One for the experts( you know who you are!)

Audaxer

jamesd said:

Amateurretiree said:

...
We are quite risk averse, currently in Defensive funds( but pretty much back where we were pre COVID.
DH retired last year at 60, gets SP at 66, I’m 58, SP at 57.
My DB pretty much covers essentials but we like long haul holidays normally a couple of times a year which can eat up a fair bit, but we want to do it while we can.

State pension deferral increases the pension by 5.8% per year deferred. Assuming DH has normal life expectancy deferring for ten years seems sensible so set aside another ten years worth to fund this. For a £9,000 initial pension this would get an extra £5,220 a year of guaranteed income with uncapped inflation increases, and actually more because that ignores the triple lock increases while deferring.

As deferring the State Pension for 10 years would make him 76 years old before claiming it, I'm just wondering how many years it would take the increased SP to recover the 10 years of missed initial pension payments?

BritishInvestor

Thrugelmir said:

BritishInvestor said:

kinger101 said:

AnotherJoe said:

kinger101 said:

HarryGray said:

Well, technically the most optimal asset allocation at a 4% withdrawal rate is 100% equity. That has the highest success rate out of any asset allocation throughout retirement. Obviously sequencing risk is a big risk, so long as take no large withdrawals you would be pretty much set. 

You do NOT want to be in a cautious asset allocation throughout retirement. 

Presumably you also think there are no black swans.

DoNt see why you'd think that since he did point out there were 3 years when it underperformed. But all you can do is go with the odds. 

Black swans aren't period of poor performance that have happened historically.  They are events that have not previously been observed outside previous range. Historic moving-window doesn't factor this in.  And undersamples much of the available data anyway.

The are different ways of using historic data, and many of these show increased SOR risk with 100% equities

I still don't understand the undersampling. If you have circa 1200 samples (100 years with 12 monthly samples) you may miss out on some of the impact of WW1, but given sequencing risks makes itself felt most in the early years of retirement, 1929 should be covered as would WW2 and the 1970s shocker.
I'm also not sure how you could use historical data in a superior way (given the limitation of MC as previously discussed).
Agreed on future events being outside historical ranges, and that must be taken into account along with the impact of fees and investor behaviour.

First time that auditors were required to sign off Annual Annual accounts was 1933 and that was in the USA. Up until then there was no standard accounting policies anywhere in the world. Companies could report whatever figures they wished. Global accounting standards are a relatively new concept that still have weaknessess. Reading too much into global consolidated naked data isn't advisable. 

Given that we are looking at global data I would be very surprised if sufficient companies fudged the books for long enough, without getting caught out by periodic recessions/depressions, to make a significant difference to historical market returns. Happy to reconsider if you have any evidence.

BritishInvestor

MarkCarnage said:

BritishInvestor said:

MarkCarnage said:

COVID is perhaps a black swan event, of as yet unknown impact. If you went back far enough, the Black Death would be one too.....there are times when you have to take a view as to what you could do in very extreme circumstances anyway. Two long haul holidays a year would be rather far down any priority list I suspect. 
 I think that there will be an inflationary impact arising from this, or the subsequent consequences of it. After all, there is a lot of debt needing to be devalued....hence my reservations about high exposures to cash and nominal bonds.

I don't see COVID as a black swan event because we have had similar events happening previously, as you mentioned the Black Death being one.

Yes, but we haven't had a pandemic in living memory, let alone one like the Black Death. It was also almost universally not forecast - and then underestimated (probably due to Chinese suppression of facts) so I think it was quite a Black Swan. Everyone was looking for problems in financial markets and/or global politics, not in this direction.  Therefore I do see it as a Black Swan, or maybe a cygnet.

Wasn't H1N1 considered a pandemic?
https://www.cdc.gov/flu/pandemic-resources/2009-h1n1-pandemic.html
"Additionally, CDC estimated that 151,700-575,400 people worldwide died from (H1N1)pdm09 virus infection during the first year the virus circulated."

bigadaj

BritishInvestor said:

MarkCarnage said:

COVID is perhaps a black swan event, of as yet unknown impact. If you went back far enough, the Black Death would be one too.....there are times when you have to take a view as to what you could do in very extreme circumstances anyway. Two long haul holidays a year would be rather far down any priority list I suspect. 
 I think that there will be an inflationary impact arising from this, or the subsequent consequences of it. After all, there is a lot of debt needing to be devalued....hence my reservations about high exposures to cash and nominal bonds.

