How to think about expected returns--continuing discussion

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

How to think about expected returns--continuing discussion

Post by larryswedroe »

robertf57
Posts: 25
Joined: Thu Jun 19, 2014 2:01 pm

Re: How to think about expected returns--continuing discussi

Post by robertf57 »

Although I am a fan of Monte Carlo simulation for stochastic processes, your article is a bit of a non-sequitur! Your intro lays out that historical results are not reliable and that starting valuations clearly matter and then you advocate a method for return prediction (Monte Carlo simulation) which requires an estimation of an expected return and its variation (presumably from history) and which is entirely independent of current valuation….. :shock:
james22
Posts: 1992
Joined: Tue Aug 21, 2007 2:22 pm

Re: How to think about expected returns--continuing discussi

Post by james22 »

Your position on matching market exposure to expected returns, Larry?
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

Robert
Don't understand at all your issue.
The research by many shows that the BEST estimate of future returns is current valuations, with similar results if use current earnings but slightly better if use a CAPE measure like Shiller's. So I recommend, and we use, the method that the literature says is the best. Because we know that the estimate is just a mean of a wide potential dispersion we use MCS to help ESTIMATE the odds of success of various AA's and withdrawal rates to help determine the right AA and spending for each person.
Now we do use historical SDs which are I think the best estimate of future SD (don't know any better one) and also importantly historical correlations, and again think that is the best estimate of future correlations (don't know any better). So in world of no clear crystal balls you use the best tools you have and understand that your outcomes are only ESTIMATES of the odds. And that is why it's also important to develop plan Bs, the actions you will take if the left tail risk shows up. It's also why an IPS and AA should be living documents, adjusting them as realized returns occur that are different than estimated returns.

James
I would not use expected returns to try and time markets at all for most people, and for the few who understand how slippery a slope it can be to only attempt it at very extreme valuations on the high side.
Now your AA should reflect estimated returns based on current valuations, so to achieve your goal they need to be considered. So say you have a low ERP estimate and a high value premium estimate, that might lead you to give more consideration to the Larry type portfolio (1999 would be good example of that). Otherwise you'll have to have a high equity allocation when ERP is low. On other hand if ERP is high perhaps you might not want to tilt as much (especially if value premium looks lower because BTM spreads between value and growth is narrow--possible). Totally ignoring valuations makes no sense to me, but one must be disciplined in implementing whatever you choose and not confuse strategy and outcome.

Larry
robertf57
Posts: 25
Joined: Thu Jun 19, 2014 2:01 pm

Re: How to think about expected returns--continuing discussi

Post by robertf57 »

LArry,

Perhaps I am reading too much into your piece or misinterpreting your statements; but, let me explain my take on it.
there isn’t any methodology that will produce highly accurate forecasts of stock returns.........
However, we do know that starting valuations clearly matter—a whole lot. We also know they are a far more accurate predictor of returns over the next decade than historical returns.
I read this to mean that you believe empirically that historical returns are not great predictors of future performance and that some assessment of valuation is better. Now you don't specifically say what valuation model you are referring to; but I'll assume CAPM/CAPE for the sake of discussion... OK SO now I am expecting to see a predictive model that is a function of valuation.. See where I am going with this? Then you write:
So how should you address the problems surrounding the difficulty of forecasting returns? We think the best way to deal with the issue of probabilistic forecasts is to use a Monte Carlo simulator
.

With MC, we assume (or generate) some distribution of returns and their variance ....
the Monte Carlo simulation program generates a sequence of random returns from which one result is applied in each year of the simulation’s time frame. The process is repeated thousands of times to calculate the likelihood of possible outcomes and their potential distributions.
Unless you are generating your distribution of returns in some novel way, this process does not include valuation at all. It is directly related to historical returns which we asserted weren't very good.


I would not argue with MC simulation methods at all. I just think the blurb (the way I read it at least ) doesn't follow itself.....or what am I missing? ..
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: How to think about expected returns--continuing discussi

Post by richard »

Valuation ratios have done a better job of predicting returns than other methods, including relying on historical returns. For example, Vanguard found "We confirm that valuation metrics such as price/earnings ratios, or P/Es, have had an inverse or mean-reverting relationship with future stock market returns, although it has only been meaningful at long horizons and, even then, P/E ratios have “explained” only about 40% of the time variation in net-of-inflation returns. Our results are similar whether or not trailing earnings are smoothed or cyclically adjusted (as is done in Robert Shiller’s popular P/E10 ratio). ... the fact that even P/Es—the strongest of the indicators we examined—leave a large portion of returns unexplained underscores our belief that expected stock returns are best stated in a probabilistic framework, not as a “point forecast,” and should not be forecast over short horizons." https://personal.vanguard.com/pdf/s338.pdf

Presumably, Larry would use the returns from the valuation ratio as input to the MC simulation.

Personally, I'd use e/p (or e10/p) as a guess for future equity returns (and current bond yields for bond returns), then add a large error range (e.g., plus or minus 5 or 10 percentage points). It's not all clear that MC does any better.
countmein
Posts: 653
Joined: Fri Dec 06, 2013 8:10 pm

Re: How to think about expected returns--continuing discussi

Post by countmein »

Speaking of expected returns, RAFI has a fancy interactive expected returns tool here: http://www.researchaffiliates.com/Asset ... rview.aspx

Rejoice, they've got US Small at 0%! That's a lot better than GMO's -5% forecast.
countmein
Posts: 653
Joined: Fri Dec 06, 2013 8:10 pm

Re: How to think about expected returns--continuing discussi

Post by countmein »

larryswedroe wrote:So say you have a low ERP estimate and a high value premium estimate, that might lead you to give more consideration to the Larry type portfolio (1999 would be good example of that).
Larry
Larry- what are your current ERP and Value premium estimates?
Rodc
Posts: 13601
Joined: Tue Jun 26, 2007 9:46 am

Re: How to think about expected returns--continuing discussi

Post by Rodc »

In order for this to be very useful the mean assumed in the simulation has to adjust over time as valuations adjust.

For example if there is a strong period of returns the valuations might then be high and thus returns assumption going forward must of course use a lower mean, or perhaps the earning have been strong and so valuations are still not high (if using a P/E model of some sort). Or one might think in terms of the Gordon equation and perhaps earnings have been above average and so the sim should assume they will mean revert.

Clearly it makes little sense, especially if one is periodically adding money or if one is rebalancing to use only one distribution based on today's valuations which exists only at a point in time when simulating a multi-year period. For example I might look at P/E10 and say that the distributions of real return going forward is 5% with a std of 20%. But then if the sim has a few down years the sim perhaps should shift towards higher returns since presumably after some down years the valuations are likely more favorable for good returns (unless of course the good returns are due to better than expected earnings rather than changes in P/E, so this is not just a simple matter of adding reversion to the mean).

Larry, how exactly do you account for changing valuations over time in your simulations and what work have you done to validate and verify your model? This is something I have given some thought to, but have not yet coded up, so I would appreciate any insight.
We live a world with knowledge of the future markets has less than one significant figure. And people will still and always demand answers to three significant digits.
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

Rod/countmein
We use the current valuation, based on CAPE 10 basically and then do the same not only for US but EM and Developed. We do it once a year, in January.
Then add historical premiums based on loadings we have for the factors.
That's the unconditional forecast--regardless of time horizon. And yes it makes sense to use one estimated return---but that gives you of course 3,000 in our run, different outcomes. As valuations change you change he expected mean and SDs and correlations (we use the longest term data for those).
Countmein
Roughly that gets you if memory serves some like nominal return of 6.5 for US. 7.5 developed and 8 for EM, then add premiums based on loadings
Larry
Rodc
Posts: 13601
Joined: Tue Jun 26, 2007 9:46 am

Re: How to think about expected returns--continuing discussi

Post by Rodc »

And yes it makes sense to use one estimated return--
Only if you have a one year or less horizon. After that the valuation has changed and your simulation had better have changed as well.
We live a world with knowledge of the future markets has less than one significant figure. And people will still and always demand answers to three significant digits.
robertf57
Posts: 25
Joined: Thu Jun 19, 2014 2:01 pm

Re: How to think about expected returns--continuing discussi

Post by robertf57 »

richard wrote:Valuation ratios have done a better job of predicting returns than other methods, including relying on historical returns. For example, Vanguard found "We confirm that valuation metrics such as price/earnings ratios, or P/Es, have had an inverse or mean-reverting relationship with future stock market returns, although it has only been meaningful at long horizons and, even then, P/E ratios have “explained” only about 40% of the time variation in net-of-inflation returns. Our results are similar whether or not trailing earnings are smoothed or cyclically adjusted (as is done in Robert Shiller’s popular P/E10 ratio). ... the fact that even P/Es—the strongest of the indicators we examined—leave a large portion of returns unexplained underscores our belief that expected stock returns are best stated in a probabilistic framework, not as a “point forecast,” and should not be forecast over short horizons." https://personal.vanguard.com/pdf/s338.pdf

Presumably, Larry would use the returns from the valuation ratio as input to the MC simulation.

Personally, I'd use e/p (or e10/p) as a guess for future equity returns (and current bond yields for bond returns), then add a large error range (e.g., plus or minus 5 or 10 percentage points). It's not all clear that MC does any better.
Using starting valuation of eacah period to condition the distribution of returns sampled for the MC simulation makes sense. That wasn't clear to me from the blurb. thanks!
User avatar
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to think about expected returns--continuing discussi

Post by siamond »

Rodc wrote:
And yes it makes sense to use one estimated return--
Only if you have a one year or less horizon. After that the valuation has changed and your simulation had better have changed as well.
Well, it depends what you use the expected returns (and the variations around it) for. Say this is for a form of TAA, then for sure, you need to reassess every year. If this is just to see the impact of a fixed annual withdrawal over 20 years, then maybe not.
User avatar
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to think about expected returns--continuing discussi

Post by siamond »

larryswedroe wrote:Rod/countmein
We use the current valuation, based on CAPE 10 basically and then do the same not only for US but EM and Developed. We do it once a year, in January
Larry, would you mind being more specific? Do you use 1/cape? X/cape? Or something like Bogle (dividends yield + earnings growth + speculative return adjustment derived from cape)? Or else? Thanks.
james22
Posts: 1992
Joined: Tue Aug 21, 2007 2:22 pm

Re: How to think about expected returns--continuing discussi

Post by james22 »

larryswedroe wrote:Now your AA should reflect estimated returns based on current valuations, so to achieve your goal they need to be considered.
I agree.

But I also believe there is an (inverse) relationship between ERP and risk (when ERP is low, risk is high).

Your AA works when ERP is high, as you can expect to achieve your goal with a low equity allocation and little risk.

But when ERP is low? To increase your equity allocation is risky.

Do you really believe it less risky than "market timing" by matching market exposure to expected returns (low when low, high when high), Larry?

Matching market exposure to expected returns maintains risk constant. Your AA does not.
Rodc
Posts: 13601
Joined: Tue Jun 26, 2007 9:46 am

Re: How to think about expected returns--continuing discussi

Post by Rodc »

siamond wrote:
Rodc wrote:
And yes it makes sense to use one estimated return--
Only if you have a one year or less horizon. After that the valuation has changed and your simulation had better have changed as well.
Well, it depends what you use the expected returns (and the variations around it) for. Say this is for a form of TAA, then for sure, you need to reassess every year. If this is just to see the impact of a fixed annual withdrawal over 20 years, then maybe not.
Well, TA is what Larry preaches (ie use valuations to determine allocation. He is not alone, Bernstein does as well).

