Leaning Against the Wind: A Cost-Benefit Analysis for an Integrated Policy Framework

Luis Brandao Marques; R. G Gelos; Machiko Narita; Erlend Nier

doi:10.5089/9781513549651.001.A001

Author: Mr. Luis Brandao Marques¹, Mr. R. G Gelos https://orcid.org/0000-0003-4362-676X, Mr. Machiko Narita, and Erlend Nier

¹ 404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Author’s E-Mail Address: LMarques@IMF.org, GGelos@IMF.org, MNarita@IMF.org, ENier@IMF.org.

Publication Date:: 07 Jul 2020

eISBN:: 9781513549651

Language:: English

Keywords:: WP; financial conditions; loss function; Monetary Policy; Macroprudential Policy; FX Interventions; Capital Flow Management; Cost-Benefit Analysis; monetary policy shock; GDP growth; quantile regression; central bank; prudential policy; monetary policy tightening; Macroprudential policy instruments; Credit; Financial conditions index; Global; quantile function; monetary policy condition; monetary policy announcement; monetary policy measure; monetary policy surprise

Download PDF (2 MB)

This paper takes a new approach to assess the costs and benefits of using different policy tools—macroprudential, monetary, foreign exchange interventions, and capital flow management—in response to changes in financial conditions. The approach evaluates net benefits of policies using quadratic loss functions, estimating policy effects on the full distribution of future output growth and inflation with quantile regressions. Tightening macroprudential policy dampens downside risks to growth stemming from loose financial conditions, and is beneficial in net terms. By contrast, tightening monetary policy entails net losses, calling for caution in the use of monetary policy to “lean against the wind.” These findings hold when policies are used in response to easing global financial conditions. Buying foreign-exchange or tightening capital controls has small net benefits.

Abstract

I. Introduction

An active debate in academia and policy circles centers on which policies should be used to address financial stability risks (e.g., Adrian and Liang 2018). Should macroprudential policy be the first line of defense? Or should monetary policy help “lean against the wind” of building financial vulnerabilities? What are the costs and benefits of using one set of policies versus the other?

In the eyes of many policymakers, macroprudential policies are preferable, since they can be better targeted at emerging risks. Monetary policy, on the other hand, is a blunt tool, which comes with potentially large costs to the economy at large (see Yellen, 2014, and references therein). Moreover, a key benefit of macroprudential policy is that it can directly strengthen resilience to future shocks, by bolstering the balance sheets of borrowers and lenders, something which monetary policy cannot do. However, as argued by many, macroprudential policy is subject to “leakage effects” (Bengui and Bianchi 2018), reducing its effectiveness, while monetary policy has the advantage of “getting into all the cracks” (Stein 2013).¹ Moreover, the macroprudential toolkit remains underdeveloped in many countries. And since macroprudential policy is relatively new, knowledge about its overall benefits and costs, and how these compare to those of monetary policy, is only gradually accumulating.

Financially integrated open economies face the additional challenge of how to respond to an easing of global financial conditions that might drive up the exchange rate and stimulate capital inflows and domestic credit. As is well acknowledged by now, flexible exchange rates typically do not fully insulate the domestic economy from the effects of such swings in global financial conditions (Obstfeld 2015, Rey 2013, Arregui and others 2018). In practice, emerging markets in particular have been using different and varying mixes of policies, also including foreign exchange intervention and capital controls, when confronted with rapidly changing global financial conditions.

This paper proposes a new empirical approach to assess the costs and benefits of using different policy tools to dampen the buildup of financial vulnerabilities from changes in domestic and global financial conditions. Building on the study by Adrian, Boyarchenko, and Giannone (2019), it estimates the policy effects on the entire distributions of future real GDP growth and inflation, and then evaluates their net benefits using quadratic loss functions. This new approach relates to and advances on various strands of the existing literature.

First, a small number of studies examine whether monetary policy should lean against the wind in a cost-benefit framework based on empirical estimates comparing the benefits of raising policy rates—lowering the probability of a future crisis—to its short-term costs— namely inducing higher unemployment (IMF 2015, Svensson 2017). These studies have generally concluded that the costs of leaning outweigh its benefits. While offering some useful insights, the approach has several limitations. For one, it assumes a binary partition of “crisis” versus “no crisis” states, while financial stability policies may be beneficial to mitigate the probability and severity of a continuum of outcomes, including standard recessions, financial recessions and all-out crises (Claessens, Kose, and Terrones, 2011). Moreover, this literature has struggled to pin down with precision the effects of changes in policy rates on crisis probabilities, since several and potentially opposing effects may be at work. Finally, it relies on assumptions on the cost of crises, and is subject to subtle arguments as to how a policy of leaning against the wind might itself affect the severity of crises—for example, does leaning create policy space for rate cuts, or does it weaken the economy, thereby exacerbating a crisis?²

Second, a growing literature examines empirically the effectiveness of macroprudential in achieving its objectives (see the studies covered in the survey by Galati and Moessner 2018). Most of these studies find that macroprudential policies can slow credit growth, with effects generally measured as stronger for borrower-based tools, such as loan to value (LTV) and debt service to income (DSTI) constraints. Some studies also consider costs, such as adverse short-run effects on consumption and growth (Alam and others 2019, Richter, Schularick, and Shim 2019). However, this literature has not yet advanced to a comprehensive assessment of the costs and benefits of macroprudential policy. For one, while increasing resilience of financial institutions and borrowers to future shocks is a primary objective of macroprudential policy, with only very few exceptions (e.g., Jiménez and others 2017) the literature has not yet attempted to quantify these effects. Second, and relatedly, empirical studies have so far mostly been confined to measuring effects on intermediate targets (e.g., credit growth) and have not arrived at quantitative assessments of effects on the ultimate objectives of macroprudential policy: to reduce tail risks to future output growth.

A third group of studies introduces a macroprudential policy tool into DSGE models, using a loss function that is typically augmented to include the volatility of credit in addition to output and inflation.³ These studies generally conclude that when macroprudential policy is available to control the volatility of credit, monetary policy is best focused on output and inflation. While intuitive, a key limitation of the approach is that the transmission of the macroprudential tool is stylized, often amounting to introducing a wedge between interbank and lending rates. In reality, by contrast, the transmission of macroprudential policy to financial stability also includes a resilience effect, and the effect on credit dynamics differs across tools. A second key limitation is that the volatility of credit is an ad-hoc addition to the loss function, whereas the ultimate objective of macroprudential policy is to avoid adverse effects of financial shocks on output (IMF 2013, Adrian and others 2018).

Lastly, a large body of literature has covered policy choices of small open economies, and in particular emerging markets, when faced with external shocks. Largely, the prescription emerging from these models has been that monetary policy should focus on domestic inflation and let the exchange rate float (see, among many others, Clarida, Galí and Gertler, 2000; and Devereux, Lane, and Xu, 2006). More recently, many models have started to build in financial frictions (e.g. Aoki, Benigno, and Kiyotaki 2016; Céspedes, Chang and Velasco, 2017; and Choi, 2020), and partly find justification for Pigovian taxes on capital inflows (e.g., Korinek, 2018), or the use of foreign-exchange intervention that builds reserves and deploys these to limit disruptive deprecation in the event of outflows.

The starting point for the novel empirical approach developed in this paper is an intertemporal trade-off where improving financial conditions today can lead to an increase in financial vulnerabilities that ultimately puts future growth at risk. As first established by Adrian and others (2018) and formalized in internally coherent models with rational expectations (e.g., Adrian et al., forthcoming), easy financial conditions tend to boost output growth in the short term but are associated with greater downside risks to growth in the medium term.⁴

Using our new framework, we ask: which policies can ameliorate the trade-off? In a first step, we employ quantile regressions to compare how policies affect the trade-off between current financial conditions and the expected distribution of future GDP growth and inflation. This analysis uses quarterly data for financial conditions, economic growth and policy indicators over the period 1990 to 2016. To identify the effect of policies on the trade-off we examine policy surprises, constructed as deviations from estimated policy rules. In a second step, to obtain estimates of the impact of policies on the entire probability distribution of future GDP growth, we fit the empirical conditional quantiles to a known distribution function (skewed-Normal). The third and final step is to feed these distributions into a standard loss function and compare loss function values with- and without policy actions.

We conduct this analysis both for changes in domestic and global financial conditions. For an easing of global financial conditions, we also consider foreign-exchange intervention and capital flow management policies that countries might want to use to ‘lean against’ the resulting appreciation and increase in capital inflows. We finally evaluate the interactions between monetary- and macroprudential policies, by examining the impact on the loss functions of a joint use of both sets of policies.

Relative to the existing literature, the novel approach developed in this paper makes at least three important advances. First, it extends beyond examining the effects of macroprudential policies on intermediate outcomes, by offering a framework that allows an assessment of effects on the ultimate objective, namely to contain tail risks to real economic outcomes. Second, it moves beyond the measurement of costs and benefits using a binary partition into crisis and non-crisis states, by considering instead the whole future distribution of output growth and measuring costs and benefits across all relevant states of the world. Third, it does not make any simplifying and potentially counterfactual assumptions about the transmission of macroprudential policy (or monetary policy). If macroprudential policy affects the ultimate objectives of the policymaker both by slowing credit growth and by bolstering the resilience of borrowers and banks to future shocks, then both these benefits would come through in the estimated impact on tail risks to GDP.

Overall, despite stark differences in the methods used, our results on the effectiveness of macroprudential policy line up with and strengthen the tentative conclusions reached by the existing literature. We find strong and new evidence that macroprudential policy lessens the trade-off between presently looser financial conditions and greater future downside risks to growth. The average tightening of macroprudential measures is associated with large loss reductions, and these benefits are found to be pronounced in particular for the tightening of borrower-based tools, such as caps on LTV and DSTI.

Moreover, tightening borrower-based macroprudential policies when vulnerabilities (as measured by credit aggregates or asset prices) are high is more decisively associated with a reduction in losses than when vulnerabilities are still low. However, financial-institutions-based tools, such as capital and liquidity tools, appear to be most useful in building resilience even when vulnerabilities are still modest.

By contrast, a tightening of monetary policy does not appear to be able to improve the tradeoff between loose financial conditions condition today and future tail risks to GDP growth. Indeed, on net, the tightening of monetary policy is associated with higher losses over the policy horizon. This confirms in our novel set-up that the economic costs of a tightening of monetary policy outweighs any financial stability benefit stemming from such a tightening.

These results largely extend when we consider potential responses to changes in global financial conditions. A tightening of macroprudential policies appears to be the most effective response to an easing of global financial conditions. Leaning against such easing with tighter monetary policy or foreign-exchange interventions entail very small (but positive) effects over the policy horizon. Likewise, tightening CFMs to counter loose global financial conditions seems to bring only a weak reduction in the loss function.

The analysis also suggests the existence of benefits from a joint use of macroprudential and monetary policies. In particular, when a tightening of macroprudential policies is accompanied by looser monetary policy, this translates into a larger reduction in the central bank’s loss function than when macroprudential tightening is conducted on its own. This finding supports the idea that accommodative monetary policy can be used at the margin to cushion any adverse effects on output of the tightening of macroprudential policy, resulting in an even larger welfare gain. The results are broadly robust to alternative specifications of loss functions, including an asymmetric loss function that penalizes downward deviations more than upward ones.

The remainder of the paper is organized as follows. In Section II we provide more detail on the econometric approach to estimating growth at risk and the data used for the analysis. Section III presents a first comparison of results comparing no-policy outcomes with outcomes where policy is used. Section IV examines impacts on the entire distribution and evaluates loss functions. Section V concludes.

II. Empirical Approach

A. Quantile Regressions

Our approach takes the stylized facts laid out by Adrian and others (2018) as a starting point: easy financial conditions tend to boost output growth in the short term but tend to be followed by greater downside risks to growth in the medium term.

A first aim is to investigate whether and how different types of policies can ameliorate this trade off. For this purpose, we first estimate fixed-effects quantile regressions (FE-QR). For each horizon h=1…H, we use the FE-QR estimator described in Kato et al. (2012):⁵

\begin{matrix} Q_{Y_{i, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ^{h}, & (1) \\ h = 1 \dots H, q = 0.05 \dots 0.95, \end{matrix}

where Y_i,t+h is a measure of economic activity in country i and quarter t+h, f_it is a financial condition index (FCI), P_it is the policy variable, x is a vector of controls including current GDP growth, inflation and credit growth, and Q is the conditional q-th quantile function of Y_i,t+h, and Z≡[f_it, P_it, x_it].

