Evaluating the Net Benefits of Macroprudential Policy
A Cookbook
Author: Mr. Nicolas Arregui, Mr. Jaromir Benes, Mr. Ivo Krznarhttps://orcid.org/0000-0003-4174-0286, Ms. Srobona Mitra, and Mr. Andre O Santos

Contributor Notes

Author’s E-Mail Address: narregui@imf.org; jbenes@imf.org; ikrznar@imf.org; smitra@imf.org; asantos2@imf.org
Publication Date:
17 Jul 2013
eISBN:
9781484335727
Language:
English
Keywords:
WP; economic activity; cost of capital; banking sector; interest rate
Download PDF (6.5 MB)

The paper proposes a simple, new, analytical framework for assessing the cost and benefits of macroprudential policies. It proposes a measure of net benefits in terms of parameters that can be estimated: the probability of crisis, the loss in output given crisis, policy effectiveness in bringing down both the probability and damage during crisis, and the output-cost of a policy decision. It discusses three types of policy leakages and identifies instruments that could best minimize the leakages. Some rules of thumb for policymakers are provided.

Abstract

The paper proposes a simple, new, analytical framework for assessing the cost and benefits of macroprudential policies. It proposes a measure of net benefits in terms of parameters that can be estimated: the probability of crisis, the loss in output given crisis, policy effectiveness in bringing down both the probability and damage during crisis, and the output-cost of a policy decision. It discusses three types of policy leakages and identifies instruments that could best minimize the leakages. Some rules of thumb for policymakers are provided.

I. Motivation

1. The main contribution of the paper is to provide a unified framework for assessing net benefits of macroprudential policy, together with guidance on assessing policy leakages. The paper suggests a simple conceptual framework for comparing the costs with benefits of macroprudential policy.2 The concept of long-run cost and benefit is adopted from BCBS (2010a), albeit with differences in detail. The costs arise from an increase in the cost of intermediation and its effect on long-run output. The benefit is derived from the resilience of the economy from the policy measure—a reduction in the probability of a crisis and output losses in the event of a crisis. As a bi-product, the paper also suggests a method for analyzing feedbacks between the financial sector and real economic activity that could be useful for policy makers in general.3

2. Existing literature has tackled some aspects of the net benefits to macroprudential policies. Most studies focus on long term costs and benefits of macroprudential regulation (higher capital and liquidity requirements, most prominently) with the exception of BCBS (2010b) that also studies transitional costs. Long term quantitative studies follow similar methodologies but with different assumptions or estimation techniques (see ICFR, 2011, for a comparative summary of different studies). Benefits are generally defined as a reduction in expected crisis costs. The emphasis is more on the effect of regulation on the probability of a crisis, and less on the effect of regulation on the depth of a crisis. In terms of costs, most papers relate regulations to credit and output through the intermediate effect of regulation on lending spreads. While a few studies mention the risks posed to financial stability by regulation-induced growth in shadow banking, such side-effects/leakages are considered only qualitatively. A review of the literature on costs and benefits is provided in Annex 1.

3. This paper takes a broader perspective on analyzing the net benefits of policy. The starting point of the analysis rests on the notion that macroprudential policies are needed to dampen the effect of three externalities, as described in De Nicolo, Favara, and Ratnovski (2012) (DFR henceforth). The paper uses indicators to describe two of the three externalities and assumes that macroprudential policies are aimed at these indicators that serve as intermediate targets to limit systemic risks.4 If systemic risk is correctly identified and policy is successful in either avoiding or reducing the effects of a crisis, systemic risk is lowered, and the economy is presumed to be on a higher level of real economic activity in the long-run. However, there are costs imposed on intermediation and output associated with (1) higher price and lower volume of intermediation during normal times, and (2) mis-identification of risk or over-regulation (“type II error”).

4. The analysis contributes to new insights on a number of fronts. So far, the cost and benefits of regulation have been discussed in the context of the Basel III rules (BCBS, 2010a and 2010b, Elliott and Santos, 2012). Even though there is some work done on analyzing the effects of macroprudential policies, there is none that uses a cross-country database of a range of macroprudential policy instruments that are currently in use by countries (except for Lim et al, 2011 and Vandenbussche et al, 2012). This paper fills the gap by using the IMF (2011c) survey (henceforth, Survey) and a new dataset compiled by Lim et al (forthcoming). Cross-country empirical evidence is essential to this analysis, because of the lack of sufficient number of crisis-episodes in individual countries. Although the structure of individual economies may vary, it is still useful to extract the common characteristics before, during and after crisis using panel data techniques. The paper also uses a time-series model for the U.S. to derive sensitivities of real GDP forecast to credit aggregates—basic macro-financial linkages—with asymmetric effects during normal and crisis times. Last, but not least, this paper analyzes leakages or unintended consequences of macroprudential policies, based on cross-country evidence and case studies.

5. The goal is to set up the components of a core semi-structural model. The model would provide real-time estimates of the medium-to-long-term trend in real economic activity, and its interactions with indicators of macrofinancial externalities (the intermediate targets). This model would bring together the various parts of the exercise—estimating probability of crisis, loss given crisis, effects of policy, forecast of real activity and intermediate targets—into one model. Although this core model is not yet ready to be discussed, all the components are introduced in this paper.

6. The strategy will be a multi-step process, providing the key ingredients that would go towards assessing the net benefits of policy. The paper will first present a conceptual framework to assess costs and benefits of policy in terms of parameters that can be estimated—probability of crisis, loss in output given crisis, policy effectiveness in bringing down both the probability of and loss during a crisis, and the cost of policy (Figure 1). Then it will provide some numbers and estimates for the building blocks in section III.

Figure 1.
Figure 1.

Structure of the Analytical Framework

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

7. The steps are as follows:

  • When to Act. Summarize the effectiveness of early warning indicators, and the role of credit aggregates in deciding when to act (Section II).

  • Concept. Set up a simple conceptual framework for understanding the net benefits of policy expressed in terms of parameters that can be estimated (Section II). These ingredients would then be part of the core analytical model.

  • Policy Effectiveness. Evaluate the effectiveness of policy in lowering systemic risk indicators (credit growth, loan/deposit ratios, asset price growth, foreign liabilities growth) (Section III).

  • Benefits. Estimate the effect of change in systemic risk indicators on the probability of a crisis and depth of the output loss during a crisis from cross-country data (Section III), by extending the analysis in IMF (2011b) to a panel-logit model.

  • Cost. Compute the cost of policy by estimating the effect of change in the intermediate target (due to policies) on the forecast for GDP. This will be done with an empirical model with asymmetric macro-financial sensitivities based on whether the banking system is in a ‘normal’ or ‘distressed’ state (Section III).

  • Leakages. While leakages from macroprudential policies are not directly included in the core analytical model, some new findings on policy side-effects will be discussed (Section IV).

Section V will conclude with takeaways and basic rules-of-thumb that would provide practical guidance for policymakers. In many cases, country circumstances and structures will require a more selective approach.

II. An Analytical Recipe: The Ingredients

A. Systemic Risk Indicators: When and Whether to Act

8. Policy instruments seek to target certain market failures or externalities. While externalities are hard to observe, some indicators could help detect their presence or their outcomes. Thus policies would target these intermediate indicators, which are seen to be manifestations of market failures. For instance, rapid credit growth during boom times could be the result of “strategic complementarities that lead banks to take excessive or correlated risks and build up vulnerabilities during the expansionary phase of the financial cycle” (DFR, 2012). Therefore, policy would be effective if it could bring down the speed of credit growth. Similarly, indicators of funding-liquidity risk and market-liquidity risk could be used as intermediate targets for externalities related to fire sales—a generalized sell-off of financial assets causing a decline in asset prices and a deterioration of the balance sheets of intermediaries, especially during the contractionary phase of a financial cycle. Policy instruments that are able to lower a liquidity risk indicator would succeed in lowering the risk of fire sales.