I don't see COVID as a black swan event because we have had similar events happening previously, as you mentioned the Black Death being one.

How is your modelling looking when backtested to between 1290 and 1665?

BritishInvestor

kinger101 said:

BritishInvestor said:

kinger101 said:

AnotherJoe said:

kinger101 said:

HarryGray said:

Well, technically the most optimal asset allocation at a 4% withdrawal rate is 100% equity. That has the highest success rate out of any asset allocation throughout retirement. Obviously sequencing risk is a big risk, so long as take no large withdrawals you would be pretty much set. 

You do NOT want to be in a cautious asset allocation throughout retirement. 

Presumably you also think there are no black swans.

DoNt see why you'd think that since he did point out there were 3 years when it underperformed. But all you can do is go with the odds. 

Black swans aren't period of poor performance that have happened historically.  They are events that have not previously been observed outside previous range. Historic moving-window doesn't factor this in.  And undersamples much of the available data anyway.

The are different ways of using historic data, and many of these show increased SOR risk with 100% equities

I still don't understand the undersampling. If you have circa 1200 samples (100 years with 12 monthly samples) you may miss out on some of the impact of WW1, but given sequencing risks makes itself felt most in the early years of retirement, 1929 should be covered as would WW2 and the 1970s shocker.
I'm also not sure how you could use historical data in a superior way (given the limitation of MC as previously discussed).
Agreed on future events being outside historical ranges, and that must be taken into account along with the impact of fees and investor behaviour.

Do you at least acknowledge that some data points are used more than others with moving window?  It's quite considerable given people are planning for 3+ decades of retirement.  That's at least 360 of the 1200 datapoints.  Including 2000 and 2008 crashes.

I don't understand what you think is so difficult about MC.  It's as hard as you want to make it.  On one level, we can just take pull random data based on monthly mean and standard deviation.  On another, you can randomly sample with replacement from old data.  

Agreed on some data points are used more than others. Regarding the number of datapoints and impact, it goes back to my point around the first decade of retirement being the most critical, and we've already passed that for 2000 and 2008.
In addition to looking at the % success rate, you can also look at the worst-case outcome (the earliest age at which the money ran out) - I use both and the latter obviously isn't impacted by the moving window.

My reservations around MC are as follows (and as I've previously mentioned, I've written them to price options and CDOs and it's something I used to enjoy!) :
1. You need to be very careful around assumptions and sensitivity of outputs to changes in inputs, and I've seen what happens when people place too much faith in models.
https://thewest.com.au/business/economy/gfc-10-years-on-a-wonky-finance-edifice-ng-b88942586z
2. If we are saying the underlying historical data is of limited value, I'm not sure what slicing and dicing/using the data in a different way would add to that. 
3. Allowing human subjectivity in the decision making process around inputs has the potential for suboptimal outcomes, IMO
4. How do you model mean reversion in MC, which (global) markets have demonstrated up to now?
5. What faith would you have in the output over what historical data has shown us?

Irrespective of which approach you use, it's obviously worth emphasising that we are dealing with uncertainty, so even though a plan shows 100% historical success rate, that's not to say we won't have to make adjustments along the way. Most people are comfortable with that. 

kinger101

BritishInvestor said:

kinger101 said:

BritishInvestor said:

kinger101 said:

AnotherJoe said:

kinger101 said:

HarryGray said:

Well, technically the most optimal asset allocation at a 4% withdrawal rate is 100% equity. That has the highest success rate out of any asset allocation throughout retirement. Obviously sequencing risk is a big risk, so long as take no large withdrawals you would be pretty much set. 

You do NOT want to be in a cautious asset allocation throughout retirement. 

Presumably you also think there are no black swans.

DoNt see why you'd think that since he did point out there were 3 years when it underperformed. But all you can do is go with the odds. 

Black swans aren't period of poor performance that have happened historically.  They are events that have not previously been observed outside previous range. Historic moving-window doesn't factor this in.  And undersamples much of the available data anyway.

The are different ways of using historic data, and many of these show increased SOR risk with 100% equities

I still don't understand the undersampling. If you have circa 1200 samples (100 years with 12 monthly samples) you may miss out on some of the impact of WW1, but given sequencing risks makes itself felt most in the early years of retirement, 1929 should be covered as would WW2 and the 1970s shocker.
I'm also not sure how you could use historical data in a superior way (given the limitation of MC as previously discussed).
Agreed on future events being outside historical ranges, and that must be taken into account along with the impact of fees and investor behaviour.