As our measure of economic activity, we consider a simple measure of the output gap given by the log level of real GDP (y_it) as a deviation from a linear time trend ${\bar{y}}_{i t}$ , that is $Y_{i, t + h} \equiv (y_{i t + h} - {\bar{y}}_{i t + h}) - (y_{i t} - {\bar{y}}_{i t})$ .⁶ We also estimate the effects on inflation with an analogous specification. A higher FCI represents looser financial conditions and a higher policy variable indicates a tightening (see Section III). Note that the specification assumes symmetric effects for tightening and loosening policy measures.

After estimating the conditional quantile functions given by (1), we are interested in testing the hypothesis that a given policy P changes the effect of f on the distribution of future GDP growth. In particular, for large h (i.e., at a long horizon) and low q (e.g., q=0.1), we expect β₁ to be negative—looser financial conditions today are associated with a fatter left tail of distribution of GDP growth. For a policy P to be effective in containing this buildup of risk, we would need β₃ to be positive at the same horizon and for the same quantiles. However, the required inference to test this hypothesis is complicated by the possibility that our data may feature significant serial correlation.⁷ We take this into account by calculating the standard errors and confidence bands of the estimators of the parameters in (1) using Hagemann’s (2017) cluster-robust bootstrap.⁸

B. Integrating and Computing Loss Functions

However, to obtain a full picture of the effects of policies, it is not sufficient to focus on how policies can change downside risk to GDP. For example, the policies may change both the left and right tails of the conditional distribution of future GDP growth. In addition, they may shift the entire distribution, not just change its dispersion or the shape of its tails.

Our analysis therefore uses summary statistics of the conditional distributions of GDP growth obtained at various horizons and aggregates them into a loss function which is intended to represent the policymaker’s preferences over economic outcomes. Here, we use a quadratic loss function as is standard in New Keynesian models. The period loss function is:

\begin{matrix} ∣_{t} = ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} . & (1) \end{matrix}

The comparison of policies is then based on the expected discounted value of future losses, written as follows:

\begin{matrix} \begin{matrix} L_{0} (Θ) & = & \sum_{h = 0}^{H} β^{h} {\hat{E}}_{t} (∣_{(t + h)} | P_{t} = 0, Θ), P_{t} \in {M P M_{t}, M P_{t}, F X I_{t}}, \\ a n d Θ & = & [θ^{1} θ^{2} Κ θ^{H}] \end{matrix} & (2) \end{matrix}

for the case of no policy action. In (3), θ^h is the vector of parameters describing the conditional distribution function of h-step ahead detrended output and π is the inflation rate. Similarly, when a given policy is being used, we have:

\begin{matrix} L_{1} (Θ) = \sum_{h = 0}^{H} β^{h} {\hat{E}}_{t} (∣_{t + h} | P_{t} = σ_{P}, Θ), & (3) \end{matrix}

where σ_p is the sample standard deviation of the relevant policy shock. The net benefit of policy P is then given by (L₁-L₀)/ L₀ from (3) and (4).

The first step is to choose the weights ω_y and ω_π in (2). We start with a simple quadratic loss function using output only (i.e., with ω_y =1 and ω_π =0). We then consider ω_y =0.542 and ω_π =1, which are Debertoli and others’ (2018) optimal weights when economic activity is measured as a deviation from trend output.⁹

The next step is to estimate the moments of the conditional distribution of GDP growth (and inflation) using the empirical quantile functions estimated in Section II A. We follow Adrian and others (2019) and use a minimum distance estimator to fit the estimated conditional quantiles to a theoretical distribution. Unlike Adrian and others, however, for computational ease, we use the Skewed-Normal distribution (Azzalini 1985) instead of the Skewed-t distribution.¹⁰ The moments θ of the skewed-Normal distribution are estimated as follows:

\begin{matrix} {\hat{θ}}^{h} = \underset{θ \in Θ}{arg m i n} \sum_{q = 1}^{19} {(E Q F_{h} (\bar{x}, q) - S k e w N Q F (θ))}^{2}, & (4) \end{matrix}

for each h=1…H, where EQF is the empirical conditional quantile function at quantiles q=5, … ,95 estimated using (1) and SkewNQF is the quantile function of a skew-normal distribution. We then calculate confidence intervals for (L₁-L₀)/ L₀ using the bootstrap.¹¹

III. Data

A. Financial and Macroeconomic Conditions

Macroeconomic conditions are measured by real GDP growth and inflation from the IMF’s World Economic Outlook database. Financial conditions broadly refer to the ease of obtaining finance (IMF 2017). A key component of this is the price of risk—the excess return required for an investor to hold one additional unit of risk (Sharpe 1964). We focus on the price of risk and disregard other variables that capture vulnerabilities such as leverage and credit growth as these will be considered explicitly in our analysis. Our financial conditions index is based on IMF (2018) and uses the same underlying data. Please see Appendix A for the details.

B. Policy Shocks

To properly measure the effects of different policies—monetary, macroprudential, foreign exchange interventions, and capital flow management policies—it is critical to obtain changes in the policy variables that do not reflect endogenous reactions to changes in the economic and the financial environment (Ramey, 2016). We achieve this by estimating policy reaction functions for each instrument, country by country, that condition on those economic and financial variables, and using the residual of such regressions as the policy shock. Tables 1 and 2 show the summary statistics of each policy shock.

^{Table 1.}

Summary Statistics

^{^*}Winsorized at the top and bottom 0.5 percent.

Note: “MPM All”, “MPM Borrower-Based”, and “MPM FI-Based” stand for the macroprudential policy shocks for all instruments, borrower-based instruments, and financial-institution-based instruments, respectively; “MP” stands for the monetary policy shock; “FXI” stands for the foreign exchange intervention shock; and “CFM” stands for the capital flow management measure shock. Please see Appendix A for details.

^{Table 1.}

Summary Statistics

Variable	Observations	Average	Standard Deviation	Median	Minimum	Maximum
Detrended output	2,876	-0.03	1.11	0.02	-8.08	7.21
Quarterly inflation	2,876	0.66	0.87	0.55	-3.07	9.84
Financial conditions index	2,876	-0.07	0.94	-0.20	-2.74	6.11
Real GDP growth^*	2,876	2.97	4.68	2.86	-22.29	26.89
YoY Inflation	2,876	2.72	2.51	2.23	-6.32	16.37
Growth in credit^*	2,876	0.64	2.46	0.50	-7.60	10.60
Credit to GDP (logs)	2,876	4.78	0.58	4.93	3.00	5.99
House prices (logs)	2,876	5.03	0.67	4.90	3.16	8.45
Output gap	2,808	0.00	0.04	0.00	-0.20	0.33
MPM All	2,876	0.01	0.48	-0.05	-2.51	2.24
MPM Borrower-Based	2,876	0.00	0.23	-0.01	-2.07	1.99
MPM FI-Based	2,876	0.00	0.40	-0.03	-2.58	2.17
MP	2,852	-0.03	0.95	-0.05	-5.74	9.05
FXI	2,259	0.00	0.01	0.00	-0.09	0.14
CFM	2,259	-0.02	0.52	0.03	-2.00	2.30

^{^*}Winsorized at the top and bottom 0.5 percent.

^{Table 1.}

Summary Statistics

Variable	Observations	Average	Standard Deviation	Median	Minimum	Maximum
Detrended output	2,876	-0.03	1.11	0.02	-8.08	7.21
Quarterly inflation	2,876	0.66	0.87	0.55	-3.07	9.84
Financial conditions index	2,876	-0.07	0.94	-0.20	-2.74	6.11
Real GDP growth^*	2,876	2.97	4.68	2.86	-22.29	26.89
YoY Inflation	2,876	2.72	2.51	2.23	-6.32	16.37
Growth in credit^*	2,876	0.64	2.46	0.50	-7.60	10.60
Credit to GDP (logs)	2,876	4.78	0.58	4.93	3.00	5.99
House prices (logs)	2,876	5.03	0.67	4.90	3.16	8.45
Output gap	2,808	0.00	0.04	0.00	-0.20	0.33
MPM All	2,876	0.01	0.48	-0.05	-2.51	2.24
MPM Borrower-Based	2,876	0.00	0.23	-0.01	-2.07	1.99
MPM FI-Based	2,876	0.00	0.40	-0.03	-2.58	2.17
MP	2,852	-0.03	0.95	-0.05	-5.74	9.05
FXI	2,259	0.00	0.01	0.00	-0.09	0.14
CFM	2,259	-0.02	0.52	0.03	-2.00	2.30

^{^*}Winsorized at the top and bottom 0.5 percent.

^{Table 2.}

Policy Shock Size by Country

Note: The table shows one standard deviation of each policy shock. “MPM All”, “MPM Borrower-Based”, and “MPM FI-Based” stand for the macroprudential policy shocks for all instruments, borrower-based instruments, and financial-institution-based instruments, respectively; “MP” stands for the monetary policy shock; “FXI” stands for the foreign exchange intervention shock; and “CFM” stands for the capital flow management measure shock. Please see Appendix A for details.

^{Table 2.}

Policy Shock Size by Country

Country	Observations	MPM All	MPM Bo rro wer-Based	MPM FI-Based	MP	FXI	CFM
Australia	103	0.38	0.00	0.19	1.03	0.13	0.58
Austria	103	0.29	0.00	0.14	0.93	0.09	0.22
Belgium	98	0.34	0.01	0.17	0.95	0.09	0.27
Brazil	59	0.96	0.13	0.47	1.01	0.17	0.95
Canada	103	0.53	0.34	0.14	0.98	0.19	0.50
Chile	55	0.13	0.13	0.00	1.04	0.19	0.69
China	42	1.23	0.55	0.41	1.04	0.05	0.96
Colombia	76	0.43	0.23	0.11	0.95
Denmark	103	0.24	0.17	0.17	1.04	0.11	0.03
Finland	103	0.34	0.14	0.17	0.95	0.09	0.25
France	103	0.22	0.00	0.14	0.97	0.09	0.31
Germany	102	0.28	0.00	0.10	0.95	0.10	0.22
Hong Kong SAR	103	0.56	0.47	0.21	1.05	0.14	0.46
Hungary	71	0.50	0.35	0.00	1.00	0.15	0.52
India	27	0.97	0.34	0.43	1.03	0.14	1.06
Indonesia	55	0.61	0.27	0.13	1.02	0.16	0.80
Ireland	78	0.30	0.16	0.16	0.94	0.16	0.09
Italy	103	0.35	0.10	0.17	0.96	0.10	0.16
Japan	103	0.24	0.00	0.10	1.22	0.12	0.27
Kazakhstan	39	0.42	0.16	0.22	1.05
Korea	103	0.74	0.58	0.19	1.04	0.20	0.98
Malaysia	102	0.58	0.22	0.14	1.02	0.19	0.82
Mexico	43	0.41	0.00	0.26	1.00	0.09	0.57
Netherlands	103	0.39	0.25	0.10	0.94	0.15	0.03
Norway	103	0.43	0.22	0.32	1.05	0.13	0.22
Peru	71	0.75	0.12	0.17	1.01
Philippines	31	0.40	0.18	0.18	0.79	0.09	0.70
Poland	68	0.59	0.24	0.21	1.04	0.16	0.51
Russia	59	0.89	0.00	0.50	0.90	0.19	0.91
Singapore	103	0.52	0.33	0.10	1.05	0.14	0.48
South Africa	103	0.24	0.00	0.19	1.03	0.11	0.85
Spain	103	0.36	0.10	0.22	0.98	0.08	0.22
Sweden	103	0.34	0.14	0.26	1.03	0.11	0.25
Switzerland	103	0.31	0.00	0.24	1.03	0.11	0.31
Turkey	23	1.00	0.29	0.50	1.03	0.19	0.64
United Kingdom	103	0.33	0.10	0.26	1.02	0.09	0.03
United States	103	0.32	0.10	0.17	1.05	0.06	0.03