But even if one does not use TA (Using valuations to set allocations), if one puts in $10,000 this year with valuations such that one expects 5% going forward and next year one puts in $10,000 when valuations say 6%, and $10,000 the next year when one expects returns to be 8% (say in a down turn like 2007, 2008, 2009) and then one keeps going year after year adding funds at different valuations, a MC run made back at the beginning (when by the way one knew they would be adding money every year and of course put that into the simulation, can't ignore that!) would be absolute garbage (as in garbage in/ garbage out).

If you use some long term average mean you have some hope that at least in the process above sometimes you will be low and sometimes will be high and it might wash out, sort of. If you start in a high valuation period and so start with a low return and use that in a 30 year sim where you are adding or subtracting funds you introduce a huge unwarranted bias because the high valuation assumption only applies to some of the money but not all of the money.

The only way what Larry is advocating makes sense if you are using a typical MC based on something like a lognormal random number generator with fixed mean and standard deviation is if you have a lump sum and do little if any trading. That is not very many of us. Otherwise the MC must track valuations (and not just returns) and apply a random number generator whose mean (and ideally standard deviation) is driven by a valuations based model of expect returns.
We live a world with knowledge of the future markets has less than one significant figure. And people will still and always demand answers to three significant digits.
Rodc
Posts: 13601
Joined: Tue Jun 26, 2007 9:46 am

Re: How to think about expected returns--continuing discussi

Post by Rodc »

Here is a more intuitive non-math explanation of the problem.

A Monte Carlo simulation is a statistical model of the stock and bond markets. It is only useful to the extent it is a reasonably decent representation of how markets behave.

If valuations are truly important to real investors in the real world, then to be a reasonably decent representation of how markets behave a model of market behavior will need to incorporate valuations in a way that is representative of the real world. Since valuations change in dramatic ways over time the model used in the Monte Carlo simulation must also change over time. And if it does not, it should at least be right on average and not wrong on average. Using a fixed model based on an extreme point must by definition be wrong on average.

Reader's digest version: If time varying valuations are critical to investors then time varying valuations are critical to simulations. If they are not critical to simulations then they are not critical to investors.
We live a world with knowledge of the future markets has less than one significant figure. And people will still and always demand answers to three significant digits.
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

two things
First, currently we use the CAPE 10 and the Gordon Model, weighting each 50%. And we apply them across US, Developed and EM.

Second, when valuations are high expected returns are low, causing need to take risk to go up. But as I have said many times that doesn't mean you SHOULD raise your equity allocation. That means you have need to do something. That could be taking higher equity allocation, or tilting more. But it could also mean you should lower your goal. Or perhaps plan on working longer, or cut spending today to invest more. The right answer depends on each person's unique ability, willingness and need to take risk (marginal utility of wealth).

Larry
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: How to think about expected returns--continuing discussi

Post by richard »

Rodc wrote:<>Reader's digest version: If time varying valuations are critical to investors then time varying valuations are critical to simulations. If they are not critical to simulations then they are not critical to investors.
larryswedroe wrote:<>First, currently we use the CAPE 10 and the Gordon Model, weighting each 50%. And we apply them across US, Developed and EM<>
While averaging CAPE 10 and Gordon seems a perfectly reasonable method of producing a point estimate of future returns, you seem to be using the estimate only as an initial input to a standard MC simulation. If so, this doesn't answer Rod's comment that you should be adjusting returns throughout the simulation to take into account valuation variations over time.

I continue to doubt MC simulations do any better than applying wide error bands to returns estimates. There's more than enough slop in the system to drown any differences between the approaches (as there is to drown any differences between p/e, CAPE and Gordon - e/p has the advantage of not having to guess future growth rates).

It might be that time varying valuations don't do any better than using an initial point return estimate for MC simulations, but if so the reason is the general problem with MC - it doesn't really add anything beyond just using wide error bands.
countmein
Posts: 653
Joined: Fri Dec 06, 2013 8:10 pm

Re: How to think about expected returns--continuing discussi

Post by countmein »

Larry,

Can you show us your math? I would think 50/50 cape10/Gordon would end up at about 0-2% for US, 4-6% International.

Also, what is your estimate of size and value premia and what are the calculations behind those?
User avatar
SimpleGift
Posts: 4477
Joined: Tue Feb 08, 2011 2:45 pm

Re: How to think about expected returns--continuing discussi

Post by SimpleGift »

larryswedroe wrote:...currently we use the CAPE 10 and the Gordon Model, weighting each 50%.
In forecasting returns, I'm beginning to see the advantage of averaging these two methods — since the CAPE earnings yield (inverse of the price-earnings ratio) is so strongly influenced by price, while the Gordon Equation is strongly influenced by the earnings/dividend growth rate. Thus the first method includes a sense of current valuations and the second a sense of expected economic growth.

I note that Antti Ilmanen (author of the book, Expected Returns) averages the two methods in much of his work.
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

Countmein
I'm simplifying here but the CAPE 10 is now at about 24.7 http://www.multpl.com/shiller-pe That produces an earnings yield of about 4.0 percent. However, we need to make an adjustment to arrive at the forecasted real return to stocks because the earnings figure from the CAPE 10 is on average a lag of 5 years. With real earnings growing about 1.5 percent a year, we need to multiply the 4.0 percent earnings yield by 1.075 percent (1.5 percent x 5 years). That produces a real expected return to stocks of about 4.4 percent.
Then depending on loading on portfolio we would add the historical premiums. So our Risk Target 3 (loading of about .3 on each of size and value) would take the premiums and multiply by 0.3. If it was the Larry Portfolio it would be RT 6, so multiply by 0.6 the historical premiums
We do the same for developed and for EM. Then weight them by the allocation in the portfolio. So each one is different. As of January 31, 2014 the current earnings yields were 5.94 for the MSCI EAFE Index and 6.64 for the MSCI Emerging Markets Index. Accounting for the lag adjustment increases those yields to 6.39 percent for the EAFE Index and 7.14 percent for the Emerging Markets Index. Then weight them depending on client allocations.
Hope that helps
Larry
countmein
Posts: 653
Joined: Fri Dec 06, 2013 8:10 pm

Re: How to think about expected returns--continuing discussi

Post by countmein »

Thanks, Larry.

