Download Procyclicality of the financial system and financial stability and more Lecture notes Financial Management in PDF only on Docsity!
Procyclicality of the financial system and financial stability:
issues and policy options
Claudio Borio, Craig Furfine and Philip Lowe 1
1. Introduction
In recent decades, developments in the financial sector have played a major role in shaping macroeconomic outcomes in a wide range of countries. Financial developments have reinforced the momentum of underlying economic cycles, and in some cases have led to extreme swings in economic activity and a complete breakdown in the normal linkages between savers and investors. These experiences have led to concerns that the financial system is excessively procyclical, unnecessarily amplifying swings in the real economy. In turn, these concerns have prompted calls for changes in prudential regulation, accounting standards, risk measurement practices and the conduct of monetary policy in an attempt to enhance both financial system and macroeconomic stability.
In this paper, we examine these concerns and discuss possible options for policy responses. It is not our intention to formally model the complex interactions between the financial system, the macroeconomy and economic policy. Rather, we have the more modest goal of stimulating discussion on some of the key linkages between developments in the financial system and the business cycle. Our main focus is on the intrinsically difficult issues of how risk moves over the course of a business cycle and on how policymakers might respond to reduce the risk of financial instability, and attendant macroeconomic costs, that can arise from the financial system’s procyclicality.
A common explanation for the procyclicality of the financial system has its roots in information asymmetries between borrowers and lenders. When economic conditions are depressed and collateral values are low, information asymmetries can mean that even borrowers with profitable projects find it difficult to obtain funding. When economic conditions improve and collateral values rise, these firms are able to gain access to external finance and this adds to the economic stimulus. This explanation of economic and financial cycles is often known as the “financial accelerator”.^2
While the financial accelerator presumably plays a role in all business cycles, it is not sufficient to generate the widespread financial instability that periodically leads to very large swings in economic activity. In this paper, we argue that an additional material source of financial procyclicality is the inappropriate responses by financial market participants to changes in risk over time. 3 These inappropriate responses are caused mainly by difficulties in measuring the time dimension of risk, but they also derive from market participants having incentives to react to risk, even if correctly measured, in ways that are socially suboptimal.
The measurement difficulties often lead to risk being underestimated in booms and overestimated in recessions. In a boom, this contributes to excessively rapid credit growth, to inflated collateral values, to artificially low lending spreads, and to financial institutions holding relatively low capital and provisions. In recessions, when risk and loan defaults are assessed to be high, the reverse tends to be the case. In some, although not all, business cycles these financial developments are powerful amplifying factors, playing perhaps the major role in extending the boom and increasing the severity and length of the downturn. We argue that the worst excesses of these financial cycles could be
(^1) We would like to thank Joe Bisignano, Bill Coen, Renato Filosa, Stefan Gerlach, Bengt Mettinger, Bill White and, in
particular, Kostas Tsatsaronis for their helpful comments. We are also grateful to the central banks that provided us with data and comments. Thanks are also due to Philippe Hainaut and Marc Klau for valuable research assistance. The views expressed are those of the authors and do not necessarily reflect those of the BIS. (^2) It has a long history, reaching back at least to Fisher (1933), and has recently been subject to extensive theoretical
modelling by, amongst others, Bernanke and Gertler (1995) and Kiyotaki and Moore (1997). For a recent survey, see Bernanke et al (1999). (^3) In recent times, although from a somewhat different perspective, the role of financial excesses has been stressed by,
amongst others, Kindleberger (1996) and (1995) and Minsky (1982).
mitigated by increased recognition of the build up of risk in economic booms and the recognition that the materialisation of bad loans in recessions need not imply an increase in risk.^4
These measurement biases, which we argue go hand in hand with economic agents being better at measuring relative than absolute risk, can arise from a variety of sources. One such source is difficulties in forecasting overall economic activity and the link with credit losses; difficulties in assessing how correlations of credit losses across borrowers and, more generally, across institutions in the financial system change over time are part and parcel of the same problem. This tends to contribute to excessively short horizons and to an extrapolation of current conditions into the future. The short-term focus is also encouraged by incentive structures that reward short-term performance, and by certain aspects of accounting and regulatory arrangements. We argue that good risk management requires both a horizon for measuring risk that is longer than one year – the typical industry practice – and a consideration of system-wide developments. Not only would such an approach contribute to the soundness of individual institutions, it would also reduce the financial amplification of economic cycles.
Looking forward, proposed changes to the way in which bank capital is regulated are likely to increase the importance of accurately measuring changes in the absolute level of risk. The proposed changes are primarily designed to rectify current problems with relative capital charges. They represent a major step forward in aligning regulatory capital charges with the relative riskiness of banks’ credit exposures (eg public sector versus private sector, high- versus low-risk corporates). As such, they significantly strengthen the soundness of individual institutions. At the same time, the proposed changes will naturally result in capital requirements on a given portfolio changing over time, as the assessed risk of that portfolio evolves. If risk is measured accurately, this has the potential to further enhance banks’ soundness and reduce the procyclicality of the financial system. However, exploiting this additional potential arguably calls for improvements in current risk measurement practices and/or greater reliance on the supervisory review process. The New Basel Capital Accord, which proposes a strengthening of the supervisory review process, could provide a sounder basis for such increased reliance.