9. Having said that, early warning indicators are imperfect at measuring the risk of a financial crisis. Previous studies have shown that rapid credit growth and credit booms signal a banking crisis (Laeven and Valencia, 2010) 2-3 years before the event (IMF, 2011b & 2012a, Lund-Jensen, 2012, Dell’Ariccia, Igan, Laeven, Tong, 2012, BCBS, 2010c, Drehmann and others, 2012). Even so, out-of-sample analysis show that credit growth (and gap measures) produced very low (but increasing) probabilities of crisis before 2007 (IMF, 2011b). Thus, while deciding on when to act, policy makers need to compare the possible benefits of avoiding a crisis and/or reducing the depth of a crisis, with the cost of tightening up on intermediation (and activity) now. They need to consider the imperfect nature of the signal and the possibility of over-regulation—that is, the risk that they would be acting upon a signal that could be false (a “false positive”) in the first place. On the other hand, they might be eager to avoid financial crisis at any cost. Thus, when they act depends upon whether they prefer to mitigate risks at the earliest time a signal detects an elevated risk, or they would rather wait it out and see if the signal is durable.

10. Credit aggregates are useful to guide policy, especially to inform tightening of policy, by using appropriate thresholds. Two studies that examine thresholds are IMF (2011b) in the Global Financial Stability Report (GFSR) and Dell’Ariccia et al (2012).5 Comparing the GFSR threshold (3-5 percentage point annual change in the credit-GDP ratio) with the Dell’Ariccia et al threshold for defining credit booms (henceforth, Boom) shows that the GFSR threshold (that tries to avoid crisis at any cost) is triggered earlier and much more often than Boom (also see Annex 2).6 If a Boom is identified, then the GFSR threshold is breached for over 70 percent of the cases (Table 1). On the other hand, if a GFSR threshold is identified, the Boom threshold is breached in about 30 percent of the cases. Expressing concern when the GFSR threshold is breached and escalating the concern and taking policy action by the time the Boom threshold is breached provides a meaningful range for policy implementation and could enhance transparency and market discipline.7

Figure 2.
Figure 2.

Policy Time Line

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

Table 1.

Thresholds for Credit Growth

(Number of occurrences)

article image
Note: The table shows the number of occurances across 183 countries from 1966 to 2010 when thresholds were breached. GFSR refers to the threshold of changes in credit/GDP ratio of 3 percentage points. “Boom” refers to the threshold in Dell’Ariccia et al (2012). See Annex 2.

11. To avoid underestimating risks, policy makers should include credit from all types of lenders, not just commercial banks. A case in point is the United States where nonbanks are significant players, especially in the consumer and mortgage lending markets (Box 1). Here, commercial bank credit alone greatly underestimates crisis risks and signals—neither the GFSR nor the Boom crosses their respective thresholds for the United States since 2000 (Annex 2). However, changes in consumer credit-GDP ratio (when measured properly) would have done so by 2003.

12. In reality, policymakers are more likely to look at several indicators to gauge building up of systemic risk. As explained in Blancher et al (2013)’s forthcoming “SysMo” project, different models and methods need to be combined to arrive at the decision on when and whether to act. For instance, indicators and models related to risks building up within the financial sector should be combined with asset price models to see if there is sectoral overheating. In addition, crisis prediction models are helpful in tracking the probability of crisis. Amplification channels could be explored to see if sectoral overheating is feeding into credit growth (for instance, rapidly increasing house prices lowering collateral constraints on consumption and lending behavior) and vice versa. A policy maker typically makes a decision partly based on these various combinations of models and partly on some judgment.

B. Conceptual Framework for Assessing Net Benefits of Policy

13. Having decided on when to act, policymakers need to assess the net benefit of any policy action on the long-term level of GDP. The paper assumes that the ultimate objective of policy is to bolster the long-term trend in real economic activity against permanent blows arising from financial crisis. The discussion of costs and benefits is, therefore, positioned in terms of the long-term forecast for GDP. At the juncture at which policymakers are taking the decision to act, they need to know the path of future output—the expected path given the probability of a crisis and the loss in output level in the event of a crisis. In Figure 3, a policy decision is taken at time t due to a signal that systemic risk is rising, and that the probability of a crisis in the not-so-distant future (as an example, within the next two years) has grown significantly. The blue line is the forecast of output (elaborated in the next section). If there is no crisis, the long-term level of output at t+2 will be Y*.8 With some probability, p, there will be a crisis with Y dropping to Yc (shown as output dropping to S). So the expected level of Yt+2, will be somewhere between Y* and YC. If p is high, the expected level will be closer to YC, otherwise it will be near Y*.

Figure 3.
Figure 3.

Cost and Benefit in terms of output level—Schematic Representation

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

14. Macroprudential policy that affects an intermediate target, like credit growth, has both costs and benefits. Policy is expected to bring down the probability of crisis from p to p* and the loss given crisis from l to l*, recognizing that it may not eliminate crisis fully but build resilience for the financial sector. At the same time, if the probability of crisis was largely overestimated, then policy is costly by bringing down the level of intermediation and the forecast for output that would have materialized without a crisis. If there is no crisis, output will be lower than Y* by a factor α—reflecting the dampening of activity from harnessing rapid credit growth. The no-policy expected output is at E (taking into account the probability and depth of crisis) and the forecast with policy is at B if the probability and depth of crisis are both reduced; thus, BE is the benefit from policy. The cost of policy is the reduction in the trend from C to O in Figure 3.

15. This paper introduces a simple analytical way of representing cost and benefit of policy. The analysis is carried out comparing the forecast for the output level 2 years from t (in the example above), with and without policy. To present the idea conceptually, the entire future path for output (and other variables) is summarized in one data point. Expected output level at t+2 is denoted by Yt+2 (or Y* in Figure 3) if there is no crisis. Crisis occurs at t+2 or at a future date with probability, p, which depends upon the levels of early warning indicators. If crisis occurs, the output forecast drops by l, which can also depend upon the intermediate targets. If the probability of crisis is nearly certain, 1, then expected output level is (1–l)Yt+2. If crisis probability is very low, say zero, then the forecast is simply Yt+2. Thus, expected output is given by (1) as:

(1)p(1−l)Yt+2+(1−p)Yt+2=(1−pl)Yt+2

When policy is introduced, both p and l can be lower. At the same time, by lowering intermediation levels (for instance, credit growth), the output forecast at t+2 is also lower by some factor α, to (1 – α)Yt+2. This factor depends upon the sensitivity of output to the intermediate targets and other second-round macro-financial linkages that are expected to differ during crisis and non-crisis times. Expected output with policy is, therefore, given by:

(2)p*(1−l*)(1−α)Yt+2+(1−p*)(1−α)Yt+2=(1−p*l*)(1−α)Yt+2

16. For policy to pay off, the ex ante net benefit should be positive. Expected output after policy, (2) should be higher than expected output in the absence of policy, (1). In other words, policy makers should act if

(3)(1−p*l*)(1−α)≥(1−pl)

Another way of representing the above is:

(3.1)1−p*l*1−pl−11−α≥0

This is a simple, static, way of understanding the concept of cost and benefit of policy, expressed in terms of things that can be estimated. As mentioned above, all the terms, p, p*, l, and l*, are functions of the intermediate targets, including credit aggregates. Once the policymaker gathers all the ingredients, it will be relatively easy to compute the cost and the benefits of policy. The first term on the left hand side (LHS) of the expression in (3.1) is the benefits of policy; it increases with the difference between p and p*, and that between l and l*. That is, the benefits rise with policy effectiveness. The second term is the cost; it increases with α that depends upon the sensitivity of output to credit supply and other intermediate targets affected by policy, when the financial sector is healthy.

