Monday, February 15, 2016

predicting long-term returns

There has been some interesting growth in the research out there of what really predicts long-term returns. And as you know, ironically--notwithstanding all the headlines about it -- it turns out that things like GDP growth in the last quarter has almost no actual predictive value for markets over any useful time period. Certainly, in the aggregate over a multidecade time period, how much the economy growth matters, but it has nothing in the way of actual predictive value.

Valuation is kind of an interesting one, though. What we're finding more and more from valuation is that it's actually still not very good at telling you what's going to happen with the markets in the next six or 12 months and where they are going in the near term. But it's perhaps better than we give it credit for in how much it actually predicts long-term market returns. I suppose in some way the recent research on this really just goes back to Benjamin Graham almost a hundred years ago, who noted that, in the short term, markets act like a voting machine; in the long term, they act like a weighing machine. In the long term, valuation eventually comes to bear and either lifts up returns if things are cheap or drags down returns when they are more expensive.

Benz: So, when you look at various time periods over which valuation tends to be most or least predictive, you said in the short term it's not so predictive, while over longer time periods it is. But very long time periods, again, maybe not so effective.

Kitces: It starts to break down. So, we've looked at lots of different valuation measures of this question of what works. Ironically, some of the ones that are most popular that we talk about most often really don't work very well. Things like forward-looking P/E ratios tend to be very poor at predicting market returns over really much of any time horizon because, unfortunately, we just tend not to get forward earnings correct. Particularly in market returns, we usually don't see the recession coming until it's too late, and then we tend to overestimate the declines and underestimate the turns when they come. So, we find that forward P/E ratios don't work very well. Things like earnings over the past year are a little bit too short term.

The measures we find that work the best are those like the Shiller P/E ratio. It's often called cyclically adjusted P/E ratios, or CAPE, where we actually take 10 years' worth of trailing earnings, adjust them for inflation, and average them out over that whole time period. So, we get something that's kind of smoothed out for all the volatile market cycles, and that turns out to actually have some very powerful predictability of future market returns; but as you said, it's only over longer time periods.

When you look over a time horizon like a year, it turns out that Shiller CAPE is only slightly more predictive than monkeys throwing darts at stocks. It's almost random. It's ever so slightly better, but it's almost random. As the time horizon stretches out, though, it becomes much, much better, and there is actually an incredibly high correlation between Shiller market valuation and returns over the next eight years or so. It's actually quite good, and it explains almost half the variation in eight-year returns. So, it doesn't necessarily tell you how you're going to get there over eight years--it just says that from high valuation points, the market returns tend to be worse over eight years and then from low valuation points, they tend to be better.

Then, when we stretch the time period out even further, it actually starts to break down again. So, we've seen a lot of people say things like, "I'm just going to drag all my retirement spending way down because it looks like the 30-year return on the market has to be bad if valuations are so high." But we actually find that the predictability of valuation for 30-year returns is hardly any better than it is for one-year returns. So, one year is too short--markets happen because they are a voting machine; 30 years is actually too long because whole economies can restructure themselves over 30 years. We really find it's that eight to 15-year time period where it's really powerful, which matters a lot for, say, retirees thinking about sequence-of-return risk and accumulators who might be in their 40s or 50s and could be 10 or 15 years away from retirement and are trying to figure out whether the market is likely to cooperate with their portfolio growth and getting them to the finish line. But you have to be careful not to either focus too short or too long.

Labels:

0 Comments:

Post a Comment

<< Home