Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Wed Aug 29, 2018 2:34 pm

One thing that always bothered me about a lot of SWR (safe withdrawal rate) stats were that they were based on historical returns. I decided to look into Monte Carlo simulations instead.

I think most of us would agree that we shouldn't expect bond returns to be as good in the next 10+ years as they were over the past few decades. In fact, I strongly suspect that if you buy a 10 year Treasury today, you'll make about 2.9% annualized between now and maturity. :) I think you'll be very lucky to make 1% over inflation. 0% is probably more like it.

And I don't think most of us expect equities to have a *real* total return above 6% in the next few decades. Bogle himself puts the number much lower.

So let's say that you have a 50/50 portfolio. Your real expected return is very optimistically about 3% with a volatility of around 7.5%. (Yes, I understand that some people argue that stock returns have been somewhat negatively autocorrelated, but I wouldn't expect that to continue. I think some of that is just survivorship bias, i.e. countries for which things didn't turn around after a bad run are no longer with us, but in any case, I think it'd be naive to count on it.)

So I went to PortfolioVisualizer.com and punched in (to keep it simple) 0% inflation, 3% returns (normally distributed), and 7.5% volatility.

A 4% withdrawal rate for 30 years only works about 82% of the time. I need to drop down to 3.4% withdrawal rate to get the success rate up to 95%.

Does this sound right to people? This is 15% less money per year than the usual rule and I'm arguably using unrealistically high expected return numbers! I'm starting to think that historical return data was making us too optimistic.

If I drop the real return of stocks down to 5%, then my portfolio return would be 2.5% real, and my SWR works out to 3.15%. This is getting depressing.

Now what if I want to retire early and use a 50 year horizon?

Things get a lot worse. That "safe" 3.4% withdrawal rate only has a 61% success rate. I would need to drop my rate down to 2.35% to get a 95% success rate.

What if I switch to a VPW method? My concern there is that these are also (AFAIK) based on historical returns. So I'm not sure if I can trust those tables. Intuitively, if I'm making 2.5% real returns, then I really can't withdraw much more than 2.5% each year in the beginning if I have a long (50 yr) retirement. So if we adjusted the VPW methodology to be based on Monte Carlo sims instead of historical returns, I don't think I'd be able to get away with withdrawals much higher than 2.5%.

Any thoughts on this?

[Please don't reply and say "nobody knows nuthin'". I understand that nobody knows what future returns will be, but we can make a very educated guess about bond returns and a decent guess about stock returns. And when people use historical sims, they're implicitly making a guess about what future returns will be anyway.]

HEDGEFUNDIE
Posts: 1257
Joined: Sun Oct 22, 2017 2:06 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by HEDGEFUNDIE » Wed Aug 29, 2018 4:02 pm

I was playing around with this tool last night, and I actually have the opposite problem. No matter what I do it tells me I have a 99%+ chance of success, median outcome of ending with $11M on my deathbed :confused

My parameters are below, maybe you can share yours?

Starting line: $600k
Annual contribution: $7k/month for next 12 years, then early retirement
AA: 100% equities gliding down to 60/40 in 18 years
Withdrawals: 4% of assets per year starting in year 12, plus $70k/year in years 18-29 for kids’ college

https://www.portfoliovisualizer.com/fin ... ftimes3=12

User avatar
patrick013
Posts: 2433
Joined: Mon Jul 13, 2015 7:49 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by patrick013 » Wed Aug 29, 2018 5:46 pm

Well you can't spend real dollars yet. I think it's better to
adjust future budgets for inflation but let things like average
returns and Monte's still use nominal dollars for their calculations.

If you can't get a TRSY10 that pays 4% you should be able to get
a Farm Credit GSE for 4% for 10 years at par. Index stocks should
return 8% nominal considering index replacement and tilting to
mid and small cap general index funds.

So.....with a 50-50 portfolio you should have yearly returns of 6%
overall foreseeable with lower SD of 12-14% approximately. Put
those numbers into Monte and see what it says. See what it says
at 4% withdrawal and 5% withdrawal as the latter would be a number
representing rather larger inflationary withdrawals, if needed.
age in bonds, buy-and-hold, 10 year business cycle

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Wed Aug 29, 2018 8:21 pm

patrick013 wrote:
Wed Aug 29, 2018 5:46 pm
Well you can't spend real dollars yet. I think it's better to
adjust future budgets for inflation but let things like average
returns and Monte's still use nominal dollars for their calculations.
I just assumed 3% inflation and subtracted that from my nominal returns. Put another way, I just assumed certain real returns, put those numbers in for nominal returns, and put in zero inflation. Doing it the "nominal way" makes almost no difference.
patrick013 wrote:
Wed Aug 29, 2018 5:46 pm
So.....with a 50-50 portfolio you should have yearly returns of 6%
overall foreseeable with lower SD of 12-14% approximately.
Even if that's true, you're still back to 3% real (if inflation runs at 3%). So I'm not sure things are any better under this scenario, but tell me if I'm missing something. :D

AlohaJoe
Posts: 3945
Joined: Mon Nov 26, 2007 2:00 pm
Location: Saigon, Vietnam

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by AlohaJoe » Wed Aug 29, 2018 8:43 pm

R2D2 wrote:
Wed Aug 29, 2018 2:34 pm
I think most of us would agree that we shouldn't expect bond returns to be as good in the next 10+ years as they were over the past few decades. [...] Now what if I want to retire early and use a 50 year horizon?
There's a pretty massive leap from "I'm expecting 3% real returns for the next decade" to "I'm expecting 3% real returns for most of the next century". You're saying "I expect the US to have equity returns like Japan has had. But not just for 30 years like Japan but for an entire half century". People aren't even that pessimistic about Japan!

Not only is that dramatically worse returns than anyone, even GMO is predicting, why are you surprised that such terrible returns -- worse than any country has ever seen outside of World Wars -- leads to extraordinarily low withdrawal rates? Think about it this way: even when France was devastated by two World Wars and had millions of dead people they still had better returns than 3% real over 50 years.

So my main thought is that simple Monte Carlo is the wrong thing to be using here. What all the experts are saying is something like "returns will be low for 10 years but then they will go back to normal because valuations would have come down dramatically". So you want a simulation that does that. Low returns for 10-15 years with things slowly reverting back to normal.

The Monte Carlo in PortfolioVisualizer isn't built for that kind of thing, instead you need to use some kind of auto-regressive model that will "return to normal".

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Wed Aug 29, 2018 9:20 pm

AlohaJoe wrote:
Wed Aug 29, 2018 8:43 pm
There's a pretty massive leap from "I'm expecting 3% real returns for the next decade" to "I'm expecting 3% real returns for most of the next century". You're saying "I expect the US to have equity returns like Japan has had. But not just for 30 years like Japan but for an entire half century". People aren't even that pessimistic about Japan!
Right, but I was assuming 6% real for stocks and 0% for bonds (giving me 3% real for a 50/50 portfolio). Do you think 6% real for stocks is Japan-like? I may have misunderstood.

I can see the argument that equity returns will pick up at some point, but I could easily imagine that low interest rates on bonds is the new normal.

Thanks for replying, AlohaJoe! Let me know if I'm understanding you correctly.

User avatar
willthrill81
Posts: 6603
Joined: Thu Jan 26, 2017 3:17 pm
Location: USA

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by willthrill81 » Wed Aug 29, 2018 9:36 pm

AlohaJoe wrote:
Wed Aug 29, 2018 8:43 pm
R2D2 wrote:
Wed Aug 29, 2018 2:34 pm
I think most of us would agree that we shouldn't expect bond returns to be as good in the next 10+ years as they were over the past few decades. [...] Now what if I want to retire early and use a 50 year horizon?
There's a pretty massive leap from "I'm expecting 3% real returns for the next decade" to "I'm expecting 3% real returns for most of the next century". You're saying "I expect the US to have equity returns like Japan has had. But not just for 30 years like Japan but for an entire half century". People aren't even that pessimistic about Japan!

