The ‘Ex-post Clean’ Paradigm¹⁶

The ‘ex-post clean’ paradigm emphasizes the difficulties faced by authorities in reliably identifying a bubble is in progress. Authorities have often set a high bar for establishing the existence of bubbles due to a perception they have less information about valuation misalignments than markets themselves (Cogley, 1999).¹⁷ If a central bank with no informational advantage over market participants believes a bubble has developed, investors—with more complete information and a vested interest in correcting irrational mispricings—will almost surely have arrived at the same realization and drive asset prices down to a level more consistent with fundamentals. To the extent that speculators stabilize markets by correcting valuation misalignments, they obviate the need for policy intervention in the first place (Friedman, 1953a). As pointed out in the late 1990s by U.S. Federal Reserve Chairman Alan Greenspan, “there is a fundamental problem with market intervention to prick a bubble: it presumes that you know more than the market … identifying a bubble in the process of inflating may be among the most formidable challenges confronting a central bank, pitting its own assessment of fundamentals against the combined judgment of millions of investors.”¹⁸ Bernanke (2002a) stressed that it was unlikely policy makers could accurately disentangle the rational and irrational component of asset prices, and expressed concern over “the effects on the long-run stability and efficiency of our financial system if the Fed attempted to substitute its judgments for those of the market.” Kohn (2006) similarly noted, “risk premiums often move in mysterious ways … we have a very poor understanding of the forces driving speculative bubbles.”

The ‘ex-post clean’ approach has also been advocated on the basis of counterfactuals and opportunity costs. For instance, policy makers cannot know ex-ante if a period of unusually elevated asset prices will plateau, benignly deflate, or be followed by a ‘catch up’ in fundamentals. As such, bubbles can only ever be identified (long) after the fact. Furthermore, a great deal of uncertainty surrounds the overall impact on the economy in the event of a sharp asset price reversal. The effects on inflation and output from booms and busts in asset prices sometimes reveal themselves with significant and variable lags (Yellen, 2009). In lieu of so much uncertainty, a more prudent course may therefore be to wait and observe what negative impact, if any, a downturn in asset prices exerts on the economy. Policies conditioned on noisy or unobserved variables can be destabilizing rather than stabilizing, rendering pre-emptive countervailing actions counterproductive (Friedman, 1953b).

If authorities believe they have the tools available to keep the harmful effects of a bursting bubble at a manageable level, then a policy of leaning against the wind may also be unnecessary. In addition to countercyclical fiscal policy, authorities have various means by which to arrest the fallout of bursting bubbles: policy rates can be taken down to the zero bound; a commitment can be made to maintain policy rates at exceptionally low levels for an extended period; risk premia can be lowered through central bank asset purchases; inflation expectations can be bolstered (lowering real interest rates) with a credible commitment to increase the supply of domestic currency; and authorities can intervene in the foreign exchange market to weaken the domestic currency and stimulate exports (Svensson, 2001; Bernanke 2002c). Recapitalization measures are also available in the event a bust threatens the broader financial sector. Prior to the global financial crisis, a number of researchers pointed out that lessons had been learned from the timidity of the Japanese policy response to the bursting Heisei bubble in the 1990s (Bernanke, 2000, 2003; Ahearne and others, 2002; Posen, 2003; Kohn, 2006). The relatively modest economic fallout associated with the stock market crashes of 1987 and 2000-2002 helped solidify the view that asset price busts, if and when they occurred, could be comfortably contained by ‘cleaning’ rather than ‘leaning.’¹⁹

Even if bubbles could be identified with confidence in real time, advocates of the ‘ex-post clean’ approach have also argued that monetary policy is not the right tool to safely diffuse them.²⁰ Implicit in this view has been the belief that the best contribution central banks could make to financial stability was to focus solely on price stability. If the sensitivity of asset markets to interest rate movements is low (and/or uncertain), curbing asset price increases may require a policy response so vigorous it risks unnecessary collateral damage to the broader economy. The crudeness of monetary policy might induce the very outcome it sought to avoid, namely a sharp recession. In this sense, the cure would be as bad, or possibly worse, than the disease (Bernanke, and others 1999; Greenspan, 2002; Bernanke, 2002a; Gruen and others, 2005; Bernanke and Kuttner, 2005; Kohn, 2006; Kuttner, 2012). For instance, invoking the Tinbergen separation principle (where to achieve a certain number of objectives, at least an equal number of instruments are required), Bernanke (2002a) argued,

“We cannot practice safe popping, at least not with the blunt tool of monetary policy. Certainly there is no way to direct the effects of monetary policy at a single class of assets while leaving other financial markets and the broader economy untouched. One might as well try to perform brain surgery with a sledgehammer. The problem of safe popping applies with double force to the aggressive bubble-popping strategy. A truly vigorous attempt by a central bank to rein in a supposed speculative bubble may well succeed but only at the risk of throttling a legitimate economic boom or, worse, throwing the whole economy into depression.”

In a related context, Greenspan (2004) suggested that in instances where a moderate first round of monetary tightening failed to elicit the desired effect, markets could become emboldened with a sense of immunity and rise to even loftier heights (thus posing an even larger threat to longer-term financial stability). A rise in interest rates could lead to rational agents boosting their assessment of the equilibrium growth rate during a speculative episode, thus negating the negative impact on fundamental asset values from a higher discount rate.

