SWR Question

Pat38493

michaels said:

Pat38493 said:

gm0 said:

There does seem to be some backtesting statistical support for the notion that when markets are highly priced (CAPE high) the demonstrably safe (for given parameters and data set) % WR (of the larger amount) should be tilted down. And conversely when markets have passed the long term trend line on the way down and overshot negatively on sentiment vs a notional fair value (in this context based on long term historic returns trend line) that a higher % of that smaller pot would be possible in anticipation of improved returns from the low base. This is one example of a "tilts" concept. 

This is potentially useful as it uses two data points you have. Start date.  CAPE now.  To help make a choice of initial %WR which you need to make in a lot of methods - to start your retirement sequence of returns and draw mix of these and capital.  Or you can ignore tilts entirely and pick a risk level or a low sub MSWR % and get on with it.   Where the balance of pot size and desired income is finely balanced (i.e. pot on the lower bound of big enough) then obviously this issue about sweating sub % details becomes more significant

I read around this subject fairly extensively on my planning journey- academic papers, ERN, several goes through McClung.  In the end I concluded that

a) some drawdown access and extraction methods are demonstrably *slightly* worse than others on a range of historic and simulated markets US and other and can be avoided on that basis. 

It is not proved that they will definitively be worse in the future.  But such evidence as does exist says A < B.  A being still not a terrible decision but since better are available.   Implication choose B.  Data beats abstract web advocacy all day long.

b) from those methods that remain after such filtering there are choices (for tested conditions) that  "work historically" or under monte-carlo or other stress market simulation without "surprises" introduced by the technique itself.  This again is helpful to sleeping easier with any given approach.

And *nobody* knows which will be "best" statistically for the specific 40 years from 2023 - 2063 for a given portfolio shape

c) A lot of this technical methods exploration is fishing for risk management gains (same income, slightly lower risk of retirement failure) or WR gains at the ~0.25% level.  Which compared to other drag factors - principally investment and fund management costs and FX is small potatoes.  Yet if an approximate target WR is ~3.5%.  0.25% is a decent proportion of the total.  0.5% would be a major win.  So perhaps worth exploring afer all.

Only after investing costs and FX factors and drag have been evaluated.  Competent tool supported IFA advice.
Or the frankly very low price of entry for the McClung book.  Whether you end up agreeing with and selecting his chosen approach or not there is confidence building value in the testing and test results presented around a range of other choices you could make.

I have found the ERN drawdown blog useful to get educated and for debunking a long list of plausible sounding things by demonstrating that they don't work in a historic series (or make little practical difference).  This is helpful. Things that demonstrably already *haven't worked* is good guidance information.  Big Six is a worse bet than Line and full odds on the craps table (due to extra drag from house edge over probability).   Similarly I could pick an arbitrary 5% WR and it "might" work - for given investments and start date.  On a range of paths it could.  Anecdotally.  And on another set of investments and start dates - depletion during the journey past working age beckons.

Occasionally a small demonstrable statistical gain is on offer if you can be bothered with the method complexity. And if you believe (this requires faith without proof - that the future will sufficiently resemble the past that squeezing the balloon into that particular shape will *again* be more optimised). 

Useful - provided you always check ERN assumptions are similar enough to those of interest to you - duration, market used, other method parameters etc.  Remembering that any evidential value/mathematical validity of what he is saying is quite strictly limited to that exact scenario as tested.  Look at what he says about Guyton-Klinger and then at what advocates of that say - and compare.  Both of these things (specific tests and analysis) can be "true" at the same time.  For the tests as run and assumptions as given.  I don't especially rate ERN's test of GK as a proper test of what they proposed.  It is GKish with shortcuts. 
The argument thus becomes about assumptions, implementation and parameters.  Which is itself a neat illustration that you can't stretch evidence much or change methods and parameters to taste without finding or running exactly equivalent tests if that validation is important to you.

Where other commentators often disagree around CAPE is around any predictive element.  Beyond the general past markets and backtesting don't predict the future.  The fact CAPE is high doesn't tell you *when* the "long term average line crossing event will come.  History suggests that it will come again as so often before.  Rough looking sinewave across along a term trend line.

There is a section on tilts and CAPE as example tilt in McClung (Living off your Money).  As I recall he quite likes the idea and post testing cautiously offers it as an optional extra tuning approach.