Not only is that dramatically worse returns than anyone, even GMO is predicting, why are you surprised that such terrible returns -- worse than any country has ever seen outside of World Wars -- leads to extraordinarily low withdrawal rates? Think about it this way: even when France was devastated by two World Wars and had millions of dead people they still had better returns than 3% real over 50 years.

So my main thought is that simple Monte Carlo is the wrong thing to be using here. What all the experts are saying is something like "returns will be low for 10 years but then they will go back to normal because valuations would have come down dramatically". So you want a simulation that does that. Low returns for 10-15 years with things slowly reverting back to normal.

The Monte Carlo in PortfolioVisualizer isn't built for that kind of thing, instead you need to use some kind of auto-regressive model that will "return to normal".
:thumbsup

Many financial advisors (who often work on an AUM basis and have a vested interest in clients not spending their portfolios) and supposed experts base an inordinate of their recommendations off of simple Monte Carlo analyses that do not explicitly account for mean reversion. Even if one isn't a 'die-hard mean reversion' proponent, most would argue that if stocks dropped by 50%, for instance, it would then be less likely that they drop another 50% than the likelihood of them dropping by 50% in the first place. But many of the 'simple' Monte Carlo analysis tools out there do not take mean reversion into account. Consequently, when you look at some of the scenarios they spit out, they look more like a Mad Max world (e.g. a 50% drop followed by another 50% drop and then another) than anything that could be reasonably expected to actually occur.

Regardless, no one knows what the future will hold. You might live to be 110 and see global stocks return 5% real for the next half century. But you're much more likely to get a terminal diagnosis at 70 and not need to worry about that 30 year retirement period that so many SWR studies are based on. Be flexible, but don't forget about your own mortality either.
“It's a dangerous business, Frodo, going out your door. You step onto the road, and if you don't keep your feet, there's no knowing where you might be swept off to.” J.R.R. Tolkien,The Lord of the Rings

AlohaJoe
Posts: 3945
Joined: Mon Nov 26, 2007 2:00 pm
Location: Saigon, Vietnam

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by AlohaJoe » Wed Aug 29, 2018 10:16 pm

R2D2 wrote:
Wed Aug 29, 2018 9:20 pm
I can see the argument that equity returns will pick up at some point, but I could easily imagine that low interest rates on bonds is the new normal.
All I'm saying is that 50 years is a ridiculously long time to make guesses about performance. Fifty years ago there were no index funds, virtually no mutual funds, and no discount stock brokers -- 50 years ago it cost 0.975% commission to trade (so buying $10,000 of an ETF would have cost you almost $100). People always look at the present and then think everything will stay that way forever. And it has never worked out that way. Financial history is full of "regime shifts" which most research & modeling does a poor job of diving into (or even mentioning). But when you're talking about a half century of returns, those regime shifts are important.

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Thu Aug 30, 2018 6:57 am

AlohaJoe wrote:
Wed Aug 29, 2018 10:16 pm
R2D2 wrote:
Wed Aug 29, 2018 9:20 pm
I can see the argument that equity returns will pick up at some point, but I could easily imagine that low interest rates on bonds is the new normal.
All I'm saying is that 50 years is a ridiculously long time to make guesses about performance. Fifty years ago there were no index funds, virtually no mutual funds, and no discount stock brokers -- 50 years ago it cost 0.975% commission to trade (so buying $10,000 of an ETF would have cost you almost $100). People always look at the present and then think everything will stay that way forever. And it has never worked out that way. Financial history is full of "regime shifts" which most research & modeling does a poor job of diving into (or even mentioning). But when you're talking about a half century of returns, those regime shifts are important.
All good points. Remember though that the regular SWR numbers are based on 100+ years of historical data, so I'm not the only one making this error. :)

But let's drop the 50 year issue and just look at 30 year retirement planning. Do you think 6% real for stocks and 0% for bonds is being too pessimistic?

GrowthSeeker
Posts: 206
Joined: Tue May 15, 2018 10:14 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by GrowthSeeker » Thu Aug 30, 2018 7:52 am

R2D2: You might find the analysis done at the Early Retirement Now website to be of interest. He looks at longer time frames, like 60 years, using monthly historical data going back to 1871. So it's not Monte Carlo like the OP's analysis, but the overall conclusion for SWR (depending on the assumptions) is more like 3.5% than the usual 4% rule. Different methodology, similar conclusion.

https://earlyretirementnow.com/2016/12/ ... t-1-intro/

He also looks at SWRs where the endpoint is not running out of money at 30 (or 60) years, but also maintaining the same real net worth at the end of that time.
Note that from a back of the envelope math point of view, the longer the retirement you plan for, the closer your SWR for "spending down to zero" number gets to the SWR number for "maintaining your net worth". So if you assume a 3% return and you want it to last forever, then your SWR is going to be about 3%.
Conclusion: the final answer is highly dependent on the underlying assumptions.
(which of course is obvious but I just thought it was worth saying)
Just because you're paranoid doesn't mean they're NOT out to get you.

randomguy
Posts: 6678
Joined: Wed Sep 17, 2014 9:00 am

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by randomguy » Thu Aug 30, 2018 9:34 am

R2D2 wrote:
Wed Aug 29, 2018 2:34 pm

I think most of us would agree that we shouldn't expect bond returns to be as good in the next 10+ years as they were over the past few decades. In fact, I strongly suspect that if you buy a 10 year Treasury today, you'll make about 2.9% annualized between now and maturity. :) I think you'll be very lucky to make 1% over inflation. 0% is probably more like it.

Sure but that has been true for pretty much the whole history outside of the 80s/90s/00s. You weren't getting big real returns from those 3% bonds you bought in the 40s,50s, and 60s. Our current rates are pretty much inline with historical norms outside of the 70s-early 00s. It isn't like we are at some levels that have never been seen before these days as we are well off the lows. Granted we are looking at small sample size issues as we have only had a couple cycles.

This is partyly why 50/50 is too conservative when you start upping the time frame past 30 years. You get better results with 70/30 or 80/20 when talking 50 year time periods.

User avatar
patrick013
Posts: 2433
Joined: Mon Jul 13, 2015 7:49 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by patrick013 » Thu Aug 30, 2018 4:02 pm

R2D2 wrote:
Wed Aug 29, 2018 8:21 pm
patrick013 wrote:
Wed Aug 29, 2018 5:46 pm
So.....with a 50-50 portfolio you should have yearly returns of 6%
overall foreseeable with lower SD of 12-14% approximately.
Even if that's true, you're still back to 3% real (if inflation runs at 3%). So I'm not sure things are any better under this scenario, but tell me if I'm missing something. :D
Well account balances are kept and compounded at nominal value.
It's the withdrawals that will have to increase when inflation and
expenses increase. The only way to beat inflation is to have a larger
principal or invest in sensible stocks to try to get a larger principal
value, just to live off earnings from it. Your AA and the estimates
used are somewhat subjective but one has to estimate something.

AlohaJoe wrote:
Wed Aug 29, 2018 8:43 pm
The Monte Carlo in PortfolioVisualizer isn't built for that kind of thing, instead you need to use some kind of auto-regressive model that will "return to normal".
Perhaps 2 separate Monte's. One for the first 15 years at a lower return
level and withdrawal rate (4%) if any. Then take those nominal account
balances and start a 2nd Monte for the second 15 years. It would have a
more mean return and also higher withdrawals (4 or 5%) to account for
inflation of expenses. The end probability of success should be a better
number then. ???
age in bonds, buy-and-hold, 10 year business cycle

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Thu Aug 30, 2018 6:01 pm

randomguy wrote:
Thu Aug 30, 2018 9:34 am
R2D2 wrote:
Wed Aug 29, 2018 2:34 pm

I think most of us would agree that we shouldn't expect bond returns to be as good in the next 10+ years as they were over the past few decades. In fact, I strongly suspect that if you buy a 10 year Treasury today, you'll make about 2.9% annualized between now and maturity. :) I think you'll be very lucky to make 1% over inflation. 0% is probably more like it.