To the extent that procyclicality stems from inappropriate responses by financial system participants to changes in risk over time, we argue that there is a case for a public policy response. Four types of responses are possible. The first is the promotion of a better understanding of risk , through the publication of risk assessments by the authorities or through supervisory reviews of risk management practices. The second is the establishment of supervisory rules that, while not explicitly contingent on the cycle, promote better measurement of the time dimension of risk and make the financial system more robust to misperceptions of risk. Examples of such rules include requiring longer horizons for risk measurement, the use of stress testing and forward-looking provisioning. The third is the use of supervisory instruments in an explicitly countercyclical fashion in an effort to limit the development and consequences of serious financial imbalances. The common element of the two supervisory responses is that they directly or indirectly encourage the building-up of a protective cushion in good times that can be drawn down in bad times. The fourth response is to use monetary policy in an effort to contain the development of financial imbalances. We see scope for the application of all four types of policies, although we argue that discretionary countercyclical adjustments in either supervisory instruments or monetary policy aimed directly at addressing financial imbalances should only occur in those cycles in which financial overextension is playing a pre-eminent role.
Throughout the paper, we stress the endogeneity of the business cycle with respect to the collective decisions of financial institutions. In other words, misperceptions of the evolution of risk over time and inappropriate responses to it, as reflected in lending and financial investment decisions, serve materially to amplify economic fluctuations. At the same time, it is important to emphasise that the appropriateness of the policy options we discuss does not hinge on this premise. These options would still apply even if the course of the economy were completely unaffected by decisions in the financial sphere. Underestimating the risk of a downturn in economic activity and its impact on credit losses, as reflected in financial institutions’ lending, provisioning and capital decisions, can be sufficient to generate financial instability. Policies designed to limit instability need to take this into account. This is true regardless of whether that instability takes the form of financial distress at individual institutions,
(^4) See also Crockett (2000a) and (2000b) and Borio and Crockett (2000). This perspective is also emphasised in Kent and
D’Arcy (2001).
To illustrate the two concepts, consider two banks of the same size. Bank A makes 100 loans for $1 million each, while Bank B makes only one loan, but for $100 million. Suppose that all loans have a 5% chance of default with no recovery in case of default, and that the correlation between defaults is zero. Both portfolios have expected credit losses of $5 million, and therefore the two banks would be viewed as equally risky using the first concept of risk. However, using the second concept, Bank B is clearly more risky. It has a 5% chance of losing the entire $100 million, while Bank A has virtually no chance (precisely (0.05) 100 ) of losing this amount. By virtue of Bank A’s diversification, there is relatively little uncertainty about its future returns.
In the course of the paper, we use the term “risk” to refer to both the level of expected losses and the potential for large unexpected losses. When the analysis refers specifically to one or the other concept, we make this distinction clear.
Whether measuring the value of expected losses or the potential for large unexpected losses, it is important to distinguish between two dimensions of risk: relative and absolute risk.
Relative risk relates to the risk, in a cross section, of a particular financial instrument, portfolio or institution. This is the dimension involved in statements such as “Bond A is riskier than Bond B” or “Institution X is more risky than Institution Y”.
Absolute risk relates to the specific value that the measure of risk takes at a particular point in time. Much of the paper relates to how the level of absolute risk varies over time. In what follows, we refer to this as the time dimension of (absolute) risk. For example, the statement “Portfolio X is more risky today than it was last year” concerns this time dimension of risk.
Focusing now on absolute risk, it is useful to distinguish between the risks of portfolios, institutions and groups of institutions (or “the system as a whole”). A bank, for instance, will be concerned with the credit risk associated with its portfolio of loans which, through the capital cushion, will map into the risk of default of the institution. This is also the risk with respect to which regulatory capital requirements are set. Crucially, the risk in the portfolio will depend on the correlation of the risk of default of the bank’s counterparties. In much the same way, the risk of a group of institutions, or the system as a whole, will depend not just on the risk of the component institutions but, importantly, on the correlation between the risk of the individual institutions.
To bring out more clearly the relationship between the risk of individual institutions and that of the system as a whole, one can think of the financial system as a portfolio of securities, with each institution representing a security. The overall risk of the portfolio is not just the sum of the risk of individual institutions but depends fundamentally on the correlation between them. This distinction will play a key role in much of what follows. For instance, if capital requirements were to be set with a view to limiting the risk of the system as a whole, they might look rather different than if they simply focused on the risk of each institution separately, as is currently the case (see Section 5). 6
Pursuing further the analogy with the portfolio of securities, the risk for each of these two types of “portfolios” can in turn be broken down into two components, the systematic and non-systematic (or idiosyncratic ) component (Box 1). 7 The systematic component is, by definition, the one associated with the correlation between the component securities. 8 Conceptually, it can be thought of as arising from exposures to common factors, such as specific industries or the business cycle. There can, of course, be several such factors. In what follows, we will focus primarily, and somewhat loosely, on the systematic risk arising from exposure to the common factor associated with the business and financial
(^6) These points are elaborated in Crockett (2000b), where the supervisory perspectives that focus on individual institutions and
the system as a whole are referred to as, respectively, “microprudential” and “macroprudential”. (^7) A useful exposition of these concepts as applied in portfolio theory can be found in standard finance textbooks, such as
Elton and Gruber (1991). (^8) This is also the component that cannot be diversified away simply by adding securities to the portfolio.
