Illustration by II
Recently I’ve been pondering the continued outperformance of growth strategies versus value — and this has led me to some startling discoveries.
Since the 2008 crisis, the Standard & Poor’s 500 Growth Total Return Index has beaten its Value Total Return counterpart in eight of 11 years, including 2019 for the year to date. Growth has put up average annualized returns of 9.4 percent in the decade since the crisis, versus just 5.6 percent for value. Worse, value has suffered through three negative years during that span, whereas the growth index has not experienced a single one.
A similar pattern of outperformance has emerged in our private equity portfolio at Texas Municipal Retirement System, where our growth equity internal rate of return since inception is about 27.8 percent. The more value-oriented buyout book has posted an IRR of just 11.5 percent over the same period.
There are likely multiple cyclical factors at play that have provided strong tailwinds for growth strategies, including simple reversion to the mean. After all, from 2000 through 2008, value outperformed substantially, winning every year but one, so maybe it’s finally growth’s turn to shine again.
However, the anomalous persistence of growth over value actually predates this. If we start in 1999 and look at the 20-year period that includes market data from both the dot-com collapse and the credit crisis, growth beats value again — albeit more modestly — annualizing at 6.1 percent, compared with 6.0 percent over those two decades. Going back even further, to the inception of the S&P Growth and Value indexes in 1994, we can see that growth has beaten value over 25 years by nearly 150 basis points annually, 10.3 percent to 8.8 percent.
If we analyze the Russell 3000 Growth and Value indexes, we see the same story. Over 30 years growth has posted returns of 10.2 percent, compared with the 9.9 percent gains posted by value stocks. I was even able to find periods beginning in the 1970s when growth outperformed over long holds. Admittedly, there’s an element of backfill bias in cherry-picking periods that show growth outperforming, but those periods when value has won have clearly been few and far between compared with the dominant performance of growth.
To someone steeped in the classical financial learnings of giants from Graham and Dodd to Fama and French, it seems value should outperform over long enough time horizons, full stop. Evidence of 20-, 30-, and 40-year periods where it did not runs contrary to what I had previously believed. Growth should have only short spurts of momentum-driven outperformance; four decades of outperformance just shouldn’t happen. If value might not work in my lifetime, let alone my career, is it really a durable factor? Could there be something more structural at work, as opposed to merely cyclical causes?
To try to resolve this conundrum, let’s look at a few other topics that may, at first blush, appear to be unrelated.
In earlier roles, I worked as an employee, service provider, researcher, and allocator at hedge funds. And I experienced firsthand what many have noted in articles and research papers over the past several years: the continued decline of hedge fund returns. Just looking at nominal returns of the HFRI Fund Weighted Composite Index, we can see that from 1990 to 2000, hedge funds generated massive returns, annualizing at 18.3 percent and beating the S&P 500’s 16.5 percent average gain by almost 200 basis points. Since then, however, the tables have turned, and hedge funds have eked out only a 5.2 percent compound rate of return, slightly trailing the 5.6 percent gain produced by stocks since the turn of the century.
Clearly, lower equity returns have been a drag on performance, but hedge funds have done even worse than would be expected based on the stock market alone. To tell the whole story, we have to look at multifactor alpha, or the excess return above and beyond all other exposures that hedge funds can take, including capitalization, momentum, and other style tilts. When these considerations are properly accounted for, the story doesn’t look any better. According to research from hedge fund analytics provider Novus, over the past decade average hedge fund alpha using its six-factor model has fallen steadily, from consistently positive before the crisis to basically zero for several years afterward, and now to persistently negative.
Some articles place the blame for this alpha erosion squarely on the shoulders of large institutional allocators, whose enormous investments into the asset class over the past 20 years may have caused the industry to exceed its carrying capacity, arbitraging alpha away as they piled in. Undoubtedly this influx of capital has been part of the equation; more assets increase efficiency, and efficiency is the enemy of alpha. But once again I couldn’t shake the persistent feeling that this formula may be incomplete. What else was I missing?
Then I read another article on the decline of alpha — but this one referenced Warren Buffett, the Oracle of Omaha. Although he boasts an unparalleled track record of beating the S&P over nearly five decades, he has struggled more recently to extend his stellar performance. From 1965 until 2008, Berkshire Hathaway, the investment vehicle controlled by Buffett, beat the S&P 500 by more than 10 percent per annum, a truly remarkable feat.
Since then, however, Berkshire Hathaway’s performance has been far more pedestrian, actually trailing the broader market by more than 100 basis points — 7.3 percent versus 8.4 percent in the S&P’s favor. Plotting the decline of the excess returns of Berkshire Hathaway Class A shares versus the market yields a familiar-looking figure.
And this comes despite outflows from active, not because of inflows to it. According to information from research firm Morningstar, active mutual funds have experienced outflows every year since 2006. It estimates that nearly 75 percent of publicly traded stocks were owned by active funds back in 2009. Today that number has fallen to roughly half as passive investment giants like Vanguard and BlackRock have gobbled up market share. Clearly, flows can’t be the only cause of alpha erosion.
