What do you use as estimate (guess) regarding future rate of return?

[ERD50]

If don't think that using static returns is very useful or insightful, other than for a reference point.

But to provide a simple reference, which I don't think I've seen commented here, for a XX year period, wouldn't zero real returns provide for a 1/XX WR (which would be 2.5% for a 40 year period)?

But that doesn't account for volatility in portfolio value, or for returns. Drawing down the portfolio while it sinks, or isn't providing above inflation returns drops it forever going forward.

Can anyone explain - what question are you trying to answer with any assumption on future returns (real or otherwise)?

edit/add - OK, I just noticed this exchange:

Quote:

Originally Posted by photoguy Quote: Originally Posted by Big_Hitter For what purpose am I making this guess? Without making an estimate of your expected returns, how do you know if your AA can meet your long term goals?

Strikes me as circular - I can't make an estimate of whether I can meet my goals unless I make an estimate of my expected returns? Either way, I'm making an estimate.

-ERD50

[Options]

Quote:

Originally Posted by ERD50 I don't use any 'number' - a single number does not include the volatility of markets and inflation, and the timing of pensions and SS. I do use the historical reporters, like FIRECalc, to better understand how my portfolio would have done in past real-world conditions. -ERD50

+1

Quote:

Originally Posted by RetireAge50 All these numbers are not meaningful unless the asset allocation is known and whether one is stating what they actually expect to happen or if they are just using a safe return.

Yes.

Quote:

Originally Posted by NanoSour 2.95% real based on "Cautious" spending using ESPlanner Monte Carlo. AA of 55/35/10.

I use the conservative mode, which assumes I earn a real return of zero (merely keeps up with inflation) in MC mode. In Upside, assuming all risky assets lose value, and safe assets earn 2%. As noted in above posts, assuming no real return is exceptionally conservative, as am I. For this reason, I'm only 40% in risky assets. Research has shown retirees are generally more interested in making their $ last a lifetime than with the size of the pot at the end (legacy goals notwithstanding). I agree with this rationale and act accordingly.

[Totoro]

Quote:

Originally Posted by ERD50 Can anyone explain - what question are you trying to answer with any assumption on future returns (real or otherwise)?

In my case: what kind of life (spending) can I afford now, and how will that impact my capacity to spend in the future?

Or differently put: what will my lifetime earnings be from this point onwards?

[photoguy]

Quote:

Originally Posted by ERD50 Can anyone explain - what question are you trying to answer with any assumption on future returns (real or otherwise)?

Here's a simple one: I have a portfolio of 500k and would like it to grow to 1M in 10 years. Will a 60-40 (equity-bond) AA get me there or do I need to take more risk and use say a 80-20 or 100-0 portfolio?

I'll come with an example for a drawdown scenario later.

Quote:

Strikes me as circular - I can't make an estimate of whether I can meet my goals unless I make an estimate of my expected returns? Either way, I'm making an estimate.

Yes and no.

There is an iterative/circular part where I check to see if my AA will meet my long terms goals -- if not then I might adjust my equity/bonds balance and recompute the expected return for my whole portfolio as per the example above.

However it's not circular in that the basis for the expectation is computed completely separately from my goals. For example, to get an expectation of the equity component I might use 1 / PE10 (Schiller PE). This part only flows forward and isn't recomputed.

[palomalou]

We have dialed stocks back a bit since semi-retirement 10 months ago. Now about 50/30/20.


Sent from my iPad using Early Retirement Forum

[Choices]

6 nominal 3 real which is what I think my friend Warren projects.

[ERD50]

Quote:

Originally Posted by ERD50 ... Can anyone explain - what question are you trying to answer with any assumption on future returns (real or otherwise)?

Quote:

Originally Posted by Totoro In my case: what kind of life (spending) can I afford now, and how will that impact my capacity to spend in the future? Or differently put: what will my lifetime earnings be from this point onwards?

And how do the responses here on what they are doing now provide any insight to that question?

Quote:

Originally Posted by photoguy Here's a simple one: I have a portfolio of 500k and would like it to grow to 1M in 10 years. Will a 60-40 (equity-bond) AA get me there or do I need to take more risk and use say a 80-20 or 100-0 portfolio? ....

Wouldn't a historical report be a better way to see what AA's have provided the returns you are looking for in that time frame?

-ERD50

[Looking4Ward]

I'll be 55 this year, and have 15 years worth of living expenses in a 65/35 taxable account.

I've got another 15 years worth of living expenses in a 75/25 IRA that I won't be touching until I start drawing SS at 70.

Anything above 0% would be fine with me.