cycle. 9 In particular, we pay special attention to its time dimension. This can refer to the systematic risk in individual portfolios or banks or, depending on the context, to the system as a whole. 10
2.2 The measurement of systematic risk: challenges and views
Measuring the time dimension of risk, especially its systematic component, is fundamentally difficult. For an individual institution it entails assessing not only how the riskiness of each individual borrower is changing over time, but also how the correlations between borrowers are changing. From the point of view of the system as a whole, a further complexity is the need to understand the correlations amongst individual financial institutions that arise from their exposure to common factors. In addition, while an individual institution might reasonably assume that the evolution of the economy is exogenous with respect to its actions, this is not true for the system as whole. The actions of individual institutions collectively affect the health of the economy, and the health of the economy affects the collective health of individual institutions.
Understanding the evolution over time of the systematic component of risk and hence the endogenous relationships between the financial sphere and the macroeconomy is central to the measurement of financial system risk. In particular, experience indicates that widespread financial system stress rarely arises from the contagion or domino effects associated with the failure of an individual institution owing to purely institution-specific factors. More often, financial system problems have their roots in financial institutions underestimating their exposure to a common factor, most notably the financial/business cycle in the economy as a whole. 11 This form of instability is also the more costly in terms of output forgone, with costs sometimes estimated to run well into double digits as a percentage of GDP. 12
Despite this, there is no consensus as to how the overall level of risk in the financial system moves over the economic cycle. As Section 4 explores, many of the risk measurement methodologies used by banks, rating agencies and bank supervisors imply that risk falls during booms and periods of financial market stability and increases only during recessions and periods of financial turmoil. The view developed here is that it is better to think of risk as increasing in booms, not recessions, and that the increase in defaults in recessions simply reflects the materialisation of risk built up in the boom.
At its heart, the difference between these two views reflects a difference in opinion about the nature of economic processes that underlie the business cycle.
The view that regards risk as moving in line with current economic conditions is more consistent with the interpretation that the economy is best characterised by a series of frequent and small unpredictable developments, or shocks, that alter the economy’s equilibrium, with the adjustment being smooth and rapid. As a result, by default, current conditions are seen as the best, if rather imprecise, guide to future conditions (Box 1). This view means that any observed cycles are mainly the result of the configuration of shocks, which by definition cannot be predicted ex ante.
(^9) By “financial cycle” we essentially mean the sequence of rapid expansion in credit and asset prices, often accompanied by a
relaxation of price and non-price terms in access to external funding, that then moves into reverse and can ultimately be followed by financial distress. See Section 3. (^10) What is the difference between systematic risk for the system as a whole and systemic risk? One possible way of thinking
about it is that systemic risk refers to the risk faced by the system as a whole, regardless of the source. For instance, systematic risk would not cover self-fulfilling crises driven by liquidity concerns (Diamond and Dybvig (1983)) or contagion from the failure of an institution due to purely idiosyncratic factors (eg gross mismanagement of operational risk) unless these causes are in turn thought of as separate common factors. In addition, from the viewpoint of the financial system in a single country , international diversification of system-wide risk is of course possible, so that a residual idiosyncratic component remains. For some related definitions and a survey on systemic risk, see De Bandt and Hartmann (1998). (^11) See Section 3. History suggests that even bank panics, or widespread bank runs, in the pre-safety net era bore a consistent
relationship to the business cycle, tending to occur close after the peak (see Gorton (1988), for the United States, and Palgrave (1894), for the United Kingdom). Wood (1999) reviews this evidence critically. (^12) For a review of episodes and associated costs, see IMF (1998) and Hoggarth et al (2000).
Second, for given correlations of asset returns, correlations of credit losses will be higher, the higher is the probability of default.^2 The intuition is that the probability of one borrower defaulting conditional on another borrower doing the same is a decreasing function of its credit quality. That is, this probability is higher the closer the value of the borrower’s assets is to that of its debt (ie the closer is the firm to its insolvency boundary). Thus, in contrast to asset returns, other things equal, higher expected losses also imply higher unexpected losses through this correlation effect. In addition, higher volatility of asset returns implies higher probabilities of default. This tightens the relationship between asset price volatility and default losses and correlations.
Third, in economic terms, there are a number of reasons why correlations of credit losses might be expected to rise in a downswing, more so than correlations of asset returns. During this phase, the actual incidence of defaults is more likely to reflect movements in the common factor than idiosyncratic elements specific to individual borrowers. For much the same reason, losses given default are likely to be more highly correlated. The changes in behavioural patterns that are typically associated with financial distress would contribute to heightening further these correlations and could lead to a stronger positive relationship between defaults and losses given default compared with normal times. In addition, credits have a strong “ageing effect”,^3 whereby default rates peak three to four years after the credits have been granted. Since more new loans and bonds are issued as the upswing gathers pace, defaults would tend to bunch up with a lag. Any systematic misperceptions and underpricing of risk in the upswing would, of course, reinforce this phenomenon, by increasing vulnerabilities in balance sheets.
