By Hyun Song Shin
This paper compares three types of early warning indicators of financial instability – those based on financial market prices, those based on normalized measures of total credit and those based on liabilities of financial intermediaries. Prices perform well as concurrent indicators of market conditions but are not suitable as early warning indicators. Total credit and liabilities convey similar information and perform better as early warning indicators, but liabilities are more transparent and the decomposition.
Finding a set of early warning indicators that can signal the vulnerability to financial turmoil has emerged as a policy goal of paramount importance in the aftermath of the global financial crisis.
There is a large literature on early warning indicators for crises, described well in Chamon and Crowe (2012). The emerging economy crises of the 1990s gave impetus to the work, which has been further developed in the aftermath of the recent global financial crisis that engulfed the advanced economies as well as emerging economies.
The literature to date could be described as being eclectic and pragmatic. It has been eclectic in that the exercise involves appeal to a wide variety of inputs, covering external, financial, real, and fiscal variables, as well as institutional and political factors and various measures of contagion. In their overview of the literature as of 1998, Kaminsky, Lizondo and Reinhart (1998) catalogue 105 variables that had been used up to that date.
The literature has also been pragmatic in that the exercise has focused on improving measures of goodness of fit, rather than focusing on the underlying theoretical themes that could provide bridges between different crisis episodes.
The distinction between emerging and advanced economies is also reflected in the work of the official sector. The IMF has added a new Vulnerability Exercise for Advanced Economies (VEA) to an existing Vulnerability Exercise for Emerging Economies (VEE), which both feed into joint early warning exercise with the Financial Stability Board (FSB).
Although the compartmentalization into emerging and advanced economies helps in improving the goodness of fit, it tends to obscure the common threads that tie together emerging and advanced economy crises. The capital flow reversals in Spain and Ireland in the European crisis have many of the features of a “sudden stop”, except that the outflow of private sector funds has been compensated by the inflow of official funds.
However, since the Eurozone crisis is taking place within a common currency area, the traditional classification of emerging market “currency crises” where currency movements play a key role do not necessarily fit easily in the empirical exercise.
Given the common threads that tie together apparently disparate crises, it can be useful to take a step back from the practical imperatives of maximizing goodness of fit and instead consider the conceptual underpinnings of early warning models. This is the purpose of my paper.
In what follows, I will suggest that the procyclicality of the financial system provides an organizing framework for selecting indicators of vulnerability to crises, especially those that are associated with banks and financial intermediaries more generally.
More specifically, I examine three broad sets of indicators for early warning purposes, and assess their relative likelihood of success. The three sets of indicators are:
*Indicators based on market prices, such as CDS spreads, implied volatility and other price-based measures of default or distress
*Gap measures of the credit to GDP ratio
*Banking sector liability aggregates, including monetary aggregates
PRICE-BASED EARLY WARNING INDICATORS
Figure 1 gives the CDS spreads of Bears Stearns and Lehman Brothers, with the right handpanel giving the longer perspective and illustrating how the spreads increase sharply with the onset of the crisis.
What is remarkable is how tranquil the CDS measure is before the crisis. There is barely a ripple in the series in the period 2004 to 2006 when the vulnerability to the financial crisis was building up. The left hand panel, which plots the CDS series for the pre-crisis period of January 2004 to January 2007 shows that CDS spreads were actually falling over the period, dipping below 20 basis points at the end of 2006. Other price-based measures, such as Value-at-Risk, implied volatility, structural models of default based on equity prices, etc. All painted the same picture.
The failure of price-based measures of early warning indicators can be traced to their implicit premise that the interaction between market signals and the decisions guided by those signals always interact in a stabilizing virtuous circle, rather than sometimes going astray and acting in concert in an amplifying vicious circle where market signals and decisions guided by those signals reinforce an existing tendency toward procyclicality. Some of the forces toward procyclicality were described in my Mundell-Fleming lecture last year (Shin (2012)).
As an illustration of the outcome of such a tendency toward procyclicality, the scatter chart in Figure 2 plots how much the change in the balance sheet size of Barclays – a typical global bank – is financed through equity and how much through debt. It also shows how much riskweighted assets change as the balance sheet grows or shrinks.
The fact that risk-weighted assets barely increase in Figure 2 even as raw assets are increasing rapidly is indicative of the lowering of measured risks (such as spreads or Valueat- Risk measures) during lending booms. Lower measured risks and lending booms thus go together. The causation in the reverse direction will also have been operating – the compression of risk spreads is induced by the rapid increase in credit supply chasing available credits. Such two-way causation lays the ground for a feedback loop in which greater credit supply and the compression of spreads feed off each other.