[photoguy]

Quote:

Wouldn't a historical report be a better way to see what AA's have provided the returns you are looking for in that time frame?

I don't think they're mutually exclusive. In general, if there's multiple ways of doing an analysis I'm a fan of doing them all because each method has different pros & cons.

However in this case there could be multiple reasons why one would prefer to compute an expected return instead of using a historical report:

(1) Historical reports don't consider important factors like valuations for equities or current yields in bond returns.

(2) A huge source of noise in returns is due to fluctuations in what investors are willing to pay for a dollar of earnings. Going to an expectation model can allow one to filter that out whereas those fluctuations are going to be in the historical data.

(3) Historical data may not be available for your asset class. Sure S&P has been around for ~100 years but what about more specialized asset classes like international reits, emerging market small cap value?

(4) Even if historical data is available there might be changes in regulations, accounting practices, etc. that lead you to think it's less relevant than before.



Here's an example of how expectations might be useful in a drawdown scenario. It's nothing earth-shattering and probably just common sense, but here goes:

Suppose I want to figure out what withdrawal rate to use for my retirement given that I'm willing to accept a 5% failure rate over a 30 year period. How do I determine what percentage I should use?

The first thing I might do is go to FIRECALC and see how different WR fared historically. FIRECALC says that historically I can pull out 4% with only a 5% failure rate. But everybody knows that FIRECALC is about the past and doesn't say anything about the future. So what should I do?

I figure that I have three different options:

(A) I decide that 4% is pessimistic for the future and I should pull out more than that.
(B) I decide that 4% is about right for the future and I should pull out that amount.
(C) I decide that 4% is optimistic for the future and I should pull out less than that.

Expectations on returns can tell us which scenario (A,B, or C) is the most likely. For example say the equity data in FIRCALC has an average return of 7% real (I'm not sure of the exact number). I compare this to the estimate of returns based on current valuations and lets say that's 4% real. This clearly tells me that I'm in scenario C. It doesn't tell me how much of a haircut I should take but I know that if my expectation is 4% real I should take a bigger reduction than if my expectation is 6% real.

[ERD50]

Quote:

Originally Posted by photoguy Quote: Wouldn't a historical report be a better way to see what AA's have provided the returns you are looking for in that time frame? I don't think they're mutually exclusive. In general, if there's multiple ways of doing an analysis I'm a fan of doing them all because each method has different pros & cons. However in this case there could be multiple reasons why one would prefer to compute an expected return instead of using a historical report: ... Suppose I want to figure out what withdrawal rate to use for my retirement given that I'm willing to accept a 5% failure rate over a 30 year period. How do I determine what percentage I should use? The first thing I might do is go to FIRECALC and see how different WR fared historically. FIRECALC says that historically I can pull out 4% with only a 5% failure rate. But everybody knows that FIRECALC is about the past and doesn't say anything about the future. So what should I do? I figure that I have three different options: (A) I decide that 4% is pessimistic for the future and I should pull out more than that. (B) I decide that 4% is about right for the future and I should pull out that amount. (C) I decide that 4% is optimistic for the future and I should pull out less than that. Expectations on returns can tell us which scenario (A,B, or C) is the most likely. For example say the equity data in FIRCALC has an average return of 7% real (I'm not sure of the exact number). I compare this to the estimate of returns based on current valuations and lets say that's 4% real. This clearly tells me that I'm in scenario C. It doesn't tell me how much of a haircut I should take but I know that if my expectation is 4% real I should take a bigger reduction than if my expectation is 6% real.

I don't see it. Looking at failures, FIRECalc is pessimistic.

OK, I kinda see where you want to go, but not how you get there.

First, probably a minor mathematical shorthand on your part, but for clarity, a 4% static real return provides for far more than a 4% WR for 30 years. One can take 3.33% (1/30) with zero static real return.

So now you want to go more pessimistic that any past period. OK fine, so where do you get the data to tell you what to use? You say that 'Historical reports don't consider important factors like valuations for equities or current yields in bond returns.' - well sure they do, they do it on the pessimistic side. The failures are due to jumping in at high valuations. So where do you get your data to make an estimate?

FIRECalc shows a 3.59% WR at 100% success for 30 years, so that indicates the worst in history was just a bit above 0% real (again, ignoring volatility, which we can't) that would be safe for a 3.33% WR (I'll do the math later).

I'm not sure where to logically go from here - it all seems so circular. I'm an Occam's Razor kind of guy. I don't know what the future holds so I look at the past, and make a guess about how much buffer I need. You are also making a guess, but you put your guess into an estimate of real return, and then you seem to want to say those resulting calculations have more value than the initial guess - but they are just the result of a guess. Just go with the guess.