Here again, any mean-reverting, predictable element in the common factor plays a significant role. As a result of this property, lengthening the horizon would “telescope” the corresponding ex ante (conditional) correlations into the upswing phase. In other words, such (conditional) correlations would increase as the upswing proceeds.
Several empirical regularities regarding financial variables seem to be broadly consistent with the picture just described. The mean-reverting element over long horizons in asset returns, including equities, has been amply documented. The same is true of the relationship between these returns and business conditions, at least over periods spanning some of the shorter postwar cycles. 4 Measured asset correlations are known to rise, alongside volatility, during bear markets, peaking during periods of financial stress.^5 There is evidence of a negative relationship between credit quality, as proxied by credit ratings, and historical correlations of default. 6 And defaults of course tend to bunch up during recessions. 7
(^2) See Zhou (1997), Gersbach and Lipponer (2000) and Erlenmaier and Gersbach (2001).
(^3) See, for instance, Saunders (1999) and references therein.
(^4) For equities, see Fama and French (1988) and, in particular, Fama and French (1989), which also examines the relationship to business conditions and corporate spreads. (^5) For correlations, see Eng et al (1994), Solnik et al (1995) and Lin et al (1994), and for volatility, see eg Schwert (1989). There is a debate about how far the well documented increase in correlations during times high volatility, and hence possibly financial stress, reflects changes in underlying behavioural patterns or is purely a statistical property (eg Forbes and Rigobon (1999) and English and Loretan (1999)). (^6) See Lucas (1995).
(^7) See also Carey (2000), who shows that simulated tail losses on portfolios, drawing randomly from “good” and “bad” years from a rating agency’s database of loss experience on bonds, differ substantially. The author also notes that the difference does not capture the range of possible outcomes, since the period available (1970-98) does not include experience with a depression or with severe stress in many specific industries.
The view we prefer emphasises the relevance of sporadic but larger unpredictable developments, such as a clustering in technological innovations, and structural behavioural patterns that result in a cyclical response in the economy. Accordingly, the forces that lead to the upswing carry the seeds of the subsequent downswing. The financial cycle supported by credit expansion, asset price developments and their interaction with expenditure decisions, in particular capital accumulation, is a prime source of the cyclical pattern. 13 Such cycles cannot be predicted exactly. And their amplitude, length and characteristics will depend in part of the nature of the original unpredictable developments or triggers and the policy response. 14 But they are in the nature of economic processes. Moreover, there are observable factors that can be relied upon to help form useful conditional judgements about the likelihood and severity of recessions and financial system problems, although their timing may be close to impossible to establish with any precision (Box 1). 15 Such factors can be used as inputs into assessments of systematic risk.
The difference in the two views is not surprising given the experience of forecasters in predicting short-term macroeconomic developments. Despite recent research suggesting that a number of financial variables are useful in predicting recessions, macroeconomic forecasters have a poor record in predicting the exact timing of recessions or turning points in the business cycle. 16 This record has led some to eschew incorporating business cycle effects into risk measurement methodologies. To the extent that these methodologies focus on risk over a one-year horizon, this might be a reasonable approach. However, if longer horizons are used, as we argue should be the case, the approach is less justifiable. While a long-running business expansion might continue for another year, it is much less likely that it will continue for another five years. Being able to predict the exact timing of a downturn is by no means necessary to design an appropriate response to it. Using longer horizons would help lessen some of the emphasis on short-term forecasting, and promote a more thorough analysis of financial vulnerabilities associated with business and financial cycles. This would promote better assessments of systematic risk.
2.3 Factors underlying misperceptions of, and inappropriate responses to, risk
More generally, it is worth standing back and examining the set of factors that can result in either misperceptions of risk per se or inappropriate responses to it. Observationally, however, it is often hard to distinguish between the two.
The first set of possible factors includes the use of the “wrong” model of the economy to interpret developments. The economics profession is now accustomed to analysing economic processes on the assumption that agents understand what drives the economy and have sufficient information to infer, up to an unbiased error, where the economy is going (so-called “rational” or “model-consistent” expectations). 17 This assumption may be helpful in capturing the behaviour of agents in a very stable environment, where economic processes are characterised by regular, recurrent patterns. 18 It is less
(^13) In this sense, this view has a long historical tradition. For example, the Austrian school, with its emphasis on waves of
technological innovation and the role of credit in generating or supporting unsustainable booms, through their interaction with capital formation, is highly relevant here (eg von Mises (1912), Hayek (1933) and Schumpeter (1939)). (^14) The potential amplitude of the financial cycle just described arguably depends on the set of institutional arrangements in the
financial and monetary spheres. For example, during the period when high inflation coexisted with less liberalised financial markets, expansions would more naturally be brought to an end by contractionary monetary policy aimed at containing inflation. The scope for the financial cycle is likely to be greater when inflation is under control and markets are liberalised. See Crockett (2000a) and Borio and Crockett (2000) for an elaboration on these issues from a historical perspective and Gertler and Lown (2000) for evidence consistent with this hypothesis in the United States since the 1980s. (^15) In the terminology of Box 1, output growth would be seen to follow a random walk, rather than having a mean-reverting
component. (^16) See, for instance, Andersen (1997), Artis (1996) and Granger (1996).
