How Much Could I Withdraw Annually

Thrugelmir

cobson said:

BritishInvestor said:

I'm not clear why a sliding window undersamples the tails of time series? 

If you have a 3 year data set and a 2 year sliding window, then you will have two data samples: years 1 and 2, and years 2 and 3. Year 2 is in both samples, but years 1 and 3 are only in one sample - so the two tail years are undersampled compared to the middle year.  The same thing happens when the numbers are scaled up.

Thanks for that. Explains why Guyton Kinger based their 30 year study on the period 1973-2004. Which is of course 31 years. Learn something everyday. 

kinger101

BritishInvestor said:

kinger101 said:

BritishInvestor said:

kinger101 said:

BritishInvestor said:

kinger101 said:

BritishInvestor said:

Linton said:

BritishInvestor said:

Linton said:

BritishInvestor said:

Linton said:

michaels said:

BritishInvestor said:

Albermarle said:

I would say the cash element is too low . Especially when you get to the point of drawing from the DC pension , you need a bigger cash buffer to get through market downturns.

Why do you "need" a bigger cash buffer? How does it improve overall outcomes?

I don't think it matters how matters how many times you point out the research on this, people still imagine you can have a nice cash pot to avoid 'drawdown' when markets are low - the same people who would claim they have no intention of ever trying to 'time the markets'....

When you need your drawdown in order to meet your basic expenses, and the market drops 40-50% what would you do?  Continue selling units but at twice the rate?

We are fortunate in that we have over a decade of historical market data to evaluate various strategies and outcomes. For example, imagine if you were an investor with a UK biased portfolio in the 1970s with the drawn out stock market slump and additional pressure being placed on the portfolio with rising interest rates. How much would you allocate to a cash buffer to cater for that?

No strategy can handle a long term fall in the markets, if that is the scenario you are proposing.  Though rising interest rates would benefit a cash buffer.  If you havent got the returns at some stage you wont be able to pay yourself the income to meet your expenses.  The best you can do is to diversify as much as possible - invest globally, invest in all sectors, invest in as many different types of asset and use as many different sources of income as you reasonably can.

The purpose of a cash buffer is to remove all risk in the medium term.  During the recent COVID mini-crash many people on the savings forum were panicking.  I was able to look on with a serene, possibly smug, smile despite my investments being essential for meeting a significant part of living expenses.

But investment strategies can only go so far, they cannot protect you from everything.  Perhaps the best they can do is to help ensure that should the world as we know it collapse one is in at least as good if not better position as everyone else.  

One detailed point - you talk about a decade of historical market data.  That means absolutely nothing.  The past decade has shown unprecedented rises in share prices and is very different to many previous decades.

Sorry, decade should be century .
) I agree that you want diversify globally and across sectors but what asset classes did you have in mind other than equities and bonds?
2) When you say no strategy can survive a long term fall in the markets, we have historical data to suggest that strategies have done and for a given retirement date we can see exactly how each strategy played out
I don't see how the cash buffer removes all risk - how do you know in advance what inflation is likely to be? What is your definition of medium term?
3) Covid (thus far) was a blip and had minimal income on a robust retirement portfolio (it's the long drawn out downturns that have the biggest impact) - we must be clear to differentiate the impact on the portfolio with the impact on the investor - that's a separate discussion.

1) Bonds covers a wide variety of different things.  In addition after a small amount of thought there is cash, direct commercial property, BTL, commodities, art, whisky, what elsewhere has been called contracts - PFI/Solar/ Toll Roads, doctors surgeries and similar infrastructure-type assets where long term agreements exist to provide a service from ownership of the assets, financial wizardry of various forms, loans, private equity/investment as opposed to public shares.  At the moment I rather like infrastructure long term income.  Certainly a better prospect than safe government bonds.
2) How can any strategy survive on long term zero or negative returns?  All expenditure would need to be financed by selling assets which are finite.  Similarly there is no way any strategy can handle global Weimar or Zimbabwe levels of inflation, except perhaps a rifle and an infinite supply of baked beans.  At some stage you must just shrug your shoulders and say that you will sort it out when it happens.  To set up an investment/retirement strategy that could handle the extremes could result in a gross distortion of asset allocation that would jeopardise your ability to handle normal events.
By medium term I am thinking 5-10 years.  This will give more than enough time in almost all cases for economic circumstances to improve or for you to re-adjust your spending requirements to correspond with the new reality.  A cash buffer cannot remove all risk, just the more immediate ones.
3) Yes the portfolio and the effect on you as an investor dependent on your investments are different facets of the equation.  However the two are inextricably linked and in practice the latter is arguably the more important.  For example having a large cash buffer could give you the confidence to invest at a higher risk/return basis than would otherwise be the case.