Sure but that has been true for pretty much the whole history outside of the 80s/90s/00s. You weren't getting big real returns from those 3% bonds you bought in the 40s,50s, and 60s. Our current rates are pretty much inline with historical norms outside of the 70s-early 00s. It isn't like we are at some levels that have never been seen before these days as we are well off the lows. Granted we are looking at small sample size issues as we have only had a couple cycles.
Great point. Thanks for the reply. I guess Treasuries historically haven't been too exciting. :)

I suppose the reason these Monte Carlo results look bad (compared to historical SWR) isn't the low bond yields we have now (as you pointed out to me) but rather just that equity returns were high and negatively autocorrelated in the past. That latter point helps a lot!

If we take 0 autocorrelation and "realistic" real equity returns, things get depressing.

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Thu Aug 30, 2018 6:04 pm

GrowthSeeker wrote:
Thu Aug 30, 2018 7:52 am
R2D2: You might find the analysis done at the Early Retirement Now website to be of interest. He looks at longer time frames, like 60 years, using monthly historical data going back to 1871. So it's not Monte Carlo like the OP's analysis, but the overall conclusion for SWR (depending on the assumptions) is more like 3.5% than the usual 4% rule. Different methodology, similar conclusion.
Thanks for mentioning this. Now that I think about it, there really are three methodologies one could use:
1. Look at past history and keep everything "in order",
2. Use bootstrapping from historical returns (in other words, do Monte Carlo on historical returns and allow things to get out of order), and
3. Use Monte Carlo with forward-looking expected returns.

This thread is making me think that I have to go out and make more money. :)

User avatar
willthrill81
Posts: 6603
Joined: Thu Jan 26, 2017 3:17 pm
Location: USA

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by willthrill81 » Thu Aug 30, 2018 6:48 pm

R2D2 wrote:
Thu Aug 30, 2018 6:04 pm
GrowthSeeker wrote:
Thu Aug 30, 2018 7:52 am
R2D2: You might find the analysis done at the Early Retirement Now website to be of interest. He looks at longer time frames, like 60 years, using monthly historical data going back to 1871. So it's not Monte Carlo like the OP's analysis, but the overall conclusion for SWR (depending on the assumptions) is more like 3.5% than the usual 4% rule. Different methodology, similar conclusion.
Thanks for mentioning this. Now that I think about it, there really are three methodologies one could use:
1. Look at past history and keep everything "in order",
2. Use bootstrapping from historical returns (in other words, do Monte Carlo on historical returns and allow things to get out of order), and
3. Use Monte Carlo with forward-looking expected returns.

This thread is making me think that I have to go out and make more money. :)
Remember that it's a balancing act between the risk of running out of money, which is far overblown IMHO, and the risk of working longer than necessary and not enjoying the fruits of your labor, saving, and investing.

"Surety you crave! Sauron gives none."
-Tolkien
“It's a dangerous business, Frodo, going out your door. You step onto the road, and if you don't keep your feet, there's no knowing where you might be swept off to.” J.R.R. Tolkien,The Lord of the Rings

marcopolo
Posts: 1299
Joined: Sat Dec 03, 2016 10:22 am

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by marcopolo » Thu Aug 30, 2018 9:48 pm

I think what you have re-discovered is that Monte Carlo simulations produce fatter tails in the distribution of outcomes than have ever occurred in real life.

Of course, Michael Kitces has an article for that:

https://www.kitces.com/blog/monte-carlo ... l-returns/
Once in a while you get shown the light, in the strangest of places if you look at it right.

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Fri Aug 31, 2018 8:27 am

willthrill81 wrote:
Thu Aug 30, 2018 6:48 pm
Remember that it's a balancing act between the risk of running out of money, which is far overblown IMHO,
Can you elaborate on that?

randomguy
Posts: 6678
Joined: Wed Sep 17, 2014 9:00 am

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by randomguy » Fri Aug 31, 2018 8:40 am

R2D2 wrote:
Fri Aug 31, 2018 8:27 am
willthrill81 wrote:
Thu Aug 30, 2018 6:48 pm
Remember that it's a balancing act between the risk of running out of money, which is far overblown IMHO,
Can you elaborate on that?
To run out of money you need both
a) bad returns
b) to be alive

The number of 60-65 year olds that make to 90 is pretty low. The number of people that retire in the worst possible year is pretty low (we are talking like 4 years in the past 100 where the 4% rule struggled).

And in reality people don't constantly spend 4%. There is a pretty natural reduction in spending that has nothing to do with lack of funds. And I am betting most people who retire in say 2008 voluntary defer a bit of spending.

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Fri Aug 31, 2018 9:07 am

marcopolo wrote:
Thu Aug 30, 2018 9:48 pm
I think what you have re-discovered is that Monte Carlo simulations produce fatter tails in the distribution of outcomes than have ever occurred in real life.

Of course, Michael Kitces has an article for that:

https://www.kitces.com/blog/monte-carlo ... l-returns/
Thanks for pointing me to this article. It was a good read.

Stuff like this makes me nervous though:
In this case, the data from 1871 to 2015 show that the annually rebalanced 60/40 portfolio had an average annual real return of 5.9%, with a standard deviation of 11.2%.
That's even more optimistic than my assumption of 6% real returns for stocks & 0% for bonds! Does anyone seriously believe that these are the returns we should expect over the next few decades? Or that we should use them for retirement planning?

marcopolo
Posts: 1299
Joined: Sat Dec 03, 2016 10:22 am

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by marcopolo » Fri Aug 31, 2018 9:21 am

R2D2 wrote:
Fri Aug 31, 2018 9:07 am
marcopolo wrote:
Thu Aug 30, 2018 9:48 pm
I think what you have re-discovered is that Monte Carlo simulations produce fatter tails in the distribution of outcomes than have ever occurred in real life.

Of course, Michael Kitces has an article for that:

https://www.kitces.com/blog/monte-carlo ... l-returns/
Thanks for pointing me to this article. It was a good read.

Stuff like this makes me nervous though:
In this case, the data from 1871 to 2015 show that the annually rebalanced 60/40 portfolio had an average annual real return of 5.9%, with a standard deviation of 11.2%.
That's even more optimistic than my assumption of 6% real returns for stocks & 0% for bonds! Does anyone seriously believe that these are the returns we should expect over the next few decades? Or that we should use them for retirement planning?
Returns to expect? I don't think anyone knows.
Use for retirement planning? I don't think anyone here credibly recommends planning based on AVERAGE historical returns.

Remember that the recommended 4% withdrawal (or maybe 3.0%-3.5% for early retirees) is not based on average returns. It is based on the worst case historical periods. In the average case a withdrawal rate closer to 5%-6% would have been fine.
So, the question you have to ask is do you think the next 30 or 40 years is likely to be worse than the worst case we have seen in the last 100 years.

This is all based of US history, so it is not unreasonable to reduce that a bit considering world wide returns.

It is certainly possible that things will be significantly worse in the future, but HOW BAD do you plan for?

My take is that if you get down to something like a 3% withdrawal rate with a reasonably well balanced portfolio, the odd are very good that you will be fine. As Willthrill pointed out you have to balance the risk of running out of money with the risk of running out of time.

YMMV :sharebeer
Once in a while you get shown the light, in the strangest of places if you look at it right.

