Bogleheads

[czeckers]

There is a lot of talk about the outlook for stocks going forward. Various models typically look at one or more of the following factors: currrent divident yield, P/E or P/B ratio, and/or the current 10-year t-bill.

I am curious how people on this forum feel about the predictive potential of the above variables as well as if there are any others. To narrow the focus, I'll say that we are looking at the performance of a total US stock market index in the 5-10 year range.

For those who are interested in these kinds of predictions, what methodology do you use?

-K
_________________
16% US LCB, 16% US SCV, 16% US REIT, 16% Developed Int'l, 16% EM, 10% Inter-term US Treasury, 10% TIPS

"A journey of a thousand miles must begin with a single step."

[pkcrafter]

For a range of 5-10 years, I think you can throw almost all predictive data out the window. If I had to pick, it would probably be the dividend discount model and P/E. But chances are high that any 5 year period will not cooperate with the models. Within 10 years chances are good that the starting model would be modified.

Paul
_________________

[nisiprius]

I think that people have been trying to make insanely overprecise predictions for many decades now.

In "Stocks for the Long Run," Siegel gives an example that illustrates the "extraordinary stability" of the long-term real return of stocks in the United States. The thing is, in his example, he uses three periods which, even if they are not cherry-picked, are all longer than fifty years.

So, just for talking purposes, maybe I'm prepared to say that on the basis of what Siegel says, you'd predict a real return of 7% plus-or-minus a percent over the next fifty years. Wait, let's make that a century just to be safe.

Any period of twenty years is quite uncertain--historically you would not ever have taken a real loss, but 1929-1948 inclusive gave you a real return of less than 1.4% per year, and 1965-1984 inclusive gave you 1.43%. So what are you going to say? Over a twenty-year period real returns have been anywhere in the range of 1.4% to 9.6% real. That's quite a range. A ten year period is a crapshoot. (And just recently we actually shot craps).

But wait, I'm taking that back. I'm not prepared to predict 7% real even for the next fifty years, because I'm not convinced there's any real reason to expect it forward. Siegel has no explanation for it. No theory. No economic model yielding that number. It's just that it's always been that way... but only in the United States. Not in other countries. In Belgium, the real return over the full century 1900-2000 was 2.5% per year.

Why do we think the next century in the U. S. will be like the last one, and not like Belgium? Is it just because the United States is kewl and deserves good fortune because we know to use ketchup instead of mayonnaise on our French fries? Or just the luck of our military fortune in the twentieth century? Or getting a gift of immigrants from Europe?

I think there's good reason for an ordinary investor to put a decent slug of his savings into stocks, along the lines of Adrian Nenu's rule or "age in bonds" or such, on the reasoning that historically U. S. stocks have had a fairly good chance of hitting a fairly decent jackpot within an investor's lifetime, but I'm becoming more and more convinced that you can't say anything more than that--and that plugging in actual numbers for expected returns is foolishness.

(That includes those Monte Carlo simulators. AFAIK they aren't based on an underlying economic model, they are just a sophisticated way of saying the past will resemble the future--statistically).

(The dividend discount model would be great if you had the next twenty years' Wall Street Journals to tell you what the future dividends and future interest rates actually were).
_________________
Annual income twenty pounds, annual expenditure nineteen nineteen and six, result happiness; Annual income twenty pounds, annual expenditure twenty pounds ought and six, result misery.

[czeckers]

I would have to say that the market, though still unpredictable, certainly becomes more predictable the longer the time horizon. If you look at one year returns, they are all over the place. Now look at 3, 5, 10, and 20 year returns and the variablility goes down significantly.

There certainly are things that correlate to varying degrees with future returns.

For example, the lower the starting P/E ratio, the higher the expected returns. This makes sense intuitively, the lower the price paid at the outset, the higher the likelyhood of profit in the long run.

Same thing with divident yield. The more your basket of stocks is paying out in dividends, the better your return in the long run assuming P/E ratios stay the same.

Finally, assuming there is an equity risk premium over the 10-year t-bill, then the higher the starting yield on the t-bill, the higher return on the stocks in nominal (though not necessarily real) terms.

Unfortunately all of these can and likely do interplay with each other. I have seen Mr. Bogle and others make long-term predictions, and I am curious what models are most commonly used for this purpose.

I am not looking for a prediction down to 3 decimal points, just a ballpark model.

-K
_________________
16% US LCB, 16% US SCV, 16% US REIT, 16% Developed Int'l, 16% EM, 10% Inter-term US Treasury, 10% TIPS

"A journey of a thousand miles must begin with a single step."

[nisiprius]

[Adrian Nenu]

http://video.yahoo.com/watch/5593260/14681081

Adrian
anenu@tampabay.rr.com

[Taylor Larimore]

[czeckers]

Mr. Bernstein's interviews are always informative, but my question is not one of risk.