Thanks for your comments.

In the meantime I exchanged a few comments with ERN on the comments section of his blog.

He acknowledged that his spreadsheet model which I think is a short cut to some extent of what he does behind the scenes, can produce odd results if you put in certain inputs - for example if you say that you want to retire today with £1 in your pocket, on the basis that an eccentric deceased relative has bequeathed you x million pounds in 30 years from now, the sheet reports that you can already start spending this money.

More realistically - the sheet doesn't throw any kind of alarm if you are trying to bridge until a DB pension kicks in with very front heavy withdrawals - if you look at the bridging period discretely you will see failures but if you look at the whole sheet everything looks ok.  The comment was that you have to use common sense when using the sheet.  

Further, I suspect that he is somehow amalgamating the present value of future cash flows with the historical model, thereby treating is with the same cumulated growth pattern as your general portfolio which is not totally accurate as it should be only treated with inflation.

However the model seems to give results in the same ballpark as other tools as long as you are within the ball park that the creator envisaged.

My approach generally is to use several different tools and sheets and compare the results to see how close they are too each other and what is causing the differences - if all of them say that my plans are looking good, then I would feel pretty confident.

Really useful.

NPVing future index linked cashflows then assuming the value increases going forward with growth of the portfolio rather than inflation sounds like a big overstatement - eg that index linked state pension payment of £800pm (in current pounds) in 40 years time might be treated as £3800 (in current terms) if it is assumed to grow by 4% pa in real terms....

So having nearly broken my brain looking at ERN's sheet again last night, I think I probably have to take back my assertion that he is wrongly inflating the supplemental cash flow by the overall portfolio return rates, at least in a sense.

His model works by calculating how much your portfolio would have grown if left fully invested for the entire retirement horizon, and then calculating the opportunity cost of each dollar of spend that you have over that period.  The latter is independent from the portfolio size so you can use it to calculate percentages based on portfolio size.

The bit I couldn't wrap my head around is why he is applying the portfolio backward return growth rates to the incoming cash flows by period.  I think I understand now that this is an "opportunity gain" - what he is saying is that each dollar of income that you received in that period, gives you the opportunity to leave a dollar invested that otherwise you needed to withdraw.

As such, using his methodology, it's correct (and quite clever).  The idea behind it I guess is that the whole thing can be done on a relatively simple spreasheet approach (depending of course how you define simple) and you don't have to iterate through every retirement period one by one.

However the issue in the end comes back to my original concern - the above maths assumes that you have dollars left in your portfolio to opportunistically leave invested.

If you have a very forward loaded withdrawal strategy (e.g. bridging 10 years until a hefty DB scheme kicks in), this model will not pick up failures that could happen before the 10 years are up.  Therefore if using this model you need to also test your bridging strategy as a discrete event.

NedS

Pat38493 said:

If you have a very forward loaded withdrawal strategy (e.g. bridging 10 years until a hefty DB scheme kicks in), this model will not pick up failures that could happen before the 10 years are up.  Therefore if using this model you need to also test your bridging strategy as a discrete event.

I think most who are doing this would probably consider the pot required to bridge a 10 year period as a separate bucket, and invest in fairly safe investments given the relatively short time horizon. I would be looking at a bond ladder for the first 5 years, and maybe a wealth preservation fund for years 5-10, gradually moving money into the bond ladder on an annual basis.
Otherwise, to require very large front-loaded withdraws from a conventional total return drawdown strategy adds additional difficult to manage risks.

SteveBLFC64

Following on from this discussion, I'd like to know if anyone uses any of the simpler online drawdown/SWR calculators, eg. https://www.2020financial.co.uk/pension-drawdown-calculator/#calculator, and, if so, do the much more detailed tools/spreadsheets (that have been discussed in this thread) give a vastly different rate? Also, if anyone does use the one I mentioned above, what percentage do they look for in the Simulations Ending Above Zero value? Obviously the higher the percentage, the better in some respects, however, having a very high value probably means you're probably having a lesser SWR just to ensure the pot doesn't run out (I realise it's a 'balancing act', and it's all down to risk appetite, etc). I do presume most people use a variety of tools/methods, just for sanity testing, and don't rely on just one, unless they're reasonably confident in the methodology they're following.

michaels

Pat38493 said:

michaels said:

Pat38493 said:

gm0 said:

There does seem to be some backtesting statistical support for the notion that when markets are highly priced (CAPE high) the demonstrably safe (for given parameters and data set) % WR (of the larger amount) should be tilted down. And conversely when markets have passed the long term trend line on the way down and overshot negatively on sentiment vs a notional fair value (in this context based on long term historic returns trend line) that a higher % of that smaller pot would be possible in anticipation of improved returns from the low base. This is one example of a "tilts" concept. 