^{Table 2.}

Policy Shock Size by Country

Country	Observations	MPM All	MPM Bo rro wer-Based	MPM FI-Based	MP	FXI	CFM
Australia	103	0.38	0.00	0.19	1.03	0.13	0.58
Austria	103	0.29	0.00	0.14	0.93	0.09	0.22
Belgium	98	0.34	0.01	0.17	0.95	0.09	0.27
Brazil	59	0.96	0.13	0.47	1.01	0.17	0.95
Canada	103	0.53	0.34	0.14	0.98	0.19	0.50
Chile	55	0.13	0.13	0.00	1.04	0.19	0.69
China	42	1.23	0.55	0.41	1.04	0.05	0.96
Colombia	76	0.43	0.23	0.11	0.95
Denmark	103	0.24	0.17	0.17	1.04	0.11	0.03
Finland	103	0.34	0.14	0.17	0.95	0.09	0.25
France	103	0.22	0.00	0.14	0.97	0.09	0.31
Germany	102	0.28	0.00	0.10	0.95	0.10	0.22
Hong Kong SAR	103	0.56	0.47	0.21	1.05	0.14	0.46
Hungary	71	0.50	0.35	0.00	1.00	0.15	0.52
India	27	0.97	0.34	0.43	1.03	0.14	1.06
Indonesia	55	0.61	0.27	0.13	1.02	0.16	0.80
Ireland	78	0.30	0.16	0.16	0.94	0.16	0.09
Italy	103	0.35	0.10	0.17	0.96	0.10	0.16
Japan	103	0.24	0.00	0.10	1.22	0.12	0.27
Kazakhstan	39	0.42	0.16	0.22	1.05
Korea	103	0.74	0.58	0.19	1.04	0.20	0.98
Malaysia	102	0.58	0.22	0.14	1.02	0.19	0.82
Mexico	43	0.41	0.00	0.26	1.00	0.09	0.57
Netherlands	103	0.39	0.25	0.10	0.94	0.15	0.03
Norway	103	0.43	0.22	0.32	1.05	0.13	0.22
Peru	71	0.75	0.12	0.17	1.01
Philippines	31	0.40	0.18	0.18	0.79	0.09	0.70
Poland	68	0.59	0.24	0.21	1.04	0.16	0.51
Russia	59	0.89	0.00	0.50	0.90	0.19	0.91
Singapore	103	0.52	0.33	0.10	1.05	0.14	0.48
South Africa	103	0.24	0.00	0.19	1.03	0.11	0.85
Spain	103	0.36	0.10	0.22	0.98	0.08	0.22
Sweden	103	0.34	0.14	0.26	1.03	0.11	0.25
Switzerland	103	0.31	0.00	0.24	1.03	0.11	0.31
Turkey	23	1.00	0.29	0.50	1.03	0.19	0.64
United Kingdom	103	0.33	0.10	0.26	1.02	0.09	0.03
United States	103	0.32	0.10	0.17	1.05	0.06	0.03

Macroprudential Policy

Macroprudential policy refers to the use of primarily prudential instruments to limit systemic risk (IMF, 2013). Depending on the nature of systemic risk, different instruments have been used across economies (e.g., Alam and others, 2019). While some macroprudential policy instruments overlap with those of other policies, the defining feature is the objective to limit system-wide financial risks.

Our indicator of macroprudential policy draws from the dummy-type policy action indicators contained in Alam and others’ (2019) new and comprehensive Integrated Macroprudential Policy (iMaPP) database. We consider three broad categories of macroprudential measures: the overall category, and its sub-categories of borrower-based measures and financial-institution-based measures. Please see Appendix A for the definitions.

The construction of macroprudential policy shocks then involves two steps. First, for each category of macroprudential measures, we estimate the following ordered probit regression:

\begin{matrix} m p m_{i t}^{*} = μ_{0 i} + μ_{1} c g a p_{i t - 1} + μ_{2} h g a p_{i t - 1} + μ_{3} \sum_{j = 1}^{4} I_{i, t - j}^{m p m} + ℰ_{i t}^{m p m}, & (6) \end{matrix}

where ${m p m}_{i t}^{*}$ is the latent variable behind the categorical macroprudential indicator $I_{i t}^{m p m}$ which takes values {-2,-1,0,1,2} if, all in net terms, there were more than one loosening actions, one loosening action, no change, one tightening action, or more than two tightening actions in the quarter t, respectively. cgap is the credit-to-gdp gap, hgap the house price gap, and µ_0i are country fixed effects.¹² For both credit and house prices, our gap measure is the deviation from the trend, using Hamilton’s (2018) approach with eight quarter lags.

The policy shock is then recovered as the difference between the actual value of the macroprudential indicator and its estimated conditional expectation:

\begin{matrix} {\hat{ε}}_{i t}^{m p m} & = & I_{i t}^{m p m} - {\hat{E}}_{t - 1} [I_{i t}^{m p m}] \\ = & I_{i t}^{m p m} - \sum_{k = - 2}^{2} {\hat{p}}_{k} (x_{i t - 1}) k, \end{matrix}

where ${\hat{E}}_{t - 1} [I_{i t}^{m p m}]$ is the sample analogue of the expected policy action indicator, conditional on the quarter t-1 information, and ${\hat{p}}_{k} (x_{i t - 1})$ the estimated probability of $I_{i t}^{m p m} = k$ , with κ∈{-2,-J,0,J,2j, conditional on the right-hand side variables (x_it-1) of equation (6).

Monetary Policy

As an indicator of monetary policy we use the policy rate as defined in the IMF’s International Financial Statistics, except for the euro area, Japan, United Kingdom, and the United States, for which we use Krippner’s (2015) shadow short rates to account for both conventional and unconventional monetary policy.¹³

Our main measure of the monetary policy shock is the residual of a Taylor-type rule for each country in the sample. Specifically, we run the following regression for each country:

\begin{matrix} \begin{matrix} Δ r_{i t} & = & α_{0 i} + α_{1 i} E_{t} Δ y_{i t + 12} + α_{2 i} E_{t} π_{i t + 12} + \sum_{j = 1}^{2} α_{3 i j} Δ y_{i t - i} + \\ \sum_{j = 1}^{2} α_{4 i j} Δ p_{i t - j} + \sum_{j = 1}^{2} α_{5 i j} Δ n e e r_{i t - j} + \sum_{j = 1}^{2} α_{6 i j} r_{i t - j} + ε_{i t}^{r}, \end{matrix} & (7) \end{matrix}

where ∆r,∆y,∆p, and ∆neer¹⁴ denote the quarter-on-quarter changes in the monetary policy rate, the log of real GDP, the log of the consumer price index (CPI), and the nominal effective exchange rate. E_t∆y_t+12 and E_t∆p_t+12 are the 12-month ahead market forecasts of GDP growth and inflation, respectively, as measured by Consensus Forecasts. For the forecasts, although we would ideally like to use central bank forecasts as in Romer and Romer (2004), these are generally not available. By using market forecasts, we implicitly assume central banks and markets have the same information set.¹⁵

For each country, the estimated residual of (7) is the monetary policy shock.¹⁶ In other words, deviations from the Taylor-type rules are intended to capture the non-systematic part of monetary policy actions. Since the overall magnitude of the shocks is very different from country to country, we standardize the residuals on a country-by-country basis. Therefore, a unit monetary policy shock signifies a one standard deviation shock in each country.

Foreign Exchange Interventions

Foreign exchange interventions (FXIs) refers to the purchases and sales of foreign exchange by central banks. As the actual intervention data is limited to some countries, we construct a proxy by taking the change in the central bank’s net foreign assets, adjusted for valuation changes and interest income flows, as suggested by Dominguez (2012) and Adler, Lisack, and Mano (2019). Please see Appendix A for the details on the FXI variables.

We construct the FXI shock using the actual data when available and the proxy variable otherwise. The shock is obtained as the residual of an OLS regression using the following linear rule for each country i:

\begin{matrix} F X I_{i t} = γ_{0 i} + {x^{'}}_{i t} γ_{1 i} + γ_{2 i} e g a p_{i t - 1} + γ_{3 i} σ {(e)}_{i t - 1} + ε_{i t}^{F X I}, & (8) \end{matrix}

where FXI_it is our measure of foreign exchange interventions in percent of GDP, which is positive (negative) when the central bank conducts net purchases (sales) of foreign currency in that quarter. The vector x includes the same covariates used in the first column of Table 2 in Forbes and Klein (2015),¹⁷ egap is country i’s dollar exchange rate deviation from the trend using Hamilton’s (2018) approach with eight quarter lags, and σ(e) is the quarterly nominal effective exchange rate volatility calculated from daily data.

Capital Flow Management Measures

Capital flow management measures (CFMs) refer to policy actions that are designed to limit capital flows. We construct the policy shock of CFMs using the changes index by Baba and others’ (forthcoming), which is based on the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER). Their CFM changes index captures all instances of tightening and easing of capital controls,¹⁸ and as such goes beyond the recording of the existence or absence of certain types of controls that underlies most existing indices of CFMs. Please also see Appendix A.

The CFM shock is obtained in two steps, as for the macroprudential policy shock. First, we estimate the following ordered probit model with country fixed effects:

\begin{matrix} C F M_{i t}^{*} = γ_{0 i} + {x^{'}}_{i t} γ_{1} + γ_{2} e g a p_{i t - 1} + γ_{3} σ {(e)}_{i t - 1} + ε_{i t}^{C F M}, & (9) \end{matrix}

where $\begin{matrix} C F M_{i t}^{*} \end{matrix}$ is the latent variable behind the categorical CFM indicator $(I_{i t}^{C F M})$ which takes value 1 (-1) for one tightening (loosening) action, 2 (-2) for more than one tightening (loosening) actions, and 0 for no change, all in net terms, in each quarter t. The controls x_it are the same as the ones used for the FXI specification. The CFM policy shocks are obtained by calculating the difference between the actual and predicted categorical indicators, as is done for the macroprudential policy shocks.

IV. Results

A. Effects of Policies on the Intertemporal Tradeoff

We start by examining the estimated effects of each policy in response to easy financial conditions on the conditional quantiles of future (detrended) GDP growth and inflation—i.e., $β_{1}^{h} (q) + β_{3}^{h} (q) \cdot P$ of equation (1) where P takes the value of one or zero. In Figure 1, we show the estimated changes in the 10^th percentile of those variables after a one standard deviation loosening in financial conditions assuming that there is no policy change (blue line) or that a given policy is tightened (red line).

First, looking at the case with no policy change (blue line), we confirm the stylized facts laid out by Adrian and others (2018). The left-hand side panels of Figure 1 show the well-documented intertemporal tradeoff between looser domestic financial conditions today and greater downside risks to economic growth in the medium term. On the other hand, as shown on the right-hand side panels, there is no apparent intertemporal tradeoff for inflation.¹⁹

Next, comparing the cases with and without policies, several interesting findings emerge. The top left panel of Figure 1 shows that tightening macroprudential policy mitigates (but does not eliminate) the tradeoff for output growth: the tail risk deterioration due to the FCI loosening is less marked in the medium term. This effect, moreover, appears to come at little cost in the short term, when tightening macroprudential policy does not offset the stimulating effects of easier financial conditions on output. However, these estimates are mostly not statistically significant. Furthermore, the policy tightening itself does have a contractionary effect on output growth: when we examine the impulse response functions to a macroprudential policy tightening shock— i.e., $β_{2}^{h} (q) + β_{3}^{h} (q) \cdot f$ of equation (1)—we tend to find contractionary effects on output (and prices, to some extent), which are reported in Figure B.2 in Appendix B, even as the magnitude of these effects does not appear large, in line with the literature. ²⁰

Turning to the effect of a monetary policy tightening, we find different results. The left panel of Figure 1 shows, if anything, a worsening of the underlying trade-off, by further worsening the tail risk statistically significantly in the medium term. The policy effects on inflation are ambiguous, though.

Finally, when we move from the tail of the distribution to median growth (see Figure B. 1 in Appendix B), we find that the intertemporal tradeoff documented by Adrian and others (2018) is absent, in that the effect of easing financial conditions on median output is positive through the entire horizon. In addition, tighter macroprudential policy does not seem to reduce that positive effect of easing financial conditions on median economic activity, while tighter monetary policy appears, if anything, to worsen the median outlook, in line with the view that macroprudential policy is effective in dampening downside risks to GDP growth whereas monetary policy is a blunt tool (Yellen, 2014).

B. Net Benefits of Policies

Changes in Domestic Financial Conditions

Next, we move beyond the effects at specific quantiles to consider the net effects of policies across the entire distribution of future output by using our loss function approach. This compares losses conditional on easing financial conditions when policies are used, versus when no policy action is taken. As our benchmark, we use the quadratic loss function (1), which measures costs and benefits using output and price volatility.

The estimates in Table 3 (left panel) suggest that tightening macroprudential policies when domestic financial conditions loosen reduces losses by about 9 percent, in the case of quadratic policymaker preferences. However, in line with Svensson’s (2017) results, we do not find that leaning against loose domestic financial conditions through monetary policy reduces losses—in fact, monetary policy tightening seems to increase losses by about 12 percent.

