Now, consider the development of derivatives pricing in the early 1970s. The key insight leading to the Merton-Black-Scholes (MBS) (Merton  and Black and Scholes ) formula was that continuous trading, creating “dynamically complete” markets, makes it possible to price options on a stock, with no need to predict the future stock price. In fact, MBS demonstrated that the expected return on a stock plays no role in the pricing of options, making it possible to manage risk (some apparently even thought, conquer risk), while sidestepping the perilous exercise of predicting whether the price of IBM stock would go up by 2%, 5%, or 8% over the next week, month, or year.9
Taken one step further, if companies—rather than stocks—are the fundamental unit of analysis, corporate debt and equity could be structurally related to one another via the same principles. With this understanding, Merton (1974) produced a model for pricing corporate debt, a truly extraordinary insight that created a wide following among researchers, who could suddenly address an endless array of new applications.
The theory now in place, derivatives markets matured quickly. The real world—powered by a healthy dose of profit motive—saw an opening at the table with the development of portfolio insurance. But as theory—based on continuous markets—ran headlong into implementation challenges in discontinuous markets, the strategies sharply exceeded their capacity, contributing to the 1987 market crash.
According to the new models, related securities ought to move in related ways. In the messy real world of financial markets, theory fell short.10 Repeated departures from these relative-pricing relationships were suggestive of relative-value trades. All of a sudden, arbitrage profit-taking opportunities could be identified and harvested over and over again. In a snowball effect, more academic papers were addressing relative-pricing identities, and more practitioners set up shop to take advantage of these alpha opportunities.
The supply of ideas can sometimes create its own demand, until the wave comes crashing down. Long Term Capital Management (LTCM) famously collapsed in 1998, losing billions for its investors, billions more in ripple effects through its trade counterparties, and allegedly putting our global financial system at significant risk. As in the case of the more-recent global financial crisis, intervention ostensibly averted a full-scale disaster, but we’ll never know whether the damage of the crisis would have exceeded the damage of the intervention. We do know that the interventions, in both cases, prevented Schumpeter’s “creative destruction,” purging the economy of the reckless facilitators of LTCM’s hyper-leverage, some of whom went on to facilitate the 2008 global financial crisis.11
Looking back, it seems that a deep and robust insight had taken on a life of its own, proliferating derivative results (forgive the pun) in the academic literature. These results were then excessively and blindly adopted in the practice of finance, with occasionally disastrous repercussions. While not perfectly related to the initial theory on dynamic replication of options, the work on pricing other types of derivatives, such as mortgage-backed securities and credit default swaps, contributed to the heroic mindset that culminated in the global meltdown of the late 2000s.
From Day Traders to Backtesters: The Empirical Finance Bubble
The 1990s gave us far more than just famous hedge funds “blowing up.” The decade paved the way for the tech bubble, democratizing computing power and allowing unprecedented access to data. Market participants convinced themselves a new world had arrived and that markets need not concern themselves with antiquated concepts of valuation, such as price-to-book or whatever fundamental they once deemed relevant. At the peak of the tech bubble, nearly 15% of the Russell 1000 Growth Index consisted of companies that had not paid dividends and had earned virtually no profits in at least five years!12 The bubble eventually burst, sensible valuations reasserted themselves, and most of the day-trading class went back to their day jobs.
For many in academia and in practice, the tech bubble and its extraordinary collapse called for a deep reassessment: Perhaps markets were not perfectly efficient! Human psychology could create deviations from textbook predictions. Thus—and this is the crucial intellectual turning point—predictable excess returns could be generated by systematically “outsmarting” the inherently flawed investing masses.
In the meantime, empirical finance work had been gaining a strong reputation, perhaps even a superiority relative to theory, with the groundbreaking work of Fama and French (1992), among others. By the 2000s, computers and data had become cheap enough that finance professors, their graduate students, and quantitative practitioners alike gained universal access to market data, the precise premise on which the tech bubble had been predicated. And so, an entire generation of academe has gone on to substitute capital (the great “backtest machine”) at the expense of labor (carefully designed theories).