Medium term @5-10 years. With a withdrawal rate of say 3% that's approx 15-20% of the portfolio in cash? Even if you take it out of the defensive portion of the portfolio it's surely got to have an impact on the "average" outcome?
https://finalytiq.co.uk/cash-reserve-buffers-withdrawal-rates-old-wives-fables-retirement-portfolio/
3) By having a combination of a large pile of cash and a riskier portfolio I would suggest you are moving away from the most efficient portfolio. It goes back to my point on theoretical vs real world SWR I made a few days back. 

It's an interesting article, but too limited a set of scenarios to draw any firm conclusions.  My understanding from a little playing around with cFireSim is that nobody would pick such a high bond allocation if they were basing their split on historic data.
There's a similar article here, which is also unsatisfactory in that many of the differences reported are unlikely to be statistically significant.
https://edrempel.com/reliably-maximize-retirement-income-4-rule-safe/

All of this is however is only historic data.  So it's based on assumption that there will be no periods during retirement worse than the dataset used.  Slightly more than a century of data suddenly doesn't seem much at all when we're taking possibly 4 decades of retirement.  I'm sure a data scientist could come up with something better.

What in particular do you feel was too limiting in the scenarios? 
I'm not clear what you mean regarding high bond allocations with reference to historical scenarios.
Historical data may not be perfect, but it's all we currently have.
I'[m sure Abraham (author) would be very interested in what a data scientist could come up with considering he employs quants etc to help him build his market-leading retirement tool

They were very limited in that they looked at 50/50 and 50/40/10 splits.  A data scientist would have written him a nice script that took every possible whole integer variant.  They might have also suggested using monte carlo simulations using historic data as a basis but decreasing the mean return or increasing standard deviation by a small margin.  They'd also report whether the differences in returns were statistically significant or not.

If they were really good, they'd also build some sort of tool that took account that someone's withdrawal rate might be higher early in retirement while they're waiting for SP and DB to kick in.  Cfiresim does this, but only seems to use historic data.

Of course, this wouldn't be perfect.  We cannot know the future.  But it would be more robust IMO than assuming the future won't be worse than the past.  

Abraham will have no doubt run it for all scenarios (he is very thorough like that ) and it would be trivial to undertaken given he has the underlying data, but a "typical" person reading a huge article with all the data in would probably fall asleep. I think it gets the point across well considering the target audience.
Given the dataset (>100 years with monthly samples), I'm not sure what Monte Carlo would bring to the party. Real data is fact, Monte Carlo inputs are opinions. Note that I'm not against Monte Carlo modelling, and have written a few for "fun", but having seen first hand how assumptions can drive suboptimal outcomes, if sufficient actual data is available I would take that.
They have a tool available that works out withdrawal rate given changing expenditure requirements.
We'll agree to disagree on our vision of perfect and the imperfect solutions we currently have

The parameters for the Monte Carlo simulations can be based solely on historic data. They can also make better use of the historic data as there won't be a bias for data in the middle if it is drawn randomly rather than using the sliding window, which undersamples the tails of time series.    

I'm not clear why a sliding window undersamples the tails of time series? 

Cobson saved me the effort of explaining the sliding window, but to answer the question does it matter?

First, lets start with the assumption that monthly returns are normally distributed, which seems correct

https://www.profoliocapital.com/single-post/2017/01/31/Are-SP-500-Returns-Normally-Distributed#:~:text=Conclusion%3A%20The%20monthly%20returns%20of,closely%20follow%20a%20normal%20distribution.

Next, we take a SP 500 dataset (well sort of since there's some retrospective fudging to account for fact that index doesn't go that far back.
https://www.officialdata.org/us/stocks/s-p-500/1900

The mean and standard deviation for the entire set is 0.53% and 4.27 %.  For the central portion which is over-represented with a sliding window, it is 0.52 % and 3.41 %.  So that middle portion of the data is more stable than the entire dataset.  Professors of economics can't seem to agree whether the stock markets are random walks or not, but if they are, then we should be using all the data with equal weighting for simulations.

numpty_dumpty

I just ran the numbers through cfiresim and firecalc and got a maximum sustainable drawdown of 23600 from cfiresim and 24000 from firecalc for 100% success over 40 years. If I run firecalc so the pot never reduces below 90 k I get 22800/year, so taking 22600 a year seems a comfortable and low risk number. That would give me 1800 a month until 60 as I would avoid tax until then and then drop after 60 as I take more taxable income. 

Thrugelmir

numpty_dumpty said:

I just ran the numbers through cfiresim and firecalc and got a maximum sustainable drawdown of 23600 from cfiresim and 24000 from firecalc for 100% success over 40 years. If I run firecalc so the pot never reduces below 90 k I get 22800/year, so taking 22600 a year seems a comfortable and low risk number. That would give me 1800 a month until 60 as I would avoid tax until then and then drop after 60 as I take more taxable income. 