17. The simple setup does not take account of two unseen ‘costs’—imperfections of the early warning indicators and policy leakages. One way to incorporate the imperfections of the signal is by including a term that represents the ‘noisiness’ of the signals or the noise-to-signal ratios (NSR)—the possibility that the signal is false or the crisis is missed (see IMF, 2011b, for NSR analysis and CGFS, 2012, for including NSR in cost-benefit analysis). But, given the wide range of thresholds for certain indicators (see Table 1 for credit aggregates), and the various models and judgment that policy makers are likely to use to take a decision, it is difficult to know how to assign NSRs to the term in (3). Costs of imperfections in signals could be lower when credit growth is starting to be rapid, simply because the source of the rapid growth might be better identified and appropriate policy applied.9 As the boom progresses, there could be cross-sectoral feedback loops (for instance, going from real estate prices to credit to real estate prices again) that might make the true source of the problem difficult to identify and to pin down by policy. The sensitivity of output to intermediate targets could be higher as the amplification channels start operating at full force, resulting in higher α. Policy leakages, on the other hand, reduce the effectiveness of the policies—reducing the differences between p and p* and l and l*. The issue of leakages is explored in Section IV.

18. The actual practical implementation of this framework will be much more complex. This is because all interactions between output, risk and policy will be dynamic, and so will be the policy criterion (e.g., a discounted sum of output over a certain medium-to-long run horizon). Besides, in this example, a binary response variable (crisis or not) at a fixed moment of time is used to describe the occurrence of a financial crisis. In a more realistic model, this will need to be replaced with a time-varying multivariate probability distribution.10

19. A few insights come out from the simple expression. For instance:

  • Net benefits do not kick in when the probability of crisis is very low, everything else constant. That is, probability is so low that even fully effective policies do not make a significant difference between p and p*.

  • Net benefits are extremely sensitive to how intermediate targets affect the output forecast, the sensitivity of which is given by α.

  • Net benefits rise with policy effectiveness. Policies that lower both the probability and the depth of a crisis, that is both p* and l*, have significantly higher net benefits.

20. Therefore, several empirical building blocks (Figures 1 and 3) are required to compare costs to benefits and these are discussed in more detail in the next section:

  • a. An early warning system that alerts policymakers 2-3 years ahead of a crisis.

  • b. Sensitivity of a forward-looking measure of output, Yt+2, to intermediate targets, a.

  • c. An estimate of the probability of crisis, p, and the loss in level of output in the event of a crisis, l.

  • d. The effectiveness of policy instruments on dampening growth in intermediate targets, and the effect of a reduction in intermediate targets on the probability of crisis, p*, and output loss given a crisis, l*.

III. Analytical Building Blocks: Measurements

A. Baseline—No-Policy Expected Output

21. To examine costs and benefits of macroprudential policies, policymakers have to start from a baseline forecast for output and the intermediate target. The baseline for expected output without policy should incorporate policymakers’ best guess of the future path of economic activity and intermediation, taking into account the risk of a crisis. First, a method is needed for estimating the real-time forecast of activity or the forecast at time t+h periods ahead, taking into account information as of time t. Second, the probability and the cost of the crisis need to be computed. Third, the forecast for activity needs to be adjusted for the risk of a crisis (see Section II, expression 1).11

Real-time forecasts and real-financial sensitivities (α)

22. For the purpose of this paper, the ‘trend’ is defined as the long-horizon forecast of the level of a series. In the previous section, this is the forecast of real activity in the two-year horizon (see Figure 3) as of time t. Since this forecast is based on actual observations or information at time t, it is a real-time estimate of the trend. It is important for the forecast at t+2 to be forward-looking. In this respect, the Beveridge-Nelson (BN) concept of the trend is intuitively appealing: an estimate of where the variable will be in the distant future, especially if there are unpredictable variations in the data (Beveridge and Nelson, 1981; Nelson, 2006). However, the choice of the mechanism to estimate the trend is left to the policymaker.

23. A buildup in financial sector risk affects the forecast of GDP and this effect changes with the state of the banking sector. A bivariate model based on U.S. data for GDP and credit-to-GDP gap is estimated in this paper (Annex 3). The model produces medium-term forecasts for output conditional on the state of the banking sector.12,13 The role of credit for the GDP forecast changes according to the state of the banking sector, and the state is proxied by a systemic financial stress index (Arsov et al, 2013).14

24. Estimates of the bivariate model point to modest direct linkages between credit and GDP forecasts during normal times, but significant linkages during distress. Results show that a 1 percent increase in the credit-to-GDP gap amount to improvements in the GDP forecasts by about 0.2 percent on a 4-6 quarter-ahead horizon if there is no banking sector distress. However, the same 1 percent increase in credit-to-GDP gap will reduce the GDP forecast by about 1 percent (based on the same forecast horizon) in times of large banking distress (Figure 4 and Annex 2). This also means that if policymakers reduce credit growth by 1 percentage point through macroprudential policies, the GDP forecast in the medium term would dip by only about 0.2 percent if there is no crisis, but will improve by 1 percentage point (that is, the output loss will be reduced) if there is indeed a crisis.15 The former is an estimate for the size of α (=0.002 for every percentage point of credit growth), introduced in the last section, for the U.S. economy.16 This finding is consistent with other papers such as Schularick and Taylor (2012) who find that excessive leverage and build-up in credit prior to the crisis results in more severe and prolonged downturn in real economic activity. This is also in line with cross-section estimates of the sensitivity of output cost during a crisis, l, to past credit developments (Annex 4).17

Figure 4.
Figure 4.

United States: Time-varying Sensitivities of Real GDP Forecast to Past Credit Growth

(in percent of 6-quarters-ahead level of GDP)1

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

1See Annex 3.

Probability of a financial crisis (p)

25. Early warning indicators are used to estimate the probability of a financial crisis. Following up on work done in IMF (2011b) and Lund-Jensen (2012), a panel logit model is used to estimate a crisis-probability 2-years ahead using credit growth and asset price growth. While equity price growth was used in IMF (2011b), real house price growth is used in this paper, together with changes in the credit-to-GDP ratio to estimate the probability of banking crisis. The analysis uses the insights that there are “good” and “bad” episodes of credit growth and that policymakers pay particular attention to fast house price growth if credit growth is also high. What started the rapid credit growth might be hard to detect but it is more likely that rapid house price growth together with rapid credit growth could end up badly. A threshold of credit-to-GDP growth of 3 percentage points is used to detect rapid credit growth. This model can itself be used as a benchmark for all countries—given limited occurrences of financial crises in any one country’s history—and is fairly robust to different sub-sample analysis (advanced versus all economies), methods and time-periods (see Annex 5 for details).

26. The probability of crisis can be conveniently located on a “heat map” for various combinations of credit and house price growth (Figure 5). The heat map shows that for credit growth below the 3 percentage point mark, house price growth does not have a marginal impact on p. However, for higher credit growth, every 1 percentage point increase in house prices increases p. For instance, a combination of 5 percentage point credit growth and 20 percent real house price growth leads to p=14 percent; a 6 percentage point credit growth and 25 percent real house price growth pushes p up to 19 percent. See Annex 5 on how to compute these probabilities from the estimated logit model.

Figure 5.
Figure 5.

Probability of Crisis and Early Warning Indicators—A Heatmap

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

Note: See Annex 5 for details on deriving the probability.

Output loss in the event of a banking crisis (l)

27. Experiences in various countries show that the GDP loss during crisis can be substantial. Cross-country data for 10 countries and 12 crisis events show that GDP on average falls about 8 percent per year from the long-term forecast for five years from the beginning of the crisis (Figure 6).18 Although, alternative cost measures (as described in Annex 1 and Annex 3) are not directly comparable, evidence shows that on average these are equivalent to 7 percent annual drop in actual output from some measure of the trend.

Figure 6.
Figure 6.

Crisis cost

(in percent of trend output)

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

Note: Average experience of 10 countries over 12 crisis episodes.