This is potentially useful as it uses two data points you have. Start date.  CAPE now.  To help make a choice of initial %WR which you need to make in a lot of methods - to start your retirement sequence of returns and draw mix of these and capital.  Or you can ignore tilts entirely and pick a risk level or a low sub MSWR % and get on with it.   Where the balance of pot size and desired income is finely balanced (i.e. pot on the lower bound of big enough) then obviously this issue about sweating sub % details becomes more significant

I read around this subject fairly extensively on my planning journey- academic papers, ERN, several goes through McClung.  In the end I concluded that

a) some drawdown access and extraction methods are demonstrably *slightly* worse than others on a range of historic and simulated markets US and other and can be avoided on that basis. 

It is not proved that they will definitively be worse in the future.  But such evidence as does exist says A < B.  A being still not a terrible decision but since better are available.   Implication choose B.  Data beats abstract web advocacy all day long.

b) from those methods that remain after such filtering there are choices (for tested conditions) that  "work historically" or under monte-carlo or other stress market simulation without "surprises" introduced by the technique itself.  This again is helpful to sleeping easier with any given approach.

And *nobody* knows which will be "best" statistically for the specific 40 years from 2023 - 2063 for a given portfolio shape

c) A lot of this technical methods exploration is fishing for risk management gains (same income, slightly lower risk of retirement failure) or WR gains at the ~0.25% level.  Which compared to other drag factors - principally investment and fund management costs and FX is small potatoes.  Yet if an approximate target WR is ~3.5%.  0.25% is a decent proportion of the total.  0.5% would be a major win.  So perhaps worth exploring afer all.

Only after investing costs and FX factors and drag have been evaluated.  Competent tool supported IFA advice.
Or the frankly very low price of entry for the McClung book.  Whether you end up agreeing with and selecting his chosen approach or not there is confidence building value in the testing and test results presented around a range of other choices you could make.

I have found the ERN drawdown blog useful to get educated and for debunking a long list of plausible sounding things by demonstrating that they don't work in a historic series (or make little practical difference).  This is helpful. Things that demonstrably already *haven't worked* is good guidance information.  Big Six is a worse bet than Line and full odds on the craps table (due to extra drag from house edge over probability).   Similarly I could pick an arbitrary 5% WR and it "might" work - for given investments and start date.  On a range of paths it could.  Anecdotally.  And on another set of investments and start dates - depletion during the journey past working age beckons.

Occasionally a small demonstrable statistical gain is on offer if you can be bothered with the method complexity. And if you believe (this requires faith without proof - that the future will sufficiently resemble the past that squeezing the balloon into that particular shape will *again* be more optimised). 

Useful - provided you always check ERN assumptions are similar enough to those of interest to you - duration, market used, other method parameters etc.  Remembering that any evidential value/mathematical validity of what he is saying is quite strictly limited to that exact scenario as tested.  Look at what he says about Guyton-Klinger and then at what advocates of that say - and compare.  Both of these things (specific tests and analysis) can be "true" at the same time.  For the tests as run and assumptions as given.  I don't especially rate ERN's test of GK as a proper test of what they proposed.  It is GKish with shortcuts. 
The argument thus becomes about assumptions, implementation and parameters.  Which is itself a neat illustration that you can't stretch evidence much or change methods and parameters to taste without finding or running exactly equivalent tests if that validation is important to you.

Where other commentators often disagree around CAPE is around any predictive element.  Beyond the general past markets and backtesting don't predict the future.  The fact CAPE is high doesn't tell you *when* the "long term average line crossing event will come.  History suggests that it will come again as so often before.  Rough looking sinewave across along a term trend line.

There is a section on tilts and CAPE as example tilt in McClung (Living off your Money).  As I recall he quite likes the idea and post testing cautiously offers it as an optional extra tuning approach.

