Submitted by Nicholas Colas of Datatrek Research
Just like the holidays, Wall Street prediction season comes earlier every year. Today we review a landmark academic paper on how to make smart forecasts. The key lesson: focus on the big, statistically validated picture, not cherry picked data anyone puts in front of you. For example, 90 years of S&P 500 return data shows the index gains an average of 11.4%/year, and the odds of a +20% gain (36%) are 3x the chances of a +10% decline (12%). Be bullish or bearish according to your beliefs, but understand the probabilities your prediction will prove accurate.
Tis the season for capital market predictions. Where will the S&P 500 be in a year? How about the 10year German bund, or the US dollar? And will 2020 finally see Value overtake Growth, or Emerging Markets stocks best US equities?
We’ll issue our own prognostications soon enough, but today we want to address a larger topic: how should one go about forecasting the future? For some guidance, we turned to seminal work on the question: “On the Psychology of Prediction”, a 1973 paper by Nobel Prize winner Daniel Kahneman and his long time collaborator Amos Tversky. With over 7,000 citations in other academic papers, it remains an important work to this day.
There is a link to the paper at the end of this section, but here are the 4 things you need to know:

People tend to lean strongly on a heuristic (mental shortcut) called “representativeness” when making predictions. They take the evidence at hand and predict the outcome that best fits that data. No matter that this information may be incomplete or suspect – what they see right in front of them is essentially the anchor for what they predict.

The problem with that approach is that it ignores statistical probability. A crude example: you are told that a fair coin has come up heads 4 times in a row on a series of flips. What are the odds it will come up heads again on the 5th flip? You know the answer is 50/50, but somewhere inside you is the nagging thought “No! Tails is due for a comeback! Just look at that weird data!”

The issue here is especially acute when considering the probability of rare/outlier events. If those possibilities are present in the data shown, individuals will key off them and overestimate the chance they will happen versus their actual, realworld possibility of occurring.

Even when presented with legitimately useful data, individuals will gravitate to easier (but incorrect) extrapolations that fit the bias to find “representative” information. They choose correlated data sets rather than uncorrelated ones because the numbers line up more easily, even though uncorrelated data yields better predictions. And they don’t consider regression to the mean even when a strange data point is clearly an outlier.
With all that in mind, we need to point out one thing before proceeding: many of the 2020 predictions we’ve seen explicitly target the reader’s representativeness heuristic rather than guide them away from it. For example:

“The US stock market at current levels is like Japan in the early 1990s or Emerging Markets in 2007/2008”. Never mind that the S&P 500 is where it is because of global tech companies (not a real estate bubble), or that interest rates are lower than in either example. “X is just like Y” is shorthand for “don’t bother with actual thinking, look at this supposedly representative example”.

“Emerging markets will outperform next year because they’ve done so poorly over the last decade.” Yes, maybe they will… But not until the companies in those regions produce better earnings and return on capital. It has nothing to do with how this asset class has performed.
With that out of the way, let’s close with a simple “no representativeness, all statistics” case study that predicts where the S&P 500 will close 2020. All we’ll do is take 90 years of return data from NYU professor Aswath Damodaran (link at the end of this section) and purposefully ignore everything else. To be clear, this is not our “prediction”, but rather a palate cleanser to refresh your thinking about where US stocks may go next year.

S&P 500 today: 3120, give or take a point or two

If the S&P performs inline with the last 90 years as measured by its arithmetic average total return of 11.4%, it should end 2020 at 3420 assuming a 1.8% dividend yield (the current payout).

Worth noting: there is only 1 Wall Street strategist (Jonathan Golub at Credit Suisse) who has a number inline with that estimate (3425). The other 13 Street estimates are all lower (see Marketwatch article below).

The odds of a doubledigit decline next year are 12% based on the last 90 years, which includes World War II, the Great Depression and Recession, 2 Presidential impeachments (one which led to resignation), and 3 largescale oil shocks.

The odds of a +20% advance for the S&P 500 in 2020 are 36% based on the historical record.

The odds that the S&P will end 2020 within 5% of today’s levels are actually pretty low: just 13% based on the historical data. Almost half the Street is expecting this outcome, for what it’s worth.
The bottom line to this analysis: it’s OK to be all bulled up about 2020 (35% chance of a +20% return) or quite downbeat (12% odds of a +10% drop). What matters is to fit your narrative to the statistical data and recognize how good/bad things have to get before your forecast comes true. As a wise old trader once told me: “Never forget, Nicky… The game is rigged to the upside”. Bad outcomes happen, of course, but less frequently than the representativeness heuristic makes us believe.
Sources:
Kahneman/Tversky paper: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.395.3759&rep=rep1&type=pdf
Damodaran data: http://pages.stern.nyu.edu/~adamodar/New_Home_Page/datafile/histretSP.html
Marketwatch article: https://www.marketwatch.com/story/wherewillthesp500goin2020herearethemostbullishandbearishstrategists20191203