Much of the skepticism over the role that monetary policy could play in safely diffusing bubbles owed to the criticism leveled at the Federal Reserve for its role in tightening monetary conditions in response to frothy stock market conditions ahead of the Great Depression (Friedman and Schwartz, 1963; Schwartz, 1981; White, 1990; and Bernanke 2002a, 2002b). As Bernanke (2002a) argued, “Early in 1928, the Fed passed into the control of a coterie of aggressive bubble-poppers … The correct interpretation of the 1920s is not the popular one – that the stock market got overvalued, crashed, and caused a Great Depression. The true story is that monetary policy tried overzealously to stop the rise in stock prices. But the main effect of tight monetary policy … was to slow the economy … The slowing economy, together with rising interest rates, was in turn a major factor in precipitating the stock market crash.”²¹

The ‘Ex-ante Lean’ Paradigm²²

The ‘ex-ante lean’ paradigm trades off the possibility of subpar macroeconomic performance in the near term, for the prospect of significantly better performance over a longer horizon, on the understanding that the repercussions of a major asset price collapse could be unmanageable. According to this view, a policy of ‘benign neglect’ toward asset price booms can store up trouble, making it difficult and costly to contain at a later date. Put differently, the long-term welfare costs associated with allowing asset bubbles to grow unchecked may be prohibitively high—although remedial policy tools may not be costless, they can be deployed effectively enough to curb the accumulation of dangerous excesses and thus aid in maximizing welfare over the longer-term. As such, authorities may need to take out insurance against the possibility of a large crash in the future by tightening financial conditions earlier than would otherwise be the case.²³ By cooling off booms before they spiral out of hand, these efforts can contribute to lengthening the economic cycle (Roubini, 2006). Barlevy (2012) suggests theoretical economic models of asset bubbles, which have tended to downplay the case for pre-emptive intervention, have yet to adequately address the long-term welfare implications associated with their bursting that the recent crisis brought to light.

Proponents of pre-emptive intervention argue that the difficulties associated with identifying bubbles in real time should not absolve policy makers from this responsibility. Although the precise magnitude of a bubble cannot be known with certainty, authorities are frequently tasked with making decisions in real time with less than perfect information.²⁴ Constructing estimates of market risk premia may be no more difficult than constructing estimates of the natural rate of interest or output (Blanchard 2000; Cecchetti, 2003). Incorporating uncertainty into the decision making process, Dupor (2005) suggested policy makers should lean against the wind of asset bubbles with an intensity linked to the confidence in which they consider asset prices to be misaligned. Others have suggested policy makers have a stronger case in leaning against credit-fuelled asset bubbles, in part because authorities have informational advantages over the market in terms of their access to timely qualitative and quantitative credit-related data (White, 2009; Mishkin, 2010).

The near-exclusive focus of monetary policy on goods and services inflation (relative to asset price inflation) might also contribute to the formation of bubbles and financial instability, particularly in instances where positive supply shocks occur in the market for goods but not assets. If positive productivity or supply shocks lower the rate of goods and services inflation below the target of the central bank, policy rates may be lowered to levels that ferment speculative risk taking. Notably, all the major U.S. stock market booms and busts over the past two centuries, including the 1920s and 1990s, have occurred against the backdrop of relatively easy monetary conditions and strong productivity growth which kept consumer price inflation low and stable (Bordo and Wheelock, 2004; Christiano and others, 2010). Bordo and Wheelock (2006, 2007) document the tendency for asset booms to occur during periods of below target inflation (or disinflation) across a host of industrialized countries over the course of the twentieth century.²⁵

From an analytical standpoint, the observation that asset valuations are mean reverting and that asset booms frequently accompany periods of benign inflation and/or robust credit growth might suggest a role for an augmented Taylor-rule to guide monetary policy settings.²⁶ This would dictate leaning against below-average risk premia and/or above-average credit growth, in addition to above-target consumer price inflation (Blanchard, 2000; Borio and Lowe, 2002, 2004; Cecchetti, 2003; Christiano and others, 2010; Acharya and Naqvi, 2012). Alternatively, because they impact the economy through the wealth effect and embed expectations about future economic conditions that can be relevant for inflation-targeting central banks, asset prices could be incorporated directly into the inflation metric that monetary authorities target (Trichet, 2003; Goodhart and Hofmann, 2007).

Finally, moral hazard considerations could also support a policy of leaning against the wind. By pre-committing to an asymmetric response to asset price booms and busts such that investors believe policy makers will ease aggressively in an asset price bust, but refrain from tightening in a boom, asset prices might take on a more dangerous speculative dynamic than would otherwise be the case (Caballero and Krishnamurthy, 2003; Roubini, 2006; Cooper, 2008; White, 2009; Issing, 2011).²⁷ At the very least, as evidence of a bubble begins to mount, a verbal policy strategy of moral suasion could have few costs relative to potential benefits. Sober assessments of ‘frothy’ market conditions by authorities in public forums, or the more aggressive announcement that policy makers stand ready to act should asset prices continue to soar, might be sufficient to prevent the emergence of destabilizing bubbles in the first place (Blanchard, 2000).

The New Consensus and the Road Ahead

The aftermath of the global financial crisis has resolved many important elements of the ‘clean vs. lean’ debate in favor of the latter. Prior to the crisis, asset bubbles and financial stability were in some quarters viewed as a “potentially dangerous distraction, risking the central bank’s attention to, and credibility in, achieving its price-stability mandate … even at the Federal Reserve, financial stability was rarely discussed” (Williams, 2014, p. 2). Yellen (2009, p. 8) noted immediately following the crisis that while most central bankers (particularly in the United States) have historically resisted the idea of leaning against asset price increases, “now that we face the tangible and tragic consequences of the bursting of the bubble, I think it is time to take another look.” And as Stein (2013a, p. 1) summarized, “Post-crisis, I think it’s safe to assume that nobody would now argue that we should be putting all our eggs in the ‘clean’ basket.”²⁸ Broadly speaking, a new consensus has emerged in policy making circles to recognize that:

• In both developed and emerging economies, the long-term welfare costs associated with allowing bubbles to grow unchecked can be more damaging than previously thought (in this sense, asset bubbles pose an ‘equal opportunity menace’);

• Long periods of stability can be destabilizing to the extent they promote ever more aggressive forms of risk-taking in asset markets (Minsky, 1986, 1992);

• Frictions in asset markets can have a pronounced, non-linear impact on the economy, especially if the imbalances to which they give rise are allowed to accumulate unchecked;²⁹

• While long-term price stability should remain a primary objective of monetary policy and can make a contribution to financial stability, it is not a sufficient condition to insure against asset bubbles or financial instability more generally;³⁰

• Highly targeted prudential measures should form the first line of defense against nascent signs of asset bubbles and concomitant threats to financial stability;’³¹

• More generally, a new macro-financial stability architecture is required at the national and global level to address the threat that asset bubbles and financial imbalances pose to economic stability.