User avatar
Hyperborea
Posts: 749
Joined: Sat Apr 15, 2017 10:31 am
Location: Osaka, Japan

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by Hyperborea » Fri Aug 31, 2018 9:55 am

R2D2 wrote:
Fri Aug 31, 2018 9:07 am
marcopolo wrote:
Thu Aug 30, 2018 9:48 pm
I think what you have re-discovered is that Monte Carlo simulations produce fatter tails in the distribution of outcomes than have ever occurred in real life.

Of course, Michael Kitces has an article for that:

https://www.kitces.com/blog/monte-carlo ... l-returns/
Thanks for pointing me to this article. It was a good read.
There are too many issues with Monte Carlo when you try to model anything as complex as investment returns. To get reasonable results you need to have a detailed enough model to simulate. If you want to model gas pressure in a box with random bouncing molecules then Monte Carlo works fine. For an economy and the resulting investing returns, the models used by (almost?) every Monte Carlo sim are too simple. They assume a simple Gaussian distribution for each your 3 variables (stock returns, bond returns, inflation) that are fully independent (no interaction between the three) and that are time independent (a 50% fall is just as likely after a 50% runup as it is after a 50% fall in the previous year). Nobody has enough knowledge to create a good simulation.

The two biggest uses for Monte Carlo in the financial world are to make your advisor seem like he has some deep insight and to show you that you need to invest more with them.

R2D2 wrote:
Fri Aug 31, 2018 9:07 am
Stuff like this makes me nervous though:
In this case, the data from 1871 to 2015 show that the annually rebalanced 60/40 portfolio had an average annual real return of 5.9%, with a standard deviation of 11.2%.
That's even more optimistic than my assumption of 6% real returns for stocks & 0% for bonds! Does anyone seriously believe that these are the returns we should expect over the next few decades? Or that we should use them for retirement planning?
I would suggest thinking about the numbers that applied to developed markets worldwide. The time period from 1871 to 2015 for the US was a fortuitous "accident" where the country went from emerging market to the latest world power. That won't be repeated going forward. Look at the numbers for Britain over that time period which went from being the then current world power to being just another developed country.
"Plans are worthless, but planning is everything." - Dwight D. Eisenhower

User avatar
willthrill81
Posts: 6603
Joined: Thu Jan 26, 2017 3:17 pm
Location: USA

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by willthrill81 » Fri Aug 31, 2018 9:58 am

randomguy wrote:
Fri Aug 31, 2018 8:40 am
R2D2 wrote:
Fri Aug 31, 2018 8:27 am
willthrill81 wrote:
Thu Aug 30, 2018 6:48 pm
Remember that it's a balancing act between the risk of running out of money, which is far overblown IMHO,
Can you elaborate on that?
To run out of money you need both
a) bad returns
b) to be alive

The number of 60-65 year olds that make to 90 is pretty low. The number of people that retire in the worst possible year is pretty low (we are talking like 4 years in the past 100 where the 4% rule struggled).

And in reality people don't constantly spend 4%. There is a pretty natural reduction in spending that has nothing to do with lack of funds. And I am betting most people who retire in say 2008 voluntary defer a bit of spending.
:thumbsup Well said.

Other factors are at work as well that reduce the risk of running out of money. First, most retirees reduce their spending in real dollars over time, by an average of 1-2% per year. Second, closely related to your last point, retirees with significant portfolios (>$500k) do not tend to spend down their portfolios, and this tendency becomes stronger as one has more wealth (a counter-intuitive result).

As an aside, when it comes to determining when to begin Social Security benefits, people think that they'll die soon (e.g. "I'll never live to 87! Start my benefits today!"). But when it comes to determining a safe withdrawal rate, they think that they'll beat the odds by a huge margin (e.g. "I may live to 100, so I need a 2% withdrawal rate!")
“It's a dangerous business, Frodo, going out your door. You step onto the road, and if you don't keep your feet, there's no knowing where you might be swept off to.” J.R.R. Tolkien,The Lord of the Rings

User avatar
willthrill81
Posts: 6603
Joined: Thu Jan 26, 2017 3:17 pm
Location: USA

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by willthrill81 » Fri Aug 31, 2018 10:06 am

R2D2 wrote:
Fri Aug 31, 2018 9:07 am
That's even more optimistic than my assumption of 6% real returns for stocks & 0% for bonds! Does anyone seriously believe that these are the returns we should expect over the next few decades? Or that we should use them for retirement planning?
I applaud your taking a hands-on approach rather than just relying on what others have said.

However, what you'll find at the end of the road is very likely to be the same as everyone else who has done work in this area: we cannot predict future returns, be they short-term or long-term. No one has been able to consistently do it with any practical degree of success.

That (largely) leaves us with historical scenarios as the best means for determining safe withdrawal rates. And as Marcopolo said, a 3% withdrawal rate is very conservative, about as close to 'bulletproof' as prudence allows. Historically, it's never even come close to failure* for a balanced portfolio (either when U.S. equities or global market cap-weighted equities were used). And in reality, you're extremely unlikely to use a fixed withdrawal rate because no one does. Everyone reduces their withdrawals when their portfolio is suffering.

*As Michael Kitces has pointed out, 'failure' in a Monte Carlo analysis sense should be thought of in terms of 'making changes to our withdrawal strategy' rather than 'running out of money'. No one blindly spends their portfolio down to the last dollar.
“It's a dangerous business, Frodo, going out your door. You step onto the road, and if you don't keep your feet, there's no knowing where you might be swept off to.” J.R.R. Tolkien,The Lord of the Rings

User avatar
patrick013
Posts: 2433
Joined: Mon Jul 13, 2015 7:49 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by patrick013 » Fri Aug 31, 2018 2:07 pm

Advisors that have begun to use Monte Carlo simulation are
still applying the same advice they applied in the past but
with access to the information Monte Carlo reports can provide.

They ask the client how much they are currently saving, what
their assets are, their risk, age of retirement and withdrawal
needs. All done with a thousand iterations instead of a class
average plus and minus an error adjustment.

They plugged these figures into their old fashioned tools that
assumed no market uncertainty on average, and they plug them into
their Monte Carlo tools to indicate how much savings shortfall
must be made up or withdrawals decreased, similarly, but with
greater concern for the result.

Critics say that even with certainty of performance replicating
over time, what a client gets from that certainty is itself
uncertain, or Monte Carlo portfolios have too much stock, or we're
dealing with ambiguity.

Advisors provide clients with a disclaimer noting that projections
generated by the tool are hypothetical.

But we do have to estimate something. Here's one.

Monte Carlo Simulation
age in bonds, buy-and-hold, 10 year business cycle

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Fri Aug 31, 2018 5:55 pm

marcopolo wrote:
Fri Aug 31, 2018 9:21 am
Returns to expect? I don't think anyone knows.
I was using the word "expected" in the mathematical sense.
marcopolo wrote:
Fri Aug 31, 2018 9:21 am
Remember that the recommended 4% withdrawal (or maybe 3.0%-3.5% for early retirees) is not based on average returns. It is based on the worst case historical periods. In the average case a withdrawal rate closer to 5%-6% would have been fine.
So, the question you have to ask is do you think the next 30 or 40 years is likely to be worse than the worst case we have seen in the last 100 years.
This is an interesting point, but I'm not sure it's quite right. If you had to pick a withdrawal rate for your retirement, would you:
  • 1. Run the portfolio over US market history going back as far as you can, and look at the worst 5% case? (This is the usual methodology around here.)
  • 2. Use a simulation based on a mean and standard deviation of returns based on past US history, and then take the worst 5% case?
  • 3. Use a simulation based on a mean and standard deviation of returns derived from finance theory (dividend rates, dividend growth rates, current yields on Treasuries, asking Mr. Bogle), and then take the worst 5% case?
I'm suggesting the we do something like #3 (in which case we want to use some pretty puny expected returns). But it sounds like you're suggesting that we do #3 but then look at the average outcome from those parameters instead of the worst x% case.

