Stock/Bond Mix in Retirement: The Surprise In The Data

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
User avatar
ram
Posts: 2281
Joined: Tue Jan 01, 2008 9:47 pm
Location: Midwest

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by ram »

A question for those mathematically more gifted than me.

Lets forget the behavioral aspects of investing for now.

From a mathematical point does past data support the following broad statement :

"During retirement the highest proportion of bonds should be on day 1 of retirement" ( Stable or upward stock glidepath during retirement)
Ram
User avatar
willthrill81
Posts: 32250
Joined: Thu Jan 26, 2017 2:17 pm
Location: USA
Contact:

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by willthrill81 »

ram wrote: Mon Oct 25, 2021 7:47 pm A question for those mathematically more gifted than me.

Lets forget the behavioral aspects of investing for now.

From a mathematical point does past data support the following broad statement :

"During retirement the highest proportion of bonds should be on day 1 of retirement" ( Stable or upward stock glidepath during retirement)
It seems that there was a very small benefit from a rising equity glidepath in the historic record, but it was not consistent enough for us to have much confidence in it occurring going forward. This was outlined in this post above.
The Sensible Steward
User avatar
Ben Mathew
Posts: 2720
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by Ben Mathew »

willthrill81 wrote: Mon Oct 25, 2021 7:11 pm
Ben Mathew wrote: Mon Oct 25, 2021 6:51 pm
willthrill81 wrote: Mon Oct 25, 2021 6:26 pm
Ben Mathew wrote: Mon Oct 25, 2021 6:12 pm
willthrill81 wrote: Mon Oct 25, 2021 10:11 am As such, we must question whether these results were mere historic artifacts that are not likely to be replicated going forward.
Northern Flicker wrote: Mon Oct 25, 2021 3:39 pm The historical bias I see in the bond tent samples is that the period that would challenge a low equity allocation with a high allocation to nominal bonds was 1967-1981. Bad retirement start years for this period in Bengen's sample in the original work on this from 1994 were 1965, 1966, 1968, and 1969. If a bond tent were used, there was a bear market in stocks 1973-1974, early in the period, followed by a not particular robust recovery. The bond tent's weakness to cope with inflation was rescued by the downturn being early in the period when the equity allocation was lower, with a weak recovery when the equity allocation was higher. The bond tent strategy led to rebalancing to a higher asset allocation during the bear market.
Monte Carlo simulations drawing randomly from the historical distribution (with or without replacement) will remove these timing artifatcts. If it hasn't been done already, it would be useful to do a Monte Carlo simulation and see if the upward sloping glidepath result survives.
Yes, MC simulations would help, although they bring their own issues to the mix, in particular, 'fat' tails. Using 'blocks' of years rather than individual years helps to correct this problem. I've not seen any such analyses with bond tent strategies.
Why would fat tails create a problem for Monte Carlo simulations?
Because the fat tails are probably very unrealistic. Not many plans could tolerate something like the Great Depression followed by 1970s stagflation followed by the Great Recession, etc., but that's precisely what some MC simulations will randomly create. Derek Tharp wrote a nice piece on Kitces' website about this issue.

There are ways to at least partly combat this problem, but they aren't always implemented.
I see. So the concern is that Monte Carlo will surface some very bad scenarios. That it will! The worst possible outcome would be the Great Depression happening every year. But the probability of extreme bad luck like that, though not zero, will still be very low. Monte Carlo planning based on the 5th or even the 1st percentile outcomes won't be impacted (rightly or wrongly) by the 1 in 100,000 events.
Total Portfolio Allocation and Withdrawal (TPAW)
User avatar
ram
Posts: 2281
Joined: Tue Jan 01, 2008 9:47 pm
Location: Midwest

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by ram »

Thanks Willthrill.
Ram
UsualLine
Posts: 70
Joined: Mon Dec 17, 2018 12:51 pm

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by UsualLine »

willthrill81 wrote: Mon Oct 25, 2021 7:55 pm
ram wrote: Mon Oct 25, 2021 7:47 pm A question for those mathematically more gifted than me.

Lets forget the behavioral aspects of investing for now.

