On the TV show "Jeopardy," the IBM computer Watson defeated two previous human champions. Self-driving cars get into accidents, but at a much lower rate than do human drivers. Computers, including computers on smart phones, routinely defeat the world's best chess players. Is investing next?
Perhaps concerned about my ability to earn a living in the emerging world of computer-driven investing, in the last couple of days, friends have been sending me articles from
The Wall Street Journal and the
New York Times outlining the shift by BlackRock away from actively managed portfolios and toward computer-based models. (See "
BlackRock Bets on Robots to Improve Its Stock Picking" and "
At BlackRock, Machines Are Rising Over Managers to Pick Stocks.") Computers are unemotional; humans are prone to worry (and sometimes panic), to bouncing back and forth between self-doubt and over-confidence, and ego-bias ("I am a genius therefore I am right and the market is wrong"). Computers can make millions of calculations a second, each based on some microscopic edge that, over thousands of transactions, can maximize risk-adjusted returns. Thirty years from now, at least 95% of all decisions in the market will be driven by computers competing against one another.
Of course, numbers of decisions is different than numbers of investors. The big money will be almost exclusively computer-run. There will still be lots of small investors around, though -- people who love the challenge of investing, and pursue that interest with varying degrees of success. Investment managers and advisors will focus on service, and advising which computer-driven fund best suits a particular investor's objectives and risk tolerance. Mutual funds will be largely replaced by ETFs because computers will prove to have both a cost advantage and a performance advantage in the time frame favored by most investors -- three months to a year.
And computers have a cost advantage. Even mediocre mutual funds -- and 85% of all funds fail to beat the market after fees over a 10-year period -- charge a lot of money. Perhaps the investor to most successfully use computer models is Jim Simons of Renaissance Technologies. A recent
Forbes article named him as the wealthiest Long Island resident, with a net worth of $18 billion. According to
Wikipedia:
Renaissance's flagship Medallion fund, which is run mostly for fund employees, is famed for one of the best records in investing history, returning more than 35 percent annualized over a 20-year span. From 1994 through mid-2014 it averaged a 71.8% annual return. Simons ran Renaissance until his retirement in late 2009. The company is now jointly run by Peter Brown and Robert Mercer, two computer scientists specializing in computational linguistics who joined Renaissance in 1993 from IBM Research.
Simons continues to play a role at the firm as non-executive chairman and remains invested in its funds, particularly the secretive and consistently profitable black-box strategy known as Medallion. Because of the success of Renaissance in general and Medallion in particular, Simons has been described as the best money manager on earth. By October 2015, Renaissance had roughly $65 billion worth of assets under management, most of which belongs to employees of the firm.
The best description I have found of the Renaissance strategy and tactics was by James Baker of Princeton and Dragon Systems on Quora, found
here. To summarize:
- Renaissance collects "all data that they believe might bear on the movement of prices of tradable instruments--news stories, analysts' reports, energy reports, crop reports, weather reports, regulatory findings, accounting data, and, of course, quotes and trades from markets around the world."
- Portfolios of long and short positions in a wide variety of markets -- currencies, futures, equities, debt, derivatives -- are created to hedge out market risk, sector risk and any other kind of risk that Renaissance can statistically predict. The extreme degree of hedging reduces rate of return, volatility and risk. Per trade, the model is profitable on average only slightly more often than not. The firm makes thousands of trades per day.
- The firm leverages up those low but reliable returns through a variety of means including options and debt.
- In order to protect their proprietary models, Renaissance employees are paid extremely well and forced to sign ironclad confidentiality agreements.
After I finish my next report, a study of the Berkshire Hathaway (NYSE:
BRK.A) portfolio, I plan a review of the 1,000 largest positions in the Renaissance Technologies portfolio, after excluding those positions it has recently reduced in size. I'm looking both for positions representing the most interesting combination of long term profitability and current value, and the common financial statement characteristics of the companies in their portfolio. Much of the investment decision, at least by Master Investors, revolves around a simple question: Can recent trends can be extrapolated? There are two main strategies:
- Mean reversion (stocks are more volatile than are the actual underlying companies because human emotion is volatile, and that disparity can generate investment returns as excesses one way or another return to norm)
- Investing in change -- company XYZ produces profits of $150 million a year, but will produce profits of $500 million next year because of some catalyst or new product or new management
Both strategies, investing in no change and investing in change, have successful adherents. Over time, most highly successful companies remain successful and most mediocre or failing enterprises continue to perform poorly, despite temporary counter-trend movements one way or another. Again, the main issue for the investor is whether or not to extrapolate recent trends. A poorly-positioned company with mediocre earnings will, from time to time, have a great quarter. An exceptional company will occasionally have a weak quarter.
