An extract from our 2014 Annual Letter
The next Warren Buffett will be a stock picking machine.
It is popular – probably too popular – to refer to Buffett every time someone wants to make an anti-efficient market argument. There is a kind of reductio ad Buffettum going on.
Perhaps we fall into this trap here, but we try to avoid it on the way to making a strong point about the need for investment processes to be systematic.
The ‘Superinvestors of Graham-and-Doddsville’ is an article by Warren Buffett promoting value investing, published in 1984.
It famously challenged the idea that equity markets are efficient through a study of nine successful investment funds generating long-term returns above the market index, all of which shared the same patriarch in Ben Graham.
“I think you will find that a disproportionate number of successful coin-flippers in the investment world came from a very small intellectual village that could be called Graham-and-Doddsville. A concentration of winners that simply cannot be explained by chance can be traced to this particular intellectual village.”
- Warren Buffett, 1984
Indeed, static strategies which focus on “value” and/or “quality” work well over the long term.
However, there is a cost: volatility is higher than the market and draw-downs are dramatic. For instance, value stocks dropped over 45% in 2008. Ouch.
While the volatility of returns to strategies like “value” has increased over the past decade (a secular trend), the capacity for investors to stomach this heightened volatility has not changed.
The result? Entrepreneurs and wealth owners with solid long term plans written in calmer times, capitulate at the worst moment (by definition, it is during these times of mass investor redemptions that the markets bottom).
Key to G2’s philosophy is the pragmatic realization that delivering good long term returns is not enough. Investment strategies must be designed to survive - if not take advantage of – short term volatility.
“The long term is nothing but a series of short terms.”
As such, it is no longer enough to rely on static buy-and-hold strategies with fixed time horizons. Dynamic and adaptive methods are required.
So what about the “Warren Buffett” way? Indeed Buffett – and his Graham-and-Doddsville peers - have delivered outstanding returns: from 1976 through 2011, Berkshire realized average annual excess return of around 14%.
But these returns also entailed more risk: annualized volatility was 24.9 percent (Sharpe ratio of 0.76), and draw-downs exceeded 30% on several occasions. For example, from July 1998 through February 2000, Berkshire lost 44 percent of its market value, while the overall stock market gained 32 percent. Not many investors can stomach a 76 percent shortfall!
The “conventional wisdom” states that Warren Buffett’s success is due to his unique skills as the chief stock picker and CEO of Berkshire Hathaway. Not so, according to a recent study “Buffett’s Alpha” by Messrs Frazzini and Kabiller of AQR, and Lasse Pedersen of NYU.
The authors found that a large part of Buffett's performance is explained by selecting stocks with exposure to the following factors:
Low earnings volatility
Profitability (high margins/ROIC)
High asset turnover (indicating efficiency)
Low financial and operating leverage
Low specific stock risk (intrinsic volatility)
In other words, it's Buffett's strategy that generated the alpha, not his stock selection skills per se.
Rather than being the “ultimate value investor”, his edge was to discover the importance of low-beta, high quality stocks decades before others recognized similar patterns.
The authors also considered whether Buffett's skill was due to his ability to buy the right stocks versus his ability as a CEO. They decomposed Berkshire's returns, finding that his public company investments performed better than Berkshire’s privately held subsidiaries. So it's not his skill as a manager that's responsible for his alpha.
For the non-Buffetts’ of the world, two lessons can be drawn: 1. Stock-picking through factor selection works: there are a variety of factors which are predictive of excess returns over the long-term; 2. Exposure to these factors must be dynamic to avoid periods of non-survivable underperformance.
It is this challenge that G2’s systems have been designed to tackle; our unique approach is briefly described as follows.
First, we integrate deep datasets, including: price/macro, fundamental, governance, management guidance and analyst expectations. Second, we apply the latest advances in applied artificial intelligence - using learning algorithms to identify predictive combinations among more than 200 factors.
To give an idea of the breadth of the challenge: the performance of each factor is conditional on other factor interrelationships as well as industry (68 different groups) and the market environment. For example, a factor ‘management buy backs’ may not be predictive per se, but for value stocks in the technology sector in an oversold environment it may be a strong predictor of outperformance.
G2’s machine learning system identifies hundreds of such pockets of micro-inefficiency. In essence we aim to combine the breadth and speed of quant strategies (1000’s of stocks assessed daily) with the depth and rigor normally reserved for the best human analysts.
In so doing, we take advantage of the quiet revolution taking place in what is called Applied AI. Huge advances are occurring quietly across 1000’s of applications. Think: search engines; the genome project; space/military robotics; facial recognition, medical diagnosis, self-driving cars, and natural language processing.
We think that just as computers now surpass humans in classification tasks like Checkers, Chess, Quiz Shows (Jeopardy), Poker … the time will come, soon, when computers surpass the best (human) stock-pickers.
Warren Buffett 2.0 will be a machine. Indeed for those who have met him, Warren Buffett 1.0 is quite machine-like in his discipline and temperament. That reminds us of a fascinating book released by Steven Greiner when G2 was launched in 2011: Ben Graham Was a Quant.