28. The output loss during the crisis is greater the higher is systemic risk prior to the crisis. It has been well documented that output loss following a financial crisis, such as asset price busts and banking crises, could be permanent (Cerra and Saxena, 2008, WEO, 2009) and usually deeper than other recessions (Claessens, Kose and Terrones, 2011a and 2011b) and that financial crises are typically associated with rapid growth in credit, house prices and residential investment (IMF, 2011b, Kannan et al, 2009). Credit growth, as one of the intermediate targets for policy, is therefore used in a cross-section model of 67 crises cases to get a sense of the magnitude of its effect on output cost during systemic banking crises, controlling for currency crises (Annex 4). The model shows that a 1 percentage point higher change in the credit to GDP ratio prior to the crisis is associated with a higher average yearly cost of a financial crisis of 0.6 percent. This is comparable to the estimate for the US in Annex 3 that showed that every percentage point higher gap in credit-GDP ratio results in 1 percent drop in the output forecast during a crisis.

Blending in all ingredients: Deriving a no-policy baseline

29. Summarizing, there are several steps involved in deriving the no-policy expected output. These steps can be illustrated by filling in the blanks in Figure 1 with the models used so far (Figure 7). Blending in all the ingredients—p and l—the baseline expected output without policy can be derived by using expression (1) in Section II(B). This baseline will be conditional on credit growth, shifting down if credit/GDP growth is more rapid.19

Figure 7.
Figure 7.

Deriving a No-Policy Baseline: Some Measurements

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

Figure 8.
Figure 8.

Baseline Expected Output

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

30. The next section looks at the effectiveness of macroprudential policies. Expected output following a policy response would take into account (a) the policy effectiveness on intermediate targets, and relate the intermediate targets to the GDP forecast; and (b) the extent to which lowering the intermediate target lowers the probability and the cost of the crisis, that is p* and l*.

B. Policy Response

31. Different policy instruments affect intermediate targets and build resilience in different ways. This section first provides an overview of how five typically-used tools are expected to affect different parts of banks’ balance sheets and lending rates, and how some tools build resilience by creating buffers in banks. Then, an empirical exercise provides evidence on the effectiveness of these tools on intermediate targets.20

Policy to Intermediate Targets: What to Expect

32. Tools can be classified according to the portion of financial institution’s balance sheets they directly affect. For instance, risk weights, the countercyclical capital buffer (CCB) and dynamic provisions have a direct impact on the economic capital, reserve requirements on the funding structure, and Loan-to-Value (LTV) and Debt-to-Income (DTI) on assets (Table 2). In addition to building buffers and creating resilience against various risks, the tools are expected to affect intermediate targets. Some tools could have immediate effect on credit growth, for instance, either through quantitative restrictions enforced through eligibility criteria or through prices like lending rates. The pricing effects are analyzed using the loan-pricing equation introduced in Elliott and Santos (2012) and described in Box 2. The underlying assumption is that a unit of loan should provide sufficient return to cover cost of equity, debt and deposits backing the loan adjusted for expected losses on the loan and other incidental costs.

Table 2.

Comparison of Policy Tools—Expected Effects on Balance Sheets and Prices

article image
Notes:

Both ‘Stock’ and ‘Flow’ are marked if there is flexibility in applying the policy tool.

House prices could be affected if capital instruments directly target mortgage loans.

Foreign assets could be lower due to higher risk weights on such credit, but foreign liabilities could either go down in line with foreign assets or remain constant.

With tightening LTV and DTI limits, some customers would be ineligible or face a prohibitively high interest rate. The eligible customers could face lower interest rates with the improvement in banks’ loan book quality.

Includes foreign bank branches and nonbank credit institutions.

33. Capital based tools. Risk weights have typically been increased on assets such as mortgage loans fearing property price booms, loans to other overheating sectors, or foreign currency loans to unhedged borrowers. These can be flexibly imposed; for instance, risk weights can be increased on the stock of existing loans or on the flow of new loans. An increase in risk-weighted assets (RWA) makes them more expensive (raising equity is costly) if capital ratios are not allowed to decline. The negative effect on credit growth of higher RWA could be through higher lending rates, as banks recuperate the cost of raising new equity. Countercyclical capital buffers (CCB) could have similar effects except that these buffers may not be as well targeted as the RWA. Some of the loan pricing effects could be offset by investors’ requiring lower return on capital due to the perception of higher resilience in the banking sector due to the higher capital buffers.

34. Reserve requirements (RR). Countries use RRs on various funding instruments to build liquidity buffers against risky sources of funds. The base for the RRs varies from certain deposits to foreign liabilities to wholesale funds, either on the stock or their flows. Since it restricts the funding available to back new loans and requires setting aside assets as reserves, there is less to lend and could have a direct effect on credit growth and loan/deposit ratio. In addition to the quantitative restriction, the increase could lead to higher lending rates as the cost of equity is higher than the cost of debt or deposits.

35. Loan-to-Value and Debt-to-Income Requirements. Both LTV and DTI tools impact the flow of new loans by tightening eligibility criteria and are quantitative tools that could limit mortgage credit growth and house price growth. Lower LTV and DTI imply that new borrowers who are not eligible are excluded from loan markets and eligible ones have to borrow less (put more down-payment). This is equivalent to a direct reduction in the flow of credit. Having more equity at stake, borrowers are discouraged from defaulting on their loan. From a risk management perspective and under the Basel II advanced approach, the reduction in borrowers’ probability of default affects the loss probability distribution of loan portfolios, lowering expected and unexpected losses. The lower expected and unexpected losses then lead to a decline in loan loss provisions—if they are forward-looking—and capital, respectively. A reduction in expected losses implies lower credit spreads and, therefore, lending rates as shown in the loan pricing equation in Box 2. In turn, a reduction in unexpected losses increases the risk-based capital ratios and would initially have no impact on lending rates. However, if banks react to the lower unexpected losses by reducing capital to keep capital ratios constant, the cost of capital funding a loan would fall, further magnifying the negative effect on lending rates arising from expected losses. It is usually expected that the negative quantitative impact on credit growth and house prices is not offset by the lowering of lending rates on eligible borrowers.

36. Dynamic provisioning. This is another tool that discourages credit growth with a pricing effect on lending rates. Dynamic provisioning rules tend to set a buffer—general provisions—against fluctuations in specific provisions for loan losses, which are currently driven by the “incurred loss” approach that recognizes loan impairment when there is objective evidence of incurred losses.21 This leads to fluctuations in provisions and an overstatement of income during the upswings in the economic cycle.22 To offset cyclical fluctuations in specific provisions, general provisions increase during the upswings and decline during the downswings. They are booked as either on the liability side of the balance sheet—associated with a reduction in net income—or part of shareholder’s equity—associated with a reduction in distributed dividends. As a result, general provisions could lead to an increase in the cost of capital and lending rates.

Policy to Intermediate Targets: Empirical Evidence

37. Actual experience with macroprudential policy instruments, featured in the Survey, shows that policy can be effective on intermediate targets. The Survey (IMF, 2011c) and its extension by Lim et al (forthcoming) provides a novel dataset to understand policy effectiveness. The instruments used from the survey are loan-to-value (LTV) limits, debt-to-income (DTI) limits, reserve requirements, provisioning and altering risk-weights on regulatory capital. Unlike previous studies (BCBS, 2010a, Roger and Vlcek, 2011), this paper estimates both the direct and indirect impact of policy instruments on intermediate targets, rather than looking only at the effect on loan-spreads.23, 24

38. The results show that certain policies could successfully bring down indicators related to two externalities. Dynamic panel regressions of the intermediate target on its own lag, some controls and the policy variable are estimated over 2000Q1-2011Q4 for countries that used the macroprudential policy measure in place (see Annex 4 for details and Figure 9 for a summary). The effects of the macroprudential instruments are over and above the effects of the policy interest rate or the lending rate and growth in aggregate demand (proxied by GDP growth).

Figure 9.
Figure 9.

