Mr. Wizard’s Epic Post (1 of 3)

By -

Preface from Tim: Our beloved Mr. Wizard composed an enormous post, long enough to be seen from space. I have broken it up into three roughly equal parts, to be published over the course of the weekend. Hearty thanks to Wiz!


Predicting the S&P

Can We Do Better Than Highly Paid Analysts?

At the end of every year, fund analysts and stock pundits fall all over themselves predicting ever higher and higher prices at where the S&P will close for the coming year. These predictions usually go something as follows: “Our analysts estimate S&P earnings for 2023 will be $240. Coupled with a conservative PE of 17.4, we believe the S&P will close 2023 at 4176.”

So we have an “estimate” for the first number, earnings, and a guess for the second number, the PE multiple. How well do these analysts do?

To be blunt: not very well.

Here’s the story about earnings estimates from Factset:

https://insight.factset.com/have-industry-analysts-overestimated-sp-500-eps-for-2023

For 2023, the bottom-up EPS estimate for the S&P 500 (which reflects an aggregation of the median EPS estimates for CY 2023 for all of the companies in the index) is $232.53. If $232.53 is the final number for the year, it will mark the highest (annual) EPS number reported by the index since FactSet began tracking this metric in 1996. However, what is the likelihood that $232.53 will be the final EPS value for the S&P 500 in 2023?

In other words, how accurate is the bottom-up EPS estimate for the S&P 500 one year in advance?
Over the past 25 years (1997 – 2021), the average difference between the bottom-up EPS estimate at the beginning of the year (December 31) and the final EPS number for that same year has been 7.0%. In other words, industry analysts on average have overestimated the final EPS number by 7.0% one year in advance.

Analysts overestimated the final value (the final value finished below the estimate) in 17 of the 25 years and underestimated the final value (the final value finished above the estimate) in the other 8 years.
For the purposes of this analysis, the final EPS number for a year is the EPS number recorded two months after the end of each calendar year (February 28) to capture the actual annual EPS results reported by most companies during the fourth quarter earnings season.“

Sounds pretty bad, doesn’t it?

Rather than just guess, and guess poorly at that, let’s see if other methods might allow us to predict the S&P. We’re going to make our prediction by using fundamental or technical methods, as opposed to earnings estimates and multiples. Let’s call these other methods valuation. Since valuation is a very weak predictor for anything other than long periods of time, we’ll predict the S&P value 10 years from now – at the end of 2032.

A common criticism of any kind of forecast is: “You can’t even predict the S&P tomorrow, how can you predict it in 10 years?” There are two replies: 1) In fact I can predict the S&P tomorrow, reasonably accurately, and 2) A 10 year forecast horizon allows the ‘signal’ to dominate the ‘noise’ and produce a more accurate prediction.

Here are six ways that we can use to predict where the S&P will close in 10 years time. And since along the way these predictions will pass through 2023, 2024, and so on, even though not very accurate, we get each individual year’s prediction basically “for free.”

Each of these valuation methods uses similar techniques. Find a dataset that relates to the stock market in a fundamental or technical way. Calculate the 10 year return of the S&P for every year from the beginning of the data series to the final, and generate a scatterplot of your preferred dataset vs the 10 year returns. Draw a regression line. Since your dataset will include current data, you can apply it to the regression and read off the 10 year annualized returns directly. Returns can then be converted to S&P closing values.

I’m going to jump to the conclusion before I even start: We’re hopelessly far above where the S&P should be trading today. Anywhere from 50% to 200%. Will this divergence ever be corrected? History says yes. “This time is different” says no.

Assuming the huge overvaluation is corrected eventually, we have a few examples to see how it’s corrected. It’s not in a straight line – the most likely scenario is we have multiple bear markets until the market becomes cheap enough to entice buyers back in. I’ve used the 70s as an example, three bears in a row, but I could also make a good case that 2000 through 2009 was just one big bear market, with a big rally in the middle.

But allow me some fun of pointing out just how far away from “fair value” the S&P is right now.

Trendlines

Let’s start with a trendline. Everybody uses them, and for good reason. They work. If they didn’t, you’d see S&P closing values literally all over the place, up 3% one year, down 10% the next, up 15% the third, down 2% the fourth; there would be no pattern, no trends, no bull or bear markets. Buy and hold wouldn’t even be a possibility. The S&P is constrained to a channel because it’s driven mostly by economic fundamentals, as opposed to simply following Gaussian distributions.

Resources:

https://www.advisorperspectives.com/dshort/updates/2022/12/05/regression-to-trend-115-above-trend-in-november

https://www.jagcapm.com/insights/the-long-term-trend-is-your-friend/

https://www.macrotrends.net/2324/sp-500-historical-chart-data – the trendline from Advisor Perspectives. The authors update it every month as new data comes out. Note this is inflation adjusted S&P prices, but doesn’t include dividends.

Perspectives Trendline

From the story: “At the beginning of January 2023, it is 115% above trend. The major troughs of the past saw declines in excess of 50% below the trend. If the current S&P 500 were sitting squarely on the regression, it would be at the 1817 level.”

They’re kidding, right? Not even the most bearish of bears dares to dream of an 1817 S&P. Overvaluation worse than in 2000. We’re probably not going to see that 1817 S&P closing price anytime soon, but from the chart we can see that it often returns to the trendline within 10 to 15 years. In 10 years that trendline projects an S&P of about 2200. About half of where it’s trading now. If they’re correct, you’ll lose badly against inflation by holding stocks.

But this is only one possible trendline we can draw – there are many others.

A Tale of Two Trendlines

The Worst of Times

OK, an S&P of 1817 is pretty outrageous, but that number is from one specific data set with one specific inflation adjusted trendline. We don’t need to use their trendline, we can draw our own.

As noted above, the data in the chart is inflation adjusted, but I’d prefer to use nominal data (not adjusted for inflation). We’ll use Professor Shiller’s data to plot our own trendline. (The same Shiller that made Irrational Exuberance famous.) Professor Shiller maintains a website that has monthly data back to 1871.

http://www.econ.yale.edu/~shiller/data.htm

Caveats: Shiller doesn’t include dividends in the S&P, and the S&P price is the mid-month average of closing prices for the entire month. We plot the S&P using a log scale on the Y axis, and then use an exponential trendline to find the best fit. Excel handily provides us the equation for the trendline, and also the R^2. We can plug future dates into the trendline equation and Excel will tell us what fair value is for any year in the future.