From a mathematical point does past data support the following broad statement :

"During retirement the highest proportion of bonds should be on day 1 of retirement" ( Stable or upward stock glidepath during retirement)
It seems that there was a very small benefit from a rising equity glidepath in the historic record, but it was not consistent enough for us to have much confidence in it occurring going forward. This was outlined in this post above.
The bond tent or rising equity glidepath is appealing and makes sense to me because it makes the worst-case scenario (large portfolio reduction early in retirement) better. Sure if stocks fall after you are out of the tent you will come out behind a fixed allocation, but it doesn't matter as much because those aren't the dangerous cases. The OP, Kitces, and ERN all show the same result with their different methodologies.

I'd also like to point out that the 20 bps increase in SWR the ERN articles show, translates into ~6% increase (3.2% -> 3.4%) in the retiree's standard of living. Maybe "very small" but certainly better than being poked in the eye with a sharp stick, as the saying goes.
User avatar
Ben Mathew
Posts: 2720
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by Ben Mathew »

Ben Mathew wrote: Sun Oct 24, 2021 1:59 pm The glidepath that minimizes sequence of return risk is downward sloping prior to retirement and fixed in retirement (putting aside pensions). The downward slope prior to retirement is to adjust for the fact that not all contributions are in. Once retirement starts, there's no more future contributions, so the asset allocation becomes fixed.

The fixed allocation during retirement may seem surprising when you consider that the age 65 portfolio is much larger than the portfolio at ages 75 or 85 will be. So how can the same stock percentage result in constant risk?

To see why, look at it like this: At age 65, the retirement account holds money that will fund many different ages: age 66, 67,...,100. At age 85, it will hold money that will fund only ages 86, 87, ..., 100.

By holding the same fixed percentage in retirement, you are simply applying the same allocation to each age bucket every year. A 10% drop in stocks at age 65 affects future consumption by the same amount as a 10% drop in stocks at age 85 would do. You are not overexposed at age 65 relative to other ages. There are just more age buckets you're funding. Age 65 returns does matter more because it impacts more ages. But each age that is impacted is impacted the same.

If instead you increase stock allocation over time (bond bridge, rising equity etc.) then the age 85 stock return will impact future ages more than the age 65 stock return. Age 90 consumption will be more dependent on age 85 return than on age 65 return. Risk is not being kept constant leading to higher overall risk for the same expected return.
I now realize that what I wrote above is correct for the variable ABW withdrawal strategy, but not for an inflexible SWR strategy. I was too quick to assume that the results from one will apply to the other. The inflexbile SWR works quite differently, and after thinking about it some more, I now believe that an increasing glide path will actually reduce risk for inflexible withdrawals.

The key difference is that while the flexible withdrawals of ABW spreads the portfolio risk across all future ages, inflexible withdrawals places all of the risk on the oldest age currently funded. So in early retirement, when the portfolio is large, say $1 million, the entire gain/loss of $1 million is being applied to the oldest funded age. Late in retirement, when the portfolio is only $200,000, only the gain/loss of $200,000 is being applied to the oldest funded age. So to equalize risk, the $1 million should have a lower AA and the $200,000 should have a higher AA. I think the rule here might be something along the lines of "to keep risk constant in an inflexible withdrawal strategy, hold a constant dollar stock allocation (present value adjusted) rather than a constant percentage stock allocation."

While the increasing glidepath result may be mathematically correct, that's only because of the inflexible nature of the withdrawals. A better approach is variable ABW withdrawals, and for that the fixed percentage allocation is the one that minimizes risk.
Last edited by Ben Mathew on Tue Oct 26, 2021 12:28 am, edited 1 time in total.
Total Portfolio Allocation and Withdrawal (TPAW)
Northern Flicker
Posts: 15363
Joined: Fri Apr 10, 2015 12:29 am

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by Northern Flicker »

Ben Mathew wrote: Mon Oct 25, 2021 6:12 pm
willthrill81 wrote: Mon Oct 25, 2021 10:11 am As such, we must question whether these results were mere historic artifacts that are not likely to be replicated going forward.
Northern Flicker wrote: Mon Oct 25, 2021 3:39 pm The historical bias I see in the bond tent samples is that the period that would challenge a low equity allocation with a high allocation to nominal bonds was 1967-1981. Bad retirement start years for this period in Bengen's sample in the original work on this from 1994 were 1965, 1966, 1968, and 1969. If a bond tent were used, there was a bear market in stocks 1973-1974, early in the period, followed by a not particular robust recovery. The bond tent's weakness to cope with inflation was rescued by the downturn being early in the period when the equity allocation was lower, with a weak recovery when the equity allocation was higher. The bond tent strategy led to rebalancing to a higher asset allocation during the bear market.
Monte Carlo simulations drawing randomly from the historical distribution (with or without replacement) will remove these timing artifatcts. If it hasn't been done already, it would be useful to do a Monte Carlo simulation and see if the upward sloping glidepath result survives.
I agree that Monte Carlo simulations will have that benefit, but they still have the limitation that we don't know the probability distribution of future returns.
User avatar
Garco
Posts: 1078
Joined: Wed Jan 23, 2013 1:04 am
Location: U.S.A.