Impact of Tightening of Macroprudential Policy Instruments on Intermediate Targets

(change in percentage points)

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

Note: See Annex 6. RR=Reserve Requirements; Cap req=Capital Requirements; Prov=Dynamic Provisioning; LTV=Loan-to-Value Ratio limits; DTI=Debt-to-Income limits. A step function variable is used for all policy instruments—takes +1 at the time the instrument is tightened, -1 when loosened. The immediate impact is the coefficient for the policy instrument in the regression for the intermediate target; the 2-year cumulative impact is (1+a+a2+a3+a4+a5+a6+a7)*immediate impact, where ‘a’ is the coefficient for the lagged dependent variable.

  • During the period when LTV limits, reserve requirements and risk weights are applied, credit growth and house price growth—intermediate targets related to the “correlated-risk-taking” externality—slow down; provisioning policy is not significant always (probably because very few countries have used it).

  • Moreover, DTI limits are powerful in slowing down loan/deposit growth and the share of foreign liabilities in foreign asset growth—intermediate targets related to “fire-sale” externality.

  • While RR also works on lowering loan-deposit ratio growth, risk weights and LTV have had on average larger effects on foreign asset-liability mismatches.

These results are similar in spirit to those in Lim et al (2011) that show how policies reduce the correlation between credit growth and GDP growth, but go further in showing the direct effects on banking aggregates that could be considered as intermediate targets for combating the two externalities (Figure 9).25

39. Moreover, the policies have prolonged impacts. The dynamic structure of the estimation method allows one to gauge both the short-term and the medium-term impacts of the policy measures on intermediate targets (Figure 9).

  • Taking the correlated-risk taking externality: higher capital requirements, on average, have historically lowered credit growth by about 1 percentage point on impact and by 5 percentage points cumulatively in two years; and lowered house price growth by 1 percentage point in the short-term and by 6 percentage points in two years.

  • For the fire sales externality, tighter DTIs seem to work towards lowering the asset-liability funding mismatches.

  • Capital requirements have cut across both the externalities, and have been effective in three of the four intermediate targets featured here. RRs have been effective in having a prolonged impact on house price growth and loan-deposit growth.

40. During the upswing, both capital and reserve requirements can build buffers but the quality of this buffer matters. This is especially true for capital requirements; capital ratios are generally buoyant during booms due to high profitability (DFR, 2012) but what could be binding for banks (going forward as Basel III takes effect) is the need to raise the quality of the loss-absorbing buffer (IMF, 2012a, BCBS, 2011b). Ensuring higher quality (as mandated in Basel III) together with the quantity would be costlier for banks and have a larger effect on intermediate targets than is shown here, creating greater resilience in the banking system.

Table 3.

Effects of Policy Tightening: An Example

(effects on credit and real house price growth, probability of crisis, loss given crisis and cost on output forecast)

article image

See Figure 5 and Annex 5 for estimates of p and p*, given credit growth and house price growth. See Annex 4 and Figure 8 for l.

See Annex 6 Table 1 for the results on changes in the credit-GDP ratio. See the note under Figure 9 for the calculation of the two-year effects.

See Annex 6 Table 2 for the results on real house price growth. See the note under Figure 9 for the calculation of the two-year effects.

See Annex 4 and Figure 8: Average loss given crisis is 0.08. With slowing credit growth, loss is lowered.

For the United States, one percentage point lower credit growth reduces the output forecast by 0.2 percent. See Annex 3.

See expression 3.1 in the text for the expression on net benefits.

From intermediate targets to lowering probability and depth of crisis: p* and l*

41. Tighter policy affects intermediate targets that, in turn, lower the probability and depth of output loss in case of crisis. Tighter capital requirements, such as through higher risk weights seems to have had a higher medium term impact on credit growth historically than other policies. However, since the policy instrument is a categorical variable in the empirical work (that is, a tightening is denoted by 1, a loosening by -1, see Annex 6), the results need to be interpreted carefully. Table 3 illustrates the effects of policy tightening on credit and house price growth, and their effects on p*, l* and α. It is assumed that the probability of crisis is 14 percent and the loss given crisis is 9.2 percent given credit and house price growth in the baseline. Tightening of capital requirements by raising risk weights, on average across time and countries, would reduce the credit-to-GDP ratio growth by a little more than 1 percentage point on impact and 5 percentage point in two years (see Figure 9). In addition, real house price growth would be reduced by 5.8 percent over two years. The reduction in credit and house price growth would together reduce the probability of crisis to 3 percent and the loss given crisis to 5 percent.

42. In reality, it depends upon how much risk weights usually change compared to other instruments. Also, higher risk weights, which are applied to the whole stock of credit in a certain category, are costly for banks and are more likely to be transferred to customers through higher credit spreads or other forms of tightening of credit conditions (including DTI limits). In contrast, DTIs have an impact on credit growth only if they are binding on the flow of new customers.

43. These responses are likely to be higher in practice. So far, the impacts have been discussed from the point of view of reducing credit growth and house price growth, which are two of the many intermediate targets. In reality, capital requirements and provisions build buffers that increase resilience of the system. They also affect other things like foreign liabilities growth, which together with credit, would have a multiplicative effect on p*. In addition, there could be positive confidence effects on the banking sector that would reduce crisis risk, which is difficult to capture in the same framework.

44. Policy costs depend upon the sensitivity of the output forecast to intermediate targets, had there been no crisis. If risks were mis-measured and crisis–risks were actually zero, then the tightened policy lowers the GDP forecast level only modestly for the United States, compared to the no-crisis, no-policy level. For instance, continuing the example in Table 3, if a crisis did not occur, the GDP forecast would be reduced by (α) 1 percent, a relatively high number since capital requirements have a large effect on intermediation.

45. Summarizing, the various ingredients can now be mixed together to get the net benefits of policy. The various components are summarized in Figure 10, filling in the missing pieces. These pieces can be used to evaluate the net benefits of policy using the concept in Section II as guidance. For instance, for the example in Table 3 (risk weights), together with the other ingredients, the net benefit of policy is still positive. Of course, these estimates need to be mixed with a lot of judgment as was noted earlier. Depending on the structure of the financial sector, policy leakages are also important considerations, which are analyzed in the next section.

Figure 10.
Figure 10.

Summary of Measurements

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

IV. Leakages

46. Macroprudential policies could have unintended side-effects, besides the intended costs and benefits, through leakages. Where incentives for risk-taking still exists, market players other than the institutions on which policies are imposed, will attempt to make full use of it. Leakages exist when there is a disconnect between the supervision and regulation of financial entities and the actual activities and the forms of intermediation of these entities. When policies clamp down on banking activities, other institutions pick up the slack. When policies restrict local operations, either the entities spread risks elsewhere or crossborder and other nonbank intermediation substitutes for the restricted or regulated activity.

47. There could be other unintended consequences of macroprudential policies. First, migration of activities to the shadow banking sector make it difficult to identify systemic risk (see Boxes 1, 3 and 4). For instance, the extent of leverage can be hidden by shadow banking activities, or could be vulnerable to long intermediation chains. Second, application of tools can lead to structural distortions in the financial system that at first sight might seem innocuous but over time could have systemic implications. For example, there are some early signs of higher securitization rates and greater reliance on wholesale funding to cover the higher costs of Basel III capital rules (IMF, 2012b). Third, the policies applied in one country to address its systemic risk can have unintended consequences of risk-shifting to other countries.

48. The focus here will be on new findings related to crossborder leakages and the shadow banking system. Crossborder arbitrage through direct crossborder lending and regulatory arbitrage through the part of the domestic financial sector falling outside the banking regulatory perimeter are two core channels of leakages. The growing literature on shadow banks in advanced economies (see Claessens et al, 2012 for a survey) and opportunities of direct crossborder credit in emerging economies (Enoch and Ötker-Robe, 2007) create good starting points for analysis. There can be many other avenues for leakages (see CGFS, 2012, Santos and Elliott, 2012, BOE, 2011), but these are beyond the scope of this paper due to data limitations.