I'm not getting it.

But to answer a previous question you had, I base my AA on the range that has provided the best portfolio survival in the worst historical periods (that's an 'investigate' option in FIRECalc). I have no basis to say that a higher or lower AA will be appropriate in a hypothetical record-setting bad future. I also notice there isn't a lot of sensitivity either - roughly 40/60 to 95/5 provides the same safety factor historically. I don't worry much about it. I think middle ground looks good, and see no basis for any other thinking.


-ERD50

[daylatedollarshort]

We use 0 - 1% real return in a spreadsheet and also look at the Fidelity RIP using very conservative AAs.

As ERD50 posted, a zero real return will let us withdraw up to 2.5% per year for 40 years, though we plan to spend less most years.

[photoguy]

Quote:

First, probably a minor mathematical shorthand on your part, but for clarity, a 4% static real return provides for far more than a 4% WR for 30 years. One can take 3.33% (1/30) with zero static real return.

The 4% is just a made up number for the example. The relevant aspect is that it is lower than the historical data in firecalc. The exact value isn't important.

Quote:

So now you want to go more pessimistic that any past period. OK fine, so where do you get the data to tell you what to use? You say that 'Historical reports don't consider important factors like valuations for equities or current yields in bond returns.' - well sure they do, they do it on the pessimistic side. The failures are due to jumping in at high valuations. So where do you get your data to make an estimate?

When I say pessimism (or optimism) I'm referring to the accuracy of the probability estimate, not whether there were good or bad sequences in the data.

More specifically, I'm saying that the FIRECALC probability number when used as an estimate of future success can either be too high (optimistic), about right, or too low (pessimistic). Obviously it has to be in one of these buckets as this covers all the possible outcomes.

Expected returns tell me which bucket we are likely to fall into in the future.

Quote:

I'm not sure where to logically go from here - it all seems so circular. I'm an Occam's Razor kind of guy. I don't know what the future holds so I look at the past, and make a guess about how much buffer I need. You are also making a guess, but you put your guess into an estimate of real return, and then you seem to want to say those resulting calculations have more value than the initial guess - but they are just the result of a guess. Just go with the guess.

The relationship between equity valuations and expected future returns is well known (See Shiller P/E or CAPE). We also know that the higher future returns, the greater withdrawal rate a portfolio will support. The relationship seems pretty clear and if I'm going to make a guess about my probability of future success, it's going to be consistent with that information.

I'm not really sure how to explain this to you differently.

[pb4uski]

Quote:

Originally Posted by ERD50 If don't think that using static returns is very useful or insightful, other than for a reference point.....

I suspect you will not agree with me in this but I'll throw it out there and we can agree to disagree if necessary.

I think deterministic analysis is quite useful for general retirement planning since you are dealing with quite long time horizons. Then it is prudent to supplement any deterministic plan with stochastic analysis to deal with sequence of returns risk. Or put another way, deterministic is easier for people to comprehend and until you have a reasonable deterministic plan that is successful, any stochastic testing is simply a waste of time.

[jkern]

I use 4% nominal with a 40/40/20 AA. The only reason I even use this is to give me a guesstimate on what the picture will look like in 5 years when I begin taking Social Security.

[ERD50]

Quote:

Originally Posted by photoguy Quote: First, probably a minor mathematical shorthand on your part, but for clarity, a 4% static real return provides for far more than a 4% WR for 30 years. One can take 3.33% (1/30) with zero static real return. The 4% is just a made up number for the example. The relevant aspect is that it is lower than the historical data in firecalc. The exact value isn't important.

I think it is important, and maybe why we are having this communication gap. The number is so far from historical data in firecalc, that I it may be distorting the view.


I went back and did the math (well, let a spreadsheet do the math). The 100% historically safe inflation adjusted WR for 30 years is the 3.59% I mentioned earlier. A static real return that that will provide that is a mere 0.52%. And to re-re-hash for reference, a 0.0% real return provides a 3.33% WR (1/30).

The difference between 4% real returns you mention, and 0.52% real returns is too huge to just hand wave. It makes it hard to talk about when we aren't in the same universe.

Quote:

More specifically, I'm saying that the FIRECALC probability number when used as an estimate of future success can either be too high (optimistic), about right, or too low (pessimistic). Obviously it has to be in one of these buckets as this covers all the possible outcomes.

Now there's some logic that I just have to agree with!

Quote:

Expected returns tell me which bucket we are likely to fall into in the future.

Well, ummm, OK. I can go along with the idea that if we are at historically high valuations (PE10 or some other measure), we are more likely to be on a path that follows the squiggly lines near the bottom of the distribution on the FIRECalc output graph.

