Nearly a third of all U.S. stock trades are now made by black-box algorithms (“algos”) that data-mine economic and non-economic data, social media communications, and the price of tea in China, looking for correlations with stock prices. Many investors fear that they cannot compete with the superhuman power of computers to scrutinize data.
For example, Brian Sozzi, co-host of Yahoo Finance’s morning show, The First Trade, recently warned that algorithmic trading has not only made value-investing and other traditional investment approaches obsolete, but has also made the stock market more volatile because algos react quickly to data blips, and multiple algos ransacking the same data are likely to discover similar buy/sell rules. The herd mentality of humans has been superseded by the herd mentality of algos. His dire warning:
The mighty algorithmic robots that have dominated global markets over the past decade spitting out insane swings in asset prices in the process will probably get even mightier during the next decade.Brian Sozzi, “Why stock market traders should be terrified of robots in the next decade” at Yahoo Finance
It is true that wild swings in asset prices make life difficult for day traders and other short-sighted speculators. Real investors should not fear but rather embrace the mindless trading of algorithmic robots.
First of all, the data blips that send data-mining algos into trading frenzies are often temporary and meaningless. Sozzi himself provides a wonderful example: a study concluded that the market usually does better when Trump tweets less. Another algo discovered that interest rate zigs and zags correlate with Trump tweets containing the words billion or great. I unleashed an algo and found several other amusing correlations involving Trump’s incessant tweeting: the Dow tends to go up when Trump tweets the word Democrat, the temperature in Moscow tends to go up when Trump tweets the phrase witch hunt, and the price of tea in China tends to go down when Trump tweets the word president.
These silly correlations are simple examples of the general principle that large data sets necessarily contain a very large number of coincidental statistical patterns. Computers excel at discovering such coincidences. While humans might laugh at such nonsense, computerized trading algorithms take them seriously. Computers cannot distinguish between meaningful relationships and silly coincidences because they are not intelligent in any meaningful sense.
One demonstration of computer unintelligence is the Winograd schema challenge—a test of whether computers can recognize the noun that an ambiguous pronoun refers to. For example, what does it refer to in this sentence?
The trophy would not fit in the brown suitcase because it was too big.
The answer is obvious to humans but computer algorithms are befuddled by such questions because they literally do not know what any of the words in the sentence mean. Computers can spellcheck words, put words in alphabetical order, retrieve matching words in other languages, and do other useful tasks, all without understanding any of the words they are processing. They are very much like New Zealander Nigel Richards, who has twice won the French-language Scrabble championship without understanding the words he is spelling.
To make the point more directly relevant, I have proposed what I modestly call the Smith Test:
Allow a computer to analyze a collection of data in any way it wants, and then report the statistical relationships that it thinks might be useful for making predictions. The computer passes the Smith test if a human panel concurs that the relationships selected by the computer make sense.
A variation on the Smith Test is to present a computer with a list of statistical correlations, some clearly meaningful and others obviously coincidental, and ask the computer to label each as either meaningful or coincidental.
Humans, having lived in the real world and accumulated a lifetime of wisdom and common sense, are much better than computers at detecting BS.
Legendary value investor Benjamin Graham asked us to imagine Mr. Market, an excitable person who comes by every day, offering to buy the stock we own or to sell us more stock. Sometimes, Mr. Market’s prices are reasonable. Other times, they are ridiculous. There is no reason for our assessment of our stocks to be swayed by Mr. Market’s prices. It is as if we owned hens producing eggs and cows producing milk and one day Mr. Market tells us that a hen is worth thousands of dollars and a cow is worth pennies. We value the hens and cows for the eggs and milk they produce, not for the numbers spouted by Mr. Market.
If we own stock in a sound company with strong earnings and satisfying dividends, we can see the earnings and cash the dividends without being troubled by Mr. Market’s hallucinations. The “insane swings in asset prices” caused by mindless algo trading can be cheerfully ignored by long-term value investors.
In fact, the growing reliance on algos should be embraced rather than feared. While insane swings in asset prices create difficulties for myopic traders trying to predict short-term movements in stock prices, they also create opportunities for investors looking to buy good stocks at reasonable prices. If Mr. Market is willing to pay thousands of dollars for our hens and sell us cows for pennies, we should rejoice at his stupidity.
If algos monitoring the words that Trump tweets cause stock prices to fall sharply, this is not a reason to panic; this is a chance to buy good stocks at bargain prices. If algos cause a bubble in stock prices, this is a chance to take advantage of Mr. Market’s delirium. Perhaps the most extreme example of this delirium occurred during the flash crash on May 10, 2010, when some algos paid more than $100,000 a share for Apple, Hewlett-Packard, and Sotheby’s, while other algos sold Accenture and other major stocks for less than a penny a share. Other instances have been less extreme but they were still lucrative opportunities.
Computer algorithms are much, much better than humans at discovering statistical patterns but much, much worse than humans at discerning whether the patterns are meaningful or meaningless. So, the algos sometimes sell at low prices and other times buy at high prices. Investors should embrace the power and fragility of computers. Their insanity is our opportunity.
If you enjoyed this discussion, don’t miss Prof. Gary Smith’s analysis of the difference a stock ticker name makes in “A BABY, A GEEK, and a COW” all walk into a bar…
Also: We see the pattern! But is it real? It’s natural to imagine that a deep significance underlies coincidences. Unfortunately, patterns are not always a source of information. Often, they are a meaningless coincidence like the 7-11 babies this summer.
Investor, AI isn’t your big fix. In investing and elsewhere, an AI label is often more effective for marketing than for performance