  • In many countries, especially in emerging Europe, there were large crossborder flows into banks and nonbanks. The flows into banks were mostly from foreign parent banks into their subsidiaries, which took advantage of the growth opportunities and competed for market share with other foreign banks. Tightening of local regulatory policies often led the parent banks to lend directly to the private sector. As in Croatia (Box 3), direct crossborder lending growth accelerated with tightening of local macroprudential policies.

  • The other source of leakage considered in this paper is the part of the financial sector that compete with the regulated banking sector but are subject to home country regulatory policies only, such as foreign bank branches in the U.K. (Box 4).

  • The third kind of leakage is through nonbank financial institutions within the domestic financial sector that are either unregulated or not given the same regulatory treatment as banks. The latter case is illustrated by the example of the United States (Box 1).

Direct Crossborder Loans

49. Empirical findings suggest that certain macroprudential instruments are more prone to leakages from crossborder loans than others. Re-estimating the panel data model on policy instruments, using data on direct crossborder loans to the private sector as the dependent variable, reserve requirements and provisioning policies tend to incentivize foreign banks to directly lend to the private sector, bypassing the subsidiaries (Annex 6).26 The evidence suggests that a tightening of reserve requirements reduces credit growth by 0.4-0.5 of a percentage point on impact but increases leakages through higher crossborder lending growth by 1.2 percentage points. Over two years, credit growth decreases by 2.5 percentage points, and direct crossborder loan grows by about 5 percentage points. Local bank provisioning policies, as was shown earlier, did not have significant impacts on credit aggregates, but does attract direct crossborder flows.

50. There could be multiple reasons for RRs to increase leakages. Reserve requirements are imposed on banks’ liabilities. If these liabilities are foreign, coming from banks’ foreign parents, additional requirements are easily circumvented by providing direct crossborder loans to the private sector instead of channeling funds through branches. RR would still have an effect (the ‘stock effect’) on credit growth from the domestic banking system, since banks are squeezed on the amount of liabilities that are available for lending. Capital requirements like higher risk weights, on the other hand, still require additional capital injections by parents or lower profit distributions that tightens the parents’ belts—the foreign parents can only direct additional funds through capital injections into subsidiaries, which are not substitutable for direct crossborder lending.

51. LTV and DTI limits, on the other hand, are part of eligibility criteria and influence loan demand. If binding, these limits help to decrease demand for certain loans, like mortgage credit. If parent banks also perceive these limits as hard limits for their own risk management, there should be less crossborder substitution of credit due to domestic regulation. Another argument for the empirical evidence is that LTV and DTI are, in most cases, applied to mortgage loans to households. Hence, it is difficult to substitute (for households) domestic loans with a foreign loan because usually households do not have an access to a foreign market and a foreign parent bank does not have a mechanism to take over the house in case of default.

52. Therefore, the findings suggest that in financially open economies, capital requirements and LTV tools could be better suited to limit buildup in credit-related risks than reserve requirements. This is especially true in economies with high foreign-ownership of the banking system and where foreign parents are willing to work with the local subsidiaries in order to get information on local clients. Having said that, there are many different kinds of RRs that the empirical work lumps into one: RRs based on the foreign currency component of liabilities, maintenance of the reserves in local versus foreign currency, marginal versus average reserve requirements, etc. Thus, some RRs might be more prone to leakages than others, but such distinction are not individually testable given current data limitations and are beyond the scope of this paper.

Foreign Branches and Nonbank Financial Institutions

53. The other leakage is related to the presence of financial institutions that are not regulated the same way as banks but offer banking services. As regulations on banks are tightened, other credit intermediaries like branches of foreign banks (that are regulated by the home country as in the United Kingdom), leasing and factoring companies and other finance companies, and investment banks (as in the United States in Box 1) take on commercial bank activities.

54. The foreign bank branches in the U.K. provide a relevant case. Empirical work shows that when capital requirements are tightened for the U.K. banks, loan supply significantly diminishes. However, about a third of the effect is offset by foreign branches in the U.K. (Box 4). Thus, any tightening in capital requirements would have to be coordinated with the home supervisor of the foreign branches, so that they tighten capital requirements for U.K. exposures as well.

55. The significant nonbank activity in the United States provides another case. In most countries, the bulk of consumer lending is done by banks. However, the United States stands out as having a large share of intermediation occurring in nonbanks or shadow banks (Box 1). As was shown earlier, measures of ‘excesses’ based on bank credit in the United States failed to signal a crisis when the two thresholds for credit growth (IMF, 2011b and Dell’Ariccia et al, 2012) were used. Using consumer credit growth from all sources of intermediation breaches atleast one threshold early in the 2000s and would have signaled a crisis. Like the case of bank branches in the United Kingdom, nonbanks in the United States are likely to circumvent macroprudential policies, if these policies are aimed just at banks.

56. In such cases, LTV and DTI limits together with capital requirements could work better. Borrower eligibility criteria can be applied to all products (rather than institutions) that are offered by any financial institution within a country. In the U.K. where there is evidence of leakage of capital requirements, LTV and DTI limits could be applied even to foreign branches (BOE, 2011). That way, at least the ‘flow’ part of the credit supply is controlled. The ‘stock’ of credit that was supplied with high LTVs, however, remains unprotected. One way to take care of the stock problem is to have higher risk weights on high-LTV loans made in the past—but to be effective, foreign parents of the branches would also need to follow the rules set by the host supervisor and regulator. For nonbank activity, close coordination with other regulators would be needed.

57. Summarizing, there are three sets of policy implications to counter leakages, depending upon the financial structure of the economy.

  • Financially open economies, especially with foreign subsidiaries: capital tools and LTV could work well for containing the “correlated risks” externality. RRs are subject to leakages through substitution of direct crossborder lending for domestic lending. Since both DTIs and RRs are usually effective at containing funding risks (loan/deposit growth)—one of the targets against the “fire sales” externality—DTI limits could be used instead or with RRs.

  • Financial systems with foreign bank branches: A combination of LTV and DTI could work well in containing risks from both externalities as they apply to the activity not the institution. Capital tools could be problematic for universal use. In this case, RRs could also help if these can be applied to branches.

  • Financial systems with a large share of nonbanks: As with branches, combinations of LTV and DTI tools could work, in close coordination with other regulators. For consumer credit, the macroprudential supervisor could coordinate with the consumer protection agency if one exists. Both capital tools and RRs would be difficult to impose in such financial systems.

58. In general, a number of standard policy prescriptions apply to plug evasions and circumventions. Regulating financial services, not sectors; strengthening (macroprudential) regulation of the (shadow) banking sector (FSB, 2013); coordinating macroprudential tools and ensuring their consistent application across borders including through reciprocity principles; monitoring arbitrage opportunities on a regular basis (CGFS, 2012) and dealing with data gaps are all potentially valuable methods of reducing systemic risks.

V. Summary and Policy Implications

59. This paper has provided a recipe for estimating cost, benefits and unintended consequences of macroprudential policies. The idea is to provide guidance to policymakers and new findings on the various components of the decision making process. In this respect, the paper provides a conceptual framework for analyzing costs and benefits. It takes forward existing studies and uses them in order to come up with simple rules-of-thumb based on additional empirical work. It provides new evidence on the effectiveness of different policy instruments using a database from a survey and provides new findings on the unintended consequences of macroprudential policies.

60. The work remains under construction. While many of the ingredients are provided in this paper, there remains work to be done on the core analytical model that blends together all the components. While the structure of the model is clear, its estimation remains complicated and subject to robustness tests. So far, the paper takes an incremental approach to integrating the various ingredients in the cost-benefit framework. The core model, which is left to another paper, would be expected to estimate the output forecast taking into account a dynamic interaction between output, risk and policy.