Do you at least acknowledge that some data points are used more than others with moving window?  It's quite considerable given people are planning for 3+ decades of retirement.  That's at least 360 of the 1200 datapoints.  Including 2000 and 2008 crashes.

I don't understand what you think is so difficult about MC.  It's as hard as you want to make it.  On one level, we can just take pull random data based on monthly mean and standard deviation.  On another, you can randomly sample with replacement from old data.  

Agreed on some data points are used more than others. Regarding the number of datapoints and impact, it goes back to my point around the first decade of retirement being the most critical, and we've already passed that for 2000 and 2008.
In addition to looking at the % success rate, you can also look at the worst-case outcome (the earliest age at which the money ran out) - I use both and the latter obviously isn't impacted by the moving window.

While SOR is indeed most likely to cause failure early in retirement, it is also possible for it to fail later on. 

My reservations around MC are as follows (and as I've previously mentioned, I've written them to price options and CDOs and it's something I used to enjoy!) :

1. You need to be very careful around assumptions and sensitivity of outputs to changes in inputs, and I've seen what happens when people place too much faith in models.
https://thewest.com.au/business/economy/gfc-10-years-on-a-wonky-finance-edifice-ng-b88942586z

The two methods I suggested for MC are only using unadulterated historic data.  The first was drawing at random based on a normal distribution using the mean and SD.  The second was drawing at random with replacement from historic data.  If you think these are incorrect assumptions, then historic window has already failed.  Once can always add their own assumptions to add an element of conservatism.  The article you're posted is a straw man.  

2. If we are saying the underlying historical data is of limited value, I'm not sure what slicing and dicing/using the data in a different way would add to that. 
We both accept it has limitations.  But why wouldn't you want to use it all equally weighted if you think it has any merit?  

3. Allowing human subjectivity in the decision making process around inputs has the potential for suboptimal outcomes, IMO
Like I said, at the very basic level, one can use actual data.  

4. How do you model mean reversion in MC, which (global) markets have demonstrated up to now?
A fundamental misunderstanding of reversion to the mean.  RTM doesn't mean if I shake hand's with a man who is 6' 5", the next man I shake hands with will be 5' 1" to even things out.  It just means there's a greater probability that his height will be closer to the national average of 5' 9".  

MC models RTM perfectly well as data is drawn based on the actual distribution.  The further a draw deviates the the mean, the higher probability the next one will have a lower deviation.

5. What faith would you have in the output over what historical data has shown us?
I'd have more faith that the moving window model as the data would tend toward suggesting a more conservative withdrawal rate.  And could also pull more dire SORs.

Irrespective of which approach you use, it's obviously worth emphasising that we are dealing with uncertainty, so even though a plan shows 100% historical success rate, that's not to say we won't have to make adjustments along the way. Most people are comfortable with that.

My original reply related to someone saying 100% equities was the safest approach because history says so.  It seems barely anyone on this thread is convinced of that as most people plan on having other assets, whether it's bonds or a cash buffer.  It would therefore seem most people have made their own decision of what they think of the moving window model, and decided it's foolhardy.  Making somewhat subjective decisions about how best to deal with SOR risk.  

It's difficult to see why you are so dismissive of a more conservative model making better use of data.

MarkCarnage

BritishInvestor said:

MarkCarnage said:

BritishInvestor said:

MarkCarnage said:

COVID is perhaps a black swan event, of as yet unknown impact. If you went back far enough, the Black Death would be one too.....there are times when you have to take a view as to what you could do in very extreme circumstances anyway. Two long haul holidays a year would be rather far down any priority list I suspect. 
 I think that there will be an inflationary impact arising from this, or the subsequent consequences of it. After all, there is a lot of debt needing to be devalued....hence my reservations about high exposures to cash and nominal bonds.

I don't see COVID as a black swan event because we have had similar events happening previously, as you mentioned the Black Death being one.

Yes, but we haven't had a pandemic in living memory, let alone one like the Black Death. It was also almost universally not forecast - and then underestimated (probably due to Chinese suppression of facts) so I think it was quite a Black Swan. Everyone was looking for problems in financial markets and/or global politics, not in this direction.  Therefore I do see it as a Black Swan, or maybe a cygnet.