^{Table 3.}

Effect of Policy Changes on Loss Functions

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t + h}} (q | z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, a n d \\ Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ 0.95, \end{matrix}

for domestic and global FCI shocks, respectively, where Q_Y(q|.) is the conditional quantile function of cumulative GDP growth for up to 14 quarters ahead, f is a country-specific financial conditions index, g is a global financial conditions index, P is a policy shock, and x is a vector of controls including current GDP growth, inflation, and credit growth. The exercise shows the results when macroprudential policy, monetary policy, and capital flow measures tighten or when the central bank intervenes in the foreign exchange market by purchasing foreign currency, when financial conditions are loose. MPM All is the shock based on Alam et al.’s (2019) index of 17 macroprudential measures. MPM Borrower-Based and MPM FI-Based are the same as MPM All but only use borrower-based and financial-institution-based prudential measures, respectively. MP is a monetary policy shock calculated as the residual of an estimated Taylor rule. FXI is a measure of FX interventions. CFM is Baba et al.’s (forthcoming) index of capital-flow-management measures. HF MP is a measure of monetary policy surprises based on high-frequency data from multiple sources. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on a country-level cluster bootstrap.

^{Table 3.}

Effect of Policy Changes on Loss Functions

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t + h}} (q | z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, a n d \\ Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ 0.95, \end{matrix}

	Domestic FCI			External FCI
	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l
MPM All	-0.089 ***	-0.085 ***	-0.083 ***	-0.112 ***	-0.107 ***	-0.104 ***
MPM Borrower-Based	-0.100 ***	-0.068 ***	-0.065 ***	-0.107 ***	-0.101 ***	-0.096 ***
MPM FI-Based	-0.053 **	-0.036 **	-0.035 **	-0.068 ***	-0.067 ***	-0.065 ***
MP	0.121 ***	0.115 ***	0.111 ***	0.038 *	0.036 *	0.036 *
FXI	-	-	-	-0.022	-0.021	-0.021
CFM	-	-	-	-0.039	-0.034	-0.030
HF MP	-0.011	-0.011	-0.011	-0.025	-0.023	-0.022

^{Table 3.}

Effect of Policy Changes on Loss Functions

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t + h}} (q | z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, a n d \\ Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ 0.95, \end{matrix}

	Domestic FCI			External FCI
	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l
MPM All	-0.089 ***	-0.085 ***	-0.083 ***	-0.112 ***	-0.107 ***	-0.104 ***
MPM Borrower-Based	-0.100 ***	-0.068 ***	-0.065 ***	-0.107 ***	-0.101 ***	-0.096 ***
MPM FI-Based	-0.053 **	-0.036 **	-0.035 **	-0.068 ***	-0.067 ***	-0.065 ***
MP	0.121 ***	0.115 ***	0.111 ***	0.038 *	0.036 *	0.036 *
FXI	-	-	-	-0.022	-0.021	-0.021
CFM	-	-	-	-0.039	-0.034	-0.030
HF MP	-0.011	-0.011	-0.011	-0.025	-0.023	-0.022

The results in favor of macroprudential policy are stronger and more statistically significant for borrower-based measures (between -6.5 and -10 percent) than for financial institution-based measures (at most -5.3 percent), even if our estimates are not sufficiently precise to establish that these two types of policies have different effects at the usual levels of statistical significance.²¹ These findings are in line with those by Alam et al. (2019), who report that the borrower-based measures have larger intended effects on credit but smaller unintended side effects on consumption.

Examining these results further, we find that the sharp contrast between the net benefits of different policies mostly comes from their opposite effects on output volatility, while all policies have little effects on price volatility. With tighter macroprudential policy, the distribution of future output gaps shrinks, as evident in the increase of its 10^th percentile (Figure 1, top left panel).²² On the other hand, leaning against the wind with monetary policy increases output volatility (Table 3).

These empirical findings are in line with arguments advanced in the literature that leaning against the wind using monetary policy can be counterproductive, and that macroprudential policy is preferred for this purpose (e.g., Bean and others, 2010; Nier and Kang, 2016; Svensson, 2017; Schularick and others, 2020). For example, Alpanda and Zubairy (2017) show that tighter borrower-based macroprudential measures can make the economy less volatile, by reducing financial accelerator effects through tighter borrowing constraints, whereas the active use of monetary policy to lean against the wind can increase the volatility of output and thereby result in welfare losses. More generally, the DSGE literature points to increases in the volatility of the output gap as a drawback of leaning against the wind using monetary policy (e.g., Gourio, Kashyap, and Sim, 2018), while borrower-based macroprudential tools are typically found more effective, by offering a better trade-off in terms of output foregone (e.g., Alpanda, Capeau and Meh, 2018).²³

Global Financial Shocks

Next, we consider shocks to global financial conditions instead of changes to the domestic FCI. We do this mainly to be able to compare macroprudential policy also to FXI and CFMs, policies that are more likely to be used to lean against external shocks than against domestic ones.²⁴ Moreover, changes in global financial conditions are exogenous for small open economies, allowing for a cleaner identification of the effects of policies.

We therefore expand the previous specification to include global financial conditions (which we proxy with financial conditions in the United States) as follows.

\begin{matrix} \begin{matrix} Q_{Δ y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, K, 0.95, \end{matrix} & (10) \end{matrix}

where g is the global FCI. We then proceed as before to estimate the effect of policies conditional on an easing of financial conditions, and compare the loss functions with and without policy actions.

We first calculate the effect of a one-standard deviation loosening in global financial conditions on the GDP-at-risk (10^th percentile of future detrended GDP). We find that the effect of macroprudential policy tightening conditional on easing of global financial conditions is by and large unchanged relative to what we obtained for domestic financial conditions, while monetary policy now appears to reduce, rather than increase, downside tail risks for GDP at the 10^th percentile (see Figure B.3 in Appendix B).

In addition, Figure 2 shows that intervening in the foreign exchange market to buy foreign currency does not affect the intertemporal tradeoff—the effect of looser global financial conditions on tails risks to GDP does not change—and the same seems to be true for adopting new measures which impede capital inflows.

Moving from specific quantiles to estimation of losses over the entire distribution, we find that loss reductions from tightening macroprudential measures are again substantial (Table 3, right panel). However, we find again that leaning against easing financial conditions with tighter monetary policy does not yield net gains when considering all the effects on the entire distribution.

Borrower-based macroprudential policies continue to have the largest beneficial effect. Loss reductions due to borrower-based macroprudential actions range between 9.6 and 10.7 percent, depending on the loss-function weights for economic activity and inflation, while tightening financial institution-based policies brings smaller gains (at most a 6.8 percent reduction).

In addition, using FX purchases (to prevent a local currency appreciation) or tightening CFMs to counter loose global financial conditions entail very small effects on the distribution of output, which are never statistically different than zero. These findings are broadly consistent with the empirical literature, which so far has found mixed evidence for these policies (e.g., Chamon and others 2019, Rebucci and Ma 2019, Bergant et al. forthcoming).

Changes in the Loss Function Over the Forecasting Horizon

How do the changes in the loss functions evolve over the forecasting horizon? To address this question, we calculate a loss differential ∆L=(L₁(Θ)-L₀(Θ))/L₀(Θ) using expressions (3) and (4) at each horizon H=1,…,14. For brevity, we do this only for a period loss with unit weight on output and zero weight on inflation and only for the case of global financial conditions. The exercise tries to show how the changes in the expected loss from a given policy tightening are distributed over the 14 quarters.

Three main patterns emerge (Figure 3). First, macroprudential policy reduces losses in a uniform way until the benefits peak around 10 quarters ahead. Therefore, while the net benefits of macroprudential policy are realized fully in the medium term, they start accruing in the short term. Monetary policy tightening, by contrast, appears to reduce losses initially, and starts to induce increases in losses from about 5 quarters out, suggesting that any initial benefits from tightening are being eroded by the effect of tighter policy on output volatility over the medium term.

Second, borrower-based macroprudential policy has a more persistent beneficial effect on the loss function than financial-institution-based macroprudential policy (not shown but available from the authors). Third, we find that for FXI and CFM (and also for financial-institution-based macroprudential policy) about half the reduction in the loss function up to 10 quarters ahead which is associated with the use of these tools is reversed after 14 quarters. This suggests that gains associated with the use of these policies are largely temporary, while the benefits of borrower-based macroprudential tools are longer lasting.

C. Effects of Policies Conditional on the Level of Financial Vulnerabilities

The effect of some policies may depend on the current level of financial vulnerabilities. For instance, the impact of a loosening shock to the FCI may depend on existing financial sector leverage. The optimal policy response to loosening financial conditions, in turn, may then be a function of the existing level of credit. Similarly, it is conceivable that the beneficial effect of policies may depend on whether asset prices are already elevated, or not.

In particular, it has been argued that macroprudential policy are needed more urgently when the level of credit is already high than when it is low (Biljanovska, Gornicka, and Vardoulakis 2019), since risks to output are then greater. Alternatively, with loosening financial conditions, tightening policy to lean against the wind may have a more beneficial effect when private sector leverage is still low, since leaning can then still reduce the further build-up of risks. We test these hypotheses with a modified version of (1) as follows.

\begin{matrix} Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) V u l_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times f_{i t} \\ + (β_{5}^{h} (q) P_{i t} + β_{6}^{h} (q) P_{i t} \times f_{i t}) V u l_{i t} + x_{i t} Γ, \\ h & = & 1, K, H, q = 0.05, K, 0. 95, \end{matrix}

where Vul is measure of financial vulnerabilities. We use alternatively the credit-to-GDP ratio (a measure of leverage), and the house price index calculated by the BIS (both in logs).

When looking at quantiles, we find that tighter macroprudential policies are more strongly associated with a reduction in tail risks to GDP growth—measured by the 10^th percentile of the distributions of future GDP growth—when credit levels or house prices are high (Figure 4). The figure also shows that using monetary policy to lean against the wind is not associated with a reduction in GDP at risk, irrespective of the level of credit.

When estimating the effects of these policies on the loss functions, the basic results are confirmed, even as an interesting nuance emerges (Table 4). The relative effectiveness of tightening macroprudential policy tools depends both on the level of vulnerability— measured by the level of credit and real estate valuations—and on the type of policy— borrower-based versus financial-institutions based.

^{Table 4.}

Effect of Policy Changes on Loss Functions by Vulnerability

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) V u l_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times f_{i t} \\ + (β_{5}^{h} (q) P_{i t} + β_{6}^{h} (q) P_{i t} \times f_{i t}) V u l_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

where Q_Y(q|.) is the conditional quantile function of cumulative GDP growth for up to 14 quarters ahead, f is a country-specific financial conditions index, g is a global financial conditions index, P is a policy shock, Vul is a measure of financial vulnerabilities (credit-to-GDP, in panel A, or house-price index, in panel B) and x is a vector of controls including current GDP growth, inflation, and credit growth. The exercise shows the results when macroprudential policy or monetary policy tighten with loose financial conditions. MPM All is the shock based on Alam et al.’s (2019) index of 17 macroprudential measures. MPM Borrower-Based and MPM FI-Based are the same as MPM All but only use borrower-based and financial-institution-based prudential measures, respectively. MP is a monetary policy shock calculated as the residual of an estimated Taylor rule. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on a country-level cluster bootstrap.

^{Table 4.}

Effect of Policy Changes on Loss Functions by Vulnerability

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) V u l_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times f_{i t} \\ + (β_{5}^{h} (q) P_{i t} + β_{6}^{h} (q) P_{i t} \times f_{i t}) V u l_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

Panel A.
	Low Credit			High Credit
	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l
MPM All	-0.089 **	-0.086 **	-0.084 **	-0.099 **	-0.094 **	-0.090 **
MPM Borrower-Based	-0.033	-0.032	-0.031	-0.083 ***	-0.078 ***	-0.075 ***
MPM FI-Based	-0.076 **	-0.072 **	-0.070 **	-0.028	-0.027	-0.026
MP	0.137 ***	0.132 ***	0.129 ***	0.126 ***	0.120 ***	0.115 ***

Panel B.
	Low Valuations			High Valuations
	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l
MPM All	-0.152 ***	-0.144 ***	-0.139 ***	-0.072 **	-0.068 **	-0.066 **
MPM Borrower-Based	-0.123 **	-0.114 **	-0.107 **	-0.074 **	-0.071 **	-0.068 **
MPM FI-Based	-0.085 **	-0.082 **	-0.080 **	-0.031	-0.031	-0.030
MP	0.150 ***	0.145 ***	0.141 ***	0.216 ***	0.208 ***	0.202 ***

^{Table 4.}

Effect of Policy Changes on Loss Functions by Vulnerability

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) V u l_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times f_{i t} \\ + (β_{5}^{h} (q) P_{i t} + β_{6}^{h} (q) P_{i t} \times f_{i t}) V u l_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

Panel A.
	Low Credit			High Credit
	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l
MPM All	-0.089 **	-0.086 **	-0.084 **	-0.099 **	-0.094 **	-0.090 **
MPM Borrower-Based	-0.033	-0.032	-0.031	-0.083 ***	-0.078 ***	-0.075 ***
MPM FI-Based	-0.076 **	-0.072 **	-0.070 **	-0.028	-0.027	-0.026
MP	0.137 ***	0.132 ***	0.129 ***	0.126 ***	0.120 ***	0.115 ***

Panel B.
	Low Valuations			High Valuations
	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l
MPM All	-0.152 ***	-0.144 ***	-0.139 ***	-0.072 **	-0.068 **	-0.066 **
MPM Borrower-Based	-0.123 **	-0.114 **	-0.107 **	-0.074 **	-0.071 **	-0.068 **
MPM FI-Based	-0.085 **	-0.082 **	-0.080 **	-0.031	-0.031	-0.030
MP	0.150 ***	0.145 ***	0.141 ***	0.216 ***	0.208 ***	0.202 ***

Table 5.