61. Summarizing, for macroprudential policy to work, the ex ante benefits should be higher than the cost. Even though this is work in progress, there are useful insights already resulting from the analysis. Net benefits would depend upon several things.

  • Most important is the performance of the early warning models. If a dependable set of indicators is not used, mis-use of macroprudential policies could be costly. As IMF (2012a) showed, there are adverse growth outcomes if financial buffers are too high, especially for emerging economies. What matters is the quality of the buffers.

  • Given the widespread role of credit in early warning, policy makers could use two thresholds provided in IMF (2011b) and Dell’Ariccia et al (2012) respectively, escalating the concern and implementing corrective policies somewhere in between the two thresholds (Table 1, section II, Annex 2). That way, policymakers use judgment together with analytical results in their policy decision. This gradual approach could also influence expectations of the private sector that could mitigate the necessity of drastic and severe policy responses.

  • Policymakers need proper measures of the intermediate target: not just bank credit but all sources of credit. The latter data would feed into the performance of the early warning models and indicators. As is shown for the U.S. case in Box 1, the out-of-sample probability of crisis in the 2000s is much higher when all sources of consumer credit are used rather than just bank credit.

  • Net benefits do not kick in when the probability of crisis is so low that even fully effective policies do not make a significant difference between pre- and post-policy probabilities of crisis (section II). In this regard, the probability of crisis increases non-linearly when, for instance, both house prices and credit are growing rapidly, especially when credit-to-GDP ratio is growing by 3 percentage points of GDP or more (Figure 5).

  • Net benefits are very sensitive to how intermediate targets affect the output forecast. If macro-financial linkages are high, output is highly sensitivity to intermediation. In such cases, policy mistakes, especially when policy over-reacts to signals (“false positives”), could be very costly. However, so far, these effects have been estimated to be very modest for the Untied States (Annex 3). If policy measures are taken and there is no crisis, the output forecast drops by 0.2 percent for every 1 percentage point tightening of credit growth.

  • Net benefits rise with policy effectiveness as long as the costs are contained For benefits to be high, the difference between pre- and post-policy probabilities of crisis (p and p*) and the pre- and post-policy loss given crisis (l and l*) need to be large (Section II).

  • The measures that are most effective and have prolonged effects on credit growth and house price growth are reserve requirements, higher risk weights on capital and LTV limits. Provisioning policies have not been tried in many countries, but where data is available, these did not have significant effects. By influencing both asset prices and credit growth, the three instruments could lower the probability of crisis and its depth substantially to increase the benefits of policy. Historically, the use of LTVs seems to have less bite than capital requirements and RRs (Section III, Annex 6).

  • Since there is evidence of leakages or unintended consequences of some policies, policymakers should pay attention to the financial structure of the economy and use the most effective set of policy instruments. For instance, reserve requirements could be subject to leakages through direct crossborder credit to the private sector especially in financially open economies, and so capital tools, LTV and DTI could be better suited. On the other hand, in economies with an active set of foreign bank branches, capital requirements could be easily circumvented, and so RRs could be better suited if these can be applied to at least broad money liabilities for foreign branches. Similarly, LTV and DTI tools could be applied across banks and nonbanks to plug leakages through nonbanks (Section IV), since they are focused on activities, not specific types of institutions.

62. A very basic recipe is provided in the paper. Policymakers are encouraged to add flavors that are country-specific and garnishes that reflect their preferences.

Nonbank Lenders in the United States: Possibilities for Underestimating Systemic Risk

There is a significant presence of nonbank lenders in the United States. The FSB estimates a growing contribution of shadow banks in the US financial sector assets, especially during 2005-2007, the run up to the 2008-2009 financial crisis (figure below, left). Shadow banks include nonbank lenders of consumer and mortgage credit to households that fall outside the regulatory regimes for commercial banks.

uA01fig02

Total Assets

(in USD trillions)

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

Source: Financial Stability Board
uA01fig01

Change in Credit-GDP Ratio: Banks versus All Sources of Credit in the United States

(annual, in percentage points)

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

Sources: IMF International Financial Statistics; Federal Reserve Board; IMF staff calculations.Note: The data on consumer credit is taken from the Federal Reserve Board’s all sources of consumer credit (G.19) and mortgage debt held by households (Flow of Funds, Z.1).

Credit aggregates used for the analysis of systemic risk should be inclusive of these nonbank lenders.

Data (figure above, right) shows that the growth in consumer credit from all sources far outpaced the growth in commercial bank credit before the advent of both the episodes of systemic banking crises in the US—1988 and 2008 (Laeven and Valencia, 2010). Moreover, credit from commercial banks could have been underestimated due to loans sold to the Government Sponsored Enterprises (conforming mortgages) and to investment banks (nonconforming mortgages). These loans, however, are captured in the flow of funds data used to estimate consumer (including mortgage) credit growth in the United States.

Crisis signals are underestimated if credit aggregates are not inclusive of all lenders. When policy makers rely on credit aggregates for a crisis signal (see Section II), credit growth from commercial banks alone do not breach either of the two thresholds used here—GFSR and Boom (Annex 2). However, the broader credit measure would breach the GFSR threshold in the period 2001-2006. Moreover, the panel logit model (Annex 5) delivers vastly different (fitted) probabilities for the two sources of credit (figure below).

uA01fig03

US - Probability of a financial crisis in two years

(in percent)

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

Source: IMF Staff estimates, see Annex 5 for methodology.
Prepared by Nicolas Arregui and Srobona Mitra.

Effects of Macroprudential Policies on Lending Rates: What to Expect

The effects on lending rates of macroprudential policies can be determined using a loan pricing equation. Any lending decision can be reduced to a pricing equation to determine whether a loan provides sufficient return:

rL*(1−t)=E*re+(D*rd+C+A−O)*(1−t),

where:

  • rL = effective interest rate on the loan, including the annualized effect of fees;

  • t = marginal tax rate for the bank;

  • E = proportion of equity backing the loan;

  • re = required rate of return on the marginal equity;

  • D = proportion of debt and deposits funding the loan, assumed to be the amount of the loan minus E;

  • rd = effective marginal interest rate on D, including indirect costs of raising funds, such as from running a branch network;

  • C = the credit spread, equal to the probability-weighted expected loss;

  • A = administrative and other expenses related to the loan;

  • O = other income and expense items related to the loan.

In simple terms, the rate on the loan needs to cover the cost of capital and other funding sources, any expected credit losses, and administrative expenses. Similar to other regulatory reforms, macroprudential policies also affect the cost of capital, other funding, and lending rates by changing the proportion of debt and equity funding a loan. However, they may also trigger offsetting adjustments to one or more variables in the loan pricing equation that reduce the impact of macroprudential policies on lending rates. In the case of a full offset where return on capital is lower as predicted by the Modigliani-Miller proposition, countercyclical capital would have no effect on lending rates. Given the empirical evidence in the next section that macroprudential policies have an effect on reducing systemic risk in the short run and that offsetting effects in the loan pricing equation may be of a long-term nature, this paper assumes that macroprudential policies have a full impact on lending rates in the short term.

The simple loan pricing equation above is an important building block in most studies described earlier, as it allows researchers to translate costs arising from regulatory reforms into loan rate increases and GDP changes.2 Among others, Elliott (2009 and 2010a), BCBS (2010a, 2010b), Slovik and Cournede (2011), and IIF (2011) use variations of this loan pricing equation (also named accounting-based approach) as it is flexible enough to capture many different attributes of credit provision. Individual country studies that build on the pricing equation above include Koop et al. (2010), Schanz et al. (2011), de-Ramon et al. (2012), for instance. The loan pricing equation is especially useful in economic and econometric models with no explicit behavior for bank capital or liquidity. The pricing equation above is also useful to analyze the interaction between capital and liquidity standards and macroprudential policies. Table 2 in the text summarizes the main effects of macroprudential policies on credit and lending rates.