Wasn't H1N1 considered a pandemic?
https://www.cdc.gov/flu/pandemic-resources/2009-h1n1-pandemic.html
"Additionally, CDC estimated that 151,700-575,400 people worldwide died from (H1N1)pdm09 virus infection during the first year the virus circulated."

In a technical sense it was, but it did not have the widespread and lasting effects globally of COVID. Bear in mind that I referred to COVID as a black cygnet too......it may or may not become a swan.

Mistermeaner

BritishInvestor said:

Mistermeaner said:

Audaxer said:

HarryGray said:

BritishInvestor said:

HarryGray said:

Well, technically the most optimal asset allocation at a 4% withdrawal rate is 100% equity. 

I'm not sure it's as clear cut as that.

Nothing is clear cut, simply stating that statistically the best retirement drawdown ignoring a clients Attitude to Risk is 100% equity at a constant withdrawal rate. There is no evidence anywhere to suggest that 'lifestyling' works 

I would have thought that if you have a portfolio of 100% equity and there is a poor sequence of returns, especially in the first decade of retirement, there would be more risk of running out of money than with a balanced portfolio?

There was a great thread somewhere recently (that I can't find) that covered this in some detail referencing various studies 

The somewhat surprising conclusion was that you are generally better off 100% in equities - or at least no worse off 

There has only been something like 2 or 3 starting years of retirement in the last 100+ years when this would have depleted your pot in 30years

Would be good to see the studies, and to be clear whether success means least (percentage of) historical failures or how long the money lasted for in the worst historical outcome.
It's also worth bearing in mind that these studies might not have included real-world issues such as various fees and investor (mis)behaviour.

Drawdown: safe withdrawal rates

jamesd

Audaxer said:

As deferring the State Pension for 10 years would make him 76 years old before claiming it, I'm just wondering how many years it would take the increased SP to recover the 10 years of missed initial pension payments?

It's about meeting their low risk objective with guaranteed income and effective longevity protection. If solely judged on average life expectancy it's too long to break even. But derisking income has value and might let them accept more volatility in their investments.

BritishInvestor

kinger101 said:

BritishInvestor said:

kinger101 said:

BritishInvestor said:

kinger101 said:

AnotherJoe said:

kinger101 said:

HarryGray said:

Well, technically the most optimal asset allocation at a 4% withdrawal rate is 100% equity. That has the highest success rate out of any asset allocation throughout retirement. Obviously sequencing risk is a big risk, so long as take no large withdrawals you would be pretty much set. 

You do NOT want to be in a cautious asset allocation throughout retirement. 

Presumably you also think there are no black swans.

DoNt see why you'd think that since he did point out there were 3 years when it underperformed. But all you can do is go with the odds. 

Black swans aren't period of poor performance that have happened historically.  They are events that have not previously been observed outside previous range. Historic moving-window doesn't factor this in.  And undersamples much of the available data anyway.

The are different ways of using historic data, and many of these show increased SOR risk with 100% equities

I still don't understand the undersampling. If you have circa 1200 samples (100 years with 12 monthly samples) you may miss out on some of the impact of WW1, but given sequencing risks makes itself felt most in the early years of retirement, 1929 should be covered as would WW2 and the 1970s shocker.
I'm also not sure how you could use historical data in a superior way (given the limitation of MC as previously discussed).
Agreed on future events being outside historical ranges, and that must be taken into account along with the impact of fees and investor behaviour.

Do you at least acknowledge that some data points are used more than others with moving window?  It's quite considerable given people are planning for 3+ decades of retirement.  That's at least 360 of the 1200 datapoints.  Including 2000 and 2008 crashes.

I don't understand what you think is so difficult about MC.  It's as hard as you want to make it.  On one level, we can just take pull random data based on monthly mean and standard deviation.  On another, you can randomly sample with replacement from old data.  

Agreed on some data points are used more than others. Regarding the number of datapoints and impact, it goes back to my point around the first decade of retirement being the most critical, and we've already passed that for 2000 and 2008.
In addition to looking at the % success rate, you can also look at the worst-case outcome (the earliest age at which the money ran out) - I use both and the latter obviously isn't impacted by the moving window.

While SOR is indeed most likely to cause failure early in retirement, it is also possible for it to fail later on. 

