credit growth 6 5 4 3 2 1 0 −2 0 2 4 6 8 10 12 Annual real credit growth (%) Figure 1: Crisis probability and credit growth for different levels of the CCyB 48 Annual probability of a crisis in the UK (%) 14 12 10 8 6 4 2 0 1985 1990 1995 2000 Year 2005 2010 2015 Figure 2: Model-implied crisis probability for the UK Period 1 welfare loss (as % of crisis loss) 5 0% cumulative 3 year real credit growth − monetary policy only 30% cumulative 3 year real credit growth − monetary policy only 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Period 2 annual crisis probability (%) Figure 3: Monetary and financial stability trade-off with monetary policy only 49 Period 1 welfare loss (as % of crisis loss) 5 0% cumulative 3 year real credit growth − monetary policy only 30% cumulative 3 year real credit growth − monetary policy only Two instruments (no supply costs of CCyB) 0% cumulative 3 year real credit growth − two instruments (supply cost) 30% cumulative 3 year real credit growth − two instruments (supply cost) 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Period 2 annual crisis probability (%) CCyB setting (%) Figure 4: Monetary and financial stability trade-off with monetary policy only and with two instruments 10 5 0 −20 −15 −10 −5 0 5 10 15 20 25 30 15 20 25 30 Annual real credit growth (%) Policy rate (%) 3 2.8 2.6 2.4 2.2 −20 −15 −10 −5 0 5 10 Annual real credit growth (%) Figure 5: CCyB and interest rate policy functions as size of credit shock varies 50 Annual crisis probability (%) After policy annual real credit growth (%) 20 10 0 −10 −20 −20 −10 0 10 20 Annual real credit growth (%) 30 2 1.5 1 0.5 0 −20 −10 0 10 20 30 Annual real credit growth (%) Annual inflation (%) Output (% deviation from efficient) 0 −0.2 −0.4 −0.6 −0.8 −20 −10 0 10 20 2.03 2.02 2.01 2 −20 30 −10 Annual real credit growth (%) 0 10 20 30 Annual real credit growth (%) Figure 6: Optimal policy outcomes as size of credit shock varies Period 1 welfare loss (as % of crisis loss) 5 0% cumulative 3 year real credit growth − zero lower bound 30% cumulative 3 year real credit growth − zero lower bound 0% cumulative 3 year real credit growth − two instruments 30% cumulative 3 year real credit growth − two instruments 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Period 2 annual crisis probability (%) Figure 7: Optimal policy under different initial credit conditions, with and without binding zero lower bound 51 10 Both instruments Zero lower bound 9 CCyB setting (%) 8 7 6 5 4 3 2 1 0 −20 −15 −10 −5 0 5 10 15 20 25 30 Annual real credit growth (%) Figure 8: CCyB policy function as credit shock varies, with and without binding zero lower bound 2.5 CCyB Policy rate Optimal policy change (pp) 2 1.5 1 0.5 0 −0.5 −1 credit demand cost−push credit and demand 1 standard deviation shock: Figure 9: Optimal policy response to different shocks 52 3.8 Benchmark calibration CCyB boosts aggregate demand CCyB has large potential supply effects (5 x Benchmark) 3.7 Optimal interest rate (%) 3.6 3.5 3.4 3.3 3.2 3.1 3 2.9 2.8 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 Optimal CCyB setting (%) 10 5 −20 0 20 CCyB setting (%) 15 10 8 6 4 2 −20 Annual real credit growth (%) 2.1 2.05 2 1.95 −20 0 0 20 Annual real credit growth (%) Policy rate (%) Inflation (Ann. %) Annual crisis prob. (%) Figure 10: Optimal CCyB and interest rate under different sets of parameters 20 3 2.8 2.6 2.4 2.2 −20 Annual real credit growth (%) 0 20 Annual real credit growth (%) No market−based finance Market−based finance sector has 75% lending share Figure 11: Optimal policy response to a credit shock under different policymaker preferences, with and without a market-based finance sector 53 CCB setting (%) 1 0.5 0 5 10 15 (1+ζ), FS weight in loss function 2.1 2 1.9 1.8 0 5 10 15 (1+ζ), FS weight in loss function 8 6 4 2 0 20 Policy rate (%) Annual crisis probability (%) Inflation (Ann. %) 2 1.5 20 0 3 2.8 2.6 2.4 2.2 0 5 10 15 (1+ζ), FS weight in loss function 5 10 15 20 20 (1+ζ), FS weight in loss function No risk−taking channel Large risk−taking channel (φis=−5000) CCyB setting (%) Figure 12: Optimal policy response to a credit shock under different policymaker preferences, with and without non-linear interaction term between policies 7 6 5 10 12 14 16 18 20 Annual real credit growth (%) Policy rate (%) 2.65 2.6 2.55 Jointly optimal policies Uncoordinated policy (Nash) Uncoordinated policy (Stackelberg) 2.5 2.45 10 12 14 16 18 20 Annual real credit growth (%) Figure 13: Optimal policy functions under coordination, Nash equlibrium and Stackelberg equilibrium 54