How To Make Better Predictions
By Nick Colas of Datatrek Research
Tis the season of Wall Street predictions. How will US, European and Asian stocks do next year? How many times will the Fed increase interest rates? Where will the 10-year Treasury trade in 2022? All this got us to thinking about the process of making predictions, and that is the subject of this week’s Story Time Thursday.
We’ll start with some wisdom from Daniel Kahneman and Amos Tversky, who coauthored a paper they called “On the Psychology of Prediction” in 1973 (link here). Despite the title, their approach to making useful predictions is math- (not psychologically-) based:
Any prediction should start with the statistical base rate associated with a given outcome. For example, if we’re trying to predict where the S&P 500 will be at year-end 2022 then the index’s long run annual rate of return should be our starting point. Let’s call that 10 percent.
Then we ask the question “how representative will 2022 be of the general environment that created that base rate?” In the case of the S&P, we might make a list of plusses and minuses with “strong earnings” in the first camp and “possible Fed policy mistake” in the latter. Let’s say the bias is to the bearish by 2-3 points.
That would leave us with a predicted return of 7-8 percent for the S&P 500 next year, which is what our DataTrek Look Ahead Survey reported was the plurality of readers’ expectations. Given the answers to several other questions in that survey (notably the coin-flip odds respondents gave to a Fed policy mistake), the “what” and the “why” of the survey’s output lines up very well.
Now, if one has a profoundly bearish outlook, this approach also allows you to consider that view in a statistical framework:
Let’s say we want to predict that the S&P 500 will be down 10 percent or more next year.
Over the last 72 years (back to 1950), the S&P has been down at least 10 percent on 6 occasions: 1957 (-10.5 pct), 1973 (-14.3 pct), 1974 (-25.9 pct), 2001 (-11.9 pct), 2002 (-22.0), and 2008 (-36.6 pct). The average of those years is a negative 20.2 percent return.
We could therefore say that there is an 8 percent chance (6 divided by 72) that we’ll get a 20 percent decline next year. That only works out to a 1.7 percentage point expected value, hardly the stuff of nightmares, at least statistically speaking. Large down years happen (obviously), but they are rare. Why? Because markets are, in general, good predictors of future business and economic conditions:
If there were significant trouble brewing in 2022, would the S&P 500 be up 26 percent this year?
History says no; the average annual return for the S&P going into 1957, 1973, 2001 and 2008 (those bad years listed above, ignoring 2-year sequences of large drawdowns) was just +5.6 percent.
Aside from Kahneman/Tversky, we also like Philip Tetlock’s research on what makes for a productive prediction framework. His original work centered on political predictions and, specifically, why so many highly accomplished people in the field were often so terribly wrong. His explanation is that there’s two sorts of people:
Those who have an overarching view of the world (dubbed “hedgehogs”, who know one thing very well) and…
Those who piece together their predictions from a wide array of information (“foxes”, who know many things, but at a more surface level).
No surprise, but “foxes” tend to make much more accurate predictions than “hedgehogs” since their worldview is more holistic and flexible. A Harvard Business Review article (link below) on Tetlock’s work fleshes out this idea:
Intelligence and genuine domain expertise are, of course, helpful in making accurate predictions.
Practice helps a lot as well, however. Making predictions is like training to play a sport. Repetition, and the feedback loop it creates, makes you a more accurate predictor over time.
Teams can make accurate predictions, but only when the participants are good predictors in their own rights. Teams of less competent individuals just become more entrenched in their pre-existing beliefs.
More open-minded people tend to make better predictions.
Training in statistics improves the quality of predictions. This is basically the Tversky/Kahneman approach.
Rushing makes for bad predictions.
The bottom line to Tetlock’s work is that making accurate predictions is a learned skill that also requires intellectual honesty and flexibility. Since investing is essentially the process of making predictions, this likely comes as no surprise to you. Nonetheless, it is always good to see one’s intuition backed up with facts from a related field.
We’ll finish out this Story Time with our own thoughts specifically about Wall Street predictions:
#1: It’s not the destination/prediction that adds value, but the journey/explanation. We don’t really care if Goldman is more bullish than Morgan Stanley or if Credit Suisse is negative on Asian stocks. What’s useful is the “why(s)”, and if the explanation fits together in a novel way that makes sense.
#2: Bearish cases always sound smarter than bullish ones. They usually sound more “in the know”, and a good negative argument on anything (Tesla, virtual currencies, whatever) invariably gets attention. All that makes for excellent entertainment, but we never mistake the headline for a useful prediction. Now, if there’s something in the “why” we don’t know – that is useful. It might not affect asset prices the way the predictor thinks, but every factual kernel of data is always worth knowing.
#3: Dogma is the enemy of predictive accuracy. Any market prediction based on politics, long-held views of the role of central banks, or other beliefs that often engender strong emotions should be treated with extreme caution. The goal of investing (i.e., predicting future asset prices) is to make a good risk-adjusted return and nothing else.
#4: The calendar is a social construct; markets don’t “change” on January 1st. Too often, year-ahead predictions implicitly assume the world will be suddenly different. Maybe they forecast that mean reversion will kick in with respect to equity valuations, or markets will suddenly pay more attention to a particular economic factor (such as inflation). The reality is that markets constantly discount new information regardless of whether it is December or June.
Summing up: we’d be remiss if we didn’t mention Yogi Berra in our list of influential thinkers about the prediction making process, for his quip “the future ain’t what it used to be”. How we think about the future absolutely should change as we see it develop. The important thing is to constantly adapt and learn, systematically incorporating what we witness into newer, better predictions. Humans crave linear, unchanging predictability, so this is no easy task. Essential, yes … But not easy.
Kahneman/Tversky paper: http://faculty.econ.ucdavis.edu/faculty/nehring/teaching/econ106/readings/kahneman-tversky-on%20the%20psychology%20of%20prediction.pdf
HBR article on Tetlock’s findings: https://hbr.org/2015/02/what-research-tells-us-about-making-accurate-predictions