Thanks for your comments.

In the meantime I exchanged a few comments with ERN on the comments section of his blog.

He acknowledged that his spreadsheet model which I think is a short cut to some extent of what he does behind the scenes, can produce odd results if you put in certain inputs - for example if you say that you want to retire today with £1 in your pocket, on the basis that an eccentric deceased relative has bequeathed you x million pounds in 30 years from now, the sheet reports that you can already start spending this money.

More realistically - the sheet doesn't throw any kind of alarm if you are trying to bridge until a DB pension kicks in with very front heavy withdrawals - if you look at the bridging period discretely you will see failures but if you look at the whole sheet everything looks ok.  The comment was that you have to use common sense when using the sheet.  

Further, I suspect that he is somehow amalgamating the present value of future cash flows with the historical model, thereby treating is with the same cumulated growth pattern as your general portfolio which is not totally accurate as it should be only treated with inflation.

However the model seems to give results in the same ballpark as other tools as long as you are within the ball park that the creator envisaged.

My approach generally is to use several different tools and sheets and compare the results to see how close they are too each other and what is causing the differences - if all of them say that my plans are looking good, then I would feel pretty confident.

Really useful.

NPVing future index linked cashflows then assuming the value increases going forward with growth of the portfolio rather than inflation sounds like a big overstatement - eg that index linked state pension payment of £800pm (in current pounds) in 40 years time might be treated as £3800 (in current terms) if it is assumed to grow by 4% pa in real terms....

So having nearly broken my brain looking at ERN's sheet again last night, I think I probably have to take back my assertion that he is wrongly inflating the supplemental cash flow by the overall portfolio return rates, at least in a sense.

His model works by calculating how much your portfolio would have grown if left fully invested for the entire retirement horizon, and then calculating the opportunity cost of each dollar of spend that you have over that period.  The latter is independent from the portfolio size so you can use it to calculate percentages based on portfolio size.

The bit I couldn't wrap my head around is why he is applying the portfolio backward return growth rates to the incoming cash flows by period.  I think I understand now that this is an "opportunity gain" - what he is saying is that each dollar of income that you received in that period, gives you the opportunity to leave a dollar invested that otherwise you needed to withdraw.

As such, using his methodology, it's correct (and quite clever).  The idea behind it I guess is that the whole thing can be done on a relatively simple spreasheet approach (depending of course how you define simple) and you don't have to iterate through every retirement period one by one.

However the issue in the end comes back to my original concern - the above maths assumes that you have dollars left in your portfolio to opportunistically leave invested.

If you have a very forward loaded withdrawal strategy (e.g. bridging 10 years until a hefty DB scheme kicks in), this model will not pick up failures that could happen before the 10 years are up.  Therefore if using this model you need to also test your bridging strategy as a discrete event.

Thanks for this - I did try putting in £1 pot and a set of future real flows and it said my safe spend was indeed the future flows rather than any other values but it is very useful that you have worked out how it works and the limitation regarding bridging periods (which is a big part of my strategy.

I am still torn between taking my DB very early with actuarial reduction as it gives certainty and also allows me to lock in 100% benefit for DW for a further 10% reduction (compared to the standard 35% until it is taken) with the suggestion that I want to maximise fully indexed DB by taking it as late as possible to reduce SWR risk and financial complexity later in life.

Pat38493

SteveBLFC64 said:

Following on from this discussion, I'd like to know if anyone uses any of the simpler online drawdown/SWR calculators, eg. https://www.2020financial.co.uk/pension-drawdown-calculator/#calculator, and, if so, do the much more detailed tools/spreadsheets (that have been discussed in this thread) give a vastly different rate? Also, if anyone does use the one I mentioned above, what percentage do they look for in the Simulations Ending Above Zero value? Obviously the higher the percentage, the better in some respects, however, having a very high value probably means you're probably having a lesser SWR just to ensure the pot doesn't run out (I realise it's a 'balancing act', and it's all down to risk appetite, etc). I do presume most people use a variety of tools/methods, just for sanity testing, and don't rely on just one, unless they're reasonably confident in the methodology they're following.

I think it gives relatively similar results but this one that you have linked seems to only allow you to put in one future pension income so you cannot model your state pension and a separate DB pension unless you are taking them both at the same time.