17 The formalisation of this very influential view goes back to Muth (1961) and Lucas (1976). After some resistance, it has become the prevailing paradigm, mainly because of the perceived intellectual ad hocery of alternative views and the difficulties of rigorously modelling looser notions involving rational learning. See also Lucas and Sargent (1979). (^18) Even then, the fact that economic outcomes in turn depend on the beliefs themselves raises daunting identification
problems. Ironically, it is precisely the endogeneity of beliefs with respect to the economic environment that has made rational expectations so influential and useful in analysing policy.
likewise. The result, however, would be a widespread reduction in the availability, and increase in the cost, of external funding, which would protract the slowdown. Analogous incentives might help lengthen the upswing. In the pursuit of long-term profits and out of fear of losing customers, lenders can face strong incentives to keep lending. 26 But if everyone does so, at some point overextension may result.
Other courses of action may appear reasonable from the perspective of individual institutions precisely as long as others do likewise. This can result in so-called “herding behaviour”, where agents conform their behaviour to that of their peers. 27 Herding may relate to the use of information, in which case it could be a direct source of misperceptions of sustainable asset values and risk.^28 More generally, it can provide fuel for lending booms and contractions, amplifying the financial cycle. Arguably, the most common factor behind herding behaviour is reward structures that limit blame in the case of collective, as opposed to individual, failure. There may be, for instance, a strong tendency not to blame individual managers for the failure of their bank if failures are widespread. Collective failure would signal homogeneous managerial skills, pointing to realistically small gains from a change in management. 29 Moreover, the authorities might be perceived as more likely to support institutions in the event of widespread financial difficulties in an attempt to limit the severity of the crisis. 30 In such situations, the pressure to conform to the norm can be quite strong. Formal compensation schedules that emphasise relative performance can exacerbate this tendency. 31
More generally, inappropriate responses to risk may derive from shortcomings in contractual arrangements. 32 Arrangements that stress short-term performance are one such example. If rewards are front-loaded in comparison with penalties, there is little incentive to take a longer-term view. 33 The problem is compounded if remuneration is not risk-adjusted. Such arrangements, in fact, are quite common. Obvious cases in point include the payment of fees up front and of bonuses related to unadjusted profitability or to the volume of business, such as to loans extended or funds under management. 34
(^26) For instance, in interviews following the Asian crisis, bankers noted that they were indeed cognisant that the spreads
“dictated” by the market underpriced risks, but that they had strong incentives to keep lending (invest in securities) owing to longer-term considerations. See CGFS (1998). (^27) Devenow and Welch (1996) provide a short review of rational theories of herding behaviour. Not all such herding behaviour
need result in undesirable collective outcomes. Herding may also reflect cognitive biases and more deep-seated traits of human nature (eg Daniel et al (1998) and Prast and Herding (forthcoming)). Evidence of herding has been documented for institutional investors (Nofsinger and Sias (1999)), for investment newsletters (Graham (1999)), for stock prices (Avery and Zemsky (1998)) and for bank lending decisions (Jain and Gupta (1987)). Welch (2000) has also found evidence that herding in the advice of securities analysts based on the prevailing consensus is more likely to take place when the outcome later turns out to be wrong, pointing to undesirable collective outcomes. (^28) The theory of so-called informational cascades is one such example. See, for instance, Bikhchandani et al (1992) and
Barnejee (1992). (^29) See, for instance, Rajan (1994).
(^30) Acharya (2000) shows formally how this can result in herding behaviour that is socially suboptimal, as systematic risk is
increased excessively. (^31) For example, formal assessment of performance in relation to the median is common in the asset management industry in
the United Kingdom (Blake et al (1997)). (^32) Note that what are referred to here as shortcomings in contractual arrangements may represent difficulties in reconciling
fundamental differences in interests and perspective. For example, a diversified shareholder would not be concerned with the idiosyncratic risk associated with the share in an individual company/bank nor, given limited liability, would it care about the loss given failure, as a regulator would. The oft-heard complaint by managers or risk controllers that shareholders are demanding returns not commensurate with risk may reflect at least in part such differences in perspective, quite apart from any overly optimistic expectations about risk/return trade-offs. 33 Herring (1999), for instance, discusses the benefits of having risk-based compensation systems. (^34) A common explanation for distorted incentives leading to increased likelihood of systemic distress is the moral hazard
associated with mispriced implicit or explicit government guarantees, such as those associated with bailout expectations or deposit insurance schemes. This can indeed fuel lending booms, sowing the seeds for subsequent crises. At the same time, the value of the corresponding implicit subsidies would, if anything, move countercyclically alongside perceptions of risk, falling in booms and increasing in recessions.
The bottom line is that several, often related and mutually reinforcing factors provide fertile ground for misperceptions of risk, or inappropriate responses to it, that can lead to excessive procyclicality in behaviour. By the same token, they can also amplify the financial and business cycles. These factors are reflected in financial quantity and price indicators that behave as if risk was perceived to decline in the upswing and rise only once it materialised. Such excessive waves of apparent optimism and pessimism in turn heighten the risk of financial instability.
3. Financial and business cycles and financial indicators of risk
Empirical evidence is generally consistent with the view that the procyclicality 35 of the financial system can be at the root of financial instability and that measures of risk behave as if risk declined during the upswing phase and rose only close to the peak or as the downswing set in. While it is beyond the scope of this paper to present new evidence, it is helpful to document in stylised terms the highly cyclical nature of the financial sector and of measures of financial system risk.