This said, old uncertainties may have been resolved, but new ones have emerged in their wake. With this in mind, the remainder of the paper is motivated by the concept that the optimal mix of policy responses—regulatory, monetary, and macro-prudential in nature—can be refined with a clearer distinction among the underlying forces that give rise to asset bubbles in capital markets.

Limits to Learning

‘Limits to learning’ models suggest that while investors process information rationally (a key assumption in efficient markets theory), large pricing errors can lead to bubbles because of incomplete and/or inaccurate information. Even if investors engage in ‘Bayesian updating’ as they rationally process new incoming information, it may not be possible to ever have complete knowledge over the (constantly changing) structure of the economy and all the factors capable of impacting the investment opportunity set. Put differently, there are limits to learning over the true data generating process which make ‘rational bubbles’ possible.⁴⁴ Markets are particularly vulnerable to bubbles where there is a low level of financial literacy among participants, and common knowledge over the existence of a bubble is absent.⁴⁵

Frictional Limits to Arbitrage

Bubbles can also arise in a rational setting because of frictional limits to arbitrage, such as trading costs and the incompleteness of markets.⁴⁶ If differences in beliefs over fundamental value for an asset wash out in aggregate (i.e., there are optimists and pessimists in roughly equal number with equal conviction), but pessimists are constrained in their ability to short or sell assets due to costs or other frictions (such as availability of inventory or instruments), then prices will disproportionately reflect the beliefs of optimists and thus rise above fair value (Miller, 1977; Chen and others, 2002).⁴⁷ Residential housing is a notable example. When short sale restrictions bind and common knowledge of a bubble is absent, even finite-lived assets can be subject to rational bubbles (Allen and others, 1993). More generally, arbitrage in the context of ‘stabilizing speculation’ is only riskless under restrictive assumptions: short sales are as seemless to implement as purchases (markets are complete); there are no funding, margin or liquidity constraints; information and transaction costs are negligible; and assets are perfectly substitutable (where two assets are highly correlated, an overpriced asset can be sold, and an underpriced asset purchased, so as to remove overall market risk).⁴⁸ However in a practical context, none of these conditions typically hold, especially during systemic liquidity shocks (as demonstrated in 1998 and 2008).

Another source of friction capable of amplifying bubbles stems from the ‘captive buying’ of securities in momentum-biased market capitalization-weighted financial benchmarks. Underlying constituents that rise most in price will see their benchmark weights increase irrespective of fundamentals,⁴⁹ inducing additional purchases from fund managers seeking to minimize benchmark tracking error.⁵⁰ As a case in point, the 1980s Heisei bubble saw Japan’s share of the MSCI World equity market capitalization soar from 21 percent in 1983 to 51 percent by 1989, while during the 1990s technology bubble, the technology sector weighting in the S&P500 rose from 5 percent in 1993 to 34 percent by 2000 (Jones, 2012).

Importantly, the captive buying phenomena is unlikely to abate given the (passive) benchmark-tracking exchange traded product industry has expanded at a decade-long annual growth rate of more than 20 percent (to US$2.5 trillion), a much faster rate than for other relatively non-benchmark constrained investors (i.e. alternative or absolute return managers).

Institutional Limits to Arbitrage

Institutional limits to arbitrage can also give rise to bubbles owing to the nature of incentives and principal-agent issues in the rapidly growing fund management industry. In the United States for instance, the proportion of stocks held by institutional investors hovered around 5 percent in the first half of the twentieth century; from 6 percent in 1952, the institutional share then rose to 45 percent in 1991, and 67 percent by 2010 (Allen and Gorton, 1993; Blume and Keim, 2012).⁵¹ While the growing presence of institutional investors might intuitively suggest markets would become more efficient over time (and thus less prone to bubbles), business risk and compensation practices in the delegated portfolio management industry can encourage institutional herding and ‘rational’ bubble riding. At the very least, incentives for investment managers can dissuade them from leaning against bubbles. Although much of the theory of asset pricing has until recently ignored the real world complication that individuals (asset owners) delegate virtually all their involvement in financial markets to professional intermediaries (asset managers), the institutional investment industry is anything but a frictionless pass-through to asset markets and the economy (Woolley, 2010).⁵² The associated principal-agent problems are still to be fully addressed.

Acting as agents, institutional fund managers are only likely to lean against bubbles where they have confidence in their ability to withstand redemptions from principals whilst waiting for a fundamental mispricing to correct. But because principals have imperfect knowledge about the ability of their investment agents—reflecting that managers operate in a high-noise, low-signal environment which makes it difficult to distinguish skill from luck—they may interpret a period of short-term underperformance against peers as a sign of manager incompetence.⁵³ As the relative underperformance grows, principals are likely to conclude their manager is unskilled and thus terminate them in favor of peers demonstrating outperformance by holding securities with strong momentum.⁵⁴ The high turnover that reflects short-term performance chasing appears to have increased over time: since the 1950s, the average holding period for a U.S. equity mutual fund has declined from 16 years to around 3 years, while annual portfolio turnover has increased from 16 percent to more than 100 percent (Bogle, 2011). As Jin (2005) points out, the “output short-termism” of investment agents can largely be explained by the “heat” they feel from the “input short-termism” of principals. Ironically, the pressure from principals to redeem on rational arbitrageurs who are long (or overweight) cheap but underperforming securities and short (or underweight) expensive but outperforming securities will typically be most intense just when the mispricing (and hence, prospective return) is greatest.⁵⁵

The impact of earlier price changes can be amplified by other institutional practices such as window dressing and fund name changes. Since holding poorly performing securities can be difficult to justify to impatient asset owners, regular reporting intervals encourage agents to ‘window dress’ their portfolios so as to appear to have shown good judgment in selecting popular securities (Lakonishok and others, 1991; Chevalier and Ellison, 1997). In order to attract client inflows, mutual funds have also resorted to name changes so as to give the impression of a close association with a fashionable sector (Cooper and others, 2005).