Effect of Coordinated Policy Changes

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ℓ_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h'}^{2} \end{matrix}

with ω_y=0.542 and ω_p=1 and which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t}^{I} + β_{3}^{h} (q) P_{i t}^{I I} + β_{4}^{h} (q) P_{i t}^{I} \times f_{i t} \\ + β_{5}^{h} (q) P_{i t}^{I I} \times f_{i t} + β_{6}^{h} (q) P_{i t}^{I} \times P_{i t}^{I I} \times f_{i t} + x_{i t} Γ, o r \\ Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{t} + β_{3}^{h} (q) P_{i t}^{I} + β_{4}^{h} (q) P_{i t}^{I I} + β_{5}^{h} (q) P_{i t}^{I} \times g_{t} \\ + β_{6}^{h} (q) P_{i t}^{I I} \times g_{t} + β_{7}^{h} (q) P_{i t}^{I} \times P_{i t}^{I I} \times g_{t} + x_{i t} Γ, h = 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

Table 5.

Effect of Coordinated Policy Changes

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ℓ_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h'}^{2} \end{matrix}

with ω_y=0.542 and ω_p=1 and which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t}^{I} + β_{3}^{h} (q) P_{i t}^{I I} + β_{4}^{h} (q) P_{i t}^{I} \times f_{i t} \\ + β_{5}^{h} (q) P_{i t}^{I I} \times f_{i t} + β_{6}^{h} (q) P_{i t}^{I} \times P_{i t}^{I I} \times f_{i t} + x_{i t} Γ, o r \\ Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{t} + β_{3}^{h} (q) P_{i t}^{I} + β_{4}^{h} (q) P_{i t}^{I I} + β_{5}^{h} (q) P_{i t}^{I} \times g_{t} \\ + β_{6}^{h} (q) P_{i t}^{I I} \times g_{t} + β_{7}^{h} (q) P_{i t}^{I} \times P_{i t}^{I I} \times g_{t} + x_{i t} Γ, h = 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l
	Expand MP
MPM All	-0.162	-0.157	-0.153
MPM Borrower-Based	-0.189	-0.180	-0.173
MPM FI-Based	-0.140	-0.136	-0.133
	Contract MP
MPM All	-0.026	-0.023	-0.021
MPM Borrower-Based	0.025	0.024	0.023
MPM FI-Based	0.023	0.023	0.024

Table 5.

Effect of Coordinated Policy Changes

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ℓ_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h'}^{2} \end{matrix}

with ω_y=0.542 and ω_p=1 and which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t}^{I} + β_{3}^{h} (q) P_{i t}^{I I} + β_{4}^{h} (q) P_{i t}^{I} \times f_{i t} \\ + β_{5}^{h} (q) P_{i t}^{I I} \times f_{i t} + β_{6}^{h} (q) P_{i t}^{I} \times P_{i t}^{I I} \times f_{i t} + x_{i t} Γ, o r \\ Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{t} + β_{3}^{h} (q) P_{i t}^{I} + β_{4}^{h} (q) P_{i t}^{I I} + β_{5}^{h} (q) P_{i t}^{I} \times g_{t} \\ + β_{6}^{h} (q) P_{i t}^{I I} \times g_{t} + β_{7}^{h} (q) P_{i t}^{I} \times P_{i t}^{I I} \times g_{t} + x_{i t} Γ, h = 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l
	Expand MP
MPM All	-0.162	-0.157	-0.153
MPM Borrower-Based	-0.189	-0.180	-0.173
MPM FI-Based	-0.140	-0.136	-0.133
	Contract MP
MPM All	-0.026	-0.023	-0.021
MPM Borrower-Based	0.025	0.024	0.023
MPM FI-Based	0.023	0.023	0.024

In particular, tightening borrower-based macroprudential policies is more strongly associated with a reduction in losses when credit and asset prices are high (high vulnerabilities) than when vulnerabilities are low. By contrast, tightening financial-institutions-based macroprudential appears to have larger benefits when credit or asset prices are still low.

This suggests that when the level of credit (or asset prices) is high, financial-institutions-based tools (such as capital requirements) that tackle risks only indirectly by affecting the supply of credit are not sufficient, and borrower-based tools that more directly increase borrowers’ resilience to shocks are required. By contrast, when credit is low, measures that reduce the further build-up of credit by affecting financial institutions can still be effective while avoiding efficiency costs from the imposition of borrower-based tools. Overall, the results are suggestive of the dual role of macroprudential policy in both leaning against the wind and building resilience.²⁵

By contrast, the effects of monetary policy appear to be independent of financial vulnerabilities. The quadratic loss functions increase by similar amounts when monetary policy is tightened, regardless of credit levels (Table 4), and the differences in the changes of the loss functions between the low and high vulnerability cases are not statistically different than zero at conventional levels of significance. This suggests that the adverse effects of tighter monetary policy on the volatility of output are independent of the level of credit, whereas the effects of tighter policy on the further supply of credit are relatively small (see again Bean and others 2010, and Gourio, Kashyap, and Sim 2018).

D. Interaction Between Monetary and Macroprudential Policies

So far, we have compared the effects of different policies assuming countries use each policy in isolation. In practice, central banks and prudential authorities are likely to want to use a combination of policies, in order to reap synergies across tools, even as the rationale and direction of such complementary use can differ (e.g., Nier and Kang 2016, Bruno and others 2017, Akinci and Olmstead-Rumsey 2018, Agur 2019, and Bodenstein and others 2019).

One school of thought has the two policies move in opposite direction in response to an easing of financial conditions (e.g., Nier and Kang, 2016, Collard and others, 2017): for instance, an easing of financial conditions calls for macroprudential policy to reduce the build-up of risky credit through financial accelerator mechanisms (Bernanke, Gertler, and Gilchrist 1999), and to increase resilience to future shocks. However, the tightening of macroprudential policy may also have undesirably strong negative effects on credit and output, which can be offset by providing more monetary accommodation at the margin (e.g., Kohn 2015, Nier and Kang 2016, and Fahr and Fell 2017).

Indeed, according to this first school, if the buildup of financial vulnerabilities caused by loose financial conditions happens mainly through a deterioration of the quality of credit, only macroprudential policy is needed because, unlike monetary policy, it can directly affect and maintain the quality of credit origination (Collard and others 2017). However, it could be optimal to tighten macroprudential requirements in response to the increase in risk-taking incentives and, at the same time, cut interest rates to mitigate the negative effects on the overall volume of credit and output (Collard and others 2017).²⁶

Another school of thought posits that it can more often be useful for both monetary and macroprudential policy to be used in the same direction when faced with a loosening of financial conditions. The idea here is that macroprudential policy is insufficient to deal with the build-up of vulnerabilities, and that monetary policy needs to ‘lend a hand’, by aiming to offset the easing of financial conditions and reduce the provision of credit to the economy (Stein 2013). However, as stated already, the counter to this argument is that such use of monetary policy may then also have strong adverse effects on output (IMF 2015, Svensson 2017).

We examine these hypotheses empirically by comparing the effects on future output growth of combining macroprudential tightening with monetary policy tightening versus combining it with monetary policy loosening. To this end, we expand our quantile regressions as follows:

\begin{matrix} Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t}^{I} + β_{3}^{h} (q) P_{i t}^{I I} + β_{4}^{h} (q) P_{i t}^{I} \times f_{i t} \\ + β_{5}^{h} (q) P_{i t}^{I I} \times f_{i t} + β_{6}^{h} (q) P_{i t}^{I} \times P_{i t}^{I I} + β_{7}^{h} (q) P_{i t}^{I} \times P_{i t}^{I I} \times f_{i t} + x_{i t} Γ, a n d \\ Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{t} + β_{3}^{h} (q) P_{i t}^{I} + β_{4}^{h} (q) P_{i t}^{I I} + β_{5}^{h} (q) P_{i t}^{I} \times g_{t} \\ + β_{6}^{h} (q) P_{i t}^{I I} \times g_{t} + β_{7}^{h} (q) P_{i t}^{I} \times P_{i t}^{I I} + β_{8}^{h} (q) P_{i t}^{I} \times P_{i t}^{I I} \times g_{t} + x_{i t} Γ, h = 1, K, H, q = 0.05, K, 0.95, \end{matrix}

for domestic financial shocks, respectively. In the quantile regressions above, PI and PII represent two different policies (e.g., macroprudential and monetary).

The results confirm the hypothesis that there are gains from combining macroprudential and monetary policies, and that moving them in opposite directions can be useful. In the face of looser domestic financial conditions, when a tightening of macroprudential policies is accompanied by looser monetary policy, this translates into a larger reduction in the central bank’s loss function (-16 percent) than what is reported in Table 3 (about -9 percent). By contrast, we find that tightening both macroprudential and monetary policies basically erases the benefits of leaning against the wind from macroprudential policy alone.

The intuition behind these results can be built from the earlier results on the individual policy effects (Section III.B). Those results show that a tightening of macroprudential policy in response to easy financial conditions generates positive net benefits by stabilizing the economy, while a tightening of monetary policy, instead, is counterproductive, generating more costs than benefits, especially in the medium term (Figure 3). Therefore, combining tight macroprudential policy with easy monetary policy can achieve larger net benefits, further stabilizing the output responses to easy financial conditions.

In sum, our results suggest that when responding to a loosening of financial conditions, it is useful to move macroprudential and monetary policy in opposite directions, since a tightening of monetary policy conditions is too costly, while a loosening of monetary policy can offset some of the undesirable effects of macroprudential policy tightening.

V. Robustness

A. Alternative Loss Functions

As an extension, we calculate the net benefit of each policy using alternative specifications for the ad-hoc loss functions. First, we consider loss functions which are linear-quadratic in output (in addition to being quadratic in inflation), to address the concern that the quadratic loss function may miss out the level effects on output by focusing on its volatility. The introduction of a level (linear) term of output in the policymaker’s preferences can be justified with a micro-founded New Keynesian model with a distorted steady state (Benigno and Woodford 2005). The loss function becomes

\begin{matrix} ∣_{t} & = & ω_{y} [{(y_{t + h} - {\bar{y}}_{t})}^{2} - (y_{t + h} - {\bar{y}}_{t})] + ω_{π} π_{t}^{2} \end{matrix} .

Including a linear term of output in the loss function yields the same results as those obtained with the pure quadratic losses in the sense that the order of loss reductions among policies is preserved (Table 6). This is likely because this linear-quadratic loss function can be rewritten as the pure quadratic loss function (without a linear term), involving only a shift of the loss function (Benigno and Woodford 2005).

^{Table 6.}

Effect of Policy Changes with Alternative Loss Functions

The table shows the estimated values of the expected loss given by the following period loss functions

\begin{matrix} ∣_{t} & = & ω_{y} [{(y_{t + h} - {\bar{y}}_{t})}^{2} - (y_{t + h} - {\bar{y}}_{t})] + ω_{π} π_{t}^{2}, a n d \\ ∣_{t} & = & \frac{(e^{a y %_{t + h}} - a y %_{t + h} - 1)}{a^{2}} w h e r e y %_{t + h} \equiv (y_{t + h} - \bar{y}), \end{matrix}

for the linear-quadratic and asymmetric policymaker preferences, respectively. The loss functions are obtained from the following quantile regressions:

\begin{matrix} Q_{Y_{i, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, a n d \\ Q_{Δ y_{_{i, t, t + h}}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

for domestic and global FCI shocks, respectively, where Q_Y(q|.) is the conditional quantile function of cumulative GDP growth for up to 14 quarters ahead, f is a country-specific financial conditions index, g is a global financial conditions index, P is a policy shock, and x is a vector of controls including current GDP growth, inflation, and credit growth. The exercise shows the results when macroprudential policy, monetary policy, and capital flow measures tighten or when the central bank intervenes in the foreign exchange market by selling foreign currency, when financial conditions are loose. MPM All is the shock based on Alam et al.’s (2019) index of 17 macroprudential measures. MPM Borrower-Based and MPM FI-Based are the same as MPM All but only use borrower-based and financial-institution-based prudential measures, respectively. MP is a monetary policy shock calculated as the residual of an estimated Taylor rule. FXI is a measure of FX interventions. CFM is Baba et al.’s (forthcoming) index of capital-flow-management measures. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on a country-level cluster bootstrap.