Nor has this underperformance appeared in the stock market alone. More research from Morningstar shows that over the past 16 years, actively managed bond funds have disappointed as well. Despite most actually outperforming their indexes on a gross return basis, once fees were accounted for, out of the 25 categories of bond funds investigated, median net returns trailed the benchmarks in all of them. Make no mistake about it — we are indeed witnessing the disappearance of alpha.
I’ve always believed that alpha migrates inexorably toward beta over time. I think the process works something like this: Initially, a select few individuals adept at pattern recognition are able to identify and implement a highly successful investment strategy. The process becomes more institutionalized as others at the firm are taught how to do it, and the business grows. Competitors externally take note as they try to chase the returns put up by the sector’s top performer.
Perhaps a few lieutenants then spin out on their own, taking their inside knowledge of the approach with them, and begin to compete with their former mentors. More and more market participants begin to imitate the strategy, academics write about it, and the skill becomes more broadly understood. Eventually, it becomes widely implemented — and ultimately turns into beta, whereby cost becomes the predominant differentiator.
This is happening faster and faster today. Perhaps there is something more structural driving the rapid acceleration in the decay function of alpha. Maybe something is essentially shortening the half-life of alpha wherever it is found, turning alpha to beta faster than has historically happened. This got me thinking about the rise of computing power and data ubiquity.
Much has been written about how artificial intelligence and machine learning are expected to revolutionize investing. Frankly, a lot of that has been pure hype. But what has been lost in that noise is the undeniable fact that the massive, muscular increase in raw computing power has automated, and obviated, most of the previously manual aggregation and analysis of financial data. Since the 1960s, Moore’s law has meant that the number of transistors on an integrated circuit has doubled roughly every 18 months or so for almost four decades. And although that rate of growth has slowed today, it still means that the most powerful semiconductor chips on the market now have nearly 25 billion transistors crammed onto them, resulting in an exponential increase in computing power.
We, on the other hand, simply cannot compete with this computational supremacy. Our human brains, about 1,260 cubic centimeters on average, have been essentially unchanged for nearly 200,000 years. We literally cannot keep up. And this does not even begin to address the numerous cognitive heuristics and biases that make us less than perfectly rational processors of data to begin with. Computers, however, are highly proficient in the synthesis and analysis of structured data.
And in a poignant example of supply creating its own demand — largely because of this increase in computing power — the amount of data available to feed these machines has exploded in nearly perfect lockstep with that increase. Today, as a society, we create roughly 2.5 exabytes of new data each day. For perspective, an exabyte is one quintillion bytes, which is a one with 18 zeroes after it. Put another way, one exabyte is one billion gigabytes, a number more familiar to most of us. My current iPhone has a storage capacity of roughly 125 gigabytes, so I would need 8 million iPhone 7s to store just one exabyte of data, and 20 million of them to warehouse the new data created every day. Data is now ubiquitous.
Although not all of this data is equally valuable — yes, Netflix knows you binge-watched “The Kardashians” last week! — the amount of information available on liquid financial markets has proliferated simultaneously. For instance, there are now about 3.7 million different indexes available on publicly traded assets globally, and more than 400,000 created in 2018 alone. Definitionally, the finer and finer you can slice beta, the harder and harder it is to find alpha.
More than 99 percent of all the data in existence has been created since 2007 — which corresponds to nearly the same date that this alpha erosion pattern started being observed across markets. Again, though I’m sure there are also powerful cyclical factors at work in all of these effects, I’m not convinced that the timing is merely coincidental, either. What if it is actually different this time?
You see, the computers are now consistently winning in other intellectual domains as well. Indeed, in 2016, Google’s AlphaGo program beat world champion Go player Lee Sedol in four games out of five. With more potential game configurations than estimated atoms in the known universe, Go is a very complicated game — ten raised to the power 100 times more complex than chess — but it’s still a board game with fixed rules. AlphaGo’s decisive victory, one year after sweeping European champion Fan Hui five games to none, means computers now reign supreme in all existing board games. Experts have long said Go would be the last game to, well, go, and it happened about ten years sooner than predicted.
So perhaps it makes sense that computers are winning in arbitrage-style hedge fund strategies, where an asset worth 95 cents is bought in one market and a quasi-identical one is sold in another market for $1.00. Most of those traditional hedge fund strategies are rules-based, and if anything, data ubiquity means computers have a huge advantage in parsing ever-increasing amounts of data not only to find that often temporary discount, but also to step in front of slower investors for a smaller spread if necessary, essentially beating them to the punch.
Which, come to think of it, is basically what value is: buying $1.00 for 80 cents, or 81, and then 82. Sifting through structured data and ordinally ranking it, then prioritizing, optimizing, and quickly executing trades is something computers are great at, and doesn’t require anywhere near the horsepower, or the complex learning behavior, of AlphaGo. But even if the specific metric used for implementation of value (say, a price-to-book or price-to-earnings ratio) changes because of market conditions, competitive dynamics, or the financial reporting tendencies of corporate executives, computers are far better positioned to identify the shift faster than we are.