Look at it this way: If a supreme being told you right now that annual real returns for equities were going to be drawn from a normal distribution with mean 2% and standard deviation 15% and real bond returns were going to be drawn from a normal distribution with mean 0% and standard deviation 2%, what would you do with that information (as far as SWR calcs go)?

I think the right thing to do would be: Figure out the mean/stdev of your 50/50 portfolio (or whatever) based on those numbers from the supreme being, run some sims, and find the highest withdrawal rate that allows your money to last 95% (or 90% or whatever) of the time.

What I wouldn't do is run sims based on last century's returns and figure that my future returns probably won't be as bad as the worst returns from last century.

Please tell me if I'm misunderstanding you. This is a great discussion and I appreciate your reply as well as all the others.

User avatar
willthrill81
Posts: 6603
Joined: Thu Jan 26, 2017 3:17 pm
Location: USA

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by willthrill81 » Fri Aug 31, 2018 6:08 pm

R2D2 wrote:
Fri Aug 31, 2018 5:55 pm
marcopolo wrote:
Fri Aug 31, 2018 9:21 am
Remember that the recommended 4% withdrawal (or maybe 3.0%-3.5% for early retirees) is not based on average returns. It is based on the worst case historical periods. In the average case a withdrawal rate closer to 5%-6% would have been fine.
So, the question you have to ask is do you think the next 30 or 40 years is likely to be worse than the worst case we have seen in the last 100 years.
This is an interesting point, but I'm not sure it's quite right. If you had to pick a withdrawal rate for your retirement, would you:
  • 1. Run the portfolio over US market history going back as far as you can, and look at the worst 5% case? (This is the usual methodology around here.)
  • 2. Use a simulation based on a mean and standard deviation of returns based on past US history, and then take the worst 5% case?
  • 3. Use a simulation based on a mean and standard deviation of returns derived from finance theory (dividend rates, dividend growth rates, current yields on Treasuries, asking Mr. Bogle), and then take the worst 5% case?
I'm suggesting the we do something like #3 (in which case we want to use some pretty puny expected returns). But it sounds like you're suggesting that we do #3 but then look at the average outcome from those parameters instead of the worst x% case.

Look at it this way: If a supreme being told you right now that annual real returns for equities were going to be drawn from a normal distribution with mean 2% and standard deviation 15% and real bond returns were going to be drawn from a normal distribution with mean 0% and standard deviation 2%, what would you do with that information (as far as SWR calcs go)?

I think the right thing to do would be: Figure out the mean/stdev of your 50/50 portfolio (or whatever) based on those numbers from the supreme being, run some sims, and find the highest withdrawal rate that allows your money to last 95% (or 90% or whatever) of the time.

What I wouldn't do is run sims based on last century's returns and figure that my future returns probably won't be as bad as the worst returns from last century.

Please tell me if I'm misunderstanding you. This is a great discussion and I appreciate your reply as well as all the others.
Why do you believe that #3 on your list is apt to be more accurate than #1? The Kitces article linked to above demonstrated that Monte Carlo simulations without mean reversion tend to overestimate the tails of the distributions compared to historical instances, presumably due to their frequent failure to incorporate mean reversion. If the simulation you're using does account for mean reversion and reasonably matches up with historical scenarios, it may be more valid, but then that begs the question why you simply aren't relying on the historical scenarios in the first place.

Try as we may, we don't know what (2) the mean returns nor (2) the standard deviations of those returns nor (3) the distribution of those returns will be going forward.
“It's a dangerous business, Frodo, going out your door. You step onto the road, and if you don't keep your feet, there's no knowing where you might be swept off to.” J.R.R. Tolkien,The Lord of the Rings

marcopolo
Posts: 1299
Joined: Sat Dec 03, 2016 10:22 am

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by marcopolo » Fri Aug 31, 2018 6:19 pm

R2D2 wrote:
Fri Aug 31, 2018 5:55 pm
marcopolo wrote:
Fri Aug 31, 2018 9:21 am
Returns to expect? I don't think anyone knows.
I was using the word "expected" in the mathematical sense.
marcopolo wrote:
Fri Aug 31, 2018 9:21 am
Remember that the recommended 4% withdrawal (or maybe 3.0%-3.5% for early retirees) is not based on average returns. It is based on the worst case historical periods. In the average case a withdrawal rate closer to 5%-6% would have been fine.
So, the question you have to ask is do you think the next 30 or 40 years is likely to be worse than the worst case we have seen in the last 100 years.
This is an interesting point, but I'm not sure it's quite right. If you had to pick a withdrawal rate for your retirement, would you:
  • 1. Run the portfolio over US market history going back as far as you can, and look at the worst 5% case? (This is the usual methodology around here.)
  • 2. Use a simulation based on a mean and standard deviation of returns based on past US history, and then take the worst 5% case?
  • 3. Use a simulation based on a mean and standard deviation of returns derived from finance theory (dividend rates, dividend growth rates, current yields on Treasuries, asking Mr. Bogle), and then take the worst 5% case?
I'm suggesting the we do something like #3 (in which case we want to use some pretty puny expected returns). But it sounds like you're suggesting that we do #3 but then look at the average outcome from those parameters instead of the worst x% case.

Look at it this way: If a supreme being told you right now that annual real returns for equities were going to be drawn from a normal distribution with mean 2% and standard deviation 15% and real bond returns were going to be drawn from a normal distribution with mean 0% and standard deviation 2%, what would you do with that information (as far as SWR calcs go)?

I think the right thing to do would be: Figure out the mean/stdev of your 50/50 portfolio (or whatever) based on those numbers from the supreme being, run some sims, and find the highest withdrawal rate that allows your money to last 95% (or 90% or whatever) of the time.

What I wouldn't do is run sims based on last century's returns and figure that my future returns probably won't be as bad as the worst returns from last century.

Please tell me if I'm misunderstanding you. This is a great discussion and I appreciate your reply as well as all the others.
A few points.

1) The track record of finance theory predicting actual returns is quite poor. I am not sure much i would want to rely on that to develop my plan

2) Classical MCS analysis overstates the best and worst case outcomes because it does not consider interaction of returns. The years after a 50% crash are usually better than random. If you believe in financial theory-based expected returns, they should be higher after a big crash. But MCS will happily create simulation instances where 50% crashes are followed by 20% and then 30% crashes.

3) If you want to use expected return in an MCS simulation, you really need a much more complex, and more accurate, model of returns of various asset classes, with a non-diagonal co-variance matrix, as well as some model of memory in the system. I am not aware of such a complex model.

What I have done is to look at both historical data based simulations and MCS. For historical, I de-rated the SWR with the expectation that future returns may be worse than the past. For the MCS, I used a 10% threshold instead of 5%. They result in values that are close enough given the uncertainty of the inputs.

You seem to be trying to use math to eliminate uncertainty. I think that is a bit of a fools errand.

In the end, it is all a bit of a crap shoot, and requires a huge leap of faith. I considered the trade-off between the risk of running out of money vs the risk of running out of time, and chose to retire earlier this year at the age 51. Only time will tell if it was the right decision.

Best of luck in your planning. :sharebeer
Once in a while you get shown the light, in the strangest of places if you look at it right.

User avatar
willthrill81
Posts: 6603
Joined: Thu Jan 26, 2017 3:17 pm
Location: USA

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by willthrill81 » Fri Aug 31, 2018 6:39 pm

marcopolo wrote:
Fri Aug 31, 2018 6:19 pm
You seem to be trying to use math to eliminate uncertainty. I think that is a bit of a fools errand.
:beer

"Surety you crave! Sauron gives none."
- The Return of the King, Tolkien
“It's a dangerous business, Frodo, going out your door. You step onto the road, and if you don't keep your feet, there's no knowing where you might be swept off to.” J.R.R. Tolkien,The Lord of the Rings

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Sat Sep 01, 2018 8:28 am

marcopolo wrote:
Fri Aug 31, 2018 6:19 pm
A few points.