Let me try to put my question in context. Part of any investment plan is a determination of how much one needs to save in order to meet one's future goals. Among the variables are the amount of money you expect to need, say, by the time you retire, as well as expected rate of return on said investments. I know that making such predictions is an imprecise exercise, but never-the-less we need to arrive at some reasonable number to enter into the forumula. From what I can tell, most sources simply use the long-term average return for whatever asset class one is considering. However, this is significantly flawed because one's future returns are greatly dictated by valuations and market conditions at the beginning of whatever time period you look at. I am simply trying to find a better alternative to just using the long-term historical return rate.

-K
_________________
16% US LCB, 16% US SCV, 16% US REIT, 16% Developed Int'l, 16% EM, 10% Inter-term US Treasury, 10% TIPS

"A journey of a thousand miles must begin with a single step."

[jeffyscott]

[Larry Johnson]

the only sage advice came this morning. A client asked his broker what he was buying. His answer: canned goods and ammunition.

[Taylor Larimore]

[Rodc]

Start with thinking about some long term horizon, maybe 30 years.

Using the median means you have a 50/50 shot of not having enough (provided the past median really predicts the future median). Similar comments if you use average returns.

If you want a pretty good shot of having enough use something like the 20th percentile return. You could use the worst we have ever seen, or even the worse we have ever seen and take a little more off as certainly the worst case could be worse than we have yet seen. But, I think something like 20th percentile is pretty good, and if worst comes to worst you’ll have to scramble a bit, but then you’ll have a lot of company eating Alpo, so maybe it won’t seem too bad.

My personal take is much like nisiprius. Especially if you are fairly young and decades way from retiring almost any prediction is going to have such huge error bars as to make the exercise all but useless. As you approach retirement you have a pretty good handle on how much you will have in the kitty, what your retirement expectations are, more will be bonds/TIPS/cash which starts to give you a floor, and the time horizon shrinks, and then just maybe some meaningful prediction is somewhat possible.

In addition to all the uncertainty in returns, if fairly young you have no real idea how your life will turn out. I would guess few in their 30s know if their career will take off and they will have more money than they dreamed, or if they will end up in a dead-end job, or face repeated bouts of unemployment, or other situation that severely limits their income.

Just as many scheme and dream of ways to beat the market, mostly to no avail, people crave certainty in predicting how much they “need to save”. That certainty is an illusion. All you can do is save a fair amount (think in terms of 20% and work up or down depending on your situation, age you start, etc.) and remain as flexible as possible (ie try to stay out of major debt or other obligation). Using something like the 20% percentile returns for a balanced portfolio as a sanity check might be a good idea, or run one of the free Monte Carlo simulators (FireCalc etc.) as a sanity check might be a good idea, just remember that they tend to put out predictions to three (or more!) significant figures, whereas in reality we might have about half of a single significant figure (ie the results are ballpark, and even then only if you remember the ballpark is really big).
_________________
"all standard caveats apply"

[dbr]

[czeckers]

So I decided to explore this a little further. Looking at s&P 500 data from 1972 to 2008 (date range of convenience), I entered data for annual returns, dividend yield, 10-year treasury note yield, and p/e ratio into a spreadsheet. I then plotted each year's treasury yield, p/e ratio, and dividend yield for the S&P 500 vs. the subsequent 10-year return of the S&P 500 (with reinvested dividends).

I then overlayed the best fit linear trend line over each plot and looked at the r-squared values.

Both the dividend yield and 10-year treasury yield plots vs. subsequent 10-year returns had an r-squared of 0.32.

The p/e ratio did better with an r-squared of 0.48.

I wish I could post the pretty picture but am not sure how to post a chart from excel on here.

I definitely see that looking 10 years into the future is pretty meaningless.

I also repeated the same exercise using my Simba's worksheet to get historical 10-year returns on my portfolio which consists of 10% TIPS, 10% interterm treasury, and then equal parts of US large blend, US small cap value, REITS, Int'l developed, and EM.

The surprising result is that the r-squared values were all much higher:
10-year treasury yield vs. portfolio r-squared=0.44
S&P P/E vs portfolio = 0.67
S&P dividend yield vs. portfolio = 0.71!

Now I would expect the S&P 500 to perhaps be predictive of the large blend portion of my portfolio, but not of the entire portfolio which is expected to have a very large amount of tracking error compared to the S&P 500. Since it is illogical that the S&P 500's dividend yield would be more predictive of the subsequent 10-year performance of my mixed portfolio vs. that of the S&P 500 itself, I can only conclude that this was an exercise in data mining and that we truly cannot predict 10-year returns with any reasonable accuracy.

Good thing I haven't quit my day job.

-K
_________________
16% US LCB, 16% US SCV, 16% US REIT, 16% Developed Int'l, 16% EM, 10% Inter-term US Treasury, 10% TIPS

"A journey of a thousand miles must begin with a single step."

[jeffyscott]

[czeckers]

I was refering to Peter Bernstein in this case as Adrian kindly provided a link to a couple of his interviews earlier in this thread. However, I think most here would agree that whether you are listening to the late Mr. Peter Bernstein or Dr. Bernstein, you will get great advice!