DeMarzo and others (2008) show that if utility is derived not on the basis of one’s own wealth, but rather relative wealth, then investors will herd into what they knowingly perceive to be a bubble for as long as others do; relative wealth concerns make them unwilling to trade against the crowd. Institutional investors may also infer information from the prior trades of highly-informed and respected peers and invest in the same direction—they disregard their private information due to the reputational risk of acting differently from other managers (Scharfstein and Stein, 1990; Cai and others, 2012). ‘Synchronization risk’ can further dissuade rational arbitrageurs from pricking a bubble because each investor is uncertain about when others will begin selling: since a single investor cannot bring down the market, coordination among a vast pool of arbitrageurs is required, and this might prove elusive for a long period (Abreu and Brunnermeier, 2002, 2003). At a minimum, these considerations reduce the amount of non-consensus risk that rational arbitrageurs might be willing to take.⁵⁶

Indeed relative performance has been established as the most important factor explaining fund survival and the continued employment of portfolio managers. In particular, because investors have a quicker tendency to ‘fire than hire,’ avoiding underperforming one’s peers in the short-term is more important than outperforming in the long-term if one’s aim is simply to survive as an institutional investment manager.⁵⁷ Feroli and others (2014, p. 10) emphasize that the high costs of underperforming peers introduces an important element of short-termism that generates negative spillover effects: “the concern of relative rankings in the payoff by the delegated agents injects an element of coordination in their portfolio choice that has the outward appearance of herding behavior.”

If principals appraise their agents on the basis of short-term performance vis-à-vis a momentum-biased benchmark, it can also be entirely rational—from the perspective of business and compensation risk—for agents to herd and knowingly participate in bubbles. Even if they suspect their benchmark to be overvalued, managers can take comfort in the belief they will be excused from poor absolute performance so long as their benchmark and peer group also perform poorly (Scharfstein and Stein, 1990; Rajan, 2005; Porter and Trifts, 2014). Herding during a bubble allows one to “share the blame” in the event of its subsequent implosion, in the same way that Keynes (1931) outlined the career risk associated with exercising prudent, non-consensus judgment: “A sound banker, alas, is not one who foresees danger and avoids it, but one who, when he is ruined, is ruined in a conventional and orthodox way along with his fellows, so that no one can really blame him.”⁵⁸ The attendant distortion of agent incentives can lead to highly inefficient and socially undesirable portfolios (Brennan, 1993; Maug and Naik, 1996; Stein, 2004; Brennan and Li, 2008).

Even in the ‘absolute return’ (benchmark agnostic) hedge fund industry, aggregate returns can increasingly be explained solely by the standard benchmark index for world equities. The correlation of hedge fund and world equity returns has almost doubled on a trend basis over the past two decades (gray line, Figure 3), while the excess returns to hedge funds (those not explained by the world equity benchmark, alternatively known as ‘alpha’) has collapsed on a trend basis from a high of 17 percent in 1996 to just 2 percent in recent years (black line, Figure 3). As the composition of investors (principals) in hedge fund vehicles has evolved from largely benchmark-agnostic high net worth investors in the 1990s to benchmark-conscious institutional investors over the past decade, so too have hedge fund returns more closely mirrored returns to benchmarks.

Benchmark Decomposition of Hedge Fund Returns

Citation: IMF Working Papers 2015, 027; 10.5089/9781475576207.001.A001

Source: Author’s estimates, BloombergNotes: Monthly returns measured over a rolling five year observation window. ‘Alpha’ is the intercept in the regression of hedge fund returns on the world equity benchmark. Hedge fund returns are based on the Hedge Fund Research Fund-Weighted Composite Index.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Benchmark Decomposition of Hedge Fund Returns

Citation: IMF Working Papers 2015, 027; 10.5089/9781475576207.001.A001

Source: Author’s estimates, BloombergNotes: Monthly returns measured over a rolling five year observation window. ‘Alpha’ is the intercept in the regression of hedge fund returns on the world equity benchmark. Hedge fund returns are based on the Hedge Fund Research Fund-Weighted Composite Index.

• Download Figure
• Download figure as PowerPoint slide

Benchmark Decomposition of Hedge Fund Returns

Citation: IMF Working Papers 2015, 027; 10.5089/9781475576207.001.A001

Source: Author’s estimates, BloombergNotes: Monthly returns measured over a rolling five year observation window. ‘Alpha’ is the intercept in the regression of hedge fund returns on the world equity benchmark. Hedge fund returns are based on the Hedge Fund Research Fund-Weighted Composite Index.