Will computers help an investor decide which trend to extrapolate? Yes, if the question is what will happen in the next quarter, and what action to take if that doesn't happen. No, I don't think so, if the question is how profitable will the company be in 10 years. That latter question depends on, as Michael Porter -- a Harvard professor and widely acknowledged expert in corporate strategy -- would say, sustainable competitive advantage. According to
a 20-year study by Credit Suisse into the best-performing public companies around the world, 51% had a probability of remaining among the best performing companies and the worst performing ones had a 56% probability of remaining the poorest performers:
- Great businesses tend to remain great or they become good businesses (combined probability of 79%). There was only a 9% chance that a great business would end up in the economic doghouse, and
- poor businesses tend to remain poor or become slightly better but still remain below average (combined probability of 83%). There was only a 6% chance that business in the economic doghouse would end up in the best category.
The returns from value investing arise from the few that exceed the market's expectations, and that thus bring up the returns of the entire group. The risks in quality investing derive mostly from the few that deteriorate significantly, bringing down the average return of that entire group. The formerly exceptional companies, which used to trade at a quality premium, suffer dramatic declines in stock (and bond for that matter) pricing when hit by the double whammy of both poor earnings and a declining P/E ratio. The most profitable investments in quality companies arise out of investments made during times of temporary adversity. Again, in investing, much depends on which trends to extrapolate.
In that decision process, made without a computer, Buffet is a master. Reportedly, he doesn't have a computer in his office -- the one he has at home he uses to send emails and play bridge. He has a TV in his office to keep on top of the news, but spends most of his time reading and thinking. Over the longer term, a company's stock price is determined by its ability to create wealth for its stockholders, which is reflected in its return on equity. You don't need a computer to assess that.
The question is: Do computers hurt or help long-term results? My conclusion is that they have no real advantage other than perhaps cost. A computer can project the odds of an exceptional company remaining exceptional, but so can a human. And both can equally easily calculate long term P/E and other value ratios over an entire business cycle, and thus make timing decisions.
I developed a computer program to detect trends in margins and capital turnover, in free cash flow and debt levels, interest coverage and liquidity. It allows me to sort through hundreds of companies in a couple of hours a week, and to conclude which ten look most interesting. Selecting which to focus on from there involves judgment. Computers can't do judgment. Or maybe they can and I don't know enough about artificial intelligence. That's always a possibility.
Faced with the choice of a Microsoft (NASDAQ:
MSFT), with a return on equity over 40%, and Coca-Cola (NYSE:
KO), with a (still exceptional) return on equity of about two-thirds of that, and knowing the CEOs of both companies, Buffett invested in Coca-Cola because it was more predictable. To him, high predictability is worth more than exceptional profitability. A fundamental concern to him is the free cash flow five and ten years down the road. His returns come mostly five to 10 years out, when the power of compound returns can develop real momentum.
Those returns don't need to be the absolute highest, they need to represent the optimum combination of predictability and profitability. He wants the most reliable compound returns so that he's safe leveraging them up and so seeks returns before leverage around 15%. And he wants to avoid taxes, to profit from the compounded returns on the money he would otherwise have paid out in taxes. He has a pretty good idea where Coca-Cola will be in 10 years. He has no idea what competition Microsoft will face in 10 years.
Although they follow vastly different strategies and employ vastly different techniques, in one sense, Buffett and Simons employ a similar tactic. They take low risk positions and leverage them up. A more complex and difficult subject -- the differences between what Buffett writes and says, and what he actually does. Now he's invested in Apple (NASDAQ:
AAPL) and IBM (NYSE:
IBM), for instance, and in a railroad which, as he said in his latest letter to shareholders, has earnings that are not entirely real. In any event, he didn't need a computer to either find Coke or to become the most successful investor in the world.
Would Warren Buffett still make investment decisions himself if he knew a computer could increase his returns? Would the world's best chess player still play chess if he knew a computer could routinely beat him? I think the answer to both is yes. For a while now, Buffett has done what he does because he loves it. He lives relatively simply; he doesn't need more money.
I too love investing, and the challenge of finding great companies and knowing them well enough that I can make a thoughtful decision on where they are in the value ebb and flow that all markets present, driven as they are by human emotion and fluctuating results. I find computers helpful in that process, but I wouldn't want one to make decisions for me. That's because I'm a long-term investor who has run businesses, who understands accounting, and who has a deep affinity for exceptional humans and their enterprises.
In the short term, though, I wouldn't even try to beat Renaissance's computer. I don't have a chance.