Three questions:

Is 4.4 earnings yield not nominal?
No accounting for speculative return / RTM?
No adjusting value or size premia expectation based on current spreads?

Here's my calculation, where am I going wrong?

CAPE10: 4.5% nominal (1/25, adjusted by lag) - 2% inflation - 1% speculative (assume PE10 goes to 15 over 10 years) = 1.5% real
Gordon: 1.5% dividend (currently on DFA core 1) + 1.5 earnings growth (per Illmanen) = 3% real

Combine them 50/50 and you get a little over 2% real. This is more in line with Bernstein, RAFI, and GMO. And subtract a percent or two for small cap since their PE's are much higher than large caps.
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

countmein
The adjusted earnings yield is the forecasted REAL return to which you add expected inflation if you want to nominal return (so your calculation is wrong, you don't subtract inflation and IMO would not subtract any speculative return either)

We assume current price is the right price (by price I mean valuations). Anything else is obviously just a guess.

We have thought about doing that, making adjustment based on spread on btm. IMO that is what should be done but no real good way to make an adjustment.

As to future earnings growth. FWIW 1.5% might be conservative for a couple of reasons
First, dividends are now currently much lower than historical with only about 30% of companies now paying them. More retention of earnings equals, or should, faster earnings growth. And you have higher buybacks than ever.
Larry
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: How to think about expected returns--continuing discussi

Post by richard »

I tend to prefer e/p over Gordon because with e/p you don't have to estimate future earnings growth.

Arnott and Asness claim that higher dividends equal higher earnings growth. http://www.rallc.com/Production%20conte ... Growth.pdf
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

Couple of things
First, here's example of why I prefer CAPE 10 over Gordon.
Say you have d/p of 2% and expect growth of 1.5%. So you have real expected return of 3.5%. Now valuations double so you have d/p of 1%. So with prices doubling you see expected returns cut just 1% to 2.5%.
Now say you started with CAPE 10 of 20 and it doubles to 40 as valuations double, Now expected returns fall from 5% to 2.5%. Much more significant drop and more realistic.

Now say half companies stop paying dividends in first example, the d/p will fall and expected returns fall while with CAPE 10 nothing has changed. Simple financial theory, Modigliani and Miller, say div policy should not matter but with Gordon it does UNLESS you adjust growth (which IMO you should)

Now Asness and Arnott showed that more retention of earnings didn't lead as theory said to higher earnings---meaning management was bad stewards of capital, engaging likely in bad acquisitions for example that just made their jobs bigger and gave them more power if you will. Perhaps no difference but today far fewer companies paying dividends and many more buying back stock instead, and that should lead to higher earnings per share growth. And that doesn't show in Gordon but does in CAPE.

Larry
209south
Posts: 528
Joined: Mon Jan 28, 2013 9:58 pm

Re: How to think about expected returns--continuing discussi

Post by 209south »

Second, when valuations are high expected returns are low, causing need to take risk to go up. But as I have said many times that doesn't mean you SHOULD raise your equity allocation. That means you have need to do something. That could be taking higher equity allocation, or tilting more. But it could also mean you should lower your goal. Or perhaps plan on working longer, or cut spending today to invest more. The right answer depends on each person's unique ability, willingness and need to take risk (marginal utility of wealth).

THIS is why projecting expected returns is important to me - in your 20s and 30s it doesn't matter much, but I'm 54 and contemplating early retirement, and it is REALLY important to have a point of view on what sort of lifestyle/legacy my portfolio will support - conservatively applying a lot of different people's estimates, I'm estimating 1.36% real returns on my portfolio, and that means I am working 5 more years to get to a comfort level. If that ends up proving conservative, I will rejoice...but discussions like this have really impacted my planning, and is another reason this forum has been invaluable for me!
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

209south
Exactly right. One should adjust their plan accordingly. Lower expected returns might mean you should take more risk, or spend less (save more), lower your goal, or plan on working longer, or any combination of these. No right answer, but one right for each person. And at least one should have estimate of odds of success while understanding that it's just an estimate and that it's important to have plans that can be adapted to outcomes left of the mean, as required. In other words, have Plan Bs already in place so not making decisions when under stress.

Larry
User avatar
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to think about expected returns--continuing discussi

Post by siamond »

larryswedroe wrote:I'm simplifying here but the CAPE 10 is now at about 24.7 http://www.multpl.com/shiller-pe That produces an earnings yield of about 4.0 percent. However, we need to make an adjustment to arrive at the forecasted real return to stocks because the earnings figure from the CAPE 10 is on average a lag of 5 years. With real earnings growing about 1.5 percent a year, we need to multiply the 4.0 percent earnings yield by 1.075 percent (1.5 percent x 5 years). That produces a real expected return to stocks of about 4.4 percent.
Even ignoring the bit about the adjustment, this appears to be a somewhat rosy way of doing things... I ran backtesting of the E/P (using CAPE10) against S&P500 over many time periods since 1871 (Shiller data), and a simple use of an E/P formula consistently over-estimates actual returns. Depending on exactly how you do it, you get from half a point to nearly a full point. Now fact is, starting from 1940 and looking at 50-years periods, the alignment is much better, but... it is hard to trust a formula that works ok half of the time, and does not half of the time.

This lead some folks on this forum to add an x% factor to the E/P formula that I could never make sense of. Worse, if you look at it over time, the x% factor has to significantly change (although it's always under 1), so who is to say what you'll need in the future...

It is one thing to have a half-decent correlation (yes, as Vanguard said in their paper, you get an R2 of 0.4 - actually better in recent times), but if the average return is overly flawed, then there is a bit of an issue here. Maybe I did something wrong though... Larry, do you have some backtesting results to share here?

PS. that being said, the Gordon way of doing things (including speculative returns RTM) tends to underestimate returns. It never crossed my mind to do 50% Gordon + 50% E/P, but I can see that this might happen to work ok, but where is the logic is doing so? :shock:
User avatar
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to think about expected returns--continuing discussi

Post by siamond »

larryswedroe wrote:First, here's example of why I prefer CAPE 10 over Gordon.
Say you have d/p of 2% and expect growth of 1.5%. So you have real expected return of 3.5%. Now valuations double so you have d/p of 1%. So with prices doubling you see expected returns cut just 1% to 2.5%.
Now say you started with CAPE 10 of 20 and it doubles to 40 as valuations double, Now expected returns fall from 5% to 2.5%. Much more significant drop and more realistic.
For sure. Just using the fundamental factors in a Gordon-like equation is not suitable for 10-to-20 years forecasts. It is only suitable for the very long run. You need to add a 'speculative' RTM factor to the equation to make it better capture the effect of valuations over the returns of the coming couple of decades. Both Bogle and Bernstein explain that very well, and backtesting tends to confirm their findings (which frankly, are very intuitive).

Now I am starting to better understand where your 50%/50% logic is coming from though. Seems quite ad hoc, but the 'overly fundamental' view of the (div yield + growth) Gordon equation combined with the 'overly speculative' view of the E/P (CAPE10) logic might very well not be far from the truth. This sounds VERY ad hoc though...
User avatar
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to think about expected returns--continuing discussi

Post by siamond »

siamond wrote:Now I am starting to better understand where your 50%/50% logic is coming from though. Seems quite ad hoc, but the 'overly fundamental' view of the (div yield + growth) Gordon equation combined with the 'overly speculative' view of the E/P (CAPE10) logic might very well not be far from the truth. This sounds VERY ad hoc though...
I was intrigued, so I gave it a quick test, and surprise, surprise, this combo does give reasonably good results. The average returns are better centered on where they should be, and the correlation is improved. I still get slightly better results by using a Gordon equation with a speculative factor, but given the margins of error on the whole thing (rodc, don't get started! :wink:), this is probably not terribly significant. Interesting. I'm still struggling to rationalize it though.
User avatar
SimpleGift
Posts: 4477
Joined: Tue Feb 08, 2011 2:45 pm

Re: How to think about expected returns--continuing discussi

Post by SimpleGift »

siamond wrote:Now I am starting to better understand where your 50%/50% logic is coming from though. Seems quite ad hoc, but the 'overly fundamental' view of the (div yield + growth) Gordon equation combined with the 'overly speculative' view of the E/P (CAPE10) logic might very well not be far from the truth. This sounds VERY ad hoc though...
We shouldn't forget that the purpose of forecasting expected returns is primarily for "broad brush" financial planning only. There will never be a high level of precision in these forecasts and there doesn't really need to be. If one can estimate that expected real stock returns will probably be in the 3%-4% range over the next 30 years — and that it's highly unlikely that they'll be close to 7%-8% historical averages — then one has useful information for long-term planning purposes, I believe.

I'm all for refining the methods of making these forward return estimates (and averaging CAPE earnings yield and the Gordon Equation seems sensible), but let's not forget the intended use of these estimates — for "back of the napkin" planning only!
Last edited by SimpleGift on Sun Oct 19, 2014 10:35 am, edited 1 time in total.
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

few thoughts
First, there is clearly no exact "science" here --the best models only explain about 40% of the future outcomes. But of course this should not be a surprise, for if the models explained 100% then there would be no risk. In other words, surprises happen, both good and bad. And when you look at the data you find the CAPE 10 does pretty good job in terms of the mean and if you add and subtract about 8% you get the full range of outcomes over 10 year periods. Now that is wide dispersion, meaning stocks are indeed very risky. And your plans should incorporate that. (Also why I favor a low beta/high tilt portfolio to cut the tails).

Second, no magic about the 50/50. We used to use only the Gordon model. And I fought for the CAPE 10 and in our committee we have varying views and in end we agreed to go with 50/50 (see the results Saimond found).

Third, I personally don't like the adjustment for speculation as that assumes you know better than the market. But very high valuations should give one great pause about potential for mean reversion at least. But there is no right ERP. And the evidence and logic is that it should be no surprise that it should be lower today than it was say in 1900 or 1950. Many good reasons for this. So that is why I've written about the Shiller average of 16 being a bad "benchmark" to expect valuations to revert to. Perhaps something more like 20 or bit higher makes more sense. Meaning stocks still look highly valued but not in bubble in anyway.

Fourth, note quite a while ago Fama and French said that they believed the true ex ante ERP was really in area of 4% and that we just got lucky outcome. So that is in line with what most financial economists think today.

Bottom line is stocks are very risky and no model can be expected to have high accuracy of outcomes looking great or there would be no risk, or very little risk, which we know is false. Models do however have use in helping to understand the possibilities, probabilities and likely range of outcomes for which one should prepare. And there is simply no way IMO to build a plan without making assumptions about returns, so you use the best tools you have (even if not perfect, and understand their limitations)

Larry
User avatar
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to think about expected returns--continuing discussi

Post by siamond »

Fair enough, Larry. Thank you for making me think about it with an angle I never thought of.

Oh, and I totally agree with your Shiller average comment. In my speculative version of the Gordon model, I use an RTM on a valuation metric (I played with PE10, or PE20/25/30, or Tobin's q), but I use a rolling average over the past 50 years to determine the 'mean' to compare to. And yes, we're closer to 20 than 16, following such logic. Some things do change, notably the accounting ways...

PS. Simplegift, you point is well taken, this is all 'black arts' for sure, but if one uses a method which is CONSISTENTLY wrong by more than 0.5% to 1% IN THE SAME DIRECTION, then something does need a bit of tuning... Take 0.75% of your portfolio, think of it as a delta on your annual withdrawals over your entire retirement period, and this will probably make you blink a tad. I know I blinked when doing so a little while ago!
User avatar
SimpleGift
Posts: 4477
Joined: Tue Feb 08, 2011 2:45 pm

Re: How to think about expected returns--continuing discussi

Post by SimpleGift »

larryswedroe wrote:And there is simply no way IMO to build a plan without making assumptions about returns, so you use the best tools you have (even if not perfect, and understand their limitations).
I'll just add that one of the beauties of these simple forecasting tools is that they are easily accessible to the average investor, using information that's commonly available. Whether one uses the earnings growth model (Bogle), the dividend growth model (Bernstein), the CAPE earnings yield, or some average of these, they are all simple, easy and quick to calculate.

If we start adding adjustments for speculative returns here, adjustments for share buybacks there, and so on, these simple tools quickly become more cumbersome than they need to be, in my view, for the level of precision in rough planning estimates. Better they be simple and easy to calculate quickly, so they'll actually be used by the average investor.
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

siamond
FYI, if unaware FAS 142 instituted around 2001 I think forced immediate writedown of goodwill vs 40-year write off. It's estimated that alone raises the figure by about 3. And the fewer companies paying dividends should lead to about another 1 adjustment. So if you look at say 19 as the average after 1960. Then add four you get about 23. And that's not that much above today's figure, and whose to say that today's is too high. As countries get wealthier and have stronger disclosures and stronger Central Banks, etc the ERP should be expected to fall, all else equal.
That's why I have never bought into the hype about Shiller CAPE 10 being too high. It's high for sure. But too high, I don't know.
Larry
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: How to think about expected returns--continuing discussi

Post by richard »

209south wrote:THIS is why projecting expected returns is important to me - in your 20s and 30s it doesn't matter much, but I'm 54 and contemplating early retirement, and it is REALLY important to have a point of view on what sort of lifestyle/legacy my portfolio will support - conservatively applying a lot of different people's estimates, I'm estimating 1.36% real returns on my portfolio, and that means I am working 5 more years to get to a comfort level. If that ends up proving conservative, I will rejoice...but discussions like this have really impacted my planning, and is another reason this forum has been invaluable for me!
Estimating returns to two decimal places implies a fair amount of false precision. The error bands around any estimate could easily be plus or minus 5 percentage points.

Vanguard reports a p/e around 20x for TSM, which suggests a real equity return of 5%. Ten year TIPS are yielding a touch over 0% real. If you had a 50/50 portfolio, that would imply around 2.5% real. Call it 2% plus or minus 5%.

The question is then how much you can withdraw from your portfolio.

Conservative is good in this context.
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: How to think about expected returns--continuing discussi

Post by richard »

larryswedroe wrote:few thoughts
First, there is clearly no exact "science" here --the best models only explain about 40% of the future outcomes. But of course this should not be a surprise, for if the models explained 100% then there would be no risk.
And for those who haven't thought this through, no risk means no risk premium. In other words, if we could forecast perfectly, returns would likely be very low.
larryswedroe wrote: In other words, surprises happen, both good and bad. And when you look at the data you find the CAPE 10 does pretty good job in terms of the mean and if you add and subtract about 8% you get the full range of outcomes over 10 year periods. Now that is wide dispersion, meaning stocks are indeed very risky. And your plans should incorporate that. (Also why I favor a low beta/high tilt portfolio to cut the tails).
Are you saying one should take the return implied by CAPE (or whatever model), then apply plus or minus 8 percentage points for the entire 10 year period? In other words, if CAPE implies 5% real, you'd expect a portfolio might lose 3%/year or gain 13%/year (or anything in between) for 10 years?

If so, is this based on history or something else?
209south
Posts: 528
Joined: Mon Jan 28, 2013 9:58 pm

Re: How to think about expected returns--continuing discussi

Post by 209south »

With respect to my returns estimate, I certainly agree, Richard - my 1.36% is the result of other 'rounded estimates' applied to different parts of my portfolio (0 for cash, 0.5% for TBM/Munis/TIPs, 3.0% for domestic equities, 3.5% for international equities, 2.5% for REITs, etc. I arrived at those estimates months ago after reviewing a variety of articles re. future returns. I tried to err on the conservative side and the 'portfolio result' is 1.36% (and shifts modestly every time the underlying values shift with market moves). I know the effort is somewhat futile, but my model shows a very affluent retirement if I assume traditional, historical returns (which I did formerly) - to Larry's point, using this lower returns outlook has caused to consider lowering my spending (not yet!), taking more risk with equities (not willing to!) or delaying retirement (my choice for now!) I hope we have another 5 years of strong returns so that I can put more money away in safe places, and retire with confidence in my late-50s!
Hank Moody
Posts: 279
Joined: Fri Oct 17, 2014 9:41 pm

Re: How to think about expected returns--continuing discussi

Post by Hank Moody »

Larry,

I'm a little new to this forum but I'm well versed in BAM and EMH topics. Can you help me understand why calculating expected returns is important in the work that you do? Are you using it to make tactical investment decisions or for other reasons?