• Download Figure
• Download figure as PowerPoint slide

A key factor motivating herding by investment managers is that their performance is usually assessed, and fees paid, on an annual rather than multi-year basis—a time horizon where momentum tends to trump valuation as a driver of returns. Bubbles are often left unattacked by rational arbitrageurs operating open-ended funds (a fund structure favored by principals) because valuation-based arbitrage requires a long investment horizon that only closed-end funds can accommodate (Stein, 2004). Given that value-based strategies have often been found to take up to 3 to 5 years to generate meaningful outperformance (in contrast to momentum strategies which are most profitable with holding and look-back periods of 6 to 12 months), abstaining from bubble riding might simply entail too much business risk (Lakonishok and others, 1992; Lux, 1995; Dasgupta and others, 2011; Vayanos and Woolley, 2013). As Grantham (2012, p.1) observes from the perspective of a value-centric manager:

“The central truth of the investment business is that investment behavior is driven by career risk. The prime directive, as Keynes knew so well, is first and last to keep your job. To do this, he explained that you must never, ever be wrong on your own. To prevent this calamity, professional investors pay ruthless attention to what other investors in general are doing. The great majority ‘go with the flow,’ either completely or partially. This creates herding, or momentum, which drives prices far above or far below fair price. There are many other inefficiencies in market pricing, but this is by far the largest … Ignoring (long-term fundamental value) may be the correct response on the part of most market players, for ignoring the volatile up and down market moves and attempting to focus on the slower burning long-term reality is simply too dangerous in career terms. Missing a big move, however unjustified it may be by fundamentals, is to take very high risk of being fired. Career risk and the resulting herding it creates are likely to always dominate investing.”

In a different context, the asymmetry associated with limited liability and incentives in delegated fund management can also impart a rational bias for bubble riding. The convex payoff structure associated with limited liability (i.e., limited downside, unlimited upside) can generate a preference for risk taking, selling disaster insurance and riding bubbles (Allen and Gorton, 1993; Allen and Gale, 2000; Rajan, 2005; Stein, 2013b).⁵⁹ If fund managers participate in the formation of a bubble, they can earn large fees on both absolute performance and the management of a swelling asset pool. If the bubble deflates while they remain invested, managers generally do not have to repay the earlier performance fees, and will continue to earn management fees (albeit off a lower base).⁶⁰ Alternatively, managers may chose to close down the existing fund and re-open another, thus lowering the hurdle level on which future compensation will be based (analogous to ‘re-striking’ a call option).⁶¹

Finally, financial sector agents involved in the dissemination of information may also have an incentive to propagate partial truths (downplaying adverse information relative to market-friendly information) which can fuel bubbles, if their business model is based on transaction volumes which rise in a strong market. Prominent examples include deal originators, securities analysts, brokers and credit rating agencies. For instance, securities analysts might fear issuing downbeat assessments of the issuers they cover for any number of reasons (i.e., they could lose favor with management and be frozen out of future communication), and both the trading and investment banking business typically stand to benefit more from bullish than bearish recommendations due to the relative difficulties in monetizing downbeat recommendations.⁶² Further, since issuers of securities have both the ability and incentive to ‘ratings shop’ for the highest possible credit ratings, and rating agencies are overwhelmingly paid by issuers, rating agencies have an incentive to upwardly bias their credit assessments.⁶³

(Irrational) Behavioral Models

The central issue in behavioral finance has been to explain why investors make systematic and irrational errors of judgment. Rationality presupposes that investors calculate the expected utility of each investment as a probability-weighted sum of utility outcomes, and then choose the option with the highest expected utility.⁶⁴ But if investors are afflicted by cognitive biases and do not process information rationally, behavioral models may offer a better guide to understanding asset price dynamics (Thaler, 2005). As far back as the 1930s, Keynes (1936, p.144) asserted that rather than rational “mathematical expectation” forming the basis of asset price dynamics, it was instead a multitude of irrational emotional influences: “Our decision making … can only be taken as a result of animal spirits—of a spontaneous urge to action rather than inaction, and not as the outcome of a weighted average of quantitative benefits multiplied by quantitative probabilities.” Yet for many years behavioral insights into asset pricing vanished from view under the influence of the efficient markets hypothesis and its macroeconomic cousin, rational expectations.

Behavioral explanations for market inefficiency and speculative bubbles derive from the principal that, due in part to their evolutionary origin, expectational errors do not cancel out across investors—they are systematic rather than idiosyncratic.⁶⁵ Market prices can become chronically distorted by widely shared, self-reinforcing misconceptions (Soros, 1987; De Bondt, 2012). Akerlof and Shiller (2009) argue the recurring nature of bubbles and panics through history suggests an irrational social-psychological explanation. Invoking the assertion by Keynes (1936) that “the facts of the existing situation enter disproportionately into the formation of our long-term expectations; our usual practice being to take the current situation and project it into the future,” Greenspan (2002, p. 6–7) noted that, “Investors too often exaggerate the extent of the improvement in economic fundamentals. Human psychology being what it is, bubbles tend to feed on themselves, and booms in their later stages are often supported by implausible projections of potential demand.” In their sweeping study of booms and busts spanning eight centuries, Reinhart and Rogoff conclude (2009b, p. 80): “The pervasive view that ‘this time is different’ is precisely why this time usually is not different, and catastrophe eventually strikes again.” The actions of investors falling prey to behavioral errors are typically painted as consistent with financially ruinous behavior.

In recent times, systematic departures from rationality have been established along two complementary lines in financial economics: experimental studies and investor surveys. Experimental studies suggest the rationality of judgment is bounded or constrained,⁶⁶ and emphasize the strong behavioral tendency of investors to herd and display overconfidence after prior vindication. Separately, survey-based studies of investor return expectations reveal a strong tendency toward extrapolation of the recent past, with direct implications for speculative bubbles and the ‘momentum effect’ in asset prices.