On the other hand, growth is perhaps harder to spot. It’s not just that doing so involves making simple forecasts about the future — computers generally beat humans at that, too, by the way — but it requires something more than that. In my experience, successful growth investors have an ability to synthesize unstructured data across multiple, often seemingly unrelated, domains.
For example, they can extrapolate broad consumer trends to specific market growth, distilling that data down to earnings potential for products that may not even exist yet by finding earlier analogies in similar markets. Or they can identify opportunities that are not obvious by slightly tweaking an existing product line to compete with legacy incumbents in a bigger or more profitable market.
Very finely tuned computing engines, like Google’s AlphaGo, are great in a specific domain with structured data and rules, and although they can improve their performance at that one task, they have yet to truly show anything remotely resembling the humanlike intelligence required for creative, unstructured thinking. AlphaGo, for example, cannot tell a cat from a table, nor can it beat me at checkers.
In a world of information asymmetry and fairly high average economic growth — as existed 40 years ago, when the ability and effort needed to dig through physical copies of opaque financial statements resulted in a clear informational edge for those who did it — value looked like alpha. But perhaps today, in a low-growth world of data ubiquity, with a Bloomberg on every desk, where computers have eaten all the value, growth looks more like alpha.
As I thought about it, this idea led me to two inescapable conclusions: First, to support the hypothesis, we should have more explicit evidence of computers winning where data is ubiquitous. Second, a lack of data ubiquity in a given arena should mean that informational advantages, and hence alpha opportunities, still exist there.
Let’s look at the latter conclusion first, where the evidence appears the strongest: private markets. In private equity, data is certainly not yet omnipresent. Despite the best efforts of firms such as Burgiss and Pitchbook to build all-encompassing private equity databases, information in private markets is still fairly limited and typically inaccessible to most; and even when available, it often requires manual scrubbing and processing to use. However, similar to hedge funds, private equity has seen massive inflows of institutional capital over the past two decades. In fact, at roughly $4.5 trillion versus the hedge fund industry’s estimated $3.2 trillion, private equity is 40 percent larger than its evergreen cousin. If it is size that erodes alpha, we might expect private equity to mirror the decline of hedge funds, or worse. If, on the other hand, that erosion is attributable more to data availability, private equity alpha should hold up better.
For a few years, it appeared it had not, leading pundits to proclaim that excess returns in private equity were a thing of the past, perhaps a bit prematurely. These assertions were based on another period of strong public equity returns — from January 2009 until December 2017, the S&P 500 annualized at 15.8 percent — and were also compared with the performance of highly immature private equity funds.
Recent research shows that as these later vintages have started to come through the so-called J-curve — where fees and expenses early in the life of a private equity fund cause paper losses before capital gains push returns into positive territory after a few years — they have posted better relative results, and almost all vintages have generated positive excess returns to public markets across multiple metrics. Comparing the ILPA Private Markets Benchmark horizon pooled return with the Cambridge Associates Modified Public Market Equivalent for the S&P 500, we can see that excess returns in private equity do appear to have revived a bit.
Private equity today is certainly more competitive as a result of the influx of institutional capital, and higher valuations are a legitimate concern. However, the sector remains transactional, relational, and less than perfectly efficient. Opportunities for informational asymmetry are still common.
Having 35 first-round bidders for a $200 million company is obviously more competitive, and hence more efficient, than having five. But that’s still nothing compared with public markets, where tens of millions of shares transact daily on typical large-cap stocks and virtually every financial data point imaginable about a company is available nearly instantaneously to the entire market.
Although consistent with my theory, private equity’s rebound in relative performance is hardly definitive support for it. But by investigating my first takeaway, explicitly testing the performance of computer versus man — not merely growth versus value or public versus private — we see far stronger evidence of what has happened to alpha.
Using the Eurekahedge series of indexes, I compared the cumulative return of the broad Eurekahedge Hedge Fund Index, an equal-weighted index of nearly 2,500 hedge funds, with the Eurekahedge AI Index, which equally weights a basket of hedge funds that use artificial intelligence and machine learning to some extent in their investment process. Though it is as representative an index as I could find, I should note that it represents a mere 14 funds.
Still, the outperformance is impressive. Since inception in 2010, the AI index has absolutely crushed the average hedge fund, with the machines annualizing at 12.8 percent per annum versus just 5.0 percent for the humans. If efficiency is the enemy of alpha, then computers are its worst nightmare, hulked out on steroids.
To be clear, I’m not willing to conclude that value will never outperform growth again, or that private equity is a portfolio panacea. However, I do think it is prudent for us all to be skeptical of our ability to harvest replicable excess returns in the future from simplistic and highly transparent rules-based strategies amid data ubiquity, where computers now have the clear advantage. As the saying goes, if you can’t spot the sucker at the table in the first ten minutes of the game, then it’s you. And if that game involves the ordinal ranking of numbers and all the other players are computers, you’re in big, big trouble.
I think it makes sense to go where they haven’t yet completely dominated the game, or to use them where they have. As another saying goes, if you can’t beat ’em, join ’em. But then again, what do I know? I’m only human.
