SWR race to the bottom

I would recommend to treat your life expectancy as a probability distribution. This is what I did in my modeling before I retired. I used a mean and a normal distribution based on actual data, although I upped the average for my unique situation and also for how much I believed that life expectancy would increase over that time frame.

This is very different than saying I am planning to live to 100 (even though that small possibility is accounted for using this technique). It is highly likely that only a small percentage of people posting on this board will ever live to see that age.

I have read detailed research overview articles that state that genetics probably accounts for a third of longevity, at most. There are many factors.
 
So, let's say you are happy with a plan that is 95% for 35 years taking you to age 95. That doesn't mean you have a 5% chance that you will run out of money before you die. Your actual chance of running out of money before you die is well below 5% because most people don't live to be 95.
Exactly.
 
Well, the 'Race to the bottom' has really bottomed out, when some folks retirement plan has them hoping to die in their 80s.......

This is why you should delay S.S. to age 70 for the increased benefit, in case you do live beyond age 85.....And this means you can have a higher withdrawal rate in your 60s as result of this 'Old Age Insurance'. It's a no-brainer if you have the Funds to afford this. Most don't have the money and have no choice, but to take S.S. at age 62.
 
So, let's say you are happy with a plan that is 95% for 35 years taking you to age 95. That doesn't mean you have a 5% chance that you will run out of money before you die. Your actual chance of running out of money before you die is well below 5% because most people don't live to be 95.


I don't think that is a useful way to look at it. Averages be damned, you either live to 95+ or you don't. I want to be prepared in case I do.

Lets make a parallel 'bet'. Let's say you live in an area that is rated as flooding once in 100 years. So that's a 1% chance in any one year, on average. Let's also say that the cost to protect your house against this kind of flooding is something you can afford - maybe making sure gutters are routed properly, some berms to deflect water around the house, a battery backed up sump pump, etc.

Would it make sense to do just 1% of those things? So if/when that flood comes, you have 9.9" of water in your basement instead of 10.0"? Of course not. We take steps to protect against things that may happen. We have to weigh the risk/reward, and maybe we decide we can't afford the flood protection, and take our chances. But we shouldn't fool ourselves about the risk we are taking, and the consequences of our decision.

-ERD50
 
For data wonks only:

I was curious (and anxious) enough about this paper to write a program that implements their model. I get completely different results than they do, and the results that I do get are more in line with some of the other calculators (Firecalc, Fidelity Retirement Income Planner) that I've tried. For example, for a 4% withdrawal rate over 30 years, and starting with the same parameters used in the paper (CAPE = 22, current 5 year bond = 2%,10000 simulations), I get a 84% success rate. Firecalc gives about a 96% success rate. If I use a more reasonable management fee (.18% instead of .5%) I get a 88% success rate. The paper gives a success rate of somewhere around 50% for the first case. I also noticed that their model shows increased volatility for a bond only allocation vs a 100% stock allocation, which doesn't make much sense.

Perhaps I screwed up something in my program, but, at least at first glance, there appear to be problems with both the model and the model calculations written up in the paper.
 
I don't think that is a useful way to look at it. Averages be damned, you either live to 95+ or you don't. I want to be prepared in case I do.

Lets make a parallel 'bet'. Let's say you live in an area that is rated as flooding once in 100 years. So that's a 1% chance in any one year, on average. Let's also say that the cost to protect your house against this kind of flooding is something you can afford - maybe making sure gutters are routed properly, some berms to deflect water around the house, a battery backed up sump pump, etc.

Would it make sense to do just 1% of those things? So if/when that flood comes, you have 9.9" of water in your basement instead of 10.0"? Of course not. We take steps to protect against things that may happen. We have to weigh the risk/reward, and maybe we decide we can't afford the flood protection, and take our chances. But we shouldn't fool ourselves about the risk we are taking, and the consequences of our decision.

-ERD50
Yes, I completely understand that (although I think your flood analogy is flawed, that is beside the point). The consequences of the negative outcome of running out of money are greater than the rewards of the positive outcome. It is why I won't be taking social security until age 70, and I just bought a big batch of 30 year TIPs that won't mature until I am 77 years old (I am currently 47 years old and retired) -- I have others maturing in my mid-70s. And I may buy annuities with that money at that time. I have established a basic income floor on my old age, no matter how long I live. But I am still not planning as if it is a LIKELY possibility that I will live to 100, as many on this board seem to do. There is a big difference.
 