I realize that there have been a lot of threads of this topic. Please give me some context for the current thread so I can follow along.

Thank you,
HM
-HM
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

Richard
Yes, exactly, and it is based on history, results from study by Asness. Even when forecast low returns still possibility of very high real returns, and vice versa.

AS to the issue of an MCS not adjusting future returns based on the first year returns. I wanted to wait before answering because we happen to be very lucky and have a world class mathematician on our staff (coached the Rumanian Math Olympiad team), who also created/taught the math of finance program at Wash U. His response was that while correct in theory it is not a major issue, at least a second order of magnitude, if not third, problem, similar to the issue of not using skewed/fat tailed distributions in that the outcomes would change very little (percent odds of success) relative to even a small change in expected returns. Errors made in effect cancel each other out. So if you have some years when returns way down and thus future should be higher, the reverse is true on other side.

I would add to this that one should not do an MCS once and forget it. Whenever realized returns are dramatically different than expected you should rerun.

Larry
james22
Posts: 1992
Joined: Tue Aug 21, 2007 2:22 pm

Re: How to think about expected returns--continuing discussi

Post by james22 »

larryswedroe wrote:But very high valuations [/lower expected returns] should give one great pause about potential for mean reversion at least.
larryswedroe wrote:Lower expected returns [/high valuations] might mean you should take more risk, or spend less (save more), lower your goal, or plan on working longer, or any combination of these.
With great respect, Larry, I just do not understand why taking on more risk is riskier than "market timing" by reducing your equity allocation when valuations are high.

Unless you expect the market to never revert (which would mean there is no risk), why not reduce and wait for a lower valuation with higher expected returns? There's little opportunity cost in a time of lower expected returns, after all.
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

James
First, not arguing one SHOULD TAKE more risk, only that one needs higher returns, but as I explained there are more than that one way to achieve that objective. But the right answer depends on each person.
Second, the reason should perhaps not choose to wait till mean reversion is that you don't know the market is "too high" or that it will ever revert to some magical mean. We don't know what the right mean is. For example, with CAPE 10 the historical average is 16, but since 1960 it's 19. And if you make adjustments for change in accounting and fewer companies paying dividends its 23. Now is 24.7 too high? And with the mean increasing over time maybe the "new" mean is really 24.7?
High valuations just mean lower future expected returns not bad returns, or negative returns.
And once you sell how will you know when to buy again? What if we never get back to 16 again? And you cannot make your goals in riskless investments.

The problem with timing is that you have to be right twice not once. And the opportunity cost isn't low as you state, it probably is about 5% in real terms if no tilt and with big tilt might be like 8-9%
Larry
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: How to think about expected returns--continuing discussi

Post by richard »

larryswedroe wrote:Richard
Yes, exactly, and it is based on history, results from study by Asness. Even when forecast low returns still possibility of very high real returns, and vice versa.
Larry, do you have a title, link or other means of finding the Asness study?

On MCS, I continue to wonder what MCS gets you compared to forecasting equity based on e/p (or Gordon), forecasting bonds based on current yield and adjusting plus or minus 8 percentage points.

Thanks for asking your math guy. I believe his response applies to a lot about forecasting - the issue doesn't much matter in the real world.
larryswedroe wrote:AS to the issue of an MCS not adjusting future returns based on the first year returns. I wanted to wait before answering because we happen to be very lucky and have a world class mathematician on our staff (coached the Rumanian Math Olympiad team), who also created/taught the math of finance program at Wash U. His response was that while correct in theory it is not a major issue, at least a second order of magnitude, if not third, problem, similar to the issue of not using skewed/fat tailed distributions in that the outcomes would change very little (percent odds of success) relative to even a small change in expected returns. Errors made in effect cancel each other out. So if you have some years when returns way down and thus future should be higher, the reverse is true on other side.

I would add to this that one should not do an MCS once and forget it. Whenever realized returns are dramatically different than expected you should rerun.

Larry
Rodc
Posts: 13601
Joined: Tue Jun 26, 2007 9:46 am

Re: How to think about expected returns--continuing discussi

Post by Rodc »

richard wrote:I continue to doubt MC simulations do any better than applying wide error bands to returns estimates. There's more than enough slop in the system to drown any differences between the approaches (as there is to drown any differences between p/e, CAPE and Gordon - e/p has the advantage of not having to guess future growth rates).

It might be that time varying valuations don't do any better than using an initial point return estimate for MC simulations, but if so the reason is the general problem with MC - it doesn't really add anything beyond just using wide error bands.
Agreed. MC is useful for making illustrations, but it is entirely unclear they have enough accuracy to be of any real value in the real world beyond a simple back of the envelop calculation. Virtually no one using them has done a proper analysis of error propagation, and they should. I have and it was a real eye opener.

Not to mention the opinion of Larry's expert which is pretty damning against the accuracy of MC below:
Larry wrote:His response was that while correct in theory it is not a major issue, at least a second order of magnitude, if not third, problem, similar to the issue of not using skewed/fat tailed distributions in that the outcomes would change very little (percent odds of success) relative to even a small change in expected returns. Errors made in effect cancel each other out. So if you have some years when returns way down and thus future should be higher, the reverse is true on other side.
I agree. In fact that is what I have already said: the time varying nature of expected returns washes out on average and at any rate the effects are likely lost in all the noise. This is precisely why it is wrong to put in a point estimate as a constant bias.

The errors in the most basic components of MC are so gross that refinements do not make sense. I am not advocating for time varying models. Rather I am pointing out the logical fallacy of using crude non-time varying models in Monte Carlo simulation within an argument for using time varying models in real life, and arguing the way to do this in real life is to use non-time varying simulations. None of that hangs together in any coherent sense.

