Michael Kitces Writes an Article Showing that, in the Very Long Term, the Effect of Valuations Diminishes

Updated on

Valuation-Informed Indexing #284

by Rob Bennett

A number of years back, I founded a discussion board called “The Safe Withdrawal Rate Research Group.” John Walter Russell and I were at the time developing the five Valuation-Informed Indexing calculators that are today housed at my web site. John was producing new research relating to the VII concept on an almost daily basis. He published his research at the board and community members would ask questions about it and he would respond to them or in some cases produce new research to explore the questions. It was an amazing place!

John has never received the credit he deserves for the many powerful insights he developed on a host of investing strategy issues that had never been explored before. So I was excited to see an article recently published by Michael Kitces at his Nerd’s Eye View blog titled Should Equity Return Assumptions In Retirement Projections Be Reduced for Today’s High Shiller CAPE Valuation? The article explores several questions that John first looked at a number of years back and affirms and expands on his findings. Things are happening! We are getting to the place where we all deep in our hearts have long wanted to go, ever so slowly but ever so surely.

Here are some words from Michael’s article:

“In today’s high-valuation environment, many retirement projections are being done with reduced long-term return assumptions… with the caveat that while many advisors agree on the need to project lower returns, there is little agreement on how much lower it should be.

“A look at the available market data suggests that realistically, it would be appropriate to reduce equity return assumptions by about 100bps (or 1 percentage point) over the next 30 years, to reflect the current valuation environment. Ironically, the reduction is not greater, because 30 years is such a long time that even if returns are bad for a period of time, there are enough subsequent years to recover as well.

“In point of fact, though, it turns out that market valuation is even more predictive of 15-year returns, and that today’s high P/E ratios imply that between now and 2030, market returns could be reduced by as much as 400bps. On the other hand, because market valuations (and the associated returns) tend to move in cycles, this also implies that in the 2030s and 2040s, market returns could be as much as 400bps above the long-term average as well.

“Ultimately, then, the ideal way to adjust return assumptions in a retirement plan in today’s environment may not be to reduce long-term returns at all, but instead to do projections with a “regime-based” approach to return assumptions. This would entail projecting a period of much lower returns, followed by a subsequent period of higher returns, in a manner that more accurately reflects the impact of market valuation on returns – and better accounts for the sequence-of-return risk along the way as well!”


This is exactly what the data says. It is a highly counter-intuitive finding. I think that explains why the experts in this field have been so slow to incorporate the implications of Shiller’s revolutionary findings into their strategy recommendations. What Kitces is saying makes perfect sense once you work the Valuation-Informed Indexing concept long enough to stop thinking like a Buy-and-Holder. These are different models proceeding from different core premises. Just as it takes some familiarity with a new language to begin thinking in the manner that people who speak that language on a daily basis think, it takes some time working with the new model to stop jumping to the conclusions that a Buy-and-Holder naturally jumps to and to appreciate what the historical stock-return data is really trying to tell us about how stock investing works.

Shiller discovered that valuations affect long-term returns. That’s an absurd finding, according to the Buy-and-Hold Model. Buy-and-Hold is rooted in a belief that the market is efficient, that stocks are always properly priced. But, if valuations affect long-term timing, mispricing is a real phenomenon. The market is NOT efficient. Any model that starts from an assumption that it is will get the numbers wrong and will suggest strategies that will not work in the real world.

In he early days of my explorations of the Buy-and-Hold concept, defenders of the old model would constantly taunt me with showings that short-term timing does not work. I of course agreed with them; Shiller never argued that short-term timing works. But, no matter how many times I would make that point, my Buy-and-Hold friends would demand that I make short-term return predictions. I can’t do it. I don’t believe that anyone can do it. But to the Buy-and-Hold mindset, the proper response to any claim that it is possible to predict stock returns is to point out that short-term timing does not work. It gets exhausting going over that same ground again and again and again and again.

Russell and I made it a practice never even to attempt short-term return predictions. There is a great hunger for such predictions. And so we were tempted from time to time to include them in our calculators. But John was ultimately always firm about ruling them out. The historical data shows that such predictions do not work. So we we never went there.

What we learned is that return predictions get better as the time-period covered grows longer. Predictions that go five years out don’t work. Predictions that go ten years out are effective so long as not too much precision is demanded. Predictions that go 20 years out are even better.

But at that point the power to make effective predictions drops sharply!

That was a Hitchcockian twist! We weren’t expecting that one!

The initial finding was that predictions got better as more time passed. But the follow-up finding was that somewhere near the 20-year mark that reality was reversed. The connection between the starting-point valuation level and the long-term return begins to get worse, not better! Go far enough out and the connection between the starting-point valuation level and the long-term return virtually disappears!

This reality is demonstrated powerfully by the results generated by The Stock-Return Predictor. The most likely annualized ten-year return on stocks purchased when prices are what they were in 1982 is 15 percent real. When stocks are priced as they were in 2000, the number is a negative 1 percent real. Valuations make a big difference when you go 10 years out.

But the most likely 60-year annualized return for stocks purchased at 1982 prices is 6.78 and the most likely 60-year annualized return for stocks purchased at 2000 prices is 6.24. That’s not nearly so big a difference. What’s going on? Why do valuations not matter much at five years out and then matter a great deal at 20 years out and then matter less at 30 years out and then matter not too much again at 60 years out? Is that not a weird reality?

I’ll explain what Russell and I concluded is going on in next week’s column.

Rob Bennett’s bio is here.

Leave a Comment