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by Garco »

I'm in my late 70's. Retired since 2014. I'm not an age-in-bonds practitioner. At this time, I'm 62% equities in my main retirement account. I see no reason to change. To me it's something of a "golden ratio." I live on my RMD's, which by formula and by market performance have increased every year of my retirement.
Last edited by Garco on Tue Oct 26, 2021 10:13 am, edited 1 time in total.
User avatar
willthrill81
Posts: 32250
Joined: Thu Jan 26, 2017 2:17 pm
Location: USA
Contact:

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by willthrill81 »

UsualLine wrote: Mon Oct 25, 2021 8:47 pm
willthrill81 wrote: Mon Oct 25, 2021 7:55 pm
ram wrote: Mon Oct 25, 2021 7:47 pm A question for those mathematically more gifted than me.

Lets forget the behavioral aspects of investing for now.

From a mathematical point does past data support the following broad statement :

"During retirement the highest proportion of bonds should be on day 1 of retirement" ( Stable or upward stock glidepath during retirement)
It seems that there was a very small benefit from a rising equity glidepath in the historic record, but it was not consistent enough for us to have much confidence in it occurring going forward. This was outlined in this post above.
The bond tent or rising equity glidepath is appealing and makes sense to me because it makes the worst-case scenario (large portfolio reduction early in retirement) better. Sure if stocks fall after you are out of the tent you will come out behind a fixed allocation, but it doesn't matter as much because those aren't the dangerous cases. The OP, Kitces, and ERN all show the same result with their different methodologies.

I'd also like to point out that the 20 bps increase in SWR the ERN articles show, translates into ~6% increase (3.2% -> 3.4%) in the retiree's standard of living. Maybe "very small" but certainly better than being poked in the eye with a sharp stick, as the saying goes.
If you want to reduce downside risk as measured by SWR, there are other strategies that have been far more effective. These include international diversification, tilting toward factors, real estate, and gold, among others.
The Sensible Steward
seajay
Posts: 1656
Joined: Sat May 01, 2021 3:26 pm
Contact:

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by seajay »

Long dated bonds can exhibit inverse correlation to stocks over the shorter term (year/few), but mid to longer term tend to correlate. stock/bonds is OK for reducing shorter term volatility but less so for the mid/longer term.

Consider a bucket approach of two assets 50/50, one that is drawn down at 8% (so 4% portfolio wide SWR), the other that is left to accumulate. Generally a bad decade/two/three for one asset servicing drawdown can be good for accumulators, so in splitting between both still accumulating and also drawing down you've lowered SWR risk. Ideally you want assets that exhibit mid/longer term inverse correlations and stock/gold tends to do that better than stock/bonds.

1980 to 2000 was a historic pretty bad time for gold, nominal price alone declined, let alone losing in real terms. Stocks did great, around 15% annualized. 50/50 stock/gold rebalanced saw around 8% combined reward, but where 10% annualized more ounces of gold were accumulated over those years. In other periods that swings the other way around, but more usually to less extremes. Measure multi-decade, 20 or 30 year period correlations and stock/gold exhibited a modest inverse correlation (around -0.5).

Rather than separate buckets however just combining the two and drawing income from the yearly better performing, leaving the other as-is (which is like still accumulating) and that tends to work out OK. Measure the yearly best and worst of stock and gold and broadly the average of the yearly best (which may be either stock or gold) and average of the yearly worst tends to exhibit figures such as +20% and 0% type averages. Which might be considered as a single asset tending to average +20% and the other averaging 0%, such that 50/50 averages 10%. Just a shame we can't identify the consistent +20% alone choice :)

The same applied to 50/50 stock/bonds, bought and held and spending bonds first and if a bad spell where bonds don't last too long then there's a risk that stock accumulation also didn't do so well i.e. the two tend to broadly correlate across multiple years.
UsualLine
Posts: 70
Joined: Mon Dec 17, 2018 12:51 pm

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by UsualLine »

willthrill81 wrote: Tue Oct 26, 2021 10:09 am
UsualLine wrote: Mon Oct 25, 2021 8:47 pm
willthrill81 wrote: Mon Oct 25, 2021 7:55 pm
ram wrote: Mon Oct 25, 2021 7:47 pm A question for those mathematically more gifted than me.