^{Table 6.}

Effect of Policy Changes with Alternative Loss Functions

The table shows the estimated values of the expected loss given by the following period loss functions

\begin{matrix} ∣_{t} & = & ω_{y} [{(y_{t + h} - {\bar{y}}_{t})}^{2} - (y_{t + h} - {\bar{y}}_{t})] + ω_{π} π_{t}^{2}, a n d \\ ∣_{t} & = & \frac{(e^{a y %_{t + h}} - a y %_{t + h} - 1)}{a^{2}} w h e r e y %_{t + h} \equiv (y_{t + h} - \bar{y}), \end{matrix}

for the linear-quadratic and asymmetric policymaker preferences, respectively. The loss functions are obtained from the following quantile regressions:

\begin{matrix} Q_{Y_{i, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, a n d \\ Q_{Δ y_{_{i, t, t + h}}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

for domestic and global FCI shocks, respectively, where Q_Y(q|.) is the conditional quantile function of cumulative GDP growth for up to 14 quarters ahead, f is a country-specific financial conditions index, g is a global financial conditions index, P is a policy shock, and x is a vector of controls including current GDP growth, inflation, and credit growth. The exercise shows the results when macroprudential policy, monetary policy, and capital flow measures tighten or when the central bank intervenes in the foreign exchange market by selling foreign currency, when financial conditions are loose. MPM All is the shock based on Alam et al.’s (2019) index of 17 macroprudential measures. MPM Borrower-Based and MPM FI-Based are the same as MPM All but only use borrower-based and financial-institution-based prudential measures, respectively. MP is a monetary policy shock calculated as the residual of an estimated Taylor rule. FXI is a measure of FX interventions. CFM is Baba et al.’s (forthcoming) index of capital-flow-management measures. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on a country-level cluster bootstrap.

	Domestic Shock			External Shock
	Linear-qua		Asymmetric	Linear-quadratic		Asymmetric
	ω_y=l, ω_p=0	ω_y=l, ω_p=1	ω_y=1, ω_p=0	ω_y=1, ω_p=0	ω_y=l, ω_p=1	ω_y=l, ω_p=0
MPM All	-0.080 ***	-0.076 ***	-0.087 ***	-0.100 ***	-0.095 ***	-0.109 ***
MPM Borrower-Based	-0.092 ***	-0.087 ***	-0.098 ***	-0.097 ***	-0.089 ***	-0.100 ***
MPM FI-Based	-0.046 **	-0.045 **	-0.053 *	-0.060 **	-0.058 **	-0.067 ***
MP	0.134 ***	0.126 ***	0.124 ***	0.046 **	0.044 **	0.040 *
FXI	-	-	-	-0.029	-0.027 *	-0.024
CFM	-	-	-	-0.040	-0.033	-0.041

^{Table 6.}

Effect of Policy Changes with Alternative Loss Functions

The table shows the estimated values of the expected loss given by the following period loss functions

\begin{matrix} ∣_{t} & = & ω_{y} [{(y_{t + h} - {\bar{y}}_{t})}^{2} - (y_{t + h} - {\bar{y}}_{t})] + ω_{π} π_{t}^{2}, a n d \\ ∣_{t} & = & \frac{(e^{a y %_{t + h}} - a y %_{t + h} - 1)}{a^{2}} w h e r e y %_{t + h} \equiv (y_{t + h} - \bar{y}), \end{matrix}

for the linear-quadratic and asymmetric policymaker preferences, respectively. The loss functions are obtained from the following quantile regressions:

\begin{matrix} Q_{Y_{i, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, a n d \\ Q_{Δ y_{_{i, t, t + h}}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

for domestic and global FCI shocks, respectively, where Q_Y(q|.) is the conditional quantile function of cumulative GDP growth for up to 14 quarters ahead, f is a country-specific financial conditions index, g is a global financial conditions index, P is a policy shock, and x is a vector of controls including current GDP growth, inflation, and credit growth. The exercise shows the results when macroprudential policy, monetary policy, and capital flow measures tighten or when the central bank intervenes in the foreign exchange market by selling foreign currency, when financial conditions are loose. MPM All is the shock based on Alam et al.’s (2019) index of 17 macroprudential measures. MPM Borrower-Based and MPM FI-Based are the same as MPM All but only use borrower-based and financial-institution-based prudential measures, respectively. MP is a monetary policy shock calculated as the residual of an estimated Taylor rule. FXI is a measure of FX interventions. CFM is Baba et al.’s (forthcoming) index of capital-flow-management measures. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on a country-level cluster bootstrap.

	Domestic Shock			External Shock
	Linear-qua		Asymmetric	Linear-quadratic		Asymmetric
	ω_y=l, ω_p=0	ω_y=l, ω_p=1	ω_y=1, ω_p=0	ω_y=1, ω_p=0	ω_y=l, ω_p=1	ω_y=l, ω_p=0
MPM All	-0.080 ***	-0.076 ***	-0.087 ***	-0.100 ***	-0.095 ***	-0.109 ***
MPM Borrower-Based	-0.092 ***	-0.087 ***	-0.098 ***	-0.097 ***	-0.089 ***	-0.100 ***
MPM FI-Based	-0.046 **	-0.045 **	-0.053 *	-0.060 **	-0.058 **	-0.067 ***
MP	0.134 ***	0.126 ***	0.124 ***	0.046 **	0.044 **	0.040 *
FXI	-	-	-	-0.029	-0.027 *	-0.024
CFM	-	-	-	-0.040	-0.033	-0.041

Second, we also consider the case of a policymaker with asymmetric preferences. That is, policymakers may have loss aversion and dislike bad outcomes more than they like good ones. A flexible loss function that accommodates such asymmetry is the linex loss function proposed by Varian (1975) and Zellner (1986) and used to model central bank preferences by Ruge-Murcia (2003) for monetary policy and Bahaj and Foulis (2017) for macroprudential policy, for example. For simplicity we focus only on output and disregard inflation. The period loss is as follows:

ℓ_{t} \begin{matrix} = \frac{(e^{a {\tilde{y}}_{t + h}} - a {\tilde{y}}_{t + h} - 1)}{a^{2}}, w h e r e {\tilde{y}}_{t + h} \equiv (y_{t + h} - \bar{y}) & (11) \end{matrix}

This loss function includes (4) as a special case when a→0 and covers the case of loss aversion when a<0. Also, for simplicity, we use a=-1 and, as before, assume that potential output grows at a constant rate. To be able to use write the loss function (11) as a function of the moments of the conditional distribution of future GDP growth, we do a fourth order Taylor expansion as follows (see Hlawitschka 1994, and Plunus and others 2015):

\begin{matrix} E_{t} (ℓ_{t + h}) & = & l (0) + l^{'} (0) E_{t} ({\tilde{y}}_{t + h}) + \frac{1}{2} {l^{'}}^{'} (0) E_{t} {({\tilde{y}}_{t + h})}^{2} + \frac{1}{3!} {l^{'}}^{'}^{'} (0) E_{t} {({\tilde{y}}_{t + h})}^{3} \\ + \frac{1}{4!} {l^{'}}^{'}^{'}^{'} (0) E_{t} {({\tilde{y}}_{t + h})}^{4} + \dots \\ \tilde{=} & \frac{1}{2} E_{t} {({\tilde{y}}_{t + h})}^{2} - \frac{1}{3!} a E_{t} {({\tilde{y}}_{t + h})}^{3} + \frac{1}{4!} a^{2} E_{t} {({\tilde{y}}_{t + h})}^{4} \end{matrix}

where b is a constant. The expression above shows that, when a is not too small, we can write (4) as a function of the conditional variance, skewness, and kurtosis of detrended GDP at each horizon h.²⁷

The results with asymmetric preferences also confirm our findings based on standard quadratic loss functions (Table 6). Borrower-based macroprudential policies remain the most effective in terms of the associated reduction in the loss functions, while monetary policy remains ineffective, and FX purchases and CFM have mostly insignificant effects.

B. Alternative Measures of Monetary Policy

Our results so far suggest that, compared to monetary policy, macroprudential policy is associated with larger reductions in the loss function when financial conditions are loose. Importantly, the results so far suggest that tightening monetary policy to work against loose financial conditions actually increases the loss functions.

However, when it comes to monetary policy shocks, it is notoriously difficult to overcome the associated identification challenges (Ramey 2016). For example, poorly identified shocks often give rise to anomalous responses in the price level (Eichenbaum 1992). A strand of literature which has dealt with this problem with some success uses the response of asset prices (money market rates, bond yields, or interest rate futures) following a monetary policy announcement (Kuttner 2001 and Gürkaynak and others 2005). Accordingly, the monetary policy shocks are measured as monetary policy surprises (i.e., the component of monetary policy which was unanticipated by financial markets).

For some advanced economies, especially for the United States and the euro area, deep and sophisticated financial markets require (and the existence of intraday data allow) the surprise to be measured within relatively short windows after the announcement (5, 15, or 30 minutes). Unfortunately, for most emerging markets and many smaller advanced economies, such data do not exist or are of dubious quality (e.g., the data only exist at most at the daily frequency and market functioning is weak). Still, we check the robustness of our findings (i.e., the relative ranking of macroprudential and monetary policies) using high-frequency measures obtained from a variety of sources (see Appendix A).²⁸

Qualitatively, the results are unchanged: as per our loss-function-based metric, macroprudential policy is still preferred to monetary policy. The last row of Table 3 shows that leaning against the wind with monetary policy is associated with a very small reduction in the loss function (1 percent for loose domestic FCI and at most 2.5 percent for loose global FCI). This finding is robust to using alternative specifications for the loss function and, for the most part, is not caused by sample composition effects.²⁹

C. Advanced- vs. Emerging Market Economies

We also check if our results are driven by our choice to pool advanced and emerging economies. There are two reasons why this choice may confound the results. First, the intertemporal tradeoff between current financial conditions and future downside risks may be different for advanced and emerging economies (see Adrian and others 2019). This suggests a potential bias caused by the assumption of homogenous slopes in the linear quantile regressions. Second, the types of macroprudential policies used differs between emerging market- and advanced economies; for example, the use of mortgage-related measures is much more common in advanced economies given the bigger role of mortgage lending in these countries. Moreover, the use of certain policies such as CFMs is more common among emerging economies than in advanced economies. Thus, the estimated beneficial effect of certain policies could be just capturing those group-specific intertemporal tradeoffs.

Our results by group of country broadly support our benchmark findings (Table 7): macroprudential policy is generally associated with larger reductions in the loss function than any other policy. Interestingly, financial-institution-based macroprudential policies and, to a smaller extent, CFMs seem to work better in emerging markets than in advanced economies, whereas FX purchases again do not reduce losses.

^{Table 7.}

Effect of Policy Changes on Loss Functions by Group of Countries

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, a n d \\ Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

for domestic and global FCI shocks, respectively, where Q_Y(q|.) is the conditional quantile function of cumulative GDP growth for up to 14 quarters ahead, f is a country-specific financial conditions index, g is a global financial conditions index, P is a policy shock, and x is a vector of controls including current GDP growth, inflation, and credit growth. The exercise shows the results when macroprudential policy, monetary policy, and capital flow measures tighten or when the central bank intervenes in the foreign exchange market by selling foreign currency, when financial conditions are loose. MPM All is the shock based on Alam et al.’s (2019) index of 17 macroprudential measures. MPM Borrower-Based and MPM FI-Based are the same as MPM All but only use borrower-based and financial-institution-based prudential measures, respectively. MP is a monetary policy shock calculated as the residual of an estimated Taylor rule. FXI is a measure of FX interventions. CFM is Baba et al.’s (forthcoming) index of capital-flow-management measures. The group advanced economies includes those considered as advanced economies as per the 2019 World Economic Outlook database; all other economies are included in the group emerging economies. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on a country-level cluster bootstrap.