1) The track record of finance theory predicting actual returns is quite poor. I am not sure much i would want to rely on that to develop my plan

2) Classical MCS analysis overstates the best and worst case outcomes because it does not consider interaction of returns. The years after a 50% crash are usually better than random. If you believe in financial theory-based expected returns, they should be higher after a big crash. But MCS will happily create simulation instances where 50% crashes are followed by 20% and then 30% crashes.
If I grant you that point, I think the easiest way around it is to do something like what William Bernstein does and turn down the standard deviation in my sims. I agree that this can help a lot.
marcopolo wrote:
Fri Aug 31, 2018 6:19 pm
3) If you want to use expected return in an MCS simulation, you really need a much more complex, and more accurate, model of returns of various asset classes, with a non-diagonal co-variance matrix, as well as some model of memory in the system. I am not aware of such a complex model.

What I have done is to look at both historical data based simulations and MCS. For historical, I de-rated the SWR with the expectation that future returns may be worse than the past. For the MCS, I used a 10% threshold instead of 5%. They result in values that are close enough given the uncertainty of the inputs.
This sounds interesting, but can you explain? I'm not sure what "de-rating" is and what the 10% vs 5% thresholds refer to.

marcopolo
Posts: 1299
Joined: Sat Dec 03, 2016 10:22 am

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by marcopolo » Sat Sep 01, 2018 11:08 am

R2D2 wrote:
Sat Sep 01, 2018 8:28 am
marcopolo wrote:
Fri Aug 31, 2018 6:19 pm
3) If you want to use expected return in an MCS simulation, you really need a much more complex, and more accurate, model of returns of various asset classes, with a non-diagonal co-variance matrix, as well as some model of memory in the system. I am not aware of such a complex model.

What I have done is to look at both historical data based simulations and MCS. For historical, I de-rated the SWR with the expectation that future returns may be worse than the past. For the MCS, I used a 10% threshold instead of 5%. They result in values that are close enough given the uncertainty of the inputs.
This sounds interesting, but can you explain? I'm not sure what "de-rating" is and what the 10% vs 5% thresholds refer to.
I can try, but i am not sure you will find it very satisfying.

For de-rating historical results, I see two possibilities:
1) Use something like the excellent work by EarlyRetireNow (https://earlyretirementnow.com/2016/12/ ... t-1-intro/). Find the appropriate SWR for your Asset Allocation and time horizon, then as a final step, reduce the resulting SWR by some factor (maybe 20%?) if you anticipate the future being that much worse than the WORST period in history. So, if the corresponding SWR based ion historical performance for your situation was 3.5%, you would de-rate that by 20% to arrive at a future WR of 2.8%.

2) Build your own simulation that uses a sequence of returns, then instead of using historical sequence of returns, reduce each years historical returns by however much (maybe 20%) that you think the future will be worse, then run the simulation across that newly created sequence.

I used method (1) for simplicity.

As far as the percentile to examine in MCS, this is really the heart of how to interpret MCS results.
When you run MCS, you are hopefully generating a large number of outcomes. I like to use at least 10,000.
Then you have to decide what worst percentage of outcomes you are willing to live with.
A typical approach is to say you will accept the 5th percentile worst outcome. So, out of 10,000 simulations, you order them from worst to best outcome, you would accept the 500th worst outcome as your baseline target plan.

Knowing that MCS applied to historical market returns tends to exaggerate the best and worst case scenarios, one could instead choose the 10th percentile outcome as the baseline. So, out of 10,000 simulation results, accept the 1000th worst one as your baseline target plan.

I recognize that there is a lot of hand waving going on here, but i am not sure there is a good closed form solution to this problem.

Best of luck.
Once in a while you get shown the light, in the strangest of places if you look at it right.

michaeljc70
Posts: 3932
Joined: Thu Oct 15, 2015 3:53 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by michaeljc70 » Sat Sep 01, 2018 4:29 pm

No, it doesn't sound right. Because you think returns may be lower over the next 10 years and you are doing a 30 year simulation. And you are really just guessing (and suppositioning that just about everyone will agree with you). What if the 10 years after that they are expected to be higher than average? I've heard people bash historical returns and then use witchcraft or a crystal ball or something else to justify some other approach.

User avatar
willthrill81
Posts: 6603
Joined: Thu Jan 26, 2017 3:17 pm
Location: USA

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by willthrill81 » Sat Sep 01, 2018 5:59 pm

michaeljc70 wrote:
Sat Sep 01, 2018 4:29 pm
I've heard people bash historical returns and then use witchcraft or a crystal ball or something else to justify some other approach.
:thumbsup

The most favorable aspect of historical data IMHO is that they were real.
“It's a dangerous business, Frodo, going out your door. You step onto the road, and if you don't keep your feet, there's no knowing where you might be swept off to.” J.R.R. Tolkien,The Lord of the Rings

Nate79
Posts: 3756
Joined: Thu Aug 11, 2016 6:24 pm
Location: Delaware

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by Nate79 » Sat Sep 01, 2018 6:15 pm

Garbage in, garbage out.

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Sun Sep 02, 2018 11:23 am

Nate79 wrote:
Sat Sep 01, 2018 6:15 pm
Garbage in, garbage out.
That's exactly my concern about using historical. Yes, we know that "historical" actually happened, but people (even on this thread) seem to agree that it's a bad guide going forward and that 4% nominal returns are more likely. But they don't seem willing to adjust appropriately.

Look, somehow or another people have to decide how much money to withdraw. I feel like people have used numbers that they admit are not realistic to come up with a SWR of something like 3.5% or 4%.

Then when someone comes along and says, "here's a different way to analyze it using numbers that we might actually believe", the response is along the lines of, "Stop trying to use math to eliminate uncertainty, it's a leap of faith, garbage in - garbage out."

Having said all that, I do really appreciate the people who have responded to this thread. Hey, maybe next I'll start a thread about whether it's good to have international equity exposure or not. :D

michaeljc70
Posts: 3932
Joined: Thu Oct 15, 2015 3:53 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by michaeljc70 » Sun Sep 02, 2018 11:28 am

Math doesn't solve every problem.....Convincing yourself you are using math to solve a non-mathematical problem is not good. If you aren't using historical data, what numbers are you using in the "math"?

I have no reason to believe real returns will be lower for equities over the long term.

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Tue Sep 04, 2018 8:27 pm

michaeljc70 wrote:
Sun Sep 02, 2018 11:28 am
Math doesn't solve every problem.....Convincing yourself you are using math to solve a non-mathematical problem is not good. If you aren't using historical data, what numbers are you using in the "math"?
I like the way we're using scare quotes on "math". :D

There are certainly alternatives to using historical data. Historical data certainly has its appeal, but if people almost universally are predicting lower equity returns (4%-ish nominal is a common forecast) based on dividend yields and dividend growth, then it certain makes sense to use something more conservative than what historical returns would suggest.

protagonist
Posts: 5487
Joined: Sun Dec 26, 2010 12:47 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by protagonist » Tue Sep 04, 2018 9:05 pm

R2D2 wrote:
Wed Aug 29, 2018 2:34 pm


A 4% withdrawal rate for 30 years only works about 82% of the time. I need to drop down to 3.4% withdrawal rate to get the success rate up to 95%.

Does this sound right to people? This is 15% less money per year than the usual rule and I'm arguably using unrealistically high expected return numbers! I'm starting to think that historical return data was making us too optimistic.

Did you remember to take into account your income from SS, pension or other sources?

I put little faith in doomsday Monte Carlo studies. I look around me. Most of my retired neighbors and 80 and 90 somethings I know seem to be doing OK. Most are not rich. I don't see a lot of them on the streets living out of cardboard boxes. Certainly not 18% of them. In fact, none of them. They seem, as a general rule, at least from my vantage point, fine. And most of them have probably never heard of a 4% withdrawal rate (which, if you have a median American retirement nest egg in your late 60s around $200K and no SS or pension would leave you about $8000/year to live off).