Lets forget the behavioral aspects of investing for now.

From a mathematical point does past data support the following broad statement :

"During retirement the highest proportion of bonds should be on day 1 of retirement" ( Stable or upward stock glidepath during retirement)
It seems that there was a very small benefit from a rising equity glidepath in the historic record, but it was not consistent enough for us to have much confidence in it occurring going forward. This was outlined in this post above.
The bond tent or rising equity glidepath is appealing and makes sense to me because it makes the worst-case scenario (large portfolio reduction early in retirement) better. Sure if stocks fall after you are out of the tent you will come out behind a fixed allocation, but it doesn't matter as much because those aren't the dangerous cases. The OP, Kitces, and ERN all show the same result with their different methodologies.

I'd also like to point out that the 20 bps increase in SWR the ERN articles show, translates into ~6% increase (3.2% -> 3.4%) in the retiree's standard of living. Maybe "very small" but certainly better than being poked in the eye with a sharp stick, as the saying goes.
If you want to reduce downside risk as measured by SWR, there are other strategies that have been far more effective. These include international diversification, tilting toward factors, real estate, and gold, among others.
Those things don't preclude implementing a glide path.
User avatar
spencydub
Posts: 142
Joined: Sat Sep 12, 2020 2:40 pm

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by spencydub »

age - 30 in bonds, stopping at 50/50 (maybe 60/40)??
tomsense76
Posts: 1428
Joined: Wed Oct 14, 2020 1:52 am

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by tomsense76 »

siamond wrote: Sat Oct 23, 2021 10:34 am
tomsense76 wrote: Fri Oct 22, 2021 9:36 pm[...] It's much easier to think of (and fairly be afraid of) a portfolio being cut in half do to a massive decline. Though it is much harder for people to conceptualize/visualize the vicious effects of high inflation like that of the 70s. This in spite of the fact we know the latter is far more dangerous to a portfolio. Especially for a retiree who no longer is earning a paycheck (though they are getting some inflation adjustment via SS that may not cover all of their spending).
...

And if you want a true heart attack, check the longer history. Yes, you got that right. Bonds didn't recover from 40+ years... And this is the US. I have the data for the UK and it is even worse. Sure, we have TIPS nowadays and they didn't exist at that time, but those remain government-controlled instruments and call me skeptical that in time of duress (e.g. war or else), things will go accordingly to plan...

Image
Wow! Thanks for sharing this. Knew bonds struggled in the 70s, but the fact the pain started in 1940 is eye opening :shock:

Yeah it will be interesting to see how TIPS and I-Bonds behave under more significant stress. Even if they do behave as investors hope, secondary effects (like taxes) can still negatively effect how they behave in practice.
"Anyone who claims to understand quantum theory is either lying or crazy" -- Richard Feynman
User avatar
siamond
Posts: 6008
Joined: Mon May 28, 2012 5:50 am

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by siamond »

tomsense76 wrote: Tue Oct 26, 2021 6:54 pmWow! Thanks for sharing this. Knew bonds struggled in the 70s, but the fact the pain started in 1940 is eye opening :shock:
Yup. And it was really a combo of inflation shock and government artificially constraining rates. I don't recall/understand the exact details, other people would be able to elaborate much more, but what I do know is that technology changes, but human (including government decision makers) behavior does not. And in time of duress, the best interest of an individual [investor] is rarely aligned with the best interest of the country... The bond crisis in the UK also stemmed very directly from WW-II (and WW-I too).
User avatar
willthrill81
Posts: 32250
Joined: Thu Jan 26, 2017 2:17 pm
Location: USA
Contact:

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by willthrill81 »

siamond wrote: Tue Oct 26, 2021 8:17 pmThe bond crisis in the UK also stemmed very directly from WW-II (and WW-I too).
Just yesterday I was learning how much wealth was transferred from Europe to the U.S. from WW1 alone. The value of exports in the U.S. alone grew by $3.8 billion from 1913 to 1917, $89 billion in today's dollars.
The Sensible Steward
BogleFan510
Posts: 1039
Joined: Tue Aug 04, 2020 2:13 pm