So far, the great backtest machine has blessed our industry with over 500 factors; according to the literature, they all work, and almost all have statistical significance over their respective period of study. Fama–French regressions demonstrate they are not just clones of the value, size, and momentum factors. The implication is that we can turn our investments over to a roster of factor strategies, all with fabulous backtested results. To be sure, some in academia and in the practitioner community resisted the temptation of data mining. For these skeptics, value, carry, momentum, and a handful of other factors remain the sole robust sources of excess returns to which investors likely ought to trust their investment dollars.
Recently, the Fama–French (1992) model has been extended to a fifth factor (low investment) (Fama and French, 2014), a testament to the reluctance of the authors to proliferate at the same pace as their disciples. The world at large has been less cautious. Every new cohort of graduate students must complete dissertations and generate significant t-stats along the way. In parallel, the purveyors of investment products must find new products to launch. Yet again, supply is creating its own demand, and a new intellectual bubble, along with a product proliferation bubble, has formed on the back of empirical finance.
In 2016, we published a series of papers, beginning with “How Can ‘Smart Beta’ Go Horribly Wrong?” In the series (Arnott, et al.  and Arnott, Beck, and Kalesnik [2016a,b]), we point out that—at least until 2016—not one of the 500-odd papers in support of new factors had bothered to ask a rather fundamental question: Did my factor (or anomaly or strategy) benefit from a tail wind of rising relative valuations, which created an illusion of alpha, and which may reverse in the case of mean reversion in valuations? We point out that some of these factors owe much (or all) of their past efficacy to rising relative valuations. We point out that some of the most popular factors worked, on average, less than half as well after publication as before. And we point out that valuation is powerfully predictive of future factor returns.
In 2017, we are publishing a series of papers (Arnott, Kalesnik, and Wu  and Arnott, Clements, and Kalesnik  with two more articles forthcoming) titled “Alice in Factorland.” In this series, we point out that factor returns earned in mutual funds are typically much smaller than the theoretical returns of academic long–short paper portfolios (although with certain exceptions, such as the size factor); that factors cannot reliably replicate other strategies without introducing higher turnover and risk, and lower capacity and returns; that factors can be used to predict which mutual funds will perform well, by buying the out-of-favor, cheap factors; and that the momentum factor has special challenges that have contributed to momentum managers’ general failure to add value.
The eventual collapse of the factor research bubble—and factor product proliferation—may cause collateral damage among respected academics, tarnishing their reputations; among practitioners who offer factor products based on backtests without careful consideration of factor valuation; and among investors who entrust their capital to the less rigorous operators of the great backtest machine.
Sifting through the Rubble
Although the negative implications of academic bubbles are a significant part of our narrative, the insight at the core of most bubbles generally has merit. Remember the tech bubble. At its core was a belief that cheap computing power and unprecedented access to data would revolutionize the way we live and work. That belief has been proven right. Some of the darlings of the tech bubble have indeed revolutionized the world of today. So let’s not throw the baby out with the bath water. Instead, we can extract a few long-shelf-life principles and put them to work in the service of investors:
- Modern Portfolio Theory. Diversification (and careful portfolio construction, more generally) can dramatically improve investment outcomes, and may be the closest example of a free lunch we’ve come across.
- Derivatives Pricing. Today, derivatives markets are mature to a degree unimaginable a generation ago, affording investors around the world access to far broader investment opportunities than ever before.
- Tech Bubble. The lessons of the tech bubble are lessons in humanity and humility. An underlying investment thesis may be right, but investors tend to be swayed to irrational extremes. In the end, valuations matter.
- Empirical Finance. Factor research has made our reality richer than many were even willing to consider only 20 years ago. Excess returns can be tied to “factors” that go beyond the market portfolio. Value, size, and carry, and perhaps a few others, have proved to be robust investment strategies, both in backtests and in live portfolios. Momentum seems powerful, if we can overcome the trading costs. Referring to the principle that “valuations matter,” the expected excess returns associated with each of these strategies will tend to vary with valuations, and likely with macroeconomic conditions.