I think the best approach is to just live sensibly, stay out of debt and not overspend. If you can do that, like your parents and their parents, you will probably be OK. Guidelines like 4% or 3% or 2% are probably useful for those with problems disciplining their spending, just as AA guidelines are useful for those with problems disciplining their drinking. The fact is, we have no idea what inflation or investment returns will look like between now and 2048 or 2068. The world may be a very different place . So if you don't have a problem with spending or budgeting, don't let these studies scare you. Just live your life sensibly and you will probably be OK, barring things you can't control anyway.

And historical market data has no scientific /statistical validity anyway- even if you believe that past performance DOES predict future results , which is a long stretch. There isn't enough of it. We rely on perhaps 90 years of historical data, during what might be the most extraordinary period of economic growth in the history of civilization, to predict what will happen for the next 30. Ninety years- even 200 years- is nothing really when looking 30 or 50 years into the future. That is the mathematical equivalent of saying "it didn't rain Monday through Wednesday so based on that information alone I assume it won't rain on Thursday". Or "The stock market went up an average of 1% per day for the last 3 days so I think it will probably go up 1% tomorrow". Think in terms of symmetry of scale.

michaeljc70
Posts: 3932
Joined: Thu Oct 15, 2015 3:53 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by michaeljc70 » Wed Sep 05, 2018 8:59 am

R2D2 wrote:
Tue Sep 04, 2018 8:27 pm
michaeljc70 wrote:
Sun Sep 02, 2018 11:28 am
Math doesn't solve every problem.....Convincing yourself you are using math to solve a non-mathematical problem is not good. If you aren't using historical data, what numbers are you using in the "math"?
I like the way we're using scare quotes on "math". :D

There are certainly alternatives to using historical data. Historical data certainly has its appeal, but if people almost universally are predicting lower equity returns (4%-ish nominal is a common forecast) based on dividend yields and dividend growth, then it certain makes sense to use something more conservative than what historical returns would suggest.
Doesn't dividend growth use historical data? And no, it certainly doesn't make more sense to me. There is nothing certain here.

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Wed Sep 05, 2018 2:12 pm

michaeljc70 wrote:
Wed Sep 05, 2018 8:59 am
R2D2 wrote:
Tue Sep 04, 2018 8:27 pm
michaeljc70 wrote:
Sun Sep 02, 2018 11:28 am
Math doesn't solve every problem.....Convincing yourself you are using math to solve a non-mathematical problem is not good. If you aren't using historical data, what numbers are you using in the "math"?
I like the way we're using scare quotes on "math". :D

There are certainly alternatives to using historical data. Historical data certainly has its appeal, but if people almost universally are predicting lower equity returns (4%-ish nominal is a common forecast) based on dividend yields and dividend growth, then it certain makes sense to use something more conservative than what historical returns would suggest.
Doesn't dividend growth use historical data? And no, it certainly doesn't make more sense to me. There is nothing certain here.
You're certainly allowed to disagree on the projections of dividend growth and thus disagree on future equity returns. The inconsistency I see if that a lot of the same people who buy Bogle's argument about future returns still want to use 20th century returns to pick their SWR.

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Wed Sep 05, 2018 2:19 pm

protagonist wrote:
Tue Sep 04, 2018 9:05 pm
I don't see a lot of them on the streets living out of cardboard boxes. Certainly not 18% of them. In fact, none of them.
I get what you're saying, but to be fair, that "18% case" would strike many people at once or not hit anybody at all. For example, if we had a really bad drawdown in equities and high inflation, then a whole lot of people would be in financial trouble at once. If equities do fine in real terms, then we could have almost everybody able to make ends meet.
protagonist wrote:
Tue Sep 04, 2018 9:05 pm
The fact is, we have no idea what inflation or investment returns will look like between now and 2048 or 2068. The world may be a very different place ... Ninety years- even 200 years- is nothing really when looking 30 or 50 years into the future.
This is a good point, and it really makes me wonder if the early retirement people are taking a much bigger risk than they believe. It's tough to plan for a 50-year (for example) retirement when the world can change so radically in just 20 years.

In fact, this makes me wonder if saying "I need to spend $x/yr and I'll adjust for inflation each year" even makes sense. What if you keep up with inflation but we keep inventing cooler and cooler stuff that you can't afford (but you can still afford your old basket of goods)? You might not be happy at all with the same real income in 20 years!!

michaeljc70
Posts: 3932
Joined: Thu Oct 15, 2015 3:53 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by michaeljc70 » Wed Sep 05, 2018 3:54 pm

R2D2 wrote:
Wed Sep 05, 2018 2:12 pm
michaeljc70 wrote:
Wed Sep 05, 2018 8:59 am
R2D2 wrote:
Tue Sep 04, 2018 8:27 pm
michaeljc70 wrote:
Sun Sep 02, 2018 11:28 am
Math doesn't solve every problem.....Convincing yourself you are using math to solve a non-mathematical problem is not good. If you aren't using historical data, what numbers are you using in the "math"?
I like the way we're using scare quotes on "math". :D

There are certainly alternatives to using historical data. Historical data certainly has its appeal, but if people almost universally are predicting lower equity returns (4%-ish nominal is a common forecast) based on dividend yields and dividend growth, then it certain makes sense to use something more conservative than what historical returns would suggest.
Doesn't dividend growth use historical data? And no, it certainly doesn't make more sense to me. There is nothing certain here.
You're certainly allowed to disagree on the projections of dividend growth and thus disagree on future equity returns. The inconsistency I see if that a lot of the same people who buy Bogle's argument about future returns still want to use 20th century returns to pick their SWR.
If you aren't going to use the 20th century, I am still not clear what data you are using (maybe the 21st century only?). John Bogle said that he thought equity returns would be lower over the next 10 years. Most people have a longer time frame than that. The original Trinity study concentrated on 30 years, and the updated one went out as much as 40 years. Is there evidence that the next 2 decades after the next one won't have higher than usual returns? Not that I see. And Bogle was talking about equities which is only a portion of most people's portfolios. It seems you are using what one guy says about 10 years to refute a study that used ~90 years of real data over 30 years. Fine. Good luck.

This is what Bogle said. It doesn't sound very scientific (or convincing) to me:

"Just for mathematical reasons, the dividend yield is 2 percent, a little under 2 percent in fact, and the long-term dividend yield on stocks is pretty close to 4 ... the earnings growth on stocks has been a little over 5, that's going to be a very tough target in the future so let's call it 4 ... 4 and 2 percent give you a 6 percent investment return, but then you have to take ... the valuations in the market. ...You take that 6 percent return and maybe knock it off a couple of points perhaps for a lower valuation, slightly lower valuation over a decade and you're talking about a 4 percent nominal return on stocks. And that's low, lower than history. History is around 6 and a half."

protagonist
Posts: 5487
Joined: Sun Dec 26, 2010 12:47 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by protagonist » Wed Sep 05, 2018 6:28 pm

R2D2 wrote:
Wed Sep 05, 2018 2:19 pm
[
This is a good point, and it really makes me wonder if the early retirement people are taking a much bigger risk than they believe. It's tough to plan for a 50-year (for example) retirement when the world can change so radically in just 20 years.

What is often ignored here is that "risk" is a double edged sword.

I had a heart attack at 62, despite being in excellent shape with no risk factors at all other than being male. I am currently 66 and healthy, having retired at 55.

I would have a lot more money now if I was still working, and by definition a more secure retirement, but would that have been the less "risky" choice?