Invested in what though?  

michaels

numpty_dumpty said:

I just ran the numbers through cfiresim and firecalc and got a maximum sustainable drawdown of 23600 from cfiresim and 24000 from firecalc for 100% success over 40 years. If I run firecalc so the pot never reduces below 90 k I get 22800/year, so taking 22600 a year seems a comfortable and low risk number. That would give me 1800 a month until 60 as I would avoid tax until then and then drop after 60 as I take more taxable income. 

That's where I got 23k from.  Alternative approach was to assume you put the money needed to top you up to your db+sp in money market funds and accept a 1% pa loss to inflation and then cfiresm the remainder of the pot over the whole of retirement and that gave the 25k

cfw1994

michaels said:

numpty_dumpty said:

I just ran the numbers through cfiresim and firecalc and got a maximum sustainable drawdown of 23600 from cfiresim and 24000 from firecalc for 100% success over 40 years. If I run firecalc so the pot never reduces below 90 k I get 22800/year, so taking 22600 a year seems a comfortable and low risk number. That would give me 1800 a month until 60 as I would avoid tax until then and then drop after 60 as I take more taxable income. 

That's where I got 23k from.  Alternative approach was to assume you put the money needed to top you up to your db+sp in money market funds and accept a 1% pa loss to inflation and then cfiresm the remainder of the pot over the whole of retirement and that gave the 25k

I think you win first prize for getting that answer ON THE FIRST PAGE, despite the many weaving paths this thread has taken from the original question!   

Well done
For my part, it was a reassuring little ‘test’ of my spreadsheet: happy with that!

BritishInvestor

cobson said:

BritishInvestor said:

I'm not clear why a sliding window undersamples the tails of time series? 

If you have a 3 year data set and a 2 year sliding window, then you will have two data samples: years 1 and 2, and years 2 and 3. Year 2 is in both samples, but years 1 and 3 are only in one sample - so the two tail years are undersampled compared to the middle year.  The same thing happens when the numbers are scaled up.

Unless the tails contain large outliers relative to the rest of the dataset (and to be fair 1915 was pretty bad which is when my data set starts, but then so were the two world wars, Great Depression, 1970s etc), I'm struggling to see how this would have much of a real-world impact.

BritishInvestor

AnotherJoe said:

BritishInvestor said:

AnotherJoe said:

BritishInvestor said:

Linton said:

BritishInvestor said:

Linton said:

michaels said:

BritishInvestor said:

Albermarle said:

I would say the cash element is too low . Especially when you get to the point of drawing from the DC pension , you need a bigger cash buffer to get through market downturns.

Why do you "need" a bigger cash buffer? How does it improve overall outcomes?

I don't think it matters how matters how many times you point out the research on this, people still imagine you can have a nice cash pot to avoid 'drawdown' when markets are low - the same people who would claim they have no intention of ever trying to 'time the markets'....

When you need your drawdown in order to meet your basic expenses, and the market drops 40-50% what would you do?  Continue selling units but at twice the rate?

We are fortunate in that we have over a decade of historical market data to evaluate various strategies and outcomes. For example, imagine if you were an investor with a UK biased portfolio in the 1970s with the drawn out stock market slump and additional pressure being placed on the portfolio with rising interest rates. How much would you allocate to a cash buffer to cater for that?

No strategy can handle a long term fall in the markets, if that is the scenario you are proposing.  Though rising interest rates would benefit a cash buffer.  If you havent got the returns at some stage you wont be able to pay yourself the income to meet your expenses.  The best you can do is to diversify as much as possible - invest globally, invest in all sectors, invest in as many different types of asset and use as many different sources of income as you reasonably can.

The purpose of a cash buffer is to remove all risk in the medium term.  During the recent COVID mini-crash many people on the savings forum were panicking.  I was able to look on with a serene, possibly smug, smile despite my investments being essential for meeting a significant part of living expenses.

But investment strategies can only go so far, they cannot protect you from everything.  Perhaps the best they can do is to help ensure that should the world as we know it collapse one is in at least as good if not better position as everyone else.  

One detailed point - you talk about a decade of historical market data.  That means absolutely nothing.  The past decade has shown unprecedented rises in share prices and is very different to many previous decades.

Sorry, decade should be century .
I agree that you want diversify globally and across sectors but what asset classes did you have in mind other than equities and bonds?
When you say no strategy can survive a long term fall in the markets, we have historical data to suggest that strategies have done and for a given retirement date we can see exactly how each strategy played out
I don't see how the cash buffer removes all risk - how do you know in advance what inflation is likely to be? What is your definition of medium term?
Covid (thus far) was a blip and had minimal income on a robust retirement portfolio (it's the long drawn out downturns that have the biggest impact) - we must be clear to differentiate the impact on the portfolio with the impact on the investor - that's a separate discussion.

Strawman. Why does it have to (and who said?) remove all risk?

See Linton's comments above

"In the medium term. "

Yep, I was referring to the medium term.