The procyclicality evident in Figure 2 poses hard challenges for traditional thinking that places faith in market discipline as an integral part of financial regulation, where prices are relied upon to issue timely warning signals. Indeed, Market Discipline was one of the three “Pillars” of the Basel II framework for international bank regulation. Economists associated with the Shadow Financial Regulatory Committee were influential in this regard. Calomiris (1999) argued for rules requiring banks to maintain a minimum amount of subordinated debt, with the rationale being that banks that take on excessive risk will find it difficult to sell their subordinated debts, and will be forced to shrink their risky assets or to issue new equity to comply with the discipline imposed by private uninsured creditors. However, the experience in the run-up to the recent crisis showed how market risk premiums erode so as to nullify market discipline.
Larry Summers’s quip (Summers (1985)) that the achievement of finance researchers is to show that “two quart bottles of ketchup invariably sell for twice as much as one quart bottles of ketchup” is related to why price-based measures of early warning indicators are likely to fail. Absence of arbitrage means that prices at a point in time are consistent, but they are liable to flip to distress mode (again, fully consistently across assets) with the onset of the crisis. If the task is to give prior warning of the onset of the crisis, price-based measures have little to say about the transition.
Since the onset of crisis is often accompanied by run-like events, the switch from a benign environment to a hostile one can be precipitous. The global games literature illustrates how the transition into financial distress – the “tipping point” – is associated with self-reinforcing effects between individual constraints and market outcomes, but how the onset of the crisis is triggered by apparently small changes in the underlying fundamentals. Outwardly, the switch has the flavor of a self-fulfilling crisis. Goldstein (2010) discusses how empirical research should take account of such tipping points, and shows how the global games framework (Morris and Shin (1998, 2000, 2008)) can be usefully invoked in the modeling exercise.
To the extent that market prices have been useful for early warning exercises at all, their usefulness comes precisely when the market price of risk is too low, rather than too high. Thus, it is when asset prices are too high relative to some benchmark that warnings signs are appropriate.
In their 2005 paper on the US housing market, Himmelberg, Mayer and Sinai (2005) argued that a high price-to-rent ratio or high price-to-income ratio need not be indicators of a housing bubble as discount rates implied by low long-term interest rates had also fallen. But since discount rates are prices, the combination of low discount rates and high housing prices is arguably the kind of point-in-time consistency in prices that Summers (1985) had in mind.
CREDIT TO GDP GAP INDICATORS
Under the Basel III framework, the ratio of credit to GDP takes a central role as the basis for the countercyclical capital buffer. This ratio has been shown to be useful as an indicator of the stage of the financial cycle, notably by Borio and Lowe (2002, 2004). To the extent that procyclicality drives financial vulnerability, detecting excessive credit growth is central.
Normalizing credit to some underlying flow fundamental measure such as GDP and detecting deviations from trend would be one way to operationalize the notion of excessive credit growth.
However, although a credit boom is clear with hindsight, there are several challenges to using the deviation of the credit to GDP ratio from trend as an early warning indicator in real time. The first is the difficulty of estimating the trend that serves as the benchmark for what is “excessive” growth. The difficulty is not unique to the credit to GDP ratio, but one shared by other macroeconomic time series. Edge and Meisenzahl (2011) find that ex-post revisions to the credit-to-GDP ratio gap in real time are sizable for the U.S. and as large as the gap itself. The source of the ex post revisions is not the revision of the underlying data, but rather from the revision of the estimated trend measured in real time.
The second difficulty is that credit growth and GDP dance to somewhat different tunes over the cycle, so that the ratio of the two may sometimes issue misleading signals. Bank lending in particular may be influenced by pre-existing contractual commitments, such as lines of credit, which are drawn down during the crisis. Ivashina and Scharfstein (2010) document the impact of such lines of credit on credit growth during the recent crisis. Therefore, lending may continue to increase for some time after the onset of the crisis.
Figure 3 is taken from Repullo and Saurina (2011) and shows the credit to GDP ratio for the UK and its HP-filtered trend (left hand panel). The HP filter parameter is set at λ=400,000 as recommended by the Basel Committee, which effectively means a linear trend. The right hand panel shows the credit-to-GDP ratio “gap” between the credit-to-GDP ratio and the trend.
From the right hand panel of Figure 3, we note that the gap measure is large even as GDP growth is falling very sharply during the crisis. Thus, the ratio of the two gives a misleadingly large credit to GDP ratio during the crisis.
Basel III discussions have given a great deal of prominence to the credit-to-GDP gap measure (BCBS (2009, 2010)). To the extent that the Basel rules are expected to be applied uniformly (or at least, in a consistent manner), finding common thresholds for the credit to GDP ratio would be a basic requirement for Basel III to apply uniformly to all Basel Committee member countries.