Most of the tools I've tried gives results that are in roughly the same ballpark but some seem to recommend starting with a kind of "median" withdrawal but then adjusting your withdrawals each year so those gives you a higher starting withdrawal with the risk that you might have to reduce it.

Ideally I would shoot for 100% but obviously that means there's a very high chance you'll end up with "too much" money at the end.  Some of the IFA videos I've watched said that many clients are comfortable with something above 90% in these type of simulations.

Pat38493

michaels said:

Pat38493 said:

michaels said:

Pat38493 said:

gm0 said:

There does seem to be some backtesting statistical support for the notion that when markets are highly priced (CAPE high) the demonstrably safe (for given parameters and data set) % WR (of the larger amount) should be tilted down. And conversely when markets have passed the long term trend line on the way down and overshot negatively on sentiment vs a notional fair value (in this context based on long term historic returns trend line) that a higher % of that smaller pot would be possible in anticipation of improved returns from the low base. This is one example of a "tilts" concept. 

This is potentially useful as it uses two data points you have. Start date.  CAPE now.  To help make a choice of initial %WR which you need to make in a lot of methods - to start your retirement sequence of returns and draw mix of these and capital.  Or you can ignore tilts entirely and pick a risk level or a low sub MSWR % and get on with it.   Where the balance of pot size and desired income is finely balanced (i.e. pot on the lower bound of big enough) then obviously this issue about sweating sub % details becomes more significant

I read around this subject fairly extensively on my planning journey- academic papers, ERN, several goes through McClung.  In the end I concluded that

a) some drawdown access and extraction methods are demonstrably *slightly* worse than others on a range of historic and simulated markets US and other and can be avoided on that basis. 

It is not proved that they will definitively be worse in the future.  But such evidence as does exist says A < B.  A being still not a terrible decision but since better are available.   Implication choose B.  Data beats abstract web advocacy all day long.

b) from those methods that remain after such filtering there are choices (for tested conditions) that  "work historically" or under monte-carlo or other stress market simulation without "surprises" introduced by the technique itself.  This again is helpful to sleeping easier with any given approach.

And *nobody* knows which will be "best" statistically for the specific 40 years from 2023 - 2063 for a given portfolio shape

c) A lot of this technical methods exploration is fishing for risk management gains (same income, slightly lower risk of retirement failure) or WR gains at the ~0.25% level.  Which compared to other drag factors - principally investment and fund management costs and FX is small potatoes.  Yet if an approximate target WR is ~3.5%.  0.25% is a decent proportion of the total.  0.5% would be a major win.  So perhaps worth exploring afer all.

Only after investing costs and FX factors and drag have been evaluated.  Competent tool supported IFA advice.
Or the frankly very low price of entry for the McClung book.  Whether you end up agreeing with and selecting his chosen approach or not there is confidence building value in the testing and test results presented around a range of other choices you could make.

I have found the ERN drawdown blog useful to get educated and for debunking a long list of plausible sounding things by demonstrating that they don't work in a historic series (or make little practical difference).  This is helpful. Things that demonstrably already *haven't worked* is good guidance information.  Big Six is a worse bet than Line and full odds on the craps table (due to extra drag from house edge over probability).   Similarly I could pick an arbitrary 5% WR and it "might" work - for given investments and start date.  On a range of paths it could.  Anecdotally.  And on another set of investments and start dates - depletion during the journey past working age beckons.

Occasionally a small demonstrable statistical gain is on offer if you can be bothered with the method complexity. And if you believe (this requires faith without proof - that the future will sufficiently resemble the past that squeezing the balloon into that particular shape will *again* be more optimised). 

Useful - provided you always check ERN assumptions are similar enough to those of interest to you - duration, market used, other method parameters etc.  Remembering that any evidential value/mathematical validity of what he is saying is quite strictly limited to that exact scenario as tested.  Look at what he says about Guyton-Klinger and then at what advocates of that say - and compare.  Both of these things (specific tests and analysis) can be "true" at the same time.  For the tests as run and assumptions as given.  I don't especially rate ERN's test of GK as a proper test of what they proposed.  It is GKish with shortcuts. 
The argument thus becomes about assumptions, implementation and parameters.  Which is itself a neat illustration that you can't stretch evidence much or change methods and parameters to taste without finding or running exactly equivalent tests if that validation is important to you.