Among the vast number of documented belief biases, the most important in the context of speculative bubbles relate to the tendency for market participants to herd, extrapolate, and fall victim to overconfidence. A typical process is as follows. The impact of a positive news shock to fundamentals generates a rational increase in asset prices. But because of anchoring and the tendency for investors to dispense with priors only slowly in the face of new information, the price increase is initially gradual and diffused, until such time the smooth upward trend attracts the attention of so-called ‘positive feedback traders.’⁶⁷ The high Sharpe ratio generated by the low-volatility uptrend begins to be extrapolated with an increasing number of positive feedback traders piling onto the trend (Cutler and others, 1991; Barsky and Delong, 1993; Lakonishok and others, 1994). With the serial correlation in returns now firmly in place, new entrants to the market with less than full confidence in their market timing ability opt to move with the herd in the belief that competitors must be acting upon superior information to their own (Scharfstein and Stein, 1990; Shiller, 2002). Reflecting expectations of a structural shift, economic commentators extrapolate the recent improvement in underlying fundamentals to persist into the indefinite future. As prices continue to rise, validation breeds overconfidence for traders and analysts alike. In certain circumstances, the positive price momentum can even reinforce the improving fundamentals that prices are supposed to reflect, such as a tightening in credit spreads lowering the interest charge (thus raising the profitability) for indebted companies. In this sense, investors can create their own sense of reality in a “reflexive” feedback process (Soros, 1987).

As serial correlation in asset returns has been documented across various asset classes and time periods, it is likely that positive feedback behavioral dynamics play at least some role in asset price determination over short to medium time horizons (Cutler and others, 1990; De Long and others, 1990b; Barberis and others, 1998, Hong and Stein, 1999). Unlike most other anomalies, there are no candidate explanations from the rational expectations school which do a sound job of explaining its pervasive existence; for this reason, momentum has been declared the “premier anomaly” in asset markets (Fama and French, 2008, p. 1653).⁶⁸ Surveys of expected returns also reveal that investors tend to anticipate strong returns in the future when returns have been strong in the recent past—despite objective mathematical estimates of expected returns (yields) having been compressed by the same rapid price rises.⁶⁹ In other words, after a boom in asset prices, subjective expectations of returns tend to be high while objective expected returns tend to be low (a finding established across stocks, bonds, exchange rates, and real estate).⁷⁰ On balance, evidence from surveys of investors’ expected returns appears “directly at odds with the implications of standard asset price theory” (Williams, 2013, p. 6). Cochrane (2011, p. 1068) similarly suggests surveys of expected returns pose an “unsettling” challenge to rational market efficiency. More constructively, Barberis and others (2013, p. 1) conclude, “Survey evidence does not need to be seen as an inconvenient obstacle to understanding the market; on the contrary, it is consistent with the facts about prices and returns, and may be the key to understanding them.”

These two behavioral concepts—extrapolation and the negative correlation between subjective surveys and ‘objective’ yield-based measures of expected returns—are illustrated in Figure 4. The left hand side panel depicts the relationship between realized annual U.S. stock returns over the prior five years (vertical axis), and expectation of annualized returns over the following ten years from a survey of U.S. based Chief Financial Officers (horizontal axis).⁷¹ These data suggest high (low) past returns are associated with high (low) expectations of future returns. The right hand side panel shows that an ‘objective’ earnings yield-based measure of expected returns (vertical axis) is negatively correlated to survey-based expectations of returns (horizontal axis):⁷² when objective measures of future stock returns are low, investors expect high returns in the future (and vice versa).⁷³ Given the ease with which extrapolative expectations (including among experienced investment professionals) can lead to speculative bubbles, the broader findings from survey data that ‘extrapolative expectations trump rational expectations’ has potentially important policy implications.

Survey-based (Subjective) Expected Returns vs. Past Returns (left) and ‘Objective’ Expected Returns (right)

Citation: IMF Working Papers 2015, 027; 10.5089/9781475576207.001.A001

Source: IMF StaffNotes: Survey return expectations for the U.S. stock market based on an average of 352 U.S. Chief Financial Officers, across 51 quarterly observations from September 2000 to March 2013, as reported in Graham and Harvey (2013). The earnings yield is shown as the inverse of the cyclically adjusted price-earnings ratio.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Survey-based (Subjective) Expected Returns vs. Past Returns (left) and ‘Objective’ Expected Returns (right)

Citation: IMF Working Papers 2015, 027; 10.5089/9781475576207.001.A001

Source: IMF StaffNotes: Survey return expectations for the U.S. stock market based on an average of 352 U.S. Chief Financial Officers, across 51 quarterly observations from September 2000 to March 2013, as reported in Graham and Harvey (2013). The earnings yield is shown as the inverse of the cyclically adjusted price-earnings ratio.

• Download Figure
• Download figure as PowerPoint slide

Survey-based (Subjective) Expected Returns vs. Past Returns (left) and ‘Objective’ Expected Returns (right)

Citation: IMF Working Papers 2015, 027; 10.5089/9781475576207.001.A001

Source: IMF StaffNotes: Survey return expectations for the U.S. stock market based on an average of 352 U.S. Chief Financial Officers, across 51 quarterly observations from September 2000 to March 2013, as reported in Graham and Harvey (2013). The earnings yield is shown as the inverse of the cyclically adjusted price-earnings ratio.

• Download Figure
• Download figure as PowerPoint slide

Limits to Learning

The main policy implication from the limits to learning-based rational bubble model is that investors should have ready access to all the information they need in assessing investment opportunities; no time-varying countercyclical policy response is warranted. As Stout (1997) points out, rational investors have little need for a paternalistic capital market policy, but require only more accurate and timely information. Authorities should therefore concentrate their efforts on strengthening corporate reporting standards and prosecuting cases of fraud and the dissemination of misleading information. Bernanke (2002a) suggests that policy makers should focus efforts not on leaning against the wind, but rather in designing initiatives to “improve the financial literacy and competence of investors.” Kirchner (2009, p. 6) similarly argues that inefficiencies associated with limits to learning provide “an argument for improving the information content and completeness of markets, not for displacing or overriding them” with countercyclical policy settings.