Prepared by Andre O. Santos.1 See also Elliott and Santos (2012) and King (2010) for a similar approach.2 This is especially true in economic and econometric models that feature neither bank capital nor bank liquidity.

Leakages through Direct Crossborder Lending: The Croatian Experience

Croatian experience with macroprudential measures over the years 2003–2008 shows how leakages make macroprudential policy less effective. The measures introduced by the authorities aimed at slowing credit growth (the credit growth cap, the marginal reserve requirement), reducing capital inflows (the marginal reserve requirement), improving bank capitalization (higher risk weight), and bolstering bank liquidity (the foreign currency liquidity requirement).1 However, most measures were accompanied by some sort of a leakage:

  • The 2003 credit growth caps: banks cut back on holdings of securities and on unused lines of credit (both items were included in the credit aggregate) and used this to offset higher growth in other categories; banks with affiliated leasing companies encouraged clients to take leases rather than loans; banks with parent banks abroad referred clients to the parent bank, with all of the screening and paperwork carried out in Croatia, but the final contracts and funding, at least formally, coming from the headquarters outside of Croatia.2

  • The 2007 version of the credit growth cap: the policy closed the line of credit loophole by limiting on-and off-balance sheet items separately and dealt with the leasing problems by capturing funding of the leasing company within the credit limit. However the issue of direct parent bank lending to clients was not resolved.

  • The marginal reserve requirement: banks attempted to fund its expansion plans by issuing bonds on the domestic market. As a response, the central bank implemented a 55 percent reserve requirement on funds raised by bond offerings.

  • The foreign currency liquidity requirement: banks tried to evade the requirement by offering local currency deposits indexed to an exchange rate. The central bank altered its rules to include FX indexed deposits, which display the same currency risk as foreign exchange deposits, in calculation of the requirement.

uA01fig04

Cred it Growth and Caps on Credit Growth

(Year-on-year percent change in nominal credit)

Citation: IMF Working Papers 2013, 167; 10.5089/9781484335727.001.A001

Note: MRR refers to Marginal Reserve Requirements on foreign liabilities. Caps on credit growth were in place 2003-2004 and 2007-2009.
Prepared by Ivo Krznar.1 See Appendix 3 and 4 in Galac (2010) for a full description of the measures and dates of their implementation.2Galac (2010) finds that the credit growth cap did slow domestic lending by Croatian banks, but that this seems to have been offset by leasing and foreign borrowing, so that the measure’s effect on total private sector debt turned out to be insignificant.

Leakages through Foreign Branches: Empirical Evidence from the UK

A unique UK database permits an evaluation of the effectiveness of macroprudential policies, showing how leakages can occur through foreign branches. Recent empirical evidence based on the database suggests that such “leakages” are substantial but fall short of a full offset.

In the 1990s and early 2000s, the Financial Services Authority (FSA) imposed time-varying minimum capital requirements—so called “trigger ratios”—at the level of individual banks. These trigger ratios were set for all banks under the FSA’s jurisdiction, i.e. all UK-owned banks and resident subsidiaries of foreign banks. The discretionary regime was intended to fill gaps in the early Basel I regime, which simply imposed a uniform minimum capital requirement of 8 percent of risk-weighted assets. Trigger ratios for each bank were reviewed every 18-36 months, and changes made if deemed necessary (incorporating judgments about, among other things, evolving market conditions as well as the quality of risk management and banks’ systems and controls). Importantly trigger ratios were not imposed on an important class of UK-resident banks: branches of foreign banks, which were subject to home rather than host country regulation.

Aiyar, Calomiris and Wieladek (2012) exploit a unique UK database to explore macroprudential leakages. The study collects quarterly data on these regulatory minimum capital requirements from the FSA, and merges it with loan data reported by individual banks to the Bank of England. The institutional set-up allows the study to ask two questions, both of which are central to the effectiveness of macroprudential policies: (i) do changes in minimum capital requirements affect loan supply by regulated banks; and (ii) is the loan supply response by regulated banks offset, partially or in full, by an opposing supply response by unregulated entities (in this case, foreign branches)?

The authors find, first, that loan supply by regulated banks responds strongly to changes in minimum capital requirements. Changes in individual bank lending to the real economy are regressed on several lags of changes in the trigger ratio. Control variables include GDP growth and a number of bank-specific balance sheet characteristics. Data on bank-specific sectoral lending patterns are used to control for demand shocks, as in Aiyar (2011). A rise in the trigger ratio of 100 basis points is estimated to induce a cumulative reduction in the growth rate of bank lending of between 6 and 8 percentage points.

Second, the study finds a robust offsetting response by domestically-unregulated foreign branches. Changes in lending by a foreign branch are regressed on several lags of the change in lending by a reference group of regulated banks. For each foreign branch, the reference group of regulated banks comprises banks that specialize in lending to the same sectors of the economy as the branch; thus the reference group captures the relevant set of competitor banks. The average branch increases lending by about 2-3 percent in response to a regulation-induced decline in lending by its reference group of 1 percent.

A UK economy-wide assessment of leakages needs to take into account that (i) foreign branches outnumber regulated banks; and (ii) the average foreign branch is much smaller than the average regulated bank. Accounting for these factors yields an estimate of aggregate leakages of about 30 percent. The fact that the offset is only partial implies that, on balance, changes in capital requirements can induce a substantial impact on aggregate credit supply. But the results also affirm the importance of cross-country cooperation on macroprudential policies, such as the reciprocity principle enshrined in the Basel III countercyclical capital buffer.

The estimate of leakages provided by the paper may be considered a lower bound, given that only one possible source of substitute credit—lending by unregulated foreign branches—is examined. But the authors argue that, in fact, this is likely to be the most potent channel of leakages. The theoretical and empirical finance literature suggests that securities offerings are not perfect substitutes for loans from intermediaries. Loans involve much more detailed contracting terms than bonds, and require monitoring and enforcement after the loan is made. Furthermore, the importance of “soft” information for limiting the screening, monitoring and enforcement costs of bank lending implies that there are limits to the ability of offshore lending to substitute for local intermediation (e.g., Aggarwal and Hauswald (2010)).

Prepared by Shekhar Aiyar.

Annex 1. Other Studies on Costs and Benefits27

Table A1.1.

Literature Review: Measuring Costs and Benefits of Macroprudential Regulation

article image
article image
article image
Notes: Additionally, see ICFR 2011 for a brief summary and comparative study of the literature.

Same paper was published by the Banque de France.

Annex 2. Signals Based on Two Different Thresholds for Bank Credit Aggregates28, 29

International experience shows that credit growth can be a powerful predictor of financial crises so monitoring “excessive” credit growth should be the starting point for assessing the build-up of systemic risks. Credit aggregates could guide policy, especially to inform a tightening of policy by using appropriate thresholds. Two studies that examine thresholds are IMF (2011b) in the Global Financial Stability Report (GFSR) and Dell’Ariccia et al (2012). Comparing the GFSR threshold (3-5 percentage point annual change in the credit-GDP ratio) with the Dell’Ariccia et al threshold for defining credit booms (henceforth, Boom) shows that the GFSR threshold is triggered much more often than Boom. Relatedly, the GFSR threshold flashes (signaling a crisis coming within two years) earlier than the Boom in most cases.

Table A2.1 allows for a visual comparison of these two methodologies to flag risks. Green and yellow cells show periods in which only one of the methodologies flags risks (green for GFSR, yellow for Dell’Ariccia and others). Red cells indicate periods in which both methodologies flag risks. The table is only shown for one decade. The high preponderance of green cells shows that the GFSR methodology flags risks more often. Moreover, if the Dell’Ariccia and others methodology is flagging risks, it is likely that the GFSR is flagging risks too, but not vice versa.

Table A2.1.

Signals based on Two Different Methodologies

article image
article image
article image
article image
article image
Note: See Box 1 for more on the U.S. case.