Where other commentators often disagree around CAPE is around any predictive element.  Beyond the general past markets and backtesting don't predict the future.  The fact CAPE is high doesn't tell you *when* the "long term average line crossing event will come.  History suggests that it will come again as so often before.  Rough looking sinewave across along a term trend line.

There is a section on tilts and CAPE as example tilt in McClung (Living off your Money).  As I recall he quite likes the idea and post testing cautiously offers it as an optional extra tuning approach.

Thanks for your comments.

In the meantime I exchanged a few comments with ERN on the comments section of his blog.

He acknowledged that his spreadsheet model which I think is a short cut to some extent of what he does behind the scenes, can produce odd results if you put in certain inputs - for example if you say that you want to retire today with £1 in your pocket, on the basis that an eccentric deceased relative has bequeathed you x million pounds in 30 years from now, the sheet reports that you can already start spending this money.

More realistically - the sheet doesn't throw any kind of alarm if you are trying to bridge until a DB pension kicks in with very front heavy withdrawals - if you look at the bridging period discretely you will see failures but if you look at the whole sheet everything looks ok.  The comment was that you have to use common sense when using the sheet.  

Further, I suspect that he is somehow amalgamating the present value of future cash flows with the historical model, thereby treating is with the same cumulated growth pattern as your general portfolio which is not totally accurate as it should be only treated with inflation.

However the model seems to give results in the same ballpark as other tools as long as you are within the ball park that the creator envisaged.

My approach generally is to use several different tools and sheets and compare the results to see how close they are too each other and what is causing the differences - if all of them say that my plans are looking good, then I would feel pretty confident.

Really useful.

NPVing future index linked cashflows then assuming the value increases going forward with growth of the portfolio rather than inflation sounds like a big overstatement - eg that index linked state pension payment of £800pm (in current pounds) in 40 years time might be treated as £3800 (in current terms) if it is assumed to grow by 4% pa in real terms....

So having nearly broken my brain looking at ERN's sheet again last night, I think I probably have to take back my assertion that he is wrongly inflating the supplemental cash flow by the overall portfolio return rates, at least in a sense.

His model works by calculating how much your portfolio would have grown if left fully invested for the entire retirement horizon, and then calculating the opportunity cost of each dollar of spend that you have over that period.  The latter is independent from the portfolio size so you can use it to calculate percentages based on portfolio size.

The bit I couldn't wrap my head around is why he is applying the portfolio backward return growth rates to the incoming cash flows by period.  I think I understand now that this is an "opportunity gain" - what he is saying is that each dollar of income that you received in that period, gives you the opportunity to leave a dollar invested that otherwise you needed to withdraw.

As such, using his methodology, it's correct (and quite clever).  The idea behind it I guess is that the whole thing can be done on a relatively simple spreasheet approach (depending of course how you define simple) and you don't have to iterate through every retirement period one by one.

However the issue in the end comes back to my original concern - the above maths assumes that you have dollars left in your portfolio to opportunistically leave invested.

If you have a very forward loaded withdrawal strategy (e.g. bridging 10 years until a hefty DB scheme kicks in), this model will not pick up failures that could happen before the 10 years are up.  Therefore if using this model you need to also test your bridging strategy as a discrete event.

Thanks for this - I did try putting in £1 pot and a set of future real flows and it said my safe spend was indeed the future flows rather than any other values but it is very useful that you have worked out how it works and the limitation regarding bridging periods (which is a big part of my strategy.

I am still torn between taking my DB very early with actuarial reduction as it gives certainty and also allows me to lock in 100% benefit for DW for a further 10% reduction (compared to the standard 35% until it is taken) with the suggestion that I want to maximise fully indexed DB by taking it as late as possible to reduce SWR risk and financial complexity later in life.

Yes me too - most of my current models assume that I will take the DB early as soon as I retire, but in reality I will monitor the ongoing situation as you can see whether something that corresponds to a previous worst case scenario seems to be evolving - for example if I look at the worst case years on some of the simulation tools I've used, you can see that the portfolio value takes a huge hit either just before retirement, or just after retirement, or both.  As time goes by you can see whether this is really happening.  Therefore I probably wouldn't trigger the DB early until I think there is an SOR scenario developing.