While few would dispute the notion that more accurate, timely and readily accessible information constitutes a societal good, the evidence that it has or could in the future offer a robust defense against asset bubbles is unpersuasive. There is little to suggest that the incidence or severity of asset booms and busts has diminished over time, despite (i) a meaningful improvement in the information disclosure practices in the financial sector; (ii) technological advances in communication which have facilitated the rapid and widespread dissemination of information throughout society; and (iii) the growth in the relative share of financial assets managed by institutional investors, who can fairly easily overcome constraints in obtaining timely and accurate information. Shiller (2000) even suggests that the free flow of information—aided and abetted by the news media which he labels “generators of attention cascades”—which acts to intensify (rather than mitigate) self-reinforcing feedback loops, investor herding, financial instability and asset bubbles.⁷⁴ Ensuring that information is more accurate, timely and widely available information would seem a natural place to begin, but not finish, a policy of ‘bubble defense.’

Frictional Limits to Arbitrage

If rational bubbles are largely the result of frictional limits to arbitrage, authorities should focus on fostering an environment in which product offerings and tradable securities (including hedging and shorting instruments) can accommodate the needs of investors with widely varying objectives and risk appetites. A system in which investors have a greater array of instruments through which to bear and diversify risk is clearly desirable, particularly in countries where economic efficiency is hampered by a lack of depth in asset markets. Tradable securities that are simple, transparent, have little embedded leverage, directly hedge economic risk (Shiller, 2003), and are amenable to trading on listed exchanges in the full glare of regulators and the investing public, have a valuable role in this regard. However, events leading up to and during the global financial crisis caution that the introduction of complex, opaque and highly leveraged financial products traded on over-the-counter markets and designed primarily for speculative purposes (rather than the hedging of economic risk) can be more a hindrance than help to economic growth and financial stability.⁷⁵ It is worth recalling that record levels of trading activity and capital market issuance prior to 2008 provided no insurance against the crisis. This suggests that while initiatives designed to ‘complete markets’ and eliminate frictions are welcome in principal, quality rather than quantity considerations need to be at the forefront of the financial deepening process.

Institutional Limits to Arbitrage

A broader suite of policy responses is likely to be required if, as I argue in this paper, incentives in the institutional investment management industry are a key factor in the growth and persistence of asset bubbles. While time-varying responses (macroprudential and monetary policy) are applicable tools once a bubble is underway, preventing their emergence in the first instance will likely necessitate time-invariant measures which alter the underlying incentives of institutional investment managers. In this context, reforms to contract and financial benchmark design could well be the natural place to start.

• Contract design. Reform proposals should reflect the notion that agents respond to the incentive structure imposed on them by principals. It is unreasonable to expect investment managers to act countercyclically (in a manner that appeals to the broader public good) when their incentive structure can instead reward, or at a minimum, fail to discourage, short-term herding behavior. A number of issues could be considered in the formulation of principal-agent contract design:

  • A stronger emphasis should be placed on long-term performance appraisal, including incentives for countercyclical investment (for instance, by weighting manager performance more heavily in difficult market conditions). This emphasis would encourage agents to lean against, rather than with, large mispricings (which can take considerable time to correct), and also better align the interests of agents and principals with the public good;

  • To discourage redemptions over short time horizons, investment fund exit sales loads could be formulated on the basis of a sliding time scale where loads reduce as a function of investment holding period. Current practice (which makes no contribution to financial stability) is that sales loads are staggered exclusively on the basis of account size;

  • Manager performance should be assessed in ways that do not exclusively involve price momentum-biased benchmarks. For instance, pension and insurance funds could place more emphasis on the degree to which a manager hedges their liability structure and thus their true source of economic risk;

  • The asymmetry that arises in agent compensation (where agents own the upside but not downside) could be addressed with more thorough use of multi-year clawback provisions (such as those now in effect in the banking industry) and the use of high watermarks;

  • In the case of financial information providers whose business model thrives on issuing upbeat assessments of issuer prospects (i.e., brokers and credit rating agencies), there should be stronger emphasis on accountability.⁷⁶

• Benchmarks. Initiatives currently aimed at reforming financial benchmarks (in the wake of interest rate and foreign exchange fixing scandals) could be broadened to examine the financial stability implications of price-based market capitalization benchmarks. These impart a procyclical bias to financial markets by forcing benchmark-constrained managers to increase (reduce) holdings of securities which have recently shown strong (weak) price momentum, irrespective of fundamentals.⁷⁷ In the case of debt markets, market-capitalization based benchmarks also force captive buying of issuers that have recently increased the amount of tradable debt outstanding. Alternative non-price based benchmark formations, for instance where objective measures of risk premia (Jones, 2013) or company fundamentals (Arnott and others, 2008) form the basis of constitute weights, can mitigate herding and be less prone to excessive concentration arising from market fads which result in heavy exposure to individual companies, sectors or countries. By forcing rebalancing to be conducted in a countercyclical manner, these approaches raise the prospect of ameliorating (rather than contributing to) financial instability, and possibly generating superior longer-term returns compared to price-based benchmarks (Figure 5).