^{Table 7.}

Effect of Policy Changes on Loss Functions by Group of Countries

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, a n d \\ Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

for domestic and global FCI shocks, respectively, where Q_Y(q|.) is the conditional quantile function of cumulative GDP growth for up to 14 quarters ahead, f is a country-specific financial conditions index, g is a global financial conditions index, P is a policy shock, and x is a vector of controls including current GDP growth, inflation, and credit growth. The exercise shows the results when macroprudential policy, monetary policy, and capital flow measures tighten or when the central bank intervenes in the foreign exchange market by selling foreign currency, when financial conditions are loose. MPM All is the shock based on Alam et al.’s (2019) index of 17 macroprudential measures. MPM Borrower-Based and MPM FI-Based are the same as MPM All but only use borrower-based and financial-institution-based prudential measures, respectively. MP is a monetary policy shock calculated as the residual of an estimated Taylor rule. FXI is a measure of FX interventions. CFM is Baba et al.’s (forthcoming) index of capital-flow-management measures. The group advanced economies includes those considered as advanced economies as per the 2019 World Economic Outlook database; all other economies are included in the group emerging economies. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on a country-level cluster bootstrap.

	Domestic FCI			External FCI
	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l,	ω_y=l, ω_p=0	ω_y=l, ω_p=1	ω_y=0.542, ω_p=1
	Advanced economies
MPM All	-0.120 **	-0.116 **	-0.113 **	-0.139 **	-0.136 **	-0.133 **
MPM Borrower-Based	-0.141 **	-0.136 **	-0.132 *	-0.142 ***	-0.139 ***	-0.136 ***
MPM FI-Based	-0.027	-0.026	-0.025	-0.046	-0.045	-0.045
MP	0.127 ***	0.124 ***	0.122 ***	0.075	0.075	0.075
FXI	-	-	-	0.051	0.049	0.047
CFM	-	-	-	0.015	0.015	0.015
	Emerging economies
MPM All	-0.081 ***	-0.078 ***	-0.075 ***	-0.143 ***	-0.062 ***	-0.038 ***
MPM Borrower-Based	-0.067 **	-0.064 **	-0.061 **	-0.136 *	-0.099 *	-0.089 *
MPM FI-Based	-0.074 **	-0.072 **	-0.070 **	-0.132 ***	-0.125 ***	-0.120 ***
MP	0.086 **	0.080 ***	0.077 ***	0.092 *	0.089 *	0.086 **
FXI	-	-	-	0.017	0.014	0.011
CFM	-	-	-	-0.065 *	-0.050	-0.040

^{Table 7.}

Effect of Policy Changes on Loss Functions by Group of Countries

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, a n d \\ Q_{Y_{i, t, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) g_{i t} + β_{3}^{h} (q) P_{i t} + β_{4}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

for domestic and global FCI shocks, respectively, where Q_Y(q|.) is the conditional quantile function of cumulative GDP growth for up to 14 quarters ahead, f is a country-specific financial conditions index, g is a global financial conditions index, P is a policy shock, and x is a vector of controls including current GDP growth, inflation, and credit growth. The exercise shows the results when macroprudential policy, monetary policy, and capital flow measures tighten or when the central bank intervenes in the foreign exchange market by selling foreign currency, when financial conditions are loose. MPM All is the shock based on Alam et al.’s (2019) index of 17 macroprudential measures. MPM Borrower-Based and MPM FI-Based are the same as MPM All but only use borrower-based and financial-institution-based prudential measures, respectively. MP is a monetary policy shock calculated as the residual of an estimated Taylor rule. FXI is a measure of FX interventions. CFM is Baba et al.’s (forthcoming) index of capital-flow-management measures. The group advanced economies includes those considered as advanced economies as per the 2019 World Economic Outlook database; all other economies are included in the group emerging economies. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on a country-level cluster bootstrap.

	Domestic FCI			External FCI
	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l,	ω_y=l, ω_p=0	ω_y=l, ω_p=1	ω_y=0.542, ω_p=1
	Advanced economies
MPM All	-0.120 **	-0.116 **	-0.113 **	-0.139 **	-0.136 **	-0.133 **
MPM Borrower-Based	-0.141 **	-0.136 **	-0.132 *	-0.142 ***	-0.139 ***	-0.136 ***
MPM FI-Based	-0.027	-0.026	-0.025	-0.046	-0.045	-0.045
MP	0.127 ***	0.124 ***	0.122 ***	0.075	0.075	0.075
FXI	-	-	-	0.051	0.049	0.047
CFM	-	-	-	0.015	0.015	0.015
	Emerging economies
MPM All	-0.081 ***	-0.078 ***	-0.075 ***	-0.143 ***	-0.062 ***	-0.038 ***
MPM Borrower-Based	-0.067 **	-0.064 **	-0.061 **	-0.136 *	-0.099 *	-0.089 *
MPM FI-Based	-0.074 **	-0.072 **	-0.070 **	-0.132 ***	-0.125 ***	-0.120 ***
MP	0.086 **	0.080 ***	0.077 ***	0.092 *	0.089 *	0.086 **
FXI	-	-	-	0.017	0.014	0.011
CFM	-	-	-	-0.065 *	-0.050	-0.040

D. Common Factors and Unobserved Heterogeneity

Macroeconomic and financial conditions are potentially correlated across countries, reflecting the role of global investors shaping global financial conditions and the global credit cycle. Our analysis, up to this point, did not explicitly control for the importance of these factors, which we now do.

First, we capture these common factors as time fixed effects, in addition to country fixed effects, as a two-way error structure. Unfortunately, under the assumption of fixed N, Kato and other’s (2012) FE-QR estimator that we used before is no longer consistent with two-way fixed effects. However, we can still estimate the country fixed effects as before (i.e., as dummy variables) and, under the assumption that the time fixed effects are constant across quantiles, use Koenker’s (2004) penalized fixed effects estimator.³⁰ The results, shown in Table 8 (panel A), are qualitatively similar to those of Section IV, with macroprudential policy reducing losses and monetary policy increasing them.

^{Table 8.}

Effect of Policy Changes on Loss Functions Accounting for Common Factors

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + λ_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

where Q_Y(q|.) is the conditional quantile function of cumulative GDP growth for up to 14 quarters ahead, α and λ are country and time fixed effects, respectively, f is a country-specific financial conditions index, g is a global financial conditions index, P is a policy shock, and x is a vector of controls including current GDP growth, inflation, and credit growth. The exercise shows the results when macroprudential policy and monetary policy measures tighten and financial conditions are loose. MPM All is the shock based on Alam et al.’s (2019) index of 17 macroprudential measures. MPM Borrower-Based and MPM FI-Based are the same as MPM All but only use borrower-based and financial-institution-based prudential measures, respectively. MP is a monetary policy shock calculated as the residual of an estimated Taylor rule. The columns in panel A (Common factors) show the results using quantile regressions with individual (country) and (quarter) fixed effects. Time fixed effects are assumed to be pure location shift parameters and estimated using Koenker’s (2004) penalized estimator (R library rqpd). The columns in panel B (Unobserved heterogeneity) show the results using quantile regressions show the results using quantile regressions with country-specific slope coefficients and five unobserved common factors. These quantile regressions are estimated using Ando and Bai’s (2019) procedure and their source code.

^{Table 8.}

Effect of Policy Changes on Loss Functions Accounting for Common Factors

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + λ_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

	Panel A Common factors
	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l
MPM All	-0.073	-0.070	-0.067
MPM Borrower-Based	-0.038	-0.035	-0.033
MPM FI-Based	-0.053	-0.051	-0.050
MP	0.142	0.135	0.130

	Panel B Unobserved heterogeneity
	ω_y=l, ω_p=0	ω_y=l, ω_p = l	ω_y=0.542, ω_p=l
MPM All	-0.104	-0.108	-0.110
MPM Borrower-Based	-0.477	-0.336	-0.228
MPM FI-Based	-0.129	-0.124	-0.120
MP	0.063	0.060	0.057

^{Table 8.}

Effect of Policy Changes on Loss Functions Accounting for Common Factors

The table shows the estimated values of the expected loss given by the following period loss function

\begin{matrix} ∣_{t} & = & ω_{y} {(y_{t + h} - {\bar{y}}_{t + h})}^{2} + ω_{π} π_{t + h}^{2} \end{matrix},

which are obtained from the following quantile regressions

\begin{matrix} Q_{Y_{i, t + h}} (q | Z_{i t}) & = & α_{0 i}^{h} (q) + λ_{0 i}^{h} (q) + β_{1}^{h} (q) f_{i t} + β_{2}^{h} (q) P_{i t} + β_{3}^{h} (q) P_{i t} \times f_{i t} + x_{i t} Γ, \\ h & = & 1, Κ, H, q = 0.05, Κ, 0.95, \end{matrix}

	Panel A Common factors
	ω_y=l, ω_p=0	ω_y=l, ω_p=l	ω_y=0.542, ω_p=l
MPM All	-0.073	-0.070	-0.067
MPM Borrower-Based	-0.038	-0.035	-0.033
MPM FI-Based	-0.053	-0.051	-0.050
MP	0.142	0.135	0.130

	Panel B Unobserved heterogeneity
	ω_y=l, ω_p=0	ω_y=l, ω_p = l	ω_y=0.542, ω_p=l
MPM All	-0.104	-0.108	-0.110
MPM Borrower-Based	-0.477	-0.336	-0.228
MPM FI-Based	-0.129	-0.124	-0.120
MP	0.063	0.060	0.057

For our benchmark results, we have also assumed that the slope coefficients of the quantile regressions are homogenous across countries and that the error structure is the only source of heterogeneity. However, assuming homogenous slopes will produce biases in the estimation if current financial conditions transmit differently to future output growth distributions in different countries. We check for the robustness of our findings against this type of misspecification using Ando and Bai’s (2019) frequentist approach to panel quantile models with unobserved heterogeneity and common factors.³¹ The model that we estimate is a slightly modified version of (3),

\begin{matrix} Q_{Δ y_{i, t, t + h}} (q | Z_{i t}) & = & φ (q)^{'} λ_{i}^{h} (q) + β_{1 i}^{h} (q) f_{i t} + β_{2 i}^{h} (q) g_{i t} + β_{3 i}^{h} (q) P_{i t} + β_{4 i}^{h} (q) P_{i t} \times g_{i t} + x_{i t} Γ_{i}, \\ h & = & 1, Κ, H, q = 0.1, Κ, 0.9, \end{matrix}

where φ and λ are quantile-specific vectors of unobservable factors and factor loadings, respectively.³² The results are in Table 8 (panel B) and are broadly in line with the benchmark results: macroprudential policy reduces losses, borrower-based measures reduce them by more, and monetary policy does not seem to do the same.

VI. Conclusions

In this paper, we have proposed a novel approach to evaluate the effects of “leaning against the wind” in response to changes in domestic and global financial conditions. We have assessed the impact of different policies on the entire future probability distributions of growth and inflation outcomes, evaluating them using a range of standard- and nonstandard loss functions.

We found that overall, the balance of trade-offs favors the use of macroprudential policies, both in response to changes in domestic financial conditions and to global financial shocks. When credit is high, tightening borrower-based measures seems more advantageous than tightening financial-institutions based macroprudential measures, and the converse seems true when credit is low. Moreover, tightening macroprudential policies has stronger benefits when they are accompanied by looser monetary policy. By contrast, the trade-off for monetary policy tightening to lean against the wind alone generally appears to be unfavorable and is associated with higher losses.

Since many emerging markets have been using CFMs and foreign-exchange intervention in response to global financial conditions, we also investigated the role of these policies. Foreign-exchange interventions and CFMs entail small reductions in the loss function, which are considerably smaller than the one associated with the use of macroprudential policy. Overall, any beneficial effects of FX purchases and CFM on future output growth are not statistically significant.

However, our results should be seen as first insights using a novel approach, rather than settling these complex issues. For instance, it is possible that FXI and CFMs develop benefits in certain circumstances even if their effects on output growth on average are small. Using these tools to mitigate adverse shocks could also have greater benefits than using them to lean against easing financial conditions. Further research needs to explore these questions further, for example by taking a closer look at cross-country differences, such as those related to the structure of their financial systems, and the nature and depth of their international financial linkages.

In addition, drawing direct policy recommendations from our findings would clearly not be immune to Lucas’s (1976) critique—that the effect of policies may come to depend on the way they are used. Overall, this paper stands in the tradition of using macroeconometric models for policy evaluation, as in Sims (1980, 1982), Jordà (2005), and Angrist and Kuersteiner (2011), maintaining the assumption that policy choices can be informed by applying regression methods to past data (Kocherlakota 2019).