The procyclicality of credit and asset prices has been amply documented and is summarised for a sample of industrial countries in Figures 1 and 2. Periods of robust economic growth tend to be associated with significant increases in the ratio of credit to GDP, and recessions with declines in this ratio. Likewise, episodes of strong credit growth tend to go hand in hand with large increases in equity and property prices, and, to varying degrees, these prices tend to decline as credit contracts in the downswing. 36 Ex post , a financial cycle is clearly apparent.
There are, of course, several possible reasons for such co-movements. And there is a debate regarding how far developments in the financial sphere cause, rather than reflect, the evolution of economic activity. This is true, in particular, for those changes in the cost and availability of external financing associated with imperfect substitutability between internal and external funding, 37 such as the easier extension of credit as the net worth and value of collateral held by borrowers increases. There are, however, good a priori grounds to believe that the process feeds on to itself. Moreover, it also stands to reason that the influence of financing constraints should become especially relevant as economic agents, suppliers and recipients of funds alike, face financial distress. In industrial countries, typical examples include the financial “headwinds” that appeared to inhibit the recovery following strains in the US banking system in the early 1990s and more recently, the serious difficulties faced by the Japanese economy following the banking crisis. 38 More generally, the recent record of financial crises, especially those in Latin America and Asia in the 1990s, amplified by boom and bust movements in international capital flows, has been interpreted as providing evidence of a sizable causal role of financial factors.^39
Above all, experience suggests that overextension in the financial system, in the form of rapid credit expansion and unusually sharp increases in asset, especially property, 40 prices during the economy’s
(^35) To avoid confusion, in what follows the movement in a financial indicator is said to be “procyclical” if it tends to amplify
business cycle fluctuations. According to this definition, for instance, provisions behave procyclically if they fall in economic upswings and rise in downswings. (^36) Formal econometric evidence on credit/asset price cycles in industrial countries can be found in Borio et al (1994). Kent and
D’Arcy (2001) examine four major credit and property price cycles and their relationship with financial stability in Australia since the 1870s. (^37) This is the “financial accelerator” hypothesis, as articulated in detail by Bernanke et al (1999) in particular. Much of the
formal statistical evidence testing this hypothesis relies on panel data and applies to the United States (Hubbarb (1998). More recently, the hypothesis has been tested with some success with time series macroeconomic data, with the spread between low- and high-quality corporate bonds being used as a proxy for the premium on external funding costs (Gertler and Lown (2000)). (^38) See Gibson (1995) and Peek and Rosengren (1997) and (2000) for evidence that Japanese banking troubles had real
effects. Hancock and Wilcox (1998) provide similar evidence for US banks. (^39) See for example the papers in Gruen and Gower (1999).
(^40) Borio et al (1994), BIS (1993) and various issues of the BIS Annual Report document in detail the behaviour of residential
and commercial real estate prices across countries. A classic reference on historical cycles in property prices in the United States is Hoyt (1933).
0
2
4
6
0
2
4
6 Total private credit/GDP ratio (lhs) Output gap (rhs) 1
0
2
4
6
0
2
4
6
0
2
4
6
0
2
4
6
0
0
2
4
6
0
3
6
9
0
4
8
12
1980 1985 1990 1995
0
2
4
6
1980 1985 1990 1995 (^1) As calculated by the OECD.
Sources: OECD Economic Outlook; national data.
United States Japan
Germany Italy
United Kingdom Spain
Australia Sweden
Finland Norway
Figure 1: Total private credit
40
60
80
100
120
140
160
0
2
4
6
40
60
80
100
120
140
160
0
2
4
6 Real aggregate asset prices (lhs) 1 Output gap (rhs) 2
70
80
90
100
110
120
130
0
2
4
6
70
80
90
100
110
120
130
0
2
4
6
40
60
80
100
120
140
160
0
2
4
6
40
60
80
100
120
140
160
0
2
4
6
70
80
90
100
110
120
130
0
2
4
6
25
50
75
100
125
150
175
0
3
6
9
25
50
75
100
125
150
175
0
4
8
12
1980 1985 1990 1995
40
60
80
100
120
140
160
0
2
4
6
1980 1985 1990 1995 (^1) Indices, 1980-99 = 100 (for Italy, 1988-99; for Spain, 1987-99); weighted average of equity and residential and commercial real estate price indices deflated by consumer prices; the weights are based on the composition of private sector wealth. (^2) As calculated by the OECD.
Sources: OECD Economic Outlook; BIS calculations.
United States Japan
Germany Italy
United Kingdom Spain
Australia Sweden
Finland Norway
Figure 2: Real aggregate asset prices
0
0
2
4
6 Provisions for loan losses/total assets (lhs) 1 Output gap (rhs) 1, 2
0
0
2
4
6
0
0
2
4
6
0
1
2
3
4
5
6
0
2
4
6
0
1
2
3
4
5
6
1986 1988 1990 1992 1994 1996 1998 – 12
0
3
6
1986 1988 1990 1992 1994 1996 1998 (^1) In percentages. 2 As calculated by the OECD. Sources: BIS survey; OECD Economic Outlook.