• Transaction Taxes. Securities transaction taxes (STT) in various guises have long been proposed to both discourage speculative investment and increase public revenues in a fair and progressive manner.^{78, 79} While lower transaction costs are undeniably welcome in markets that efficiently intermediate long-term savings and investment, the case becomes less clear in markets dominated by short-term herding and speculation. Woolley (2010, p. 133) suggests that in inefficient markets, “lowering the frictional costs of trading opens the door to short-termism and momentum trading which distorts prices … both contribute to the loss of social utility.”⁸⁰ However, all STT will impede price discovery to some degree by discouraging new information from being incorporated in prices, thus reducing the informational efficiency of markets and raising the cost of capital for issuers. Rogoff (2011) argues the latter impact will in turn lower corporate investment (with no concomitant increase in financial stability) and thus prove counterproductive. More specifically, a blanket STT does not distinguish motives or holding periods (i.e. short-term momentum-based speculation from long-term countercyclical investment or standard portfolio rebalancing) and so is a crude response to mitigating the threat of asset bubbles. Nor is there empirical evidence to suggest that such measures reduce short-term price volatility (see Habermeier and Kirilenko, 2003; and Matheson, 2011).⁸¹ Taxes based on holding periods and the frequency of portfolio turnover are more directly targeted toward discouraging short-term speculation (for instance, taxes can be set as a declining function of holding period, and a rising function of portfolio turnover).⁸² Taxes on derivatives (which have built-in leverage) that are used for non-hedging purposes could be more thoroughly examined, though instruments that explicitly target leverage, such as higher margin and collateral requirements, and more generally, the removal of tax subsidies on corporate and financial sector balance sheet debt, could be more effective in addressing leverage and procyclicality (Matheson, 2011).⁸³ Many tax codes allow interest payments to be deducted as a business expense, yet there is no equivalent allowance for payments to equity investors. The application of tax breaks on debt but not equity is counterintuitive from the perspective of financial stability.

• Time-varying Macroprudential and Monetary Policies. If incentives and principal-agent frictions in the institutional investment industry give rise to unjustifiably low risk premia and sharp increases in leverage which could pose a systemic threat to economic stability, macroprudential policies and tighter monetary policy will have a role to play in warding off bubbles. Given they can be highly targeted, macroprudential measures should offer the first line of defense. But just as monetary policy can contribute to easing financial conditions by compressing risk premia—a central aim of unconventional monetary policy in recent years—logic dictates that monetary policy can (and should) also lean in the opposite direction when circumstances warrant (tighter monetary policy is the only instrument that can instantaneously “get into all the cracks”). Nevertheless my reading of the evidence suggests that in the absence of fundamental reforms to incentives, contracts and benchmarks, both macroprudential and monetary policies will be left to deal with the symptoms (rather than underlying causes) of deeper flaws in the financial system. If the institutional contribution to asset bubbles is left unaddressed, and the institutional investment industry continues to grow in absolute terms and as a share of total investment activity, this raises the unappealing prospect of indefinitely recurring waves of boom and bust in the future.

Relative 10-year Annualized Outperformance of Fundamental-based Indices vs. Industry Standard Market Capitalization-based Indices

Citation: IMF Working Papers 2015, 027; 10.5089/9781475576207.001.A001

Source: IMF Staff, Research AffiliatesNotes: Average annualized return differential between fundamental-based indices and industry standard market capitalization-weighted benchmarks. Fundamental Index constituents are weighted using a composite of fundamental factors, including cash payouts, free cash flow, sales and book value of asset (prices and market values are not determinants of index weights). The 10-year performance assessment period is up to July 2014.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Relative 10-year Annualized Outperformance of Fundamental-based Indices vs. Industry Standard Market Capitalization-based Indices

Citation: IMF Working Papers 2015, 027; 10.5089/9781475576207.001.A001

Source: IMF Staff, Research AffiliatesNotes: Average annualized return differential between fundamental-based indices and industry standard market capitalization-weighted benchmarks. Fundamental Index constituents are weighted using a composite of fundamental factors, including cash payouts, free cash flow, sales and book value of asset (prices and market values are not determinants of index weights). The 10-year performance assessment period is up to July 2014.

• Download Figure
• Download figure as PowerPoint slide

Relative 10-year Annualized Outperformance of Fundamental-based Indices vs. Industry Standard Market Capitalization-based Indices

Citation: IMF Working Papers 2015, 027; 10.5089/9781475576207.001.A001

Source: IMF Staff, Research AffiliatesNotes: Average annualized return differential between fundamental-based indices and industry standard market capitalization-weighted benchmarks. Fundamental Index constituents are weighted using a composite of fundamental factors, including cash payouts, free cash flow, sales and book value of asset (prices and market values are not determinants of index weights). The 10-year performance assessment period is up to July 2014.

• Download Figure
• Download figure as PowerPoint slide

Behavioral Models

Behavioral models offer tremendous promise in enriching our understanding of various aspects of economic life. And survey data strongly suggests that policy makers should stay attuned to the possibility of large scale behavioral errors in the pricing of assets. To the extent that (i) behavioral errors combine with other explanations to amplify bubbles, and (ii) there is good reason to suspect that a reversal in irrationally low risk premia could exert a large negative impact on the economy, a full suite of policy responses (the proverbial “kitchen sink”) may be required. These would include the aforementioned time-invariant measures to address information asymmetries, the incompleteness of markets, and incentives and principal-agent frictions in the institutional fund management industry. If fickle and irrational behavior is the driving force behind asset prices, it also raises questions over whether entities should be able to lever up or consume out of unrealized capital gains. On top of such measures, an even stronger emphasis on countercyclical macroprudential and monetary measures would likely be required in order to temper the impact of extrapolation and overconfidence errors that can characterize late cycle bubbles.

Nevertheless, when viewed in isolation, behavioral models of widespread irrationality tend to leave open key issues. For instance: if all investors are irrational due to our biological hardwiring, who is on the other side of the irrational trading activity? If widespread irrationality leads to financially ruinous behavior, shouldn’t the process of natural selection wipe out such investors, and thus bias markets toward a more rational and efficient state over time? If irrationality is so pervasive, shouldn’t it be straightforward to outperform the market?⁸⁴ Finally, how is it possible to defend against the same criticism leveled at ‘rational’ explanations of asset pricing that the pliability of behavioral finance is both its key strength and Achilles heel? Because behavioral explanations have proliferated in recent times, they can be retrospectively selected (from what is now a large menu) to justify nearly any empirical finding.

On balance, these considerations suggest that widespread investor irrationality is unlikely to be the only factor at play in the formation and persistence of asset bubbles.