References

Adam, Klaus, and Michael Woodford. 2018. “Leaning Against Housing Prices as Robustly Optimal Monetary Policy,” NBER Working Paper No. 24629
- Search Google Scholar
- Export Citation
Adler, Gustavo, Noemie Lisack, and Rui Mano. 2019. “Unveiling the effects of foreign exchange intervention: A panel approach.” Emerging Markets Review, Vol. 40: 100620.
- Search Google Scholar
- Export Citation
Adrian, Tobias, Nina Boyarchenko, and Domenico Giannone. 2019. “Vulnerable Growth.” American Economic Review, Vol. 109(4): 1263–89.
- Search Google Scholar
- Export Citation
Adrian, Tobias, Fernando Duarte, Nellie Liang, and Pawel Zabcyk. Forthcoming. “Monetary and Macroprudential Policy with Endogenous Risk.” IMF Working Paper.
- Search Google Scholar
- Export Citation
Adrian, Tobias, Federico Grinberg, Nellie Liang, and Sheheryar Malik. 2018. “The Term Structure of Growth-at-Risk,” IMF Working Paper No. 18/180. (Washington: International Monetary Fund).
- Search Google Scholar
- Export Citation
Adrian, Tobias, and Nellie Liang. 2018. “Monetary Policy, Financial Conditions, and Financial Stability.” International Journal of Central Banking 14(1): 73–131.
- Search Google Scholar
- Export Citation
Agénor, Pierre-Richard, Koray Alper, and Luiz Pereira da Silva. 2013. “Capital Regulation, Monetary Policy, and Financial Stability,” International Journal of Central Banking 9(3): 198–243.
- Search Google Scholar
- Export Citation
Agur, Itai. 2019. “Monetary and Macroprudential Policy Coordination Among Multiple Equilibria.” Journal of International Money and Finance 96: 192–209.
- Search Google Scholar
- Export Citation
Ajello, Andrea, Thomas Laubach, David López-Salido, and Taisuke Nakata. 2019. “Financial Stability and Optimal Interest Rate Policy,” International Journal of Central Banking 15(1): 279–326.
- Search Google Scholar
- Export Citation
Akinci, Ozge, and Jane Olmstead-Rumsey. 2018. “How effective are macroprudential policies? An empirical investigation,” Journal of Financial Intermediation 33:33–57
- Search Google Scholar
- Export Citation
Alam, Zohair, Adrian Alter, Jesse Eiseman, Gaston Gelos, Heedon Kang, Machiko Narita, Erlend Nier, and Naixi Wang. 2019. “Digging Deeper—Evidence on the Effects of Macroprudential Policies from a New Database.” IMF Working Paper No. 19/66. (Washington: International Monetary Fund).
- Search Google Scholar
- Export Citation
Alpanda, Sami, and Sarah Zubairy. 2017. “Addressing Household Indebtedness: Monetary, Fiscal or Macroprudential Policy?” European Economic Review 92: 47–73.
- Search Google Scholar
- Export Citation
Alpanda, Sami, Gino Cateau, and Cesaire Meh. 2018. “A policy model to analyze macroprudential regulations and monetary policy.” Canadian Journal of Economics, 51 (3), 828–863
- Search Google Scholar
- Export Citation
Amengual, Dante and Mark W Watson. 2007. “Consistent Estimation of the Number of Dynamic Factors in a Large N and T Panel,” Journal of Business & Economic Statistics, 25(1): 91–96.
- Search Google Scholar
- Export Citation
Ando, Tomohiro, and Jushan Bai. 2019. “Quantile co-movement in financial markets: A panel quantile model with unobserved heterogeneity.” Journal of the American Statistical Association, article in press.
- Search Google Scholar
- Export Citation
Angelini, Paolo, Stefano Neri, and Fabio Panetta. 2014. “The Interaction between Capital Requirements and Monetary Policy.” Journal of Money, Credit, and Banking 46 (6): 1073–1112.
- Search Google Scholar
- Export Citation
Angrist, Joshua D., and Guido M. Kuersteiner. 2011. “Causal Effects of Monetary Shocks: Semiparametric Conditional Independence Tests with a Multinomial Propensity Score.” Review of Economics and Statistics 93, no. 3: 725–747.
- Search Google Scholar
- Export Citation
Aoki, Kasuke, Gianluca Benigno, and Nobuhiro Kiyotaki, 2016. “Monetary and Financial Policies in Emerging Markets,” unpublished manuscript.
- Search Google Scholar
- Export Citation
Arregui, Nicolás, Selim Elekdag, Gaston Gelos, Romain Lafarguette, and Dulani Seneviratne. 2018. “Can Countries Manage Their Financial Conditions Amid Globalization?”, IMF Working Paper 18/15.
- Search Google Scholar
- Export Citation
Auerbach, Alan J. and Yuriy Gorodnichenko, 2013. “Measuring the Output Responses to Fiscal Policy.” American Economic Journal: Economic Policy, 4 (2): 1–27.
- Search Google Scholar
- Export Citation
Avdjiev, Stefan, Valentina Bruno, Catherine Koch, and Hyun Song Shin. 2019. “The Dollar Exchange Rate as a Global Risk Factor: Evidence from Investment.” IMF Economic Review 67, no. 1: 151–173.
- Search Google Scholar
- Export Citation
Azzalini, Adelchi. 1985. “A Class of Distributions Which Includes the Normal Ones.” Scandinavian Journal of Statistics 12, no. 2: 171–178.
- Search Google Scholar
- Export Citation
Azzalini, Adelchi. 2013. The Skew-Normal and Related Families. Vol. 3. Cambridge University Press.
- Search Google Scholar
- Export Citation
Baba, Chikako, Ricardo Cervantes, Salim Darbar, Annamaria Kokenyne, and Viktorya Zotova. Forthcoming. “New Measures of Capital Flow Restrictions — AREAER Indices.” IMF Working Paper. (Washington: International Monetary Fund).
- Search Google Scholar
- Export Citation
Backus, David, and Patrick Kehoe. 1989. “On the Denomination of Government Debt: A Critique of the Portfolio Balance Approach,” Journal of Monetary Economics, 23 (3), 359–76.
- Search Google Scholar
- Export Citation
Bahaj, Saleem, and Angus Foulis. 2017. “Macroprudential Policy Under Uncertainty.” International Journal of Central Banking 13, no. 3: 119–154.
- Search Google Scholar
- Export Citation
Bai, Jushan and Serena Ng. 2002. “Determining the Number of Factors in Approximate Factor Models,” Econometrica, 70: 191–221.
- Search Google Scholar
- Export Citation
Bean, Charles, Matthias Paustian, Adrian Penalver, and Tim Taylor, 2010. “Monetary Policy after the Fall”, Federal Reserve Bank of Kansas City in Macroeconomic Challenges: The Decade Ahead: Proceedings of the Federal Reserve Bank of Kansas City Economic Symposium, Jackson Hole.
- Search Google Scholar
- Export Citation
Bengui, Julien, and Javier Bianchi. 2018. “Macroprudential Policy with Leakages.” NBER Working Paper No. 25048. National Bureau of Economic Research.
- Search Google Scholar
- Export Citation
Benigno, Pierpaolo, and Michael Woodford. 2005. “Inflation stabilization and welfare: The case of a distorted steady state.” Journal of the European Economic Association 3, no. 6: 1185–1236.
- Search Google Scholar
- Export Citation
Bergant, Katharina, Francesco Grigoli, Niels-Jakob Hansen and Damiano Sandri. Forthcoming. “Dampening Global Financial Shocks: Can Macroprudential Regulation Help (More than Capital Controls)?” IMF Working Paper.
- Search Google Scholar
- Export Citation
Bernanke, Ben S., Mark Gertler, and Simon Gilchrist. 1999. “The Financial Accelerator in a Quantitative Business Cycle Framework.” Handbook of Macroeconomics 1: 1341–1393.
- Search Google Scholar
- Export Citation
Biljanovska, Nina, Lucyna Gornicka, and Alexandros Vardoulakis. 2019. “Optimal Macroprudential Policy and Asset Price Bubbles.” IMF Working Paper No. 19/184. International Monetary Fund.
- Search Google Scholar
- Export Citation
Bodenstein, Martin, Luca Guerrieri, and Joe LaBriola. 2019. “Macroeconomic Policy Games.” Journal of Monetary Economics 101: 64–81.
- Search Google Scholar
- Export Citation
Bruno, Valentina, Ilhyock Shim, and Hyun Song Shin. 2017. “Comparative assessment of macroprudential policies.” Journal of Financial Stability, Vol. 28: 183–202.
- Search Google Scholar
- Export Citation
Caldera Sánchez, Aida, and Oliver Röhn. 2016. “How do policies influence GDP tail risks?” OECD Economics Department Working Papers No. 1339.
- Search Google Scholar
- Export Citation
Canay, Ivan A., 2011. “A Simple Approach to Quantile Regression for Panel Data.” The Econometrics Journal 14, no. 3: 368–386.
- Search Google Scholar
- Export Citation
Carrillo, Juan, Enrique G. Mendoza, Victoria Nuguer, and Jessica Roldán-Peña. 2018. “Tight Money-Tight Credit: Coordination Failure in the Conduct of Monetary and Financial Policies.” ECB Working Paper 2129.
- Search Google Scholar
- Export Citation
Cavallino, Paolo, 2019. “Capital Flows and Foreign Exchange Intervention.” American Economic Journal: Macroeconomics, Vol. 11(2), 127–70.
- Search Google Scholar
- Export Citation
Céspedes, Luis Felipe & Chang, Roberto & Velasco, Andrés, 2017. “Financial intermediation, real exchange rates, and unconventional policies in an open economy,” Journal of International Economics, Vol. 108(S1): 76–86.
- Search Google Scholar
- Export Citation
Chamon, Marcos, David Hofman, Sergi Lanau, Umang Rawat, and Miklos Vari. 2019. “The Effectiveness of Intervention,” Chapter 4: 43–61, in Foreign Exchange Intervention in Inflation Targeters in Latin America, edited by Marcos Chamon, David Hofman, Nicolás E. Magud, and Alejandro Werner. International Monetary Fund.
- Search Google Scholar
- Export Citation
Chang, Roberto. 2018. “Foreign Exchange Intervention Redux,” NBER Working Paper No. 24463.
- Search Google Scholar
- Export Citation
Chinn, Menzie D. and Hiro Ito, 2006. “What Matters for Financial Development? Capital Controls, Institutions, and Interactions.” Journal of Development Economics, Volume 81, Issue 1, Pages 163–192.
- Search Google Scholar
- Export Citation
Choi, Jae Hoon. 2020. “Capital Controls and Foreign Exchange Market Intervention.” Journal of International Money and Finance 101: 102098.
- Search Google Scholar
- Export Citation
Claessens, Stijn, M. Ayhan Kose, and Marco E. Terrones. 2011. “Financial Cycles: What? How? When?” In International Seminar on Macroeconomics, Volume 7, No. 1: 303–344. Chicago, IL: University of Chicago Press.
- Search Google Scholar
- Export Citation
Clarida, Richard, Jordí Gali and Mark Gertler, 2000. “Monetary Policy Rules and Macroeconomic Stability: Evidence and Some Theory”, The Quarterly Journal of Economics, Vol 115 (1), 147–180.
- Search Google Scholar
- Export Citation
Coibion, Olivier, and Yuriy Gorodnichenko, 2012. “What Can Survey Forecasts Tell Us about Information Rigidities?” Journal of Political Economy, Volume 120, No. 1: 116–159.
- Search Google Scholar
- Export Citation
Collard, Fabrice, Harris Dellas, Behzad Diba, and Olivier Loisel. 2017. “Optimal Monetary and Prudential Policies.” American Economic Journal: Macroeconomics 9, no. 1: 40–87.

Topics

Countries

Series

About

Resources

Contributor Notes

Abstract

I. Introduction

II. Empirical Approach

A. Quantile Regressions

B. Integrating and Computing Loss Functions

III. Data

A. Financial and Macroeconomic Conditions

B. Policy Shocks

Macroprudential Policy

Monetary Policy

Foreign Exchange Interventions

Capital Flow Management Measures

IV. Results

A. Effects of Policies on the Intertemporal Tradeoff

B. Net Benefits of Policies

Changes in Domestic Financial Conditions

Global Financial Shocks

Changes in the Loss Function Over the Forecasting Horizon

C. Effects of Policies Conditional on the Level of Financial Vulnerabilities

D. Interaction Between Monetary and Macroprudential Policies

V. Robustness

A. Alternative Loss Functions

B. Alternative Measures of Monetary Policy

C. Advanced- vs. Emerging Market Economies

D. Common Factors and Unobserved Heterogeneity

VI. Conclusions

References