But if the a FIRECalc 100% SWR ends up being too high, it just means the actual future ended up being worse than the worst in the data set. I've said that all along.


Are we more likely to see relatively high stresses on a portfolio when we are at a peak (which we don't know until it passes), or even when we seem to be nearing a peak? Sure, if we continue to see economic cycles similar to the past.

But once again, only if they are worse than the worst in the data set. Past high valuations are already in there. And now we are back to square one - how do you estimate how much worse we might get?

And I don't see where creating some fictitiously low real return number is any different than any other guestimate. But you seem to be implying that some how it is more logical, rigorous, methodological, or something?

-ERD50

[Gatordoc50]

Quote:

Originally Posted by palomalou I have read a few people on this board who say they use 0-2%. Is that what most of you do, and if not, what do you estimate and why?

My plan is that our portfolio keeps pace with inflation. If I believed that the only risk to my plan involved shallow risks as Bernstein described them, I would probably be invested totally in the market at about 75/25. I do believe, however, there are threats that could cause deep risks ( Bernstein) and a permanent loss of capital. Tax increases or means tested entitlements would be examples. Those don't necessarily hamper market returns but may have a profound effect on ones personal rate or return. So, I try to hold assets like real estate which are harder ( maybe/ hopefully) to means tested but still provide a real return. Only time will tell.

[Live And Learn]

Quote:

Originally Posted by ERD50 If don't think that using static returns is very useful or insightful, other than for a reference point. ...<snip>... Strikes me as circular - I can't make an estimate of whether I can meet my goals unless I make an estimate of my expected returns? Either way, I'm making an estimate.

I don't use a static return per se, but being a spreadsheet junkie I do have one for that. At a 1.5% real return I am good to go. Since the real world doesn't give static returns I only use this spreadsheet to see if the "needed" annual real return seems reasonable for my 50/40/10 AA.

[photoguy]

Quote:

Originally Posted by ERD50 I think it is important, and maybe why we are having this communication gap. The number is so far from historical data in firecalc, that I it may be distorting the view.

I was trying to give a qualitative example to keep things simple. It's possible to use expectations in a quantitative manner (see my comments at the end).

Quote:

Well, ummm, OK. I can go along with the idea that if we are at historically high valuations (PE10 or some other measure), we are more likely to be on a path that follows the squiggly lines near the bottom of the distribution on the FIRECalc output graph.

I think we have agreement here. Your statement is

high valuations --> lower lines more likely on firecalc

I'm just saying it's because of expectations:

high valuations --> lower expected returns --> lower lines more likely on firecalc

To take this one step further, "lower lines more likely" means that the success probability that comes out of firecalc is too high (when used as a future estimate). Firecalc computes an equal weighted probability but if we agree that some lines are more likely, then they should have greater weight in the calculation

If we have more weight on the lower lines, then we are in the following situation:

(A) if we keep the withdrawal rate constant, the probability of success decreases

(B) if we keep the probability of succcess constant, we have to reduce the withdrawal rate

Quote:

But if the a FIRECalc 100% SWR ends up being too high, it just means the actual future ended up being worse than the worst in the data set. I've said that all along.

I never said that a withdrawal rate of 3.59% (for 100% success in FIRECALC) is too high or too low. I'm saying that probability of future success for a 3.59% w.r. is not 100%. And it's going to be lower when current valuations are higher (expectations are lower).

I'm not sure why it's necessary to draw a distinction with the future being worse than past. The future can easily be worse than the past. Consider that we have only about 4 independent (non-overlapping) 30 year periods in Firecalc. What's the chance that the next (5th) period is going to be worse?

Quote:

Are we more likely to see relatively high stresses on a portfolio when we are at a peak (which we don't know until it passes), or even when we seem to be nearing a peak? Sure, if we continue to see economic cycles similar to the past.

Quote:

But once again, only if they are worse than the worst in the data set. Past high valuations are already in there.

The issue is not whether high valuations are in the data used by FIRECALC but whether they receive the right weight (for forward estimates of success). If you agree that "some lines are more likely" then by definition the equal weight calculation FIRECALC uses is not appropriate.

Quote:

And now we are back to square one - how do you estimate how much worse we might get?

In this thread, I've been trying to stick to a qualitative example to keep things simple. But if you want a quantitative number, you can get this through simulation or monte carlo methods.

(I know you don't like monte carlo and I definitely don't want to rehash their pros and cons here)

Edit: I guess an alternative to using MC methods would be to re-weight the lines in Firecalc according how likely you think each path is based on current valuations.