Now Larry is arguing that among all the various market effects time varying valuations wash out in the noise while at the same time saying that somehow time varying effects must be modeled (incorrectly) at one and only one point in time and moreover this is actually important.

I for one am going with Larry's expert: the valuation effects are not all that important since they wash out in the noise of everything else going on, and at any rate we can't model them with enough accuracy to care.

The level of false precision in all this is just huge.

FWIW: I don't have a problem with the argument that one should plan on low returns for a while going forward since valuations of stocks (and bonds) are not favorable. I just have a problem with all the false precision of what that means (for example save more, wait a bit to retire, etc. vs fiddle around with asset allocations where all you are doing is playing in the noise).
We live a world with knowledge of the future markets has less than one significant figure. And people will still and always demand answers to three significant digits.
User avatar
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to think about expected returns--continuing discussi

Post by siamond »

larryswedroe wrote:siamond
FYI, if unaware FAS 142 instituted around 2001 I think forced immediate writedown of goodwill vs 40-year write off. It's estimated that alone raises the figure by about 3. And the fewer companies paying dividends should lead to about another 1 adjustment. So if you look at say 19 as the average after 1960. Then add four you get about 23. And that's not that much above today's figure, and whose to say that today's is too high. As countries get wealthier and have stronger disclosures and stronger Central Banks, etc the ERP should be expected to fall, all else equal.
That's why I have never bought into the hype about Shiller CAPE 10 being too high. It's high for sure. But too high, I don't know.
Larry
Thanks for those thoughts, Larry. Yes, I was aware of those considerations, although I would tend to think that so many pieces are in flight with earnings, that any point reasoning on one given topic (e.g. FAS 142, just to pick one) might lead to bigger mistakes when trying to quantify than better accuracy - you know, the tree and the forest... When reading Siegel's recent pieces on the matter, I fear he's falling in such trap, just fact-picking in a selective manner. This is why I like a 50-years rolling average, I hope this will capture the sum of a bunch of micro-trends without me having to think about the detailed reasoning and make mistakes about it. For sure, this is a very blunt instrument though.

Now one more question for you while we're on this... Do you solely rely on the CAPE 10 as a valuation measure? I played around with Tobin's q (either as such, or a variation of it without hindsight, e.g. rolling mean as the RTM reference), and well, there are many periods of history where those two measures were basically saying the same thing, and some periods where this isn't quite the case. I also played around with PE20/25/30. Some like Price to Book. Etc. Any thought on this?
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

saimond
The research seems to show that no magic about CAPE 10, 8 works about as well, and others likely would too. In fact current works about as well. Not fan of Tobin's Q for various reasons.

As to MCS, one of the biggest benefits is that it allows you to do MARGINAL analysis very easily --seeing how small changes or large ones in terms of the AA or spending/saving and withdrawal rates impact the odds of success. That is missing in this entire discussion, and no way IMO to do that on back of envelop, and see how those changes interact, like I want to cut equity risk and then cut my withdrawal rate to compensate and/or extend working horizon (delay retirement). MCS allows you to do these things very easily.

The key is to be humble about the outcomes, meaning that they are only ESTIMATES of the real odds, we don't know the real odds. But we can analyze the potential dispersion of outcomes and develop plans that give best odds of success in the particular situation.

For what it's worth related to Doc's comments, our math person highly values the tool and uses it in client discussions all the time. Just have to know its limitations

Larry
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: How to think about expected returns--continuing discussi

Post by richard »

Rodc wrote:<>FWIW: I don't have a problem with the argument that one should plan on low returns for a while going forward since valuations of stocks (and bonds) are not favorable. I just have a problem with all the false precision of what that means (for example save more, wait a bit to retire, etc. vs fiddle around with asset allocations where all you are doing is playing in the noise).
Even if people are warned about false precision and that there is massive uncertainty associated with these estimates, people still focus unduly on the estimates. This is an area rife with behavioral biases, including anchoring and the general issue of ignoring disclaimers.
Rodc
Posts: 13601
Joined: Tue Jun 26, 2007 9:46 am

Re: How to think about expected returns--continuing discussi

Post by Rodc »

As to MCS, one of the biggest benefits is that it allows you to do MARGINAL analysis very easily --seeing how small changes or large ones in terms of the AA or spending/saving and withdrawal rates impact the odds of success.
Yes, as already noted it is useful in making illustrations, IF all sorts of assumptions we don't really know hold and parameter p1 changes by X% and nothing else changes, then result r1 changes by Y%. just not in really doing a lot else. It is like MPT in that regard, interesting for understanding certain theoretical aspect of how markets might behave, just too numerically unstable to be directly usable.
For what it's worth related to Doc's comments, our math person highly values the tool and uses it in client discussions all the time.
Yes, Edward Jones finds it to be a fine sales tool as well, so I will grant you that one. Indeed MC results seems to be a very ubiquitous glossy brochure item. That is only a little tongue in cheek. Again though, I can see it being a useful tool for helping people understand the down stream effects of uncertain returns or other uncertainties. But that is back to using it for illustrations which has nothing to do with why you brought up MC in this thread as far as I can tell.
We live a world with knowledge of the future markets has less than one significant figure. And people will still and always demand answers to three significant digits.
Topic Author
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: How to think about expected returns--continuing discussi

Post by larryswedroe »

Rod
Very much disagree about its usefulness. Example, if you see reducing your spending rate by 5% increases your odds of success by 10% that is helpful. If decreasing your equity allocation by 10% increases your odds of not running out of money by 10% but decreases your odds of leaving an estate of say $1mm by 20%, that is useful information. And similarly seeing how tilting impacts your odds of success, or allocation more to EM or international. Models are just that, like 3 factor or five factor or 7 factor models, they help us make more informed decisions in the presence of uncertainty. Understanding the flaws is also important.
You use the best tools that you have
Larry
Post Reply