My reservations around MC are as follows (and as I've previously mentioned, I've written them to price options and CDOs and it's something I used to enjoy!) :

1. You need to be very careful around assumptions and sensitivity of outputs to changes in inputs, and I've seen what happens when people place too much faith in models.
https://thewest.com.au/business/economy/gfc-10-years-on-a-wonky-finance-edifice-ng-b88942586z

The two methods I suggested for MC are only using unadulterated historic data.  The first was drawing at random based on a normal distribution using the mean and SD.  The second was drawing at random with replacement from historic data.  If you think these are incorrect assumptions, then historic window has already failed.  Once can always add their own assumptions to add an element of conservatism.  The article you're posted is a straw man.  

2. If we are saying the underlying historical data is of limited value, I'm not sure what slicing and dicing/using the data in a different way would add to that. 
We both accept it has limitations.  But why wouldn't you want to use it all equally weighted if you think it has any merit?  

3. Allowing human subjectivity in the decision making process around inputs has the potential for suboptimal outcomes, IMO
Like I said, at the very basic level, one can use actual data.  

4. How do you model mean reversion in MC, which (global) markets have demonstrated up to now?
A fundamental misunderstanding of reversion to the mean.  RTM doesn't mean if I shake hand's with a man who is 6' 5", the next man I shake hands with will be 5' 1" to even things out.  It just means there's a greater probability that his height will be closer to the national average of 5' 9".  

MC models RTM perfectly well as data is drawn based on the actual distribution.  The further a draw deviates the the mean, the higher probability the next one will have a lower deviation.

5. What faith would you have in the output over what historical data has shown us?
I'd have more faith that the moving window model as the data would tend toward suggesting a more conservative withdrawal rate.  And could also pull more dire SORs.

Irrespective of which approach you use, it's obviously worth emphasising that we are dealing with uncertainty, so even though a plan shows 100% historical success rate, that's not to say we won't have to make adjustments along the way. Most people are comfortable with that.

My original reply related to someone saying 100% equities was the safest approach because history says so.  It seems barely anyone on this thread is convinced of that as most people plan on having other assets, whether it's bonds or a cash buffer.  It would therefore seem most people have made their own decision of what they think of the moving window model, and decided it's foolhardy.  Making somewhat subjective decisions about how best to deal with SOR risk.  

It's difficult to see why you are so dismissive of a more conservative model making better use of data.

Probably easier to reply to individual parts
"The two methods I suggested for MC are only using unadulterated historic data.  The first was drawing at random based on a normal distribution using the mean and SD."

"A fundamental misunderstanding of reversion to the mean.  RTM doesn't mean if I shake hand's with a man who is 6' 5", the next man I shake hands with will be 5' 1" to even things out.  It just means there's a greater probability that his height will be closer to the national average of 5' 9".  

"MC models RTM perfectly well as data is drawn based on the actual distribution.  The further a draw deviates the the mean, the higher probability the next one will have a lower deviation."

By reversion to the mean, I am referring to the market returning to its long term upward average. In order to return to its long term average, it must grow by more than the long term average after periods of underperformance which is how global markets have historically behaved.
By drawing at random, and given enough simulations, you are going to get a outputs(s) where the result does not reflect historical market behaviour (for the data we have). 
To take your, example, if we have a series of 5' 1" people, we really need to get some 6' 5" samples to pull us back to the long term 5' 9" mean -  a greater probability of getting someone close to 5' 9" isn't sufficient.

"My original reply related to someone saying 100% equities was the safest approach because history says so. "

It's not as clear cut as that in the data that I have, and even if it were, I don't think many people would be happy living with the potential volatility.

"It would therefore seem most people have made their own decision of what they think of the moving window model, and decided it's foolhardy."

I'm not entirely clear of the link you are making between asset allocation and the moving window model, but "foolhardy" is not a view I see expressed very often, and tools such as Timeline (which to be fair also offer MC simulations) has seen a good uptake. 

"It's difficult to see why you are so dismissive of a more conservative model making better use of data."

Two reasons:
1. Either the moving window approach is too risky, and an MC approach you describe is superior. Given the real-world SWR of the moving window approach, and the reduction in SWR with the MC approach,  the safety-first approach (securing more of your income) rather than probability-based might make more sense.

If you have some example of the difference in SWR/success rate between MC and sliding window it would be useful to see them to compare with what I'm seeing in Timeline.

2. Or, the MC approach is too cautious which could mean people delaying their retirement, starting off with a lower withdrawal rate at retirement and/or dying with far too much.