[ERD50]

Quote:

Originally Posted by Live And Learn I don't use a static return per se, but being a spreadsheet junkie I do have one for that. At a 1.5% real return I am good to go. Since the real world doesn't give static returns I only use this spreadsheet to see if the "needed" annual real return seems reasonable for my 50/40/10 AA.

I also have some spreadsheets with static returns. As I said earlier, I think it makes sense to use them for a reference point, keeping in mind their limitations.

I actually use ZERO real returns in those spreadsheets, for two reasons. First, it keeps things simple, I'm far less likely to make an error, as I don't need formulas for inflation adjustments. The only adjustment is that I deflate my non-cola pension, but that's one line and fairly small impact anyhow. The second is that zero real return is a bit worse than the historical worst case that I showed in the previous post, so I get a simple, conservative SS.

I agree with you, I think it is useful as just another data point, to see if things seem within reason when looked at from another angle. I've always found it useful to view things from different angles, as a check (it applies to pretty girls as well! It's a universal constant! ).




photoguy, I'm not going to quote it all, but I think there is just a semantic difference here, a gap that keeps this conversation going back and forth. I think I said this before, but I can't find it off-hand so I'll repeat it here:

A) FIRECalc provides a report. Barring calculation or data errors, and within rounding, it is 100% accurate. There were references to this being 'scientific' - OK, but I don't care to debate the meaning(s) of that word, I don't think it is important. We could say it is deterministic. Different people with the same inputs get the same results.

B) What you do with those results is up to you. Looking to the future, and devising a plan is not deterministic. Different people with the same inputs will offer different approaches.

And I think you continue to blend the two, and that's where our conversations get twisted and difficult.


I'll try a few excerpts:

Quote:

Originally Posted by photoguy ... To take this one step further, "lower lines more likely" means that the success probability that comes out of firecalc is too high (when used as a future estimate). ...

Here's where we breakdown (cue Led Zeppelin...) Using 100% HSWR (Historically Safe Withdraw Rates), none of those lines fail. The probability of success is 100%. Period.

(when used as a future estimate) - now you are into 'B'. It's a different conversation. It isn't FIRECalc that reported anything that is 'too high', it just reported the facts. They are what they are.

Quote:

Firecalc computes an equal weighted probability but if we agree that some lines are more likely, then they should have greater weight in the calculation If we have more weight on the lower lines, then we are in the following situation: (A) if we keep the withdrawal rate constant, the probability of success decreases (B) if we keep the probability of success constant, we have to reduce the withdrawal rate

No, the lower lines are the worst case, and with a HSWR, they don't fail. You don't have to make any changes, they don't fail. Period.


bold mine:

Quote:

I'm saying that probability of future success for a 3.59% w.r. is not 100%.

And I've never disagreed with that. I have no Crystal Ball. But don't confuse/mix that and try to rewrite history. It is 100% for the past. Period. It confuses things when we don't accept that.

Quote:

And it's going to be lower when current valuations are higher (expectations are lower).

Sure, FIRECalc (and common sense) tells us that it's better to have $1M at the trough of an economic cycle than to have $1M at the peak.

Quote:

I'm not sure why it's necessary to draw a distinction with the future being worse than past.

Because we run into these endless semantic differences if we don't!

Quote:

The future can easily be worse than the past. Consider that we have only about 4 independent (non-overlapping) 30 year periods in Firecalc. What's the chance that the next (5th) period is going to be worse?

I don't know, do you? It is certainly a possibility.

My point is, take history for what it is, and then apply your fudge factor, don't mix the two.

Take the case of ten engineers reporting to their boss on the probability of failure of their subsystem. The boss wants to calculate the probability of failure of the overall system. If each of the engineers includes their own fudge factor, the boss is lost when trying to apply some fudge. Better to take the numbers straight, and apply one overall fudge factor. It's cleaner and more transparent.

That's all I'm saying.

One more analogy - what are the odds of rolling a "1" with a single, fair die? Clearly 1 out of 6. No one can correctly dispute that. Do you want to take the bet with a fair payout (6:1)? Do you want to take the bet with $100,000 at stake with a generous payout (say 12:1)? Different people will give different answers, but the 1 out of 6 odds remain the same. They are fact. They don't change based on the decision of taking the bet or not. It's the A/B I posted above. Get it?


-ERD50

[ERD50]

photoguy - after my last post, let me try it from a positive angle, where we kind of come together....

I think we would agree that if we took the 10 starting years with the highest valuations, and ran FIRECalc for just those years, the average ending portfolios would very likely be lower than the averages for all years in the data set. Agreed?

But with a HSWR, they still all succeed. The HSWR tells you were the 'edge' of passing/failing has been. Now, go to "B", and decide what to do with that information.


-ERD50