-K
_________________
16% US LCB, 16% US SCV, 16% US REIT, 16% Developed Int'l, 16% EM, 10% Inter-term US Treasury, 10% TIPS

"A journey of a thousand miles must begin with a single step."

[Taylor Larimore]

Hi Jeffy:

Bernstein interview:

[Adrian Nenu]

Gibson's "Asset Allocation" gives an average annual return for US large company stocks of 11.2% between 1926 and 1998 with a 20.3 standard deviation. Let's use a 20 year time period:

20.3/20^2 = 4.54 (one standard deviation)

Expected returns = 11.2% +/-4.54 (6.66% to 15.74%)

So $1,000 invested could end up between $3,631 and $18,607 in 20 years (68% of the time)




For intermediate term government bonds:

5.7/20^2 = 1.28 (one standard deviation)

Expected returns = 5.3% +/-1.28 (4.10% to 6.58%)

So $1,000 invested could end up between $2,234 and $3,578 in 20 years (68% of the time)


This is of course only one standard deviation (68% of the time). The variability of returns is much wider the other 32% of the time. And hopefully you don't have to deal with a bear market during the first few years of the 20 year time period or all bets are off. That is something standard distribution can't show or account for because such events happen more often than normal distribution stats predict.

If you earn the average return number or higher, great. But what happens if stock returns end up at the low end or worse, more than one standard deviation towards the low end? That's why diversification with bonds is important because it provides more consistent returns.

Tolerable Loss x 2 = Equity Allocation < 50%

Adrian
anenu@tampabay.rr.com

[czeckers]

This really shows how difficult it is to plan for retirement. I see a number of threads discussing the replacement of various percentages of one's pre-retirement income. However, even if I calculate with some precision based on expected social security income, tax bracket, expected fixed and variable expenses, etc. that I want to replace 70% of my pre-retirement income, it still doesn't really help me come up with a saving and investment plan since it is hard to estimate what returns I should expect. It sort-of makes the income replacement formulas a moot point as well all those retirement planning calculators out there.

We are simply left with the advice of: Save as much as you can, starting as early as you can, stick to low-cost index funds, and only take on as much risk as you can handle, perhaps erring a bit on the side of caution. There you have it.

-K
_________________
16% US LCB, 16% US SCV, 16% US REIT, 16% Developed Int'l, 16% EM, 10% Inter-term US Treasury, 10% TIPS

"A journey of a thousand miles must begin with a single step."

[Rodc]

[Rodc]

[pkcrafter]

Taylor wrote:

[Adrian Nenu]

The inverted yield curve has a nearly 100% accuracy record of predicting recessions but it's still not good enough for some people.

Adrian
anenu@tampabay.rr.com

[wbond]

I've just started the new WB "Manifesto" book, and I don't have it in front of me, but he emphasizes the Gordon equation for estimating long-term (several decade) returns, as he has in his prior writing.

He actually goes so far as to say something like (paraphrasing from memory): the investor who cannot estimate long-term returns correctly might as well stuff half of his capital in a mattress and burn the other half.

It's simple: take the dividend yield and add 1.3% real growth (the historical average, that mirrors the per capita growth of the economy).

I have never heard a good argument as to why his is a poor approach.

[Adrian Nenu]

[jeffyscott]

[fluffyistaken]

[Rodc]

[nisiprius]

[Adrian Nenu]

[jeffyscott]

[peter71]

Hi All,

A few general thoughts . . .

Agreed that the Buffet quote about the mattress/matches allocation is weirdly hyperbolic.

I really like the CXO Advisory Group site and beyond the "Guru Grades" there's a lot of good stuff there about using more than one predictive indicator at a time . . .

In that vein, I'm personally having fun tracking a three-way race in football pick-making between a) gurus, b) quantitative models and c) the wisdom of crowds on ESPN this season . . . pretty small sample to date, but at this point it looks like the gurus are worse than EITHER the models (accuscore) or the crowds (pick 'em).

http://espn.go.com/nfl/picks

I'm personally open to the possibility that, while even multivariate models are never going to be precise as some would like (leading to dismissals of the form, "this model didn't exactly predict the score of the Vikes-Packers game so it's useless), they may nonetheless add a LITTLE value beyond the wisdom of crowds (at least until the crowds get wiser still
) . . .

All best,
Pete

[fluffyistaken]

[Rodc]

Hi Jeffy,

My data were a couple of years old, so I guess I'm not too surprised to see a shift given the market gyrations of the last couple of years. Might be shifting noticeably nearly monthly. But the point remains regardless of minor details.
_________________
"all standard caveats apply"

[Adrian Nenu]

[jeffyscott]

[wbond]

[Adrian Nenu]

Interesting that 10 years ago nobody using the Gordon model or any other predicted negative returns for the S&P 500 index or even that bonds would outperform it. I guess it all depends on the after the fact input to get the formula to match the past. It's still a poor predictor of future stock returns no matter how the numbers are tortured and it's pure luck if it works.

Adrian
anenu@tampabay.rr.com

[Verde]

wbond, I want to congratulate you on a great post.
Thanks for that.

[jeffyscott]

[cjking]

[Rodc]