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by BogleFan510 »

Northern Flicker wrote: Tue Oct 26, 2021 12:27 am
Ben Mathew wrote: Mon Oct 25, 2021 6:12 pm
willthrill81 wrote: Mon Oct 25, 2021 10:11 am As such, we must question whether these results were mere historic artifacts that are not likely to be replicated going forward.
Northern Flicker wrote: Mon Oct 25, 2021 3:39 pm The historical bias I see in the bond tent samples is that the period that would challenge a low equity allocation with a high allocation to nominal bonds was 1967-1981. Bad retirement start years for this period in Bengen's sample in the original work on this from 1994 were 1965, 1966, 1968, and 1969. If a bond tent were used, there was a bear market in stocks 1973-1974, early in the period, followed by a not particular robust recovery. The bond tent's weakness to cope with inflation was rescued by the downturn being early in the period when the equity allocation was lower, with a weak recovery when the equity allocation was higher. The bond tent strategy led to rebalancing to a higher asset allocation during the bear market.
Monte Carlo simulations drawing randomly from the historical distribution (with or without replacement) will remove these timing artifatcts. If it hasn't been done already, it would be useful to do a Monte Carlo simulation and see if the upward sloping glidepath result survives.
I agree that Monte Carlo simulations will have that benefit, but they still have the limitation that we don't know the probability distribution of future returns.
Agreed. The problem is that we only get one outcome. If the economy tanks or the US goes into a crisis competitively or politically, low probability notwithstanding, we dont get another shot at wealth building. Models are great, but life is not a model, hence some value to conservative views.

The upside tail case of more unneeded wealth may be asymetric to the downside tail case of equity crash poverty.
User avatar
willthrill81
Posts: 32250
Joined: Thu Jan 26, 2017 2:17 pm
Location: USA
Contact:

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by willthrill81 »

Ben Mathew wrote: Mon Oct 25, 2021 8:14 pm
willthrill81 wrote: Mon Oct 25, 2021 7:11 pm
Ben Mathew wrote: Mon Oct 25, 2021 6:51 pm
willthrill81 wrote: Mon Oct 25, 2021 6:26 pm
Ben Mathew wrote: Mon Oct 25, 2021 6:12 pm Monte Carlo simulations drawing randomly from the historical distribution (with or without replacement) will remove these timing artifatcts. If it hasn't been done already, it would be useful to do a Monte Carlo simulation and see if the upward sloping glidepath result survives.
Yes, MC simulations would help, although they bring their own issues to the mix, in particular, 'fat' tails. Using 'blocks' of years rather than individual years helps to correct this problem. I've not seen any such analyses with bond tent strategies.
Why would fat tails create a problem for Monte Carlo simulations?
Because the fat tails are probably very unrealistic. Not many plans could tolerate something like the Great Depression followed by 1970s stagflation followed by the Great Recession, etc., but that's precisely what some MC simulations will randomly create. Derek Tharp wrote a nice piece on Kitces' website about this issue.

There are ways to at least partly combat this problem, but they aren't always implemented.
I see. So the concern is that Monte Carlo will surface some very bad scenarios. That it will! The worst possible outcome would be the Great Depression happening every year. But the probability of extreme bad luck like that, though not zero, will still be very low. Monte Carlo planning based on the 5th or even the 1st percentile outcomes won't be impacted (rightly or wrongly) by the 1 in 100,000 events.
If you read the Tharp piece I linked to above, you'll see that MC simulations can produce 'fatter' tails than that. The worst case historic scenario in the U.S. was about the 7th percentile from a MC simulation using historic parameters. And the right tails are similarly overly optimistic compared to history as well.

Image
https://www.kitces.com/blog/monte-carlo ... l-returns/

The likely reason for these fat tails is because most MC simulations don't incorporate any mean reversion, what Tharp refers to as "the natural rebound effects of markets."
Last edited by willthrill81 on Wed Oct 27, 2021 1:07 pm, edited 1 time in total.
The Sensible Steward
User avatar
Topic Author
martincmartin
Posts: 900
Joined: Wed Jul 02, 2014 3:04 pm
Location: Boston, MA USA