United States
Japan United Kingdom
Italy Spain
Australia Sweden
Finland Norway
Figure 5: Bank provisioning
In large part, the behaviour of provisions translates into a clear procyclical pattern in bank profitability, which further encourages procyclical lending practices. This pattern appears to be strongest in those countries that experienced banking system problems in the 1990s (Table 1). German banks are the only exception to this procyclical behaviour, given their ability to smooth profits through hidden reserves. The procyclical nature of bank profits has arguably also contributed to bank equity prices being positively correlated with the business cycle, although the correlation is typically somewhat weaker than that for profitability, reflecting the forward-looking nature of the equity market.
The relationship between the business cycle and bank capital is less obvious (Table 1 and Figure 6). While it is clear that the level of bank is positively correlated with the economic cycle, there does not appear to be a robust relationship between measured capital ratios and the business cycle. To some extent, the task of detecting any relationship is made difficult by the introduction of the Capital Accord in 1988, which some have argued caused a structural change in capital ratios in some countries. 47 Analysis is also complicated by the fact that government support schemes have influenced capital ratios. Nevertheless, long-run historical time series do not suggest a strong business cycle effect, with the main stylised fact being a steady decline in capital ratios over the 20th century, before a slight increase over the past decade or so.
At the same time, there are two important qualifications to the conclusion that capital ratios tend to be acyclical. The first is that, to the extent that provisions underestimate expected losses in expansions, measured capital ratios overstate true capital ratios in expansions. This effect can be significant. For example, if the ratio of provisions to total assets is 1 percentage point below where it should be, then the measured capital ratio is likely to overstate the true capital ratio by at least 10%. If adjustments were made to capital for underprovisioning in economic booms, it is likely that, all else constant, measured capital ratios would fall during expansions and increase during downswings.
The second qualification is that there has been a pronounced cycle in aggregate capital ratios over the 1990s in those countries that experienced problems early in the decade. In the years immediately after the crisis, when conditions were still relatively depressed, banks made a concerted effort, not only to rebuild their capital ratios, but also to substantially increase them above previous levels. Then, starting in the middle of the decade, when economic expansions were firmly entrenched, some of the increase in capital ratios was unwound. This pattern is evident in Australia, Sweden and Norway, and to a lesser extent in Finland. 48
Further, the cycle in the ratio of capital to risk-weighted assets was much more pronounced than the cycle in the ratio of capital to total assets. This reflects the fact that, in the aftermath of the banking crises, risk-weighted assets fell more strongly than total assets, as banks shifted their portfolios away from commercial lending (which has a relatively high risk weight) towards residential mortgages and public sector securities (both of which have relatively low risk weights).
Overall, these various indicators suggest that perceived risk in the financial system does not increase in business cycle expansions, and that it may actually decline during periods of robust economic growth. They also suggest that typically risk is assessed to have increased only when credit losses materialise, rather than when the problems that underlie the losses are building up. 49 These risk assessments sit uncomfortably alongside recent experiences, which suggest that business cycle expansions underpinned by rapid credit growth, large increases in asset, especially property, prices and high levels of investment (particularly in the property sector) can sow the seeds of subsequent financial system problems. While incorporating the lessons from these experiences into risk measurement systems is not straightforward, doing so would help improve the way in which risk is assessed to evolve over time.
(^47) For a survey of the impact of the Basel Capital Accord, see BCBS (1999a).
(^48) In Section 5, we consider the reasons for this.
(^49) One set of alternative explanations of analogous patterns in asset price behaviour seeks to explain them as reflecting
changes in equilibrium required returns by risk averse investors. For instance, Fama and French (1989) document how corporate bond spreads and dividend-price ratios are comparatively high in weak business conditions and low in strong conditions and how these variables can help to forecast future corporate bond and equity returns. Low dividend yields or spreads tend to be followed by sup-par asset price performance. This is interpreted as reflecting comparatively low (high) required returns in good (bad) times. See Campbell et al (1999) for an overview of attempts to explain asset returns along these lines.
4. Risk measurement methodologies
As argued in Section 2, a number of factors could explain possible misperceptions of, and inappropriate responses to, risk and contribute to the observed procyclical behaviour in market indicators of risk documented in Section 3. Here, we review in more detail the main risk measurement methodologies actually employed in the financial system, including banks’ internal rating systems, ratings by credit rating agencies, credit risk models and the approaches used by bank supervisors and other policymakers. We argue that while, in general, these methodologies are well suited to addressing relative risk, they have difficulty in measuring the systematic component of risk associated with financial and business cycles. This difficulty stems from the relatively short horizons that are often used to assess expected and unexpected losses and from insufficient attention being paid to the movement of correlations over time.
4.1 Commercial banks’ internal rating systems
Many banks have recently increased the attention paid to the quantification of risk. This has typically involved the development and implementation of internal grading systems, which classify loans into specific risk categories or ratings. 50 These internal ratings are used as inputs into decisions regarding pricing, capital allocation and provisioning. 51
Most internal rating systems have a “point-in-time” focus and use a one-year horizon for measuring risk. This means that the systems are designed around the idea of measuring the probability of default over the next year. The choice of a one-year horizon is driven by a variety of factors, including the availability of data, the internal budgeting cycle of the bank, and the interval in which new capital can be raised or loss mitigation action taken. 52
The nature of the internal rating systems means that the average rating of a bank’s loan portfolio is likely to change over the course of the business cycle. When economic conditions are strong, loans are likely to move up the rating scale (to lower-risk ratings) given that the probability of default in the next year is relatively low. Conversely, in an economic downturn the average rating is likely to decline, given the increased probability of default in the short run. As a result, measured risk, as revealed by average internal ratings, is likely to be negatively correlated with the economic cycle - that is, it falls in booms and increases in recessions.