I could have died at 62- I almost did- and "risked" spending the last seven years of my life (55 to 62) working for my retirement until the day I was rushed into the intensive care unit. I would have died with a pile of money I never got to enjoy. Instead I spent it windsurfing, playing and honing my skills at an instrument, cementing tight relationships with family and friends (difficult when you have little free time), sharing quality time with my partner, and living life to its fullest. At 66 I'm happier than I have ever been.

If you retire early you run greater risk of running out of money. If you retire late (or never) you run greater risk of never getting to enjoy your money. Either way, you are taking risk against undefinable odds in an undefinable future. That's called "life". He who dies with the most toys does not necessarily win the game.

The bottom line is , to paraphrase Shakespeare, all the world is a play.

(By the way, your point about the 18% - hitting them all at once- is a good one. I would still ignore the result, however, and just hope for the best in an uncertain world. Life is too short to climb a wall of worry.)
Last edited by protagonist on Wed Sep 05, 2018 6:58 pm, edited 1 time in total.

randomguy
Posts: 6678
Joined: Wed Sep 17, 2014 9:00 am

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by randomguy » Wed Sep 05, 2018 6:49 pm

protagonist wrote:
Tue Sep 04, 2018 9:05 pm
There isn't enough of it. We rely on perhaps 90 years of historical data, during what might be the most extraordinary period of economic growth in the history of civilization, to predict what will happen for the next 30.
Or it might be pretty historically bad 90 years. What are the odds the next 90 or so years will have 2 World wars, a great depression caused by poor government action, another 3 police action/wars, an oil shock/stagflation with poor government response, and then some more deregulation to cause a real estate collapse?:) I have no clue if that is a above average, below average or about average number of screw ups:)

There is a long history of doomsday predications. They are pretty much always wrong. Except for the one time they are right;)

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Fri Sep 07, 2018 9:15 pm

randomguy wrote:
Wed Sep 05, 2018 6:49 pm
Or it might be pretty historically bad 90 years. What are the odds the next 90 or so years will have 2 World wars, a great depression caused by poor government action, another 3 police action/wars, an oil shock/stagflation with poor government response, and then some more deregulation to cause a real estate collapse?:) I have no clue if that is a above average, below average or about average number of screw ups:)
Wow, I think that actually made me feel a little better. :)

jmk
Posts: 383
Joined: Tue Nov 01, 2011 7:48 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by jmk » Fri Sep 07, 2018 11:54 pm

R2D2 wrote:
Thu Aug 30, 2018 6:04 pm
Now that I think about it, there really are three methodologies one could use:
1. Look at past history and keep everything "in order",
2. Use bootstrapping from historical returns (in other words, do Monte Carlo on historical returns and allow things to get out of order), and
3. Use Monte Carlo with forward-looking expected returns
By the way, you can pretty easily do these bootstrapping experiments with Simba spreadsheet and excel running eg 10000 runs with "Data Table" function. I went down a similar path with 50/50 for 40y and came out with 3.5% SWR. I assumed 3% for stocks for the next 15 years and one percent for Bonds (real). And then I did a normal bootstrap 1871 to 2017 for the other 25 years. I used 6-year samples of returns In order to get away from the “mad Max” effect that was described above. The average SWR was pretty similar with all the sample sizes (not surprising since same source data), but chances of success varied with sample size: When I ran with random one year bootstrapping, the returns or more diverse, with more low and high results. 12y samples had higher probabilities of success, due to reversion to mean. I finally settled on 6y sample as a compromise between revert to mean and random.

I didn't use parametric monte carlo, I used actual data for a given six years (maintaining equity-bond-inflation for a given year)-- this was to preserve macro economic relationships between stocks and bonds and inflation.

Oddly, even with this primitive method I got the same results as posted by Glau and others using more robust monte carlo methods.
Last edited by jmk on Mon Sep 10, 2018 7:46 pm, edited 3 times in total.

User avatar
Nestegg_User
Posts: 1145
Joined: Wed Aug 05, 2009 1:26 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by Nestegg_User » Sat Sep 08, 2018 12:09 am

R2D2

you might want to peruse some of the work of Pfau (and others) regarding international SWR’s, which includes some countries either overrun in the world wars or direct participants...not including the US. It shows just how low things can get, especially when you weren’t the “lucky” country that didn’t have significant losses of capacity, capital, or population. By viewing these cases, it helps inform on the very real case if the US becomes just another of the developed market instead of the leading market (as noted above that happened to the UK).

am
Posts: 2794
Joined: Sun Sep 30, 2007 9:55 am

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by am » Sat Sep 08, 2018 1:19 pm

R2D2 wrote:
Fri Sep 07, 2018 9:15 pm
randomguy wrote:
Wed Sep 05, 2018 6:49 pm
Or it might be pretty historically bad 90 years. What are the odds the next 90 or so years will have 2 World wars, a great depression caused by poor government action, another 3 police action/wars, an oil shock/stagflation with poor government response, and then some more deregulation to cause a real estate collapse?:) I have no clue if that is a above average, below average or about average number of screw ups:)
Wow, I think that actually made me feel a little better. :)
But a good 21st century may not mean better returns, just like good companies don’t always have the best returns.

Seems like a balancing act between running out of money or time. Once you get to middle age and beyond, your chances of getting a debilitating sickness or death are exponentially higher. This can come at any time even if your currently healthy and even if your parents lived to 100. I see this all the time in my profession. Therefore guaranteeing a 99.9% successful 50 yr retirement by working more years may not pay off. At some point you just have to be reasonable and flexible and let go if you don’t want to drop dead at work or shortly after (have seen this also).

KyleAAA
Posts: 6805
Joined: Wed Jul 01, 2009 5:35 pm
Contact:

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by KyleAAA » Sat Sep 08, 2018 3:08 pm

You are using unrealistically pessimistic assumptions. People have been predicting low equity returns over the next decade for well over a decade. I haven't seen a compelling argument to expect less than 8-10% going forward, maybe a bit more for a heavily tilted portfolio.The pessimistic predictions mostly seem to be bases on bad logic, and i invest globally, besides. I think you're being similarly too pessimistic about long term bond returns.

But then, I also think 50/50 is a reckless asset allocation if you plan yo live 50 years.

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Mon Sep 10, 2018 12:22 pm

jmk wrote:
Fri Sep 07, 2018 11:54 pm
By the way, you can pretty easily do these bootstrapping experiments with Simba spreadsheet and excel running eg 10000 runs with "Data Table" function. I went down a similar path with 50/50 for 40y and came out with 3.5% SWR. I assumed 3% for stocks for the next 15 years and one percent for Bonds (real). And then I did a normal bootstrap 1871 to 2017 for the other 25 years. I used 6-year samples of returns In order to get away from the “mad Max” effect that was described above. The average SWR was pretty similar, but chances of success varied with sample size: When I ran with random one year bootstrapping, the returns or more diverse, with more low and high results. 14y samples had higher probabilities of success. I finally settled on 6y sample as a compromise between revert to mean and random.
Wow, that is very, very helpful. I'm probably looking at a 40 yr horizon at the least and these assumptions seem pretty reasonable. Thanks for posting this.

R2D2
Posts: 266
Joined: Tue Aug 19, 2014 4:37 pm

Re: Historical SWR vs. Monte Carlo on PortfolioVisualizer.com

Post by R2D2 » Mon Sep 10, 2018 12:23 pm

KyleAAA wrote:
Sat Sep 08, 2018 3:08 pm
You are using unrealistically pessimistic assumptions. People have been predicting low equity returns over the next decade for well over a decade. I haven't seen a compelling argument to expect less than 8-10% going forward, maybe a bit more for a heavily tilted portfolio.The pessimistic predictions mostly seem to be bases on bad logic, and i invest globally, besides. I think you're being similarly too pessimistic about long term bond returns.

But then, I also think 50/50 is a reckless asset allocation if you plan yo live 50 years.
Can you elaborate? Are you saying that I should have at least 60% equities?

Post Reply