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by martincmartin »

willthrill81 wrote: Wed Oct 27, 2021 10:49 am The likely reason for these fat tails is because most MC simulations don't incorporate any mean reversion, what Tharp refers to as "the natural rebound effects of markets."
Autoregressive models do. But of course, they're still making lots of assumptions.
User avatar
Ben Mathew
Posts: 2720
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by Ben Mathew »

willthrill81 wrote: Wed Oct 27, 2021 10:49 am
Ben Mathew wrote: Mon Oct 25, 2021 8:14 pm
willthrill81 wrote: Mon Oct 25, 2021 7:11 pm
Ben Mathew wrote: Mon Oct 25, 2021 6:51 pm
willthrill81 wrote: Mon Oct 25, 2021 6:26 pm

Yes, MC simulations would help, although they bring their own issues to the mix, in particular, 'fat' tails. Using 'blocks' of years rather than individual years helps to correct this problem. I've not seen any such analyses with bond tent strategies.
Why would fat tails create a problem for Monte Carlo simulations?
Because the fat tails are probably very unrealistic. Not many plans could tolerate something like the Great Depression followed by 1970s stagflation followed by the Great Recession, etc., but that's precisely what some MC simulations will randomly create. Derek Tharp wrote a nice piece on Kitces' website about this issue.

There are ways to at least partly combat this problem, but they aren't always implemented.
I see. So the concern is that Monte Carlo will surface some very bad scenarios. That it will! The worst possible outcome would be the Great Depression happening every year. But the probability of extreme bad luck like that, though not zero, will still be very low. Monte Carlo planning based on the 5th or even the 1st percentile outcomes won't be impacted (rightly or wrongly) by the 1 in 100,000 events.
If you read the Tharp piece I linked to above, you'll see that MC simulations can produce 'fatter' tails than that. The worst case historic scenario in the U.S. was about the 93rd percentile from a MC simulation using historic parameters. And the right tails are similarly overly optimistic compared to history as well.

Image
https://www.kitces.com/blog/monte-carlo ... l-returns/

The likely reason for these fat tails is because most MC simulations don't incorporate any mean reversion, what Tharp refers to as "the natural rebound effects of markets."
Yes, you and Tharp are right that not incorporating mean reversion into the Monte Carlo simulation will lead to a wider dispersion in the simulated data over horizons longer than the sampled periods. MC simulations can adjust for this, but yes, most don't. At the same time, not assuming mean reversion is the more conservative modeling option. So that might still be the right choice for someone seeking conservative financial planning. See this post for my thoughts on mean reversion in financial planning.
Total Portfolio Allocation and Withdrawal (TPAW)
Northern Flicker
Posts: 15363
Joined: Fri Apr 10, 2015 12:29 am

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by Northern Flicker »

A Monte Carlo simulation is simulating the outcome of a random variable or time series of random variables.

A Monte Carlo simulation is only susceptible to a problem with fat tails if the probability distribution for the time series of the random variable(s) being simulated is wrong.

If mean reversion needs to be modeled, the time series includes state, which would require a Markov chain or process to be simulated. In the absence of state, the tendency toward mean reversion is the Law of Large Numbers.
Last edited by Northern Flicker on Wed Oct 27, 2021 11:13 pm, edited 3 times in total.
User avatar
willthrill81
Posts: 32250
Joined: Thu Jan 26, 2017 2:17 pm
Location: USA
Contact:

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by willthrill81 »

Ben Mathew wrote: Wed Oct 27, 2021 1:33 pm Yes, you and Tharp are right that not incorporating mean reversion into the Monte Carlo simulation will lead to a wider dispersion in the simulated data over horizons longer than the sampled periods. MC simulations can adjust for this, but yes, most don't. At the same time, not assuming mean reversion is the more conservative modeling option. So that might still be the right choice for someone seeking conservative financial planning. See this post for my thoughts on mean reversion in financial planning.
Many believe that planning for the worst case in U.S. history is conservative enough already.

We must be careful to weigh the potential benefits of lower withdrawal rates against the certainty of having to delay retirement in order to achieve them. For some, the decision to delay retirement is easy, while for others, the opposite is true.
The Sensible Steward
3feetpete
Posts: 613
Joined: Sun Dec 14, 2014 6:30 pm

Re: Stock/Bond Mix in Retirement: The Surprise In The Data

Post by 3feetpete »

One mix doesn't fit all. Think of a man with a 1 million dollar portfolio who needs 40k per year from it. Then think of a man with a 5 million dollar portfolio who needs 40k per year from it. The guy with 1 million should have a more conservative approach than the guy with 5 million.
Post Reply