The correlation issue is not relevant for simple rating schemes, although it is critical in assessments of overall portfolio risk; see Section 4.3.
4.2 External rating agencies
The approach used by most credit rating agencies attempts to rate borrowers “through the cycle”. This means that ratings are less likely to move over the course of the business cycle, with borrowers being rated on their probability of defaulting in a constant hypothetical downside scenario. 53 Ratings will only change over time if the rating agency changes its assessment of the probability of default in the downside scenario, or changes the scenario itself. 54
(^50) For a recent survey of banks’ internal rating systems, see BCBS (2000a).
(^51) For example, English and Nelson (1998) use survey information for United States banks to document the fact that
lower-rated credits are charged a higher price (both in terms of interest rates and non-price terms of credit). (^52) See the survey by the BCBS (2000a). The survey also notes that banks that use a longer horizon do so because their
exposures are typically held to maturity and because of a lack of markets in which their credits can be traded. 53 For a comprehensive overview of the credit rating industry, its approach to measuring risk, and its successes and failures, see BCBS (2000b). (^54) Carty and Fons (1994) report that during the period 1980-1993, 88% of all ratings remained unchanged over a one-year
horizon. This number is lower than the 95% stability of ratings reported for the 1950-1980 period. Lucas and Lonski (1992) have also documented the volatility of ratings, reporting that 1% of issues rated AAA and 9% of issues rated Baa were downgraded to speculative grade within five years.
The through-the-cycle approach does not guarantee that the ratings will be acyclical. In particular, an economic downturn that is worse than expected is likely to lead to ratings being downgraded. Table 2 summarises evidence from a recent study that documents this empirical fact. 55 The authors’ estimates show that the probability of being downgraded, particularly for bond issues at either end of the rating scale, rises during recessions and falls during booms. Despite this, it remains likely that these ratings are less procyclical than internal bank ratings (Box 2).
Historically, the agencies have been relatively successful at measuring the cross-sectional dimension of risk.^56 As discussed in Section 3, however, they have been less successful in downgrading ratings prior to a borrower defaulting. 57
4.3 Quantitative credit risk models^58
Given that the focus of internal and external ratings is on measuring the risk of individual instruments or borrowers, such systems do not explicitly consider the correlations between ratings and how these correlations change over time. Thus, such ratings by themselves cannot easily be used to address the credit risk of large and complicated portfolios. As a result, a number of financial institutions have recently developed, or purchased, quantitative credit risk models.
While the various models have different structures, most tend to extrapolate recent history in one way or another, so that good current economic conditions signal good future prospects. 59 Moreover, while the various approaches incorporate correlations, the treatment is often simplistic, with correlations either fixed or dependent on the recent history of financial markets.
One of the most commonly used approaches relies on equity price data and option pricing theory to construct measures of risk. In these “Merton-type” models, a rise in a firm’s indebtedness, a fall in its equity price or an increase in the volatility of its equity price leads to an increase in the measured probability of default (all else constant) of the firm over the next year. 60 Even where estimates can be calculated over a multi-year horizon, the assumptions made result in simple extrapolations that effectively rule out business cycle effects. 61 The correlations between firms are derived on the basis of past movements in equity prices.
(^55) See Nickell et al (2000).
(^56) See Moody’s (2000). A more scientific documentation of the success that rating agencies have had in distinguishing the
cross-sectional dimension of risk can be found in Brand and Bahar (1999) and Keenan (1999). (^57) More generally, credit ratings are less successful at measuring the time dimension of risk. For instance, Cantor and Packer
(1994) document that the default rates associated with each rating change significantly over time. For example, over the period 1970 to 1989, the five-year default rates associated with a BBB rating ranged anywhere from 0.8% to nearly 5%. 58 A review of the historical developments in credit risk modelling can be found in Altman and Saunders (1997). A summary of the most popular models currently in use can be found in Saunders (1999), which also contains an extensive bibliography on the measurement of credit risk. Crouhy et al (2000) and Gordy (2000) provide a more technical and detailed comparative analysis of the models. (^59) While there are a number of papers that compare and contrast the different models, we know of no literature that compares
how aggregate measures of risk generated by these models are likely to move over the course of the business cycle. Given the increasing importance of these models, this seems a useful area for future research. 60 The original papers on which this methodology is based are Merton (1973) and (1974). (^61) One such model employing the Merton approach and allowing risk to be measured over different horizons has been
developed by KMV. Since KMV forecasts equity price movements by essentially assuming that stock prices follow a random walk, increasing the horizon over which KMV forecasts expected default probabilities leads to a mechanical adjustment rather than to a thorough assessment of longer-term vulnerabilities. Moreover, the assuption fails to capture the mean reverting properties associated with equity returns over longer horizons (Box 1).