Also, don't forget William Bernstein's advice in the "Retirement Calculator from Hell, Part III" about the futility of low failure rates on long term withdrawals from a portfolio. Although I think he is too pessimistic in the article, it is general advice well worth noting:

The Retirement Calculator from Hell, Part III

For this to be a useful estimate of your true chance of not running out of money, the "success rate" of your ambient political, economic, and military environment must be at least 97% over this 40-year period. Do you think that this is likely? Only if you are an historical illiterate (which, I’m afraid, subsumes many finance academics).

Let’s examine a small sampling of possible political, economic, and military failure modes:

The mildest scenario is that of catastrophic inflation, as experienced in Germany and Hungary in the 1920s or, more recently, in much of the developing world.

Political failures are slightly worse, since these threaten the basic human motivation to work and produce. The state, for whatever reason, can decide to confiscate your assets or, worse, society’s means of production. Anyone who judges this unlikely should turn on CNN during any G-8 or WTO conference.

Local military action. Probably the lowest-probability item on this list, but something to think about on other continents.

The Big One: Some deranged prime minister or colonel in central Russia, Pyongyang, or South Asia could let loose the four horsemen upon the planet.

So, think about what a 97% 40-year success rate means: the absence of all of the above for approximately the next 1,200 years. (A 97% success rate means a 3% failure rate; those 40 years divided by 0.03 is 1,200 years.) Ignore for a minute the uncertainties of the less-developed world and think only about the winners: Germany—in this century alone, three episodes of military and/or economic disaster, the first two associated with mass starvation. Japan—wartime devastation even worse than Germany’s. England—near brushes with disaster in 1812-1814 and in both world wars. And even the United States—repeated banking failures, civil war, and the near-bankruptcy of the Treasury in the 19th century. The near collapse of the capitalist economy in the 1930s. And oh yes, I almost forgot—the entire globe barely missed mass incineration in October 1962.

History’s best-case scenario was the Roman Empire, which survived more or less intact for about seven centuries (if you ignore the odd sackings of the capital after 200 A.D.).

A wildly optimistic historian might give us another few centuries of economic, political, and military continuity. Back-of-the-envelope, that’s about an 80% survival rate over the next 40 years. Thus, any estimate of long-term financial success greater than about 80% is meaningless.
 
For data wonks only:

I was curious (and anxious) enough about this paper to write a program that implements their model. I get completely different results than they do, and the results that I do get are more in line with some of the other calculators (Firecalc, Fidelity Retirement Income Planner) that I've tried. For example, for a 4% withdrawal rate over 30 years, and starting with the same parameters used in the paper (CAPE = 22, current 5 year bond = 2%,10000 simulations), I get a 84% success rate. Firecalc gives about a 96% success rate. If I use a more reasonable management fee (.18% instead of .5%) I get a 88% success rate. The paper gives a success rate of somewhere around 50% for the first case. I also noticed that their model shows increased volatility for a bond only allocation vs a 100% stock allocation, which doesn't make much sense.

Perhaps I screwed up something in my program, but, at least at first glance, there appear to be problems with both the model and the model calculations written up in the paper.

Fred I am curious how did you model the equity returns? In particular did you decrease the average returns by 2% like they did, but keep the variance the same.

One of the things that I find most suspicious about the paper is this. The found the optimum AA was 40% equities and 60% bonds. At time when bonds are at historic lows (when the paper was written) such an AA makes no sense even if you believe stocks are overpriced.

Does anybody know if the Journal of Financial Planning articles are per reviewed? Perhaps that is what he is getting now.
 
Last edited:
Fred I am curious how did you model the equity returns? In particular did you decrease the average returns by 2% like they did, but keep the variance the same.

One of the things that I find most suspicious about the paper is this. The found the optimum AA was 40% equities and 60% bonds. At time when bonds are at historic lows (when the paper was written) such an AA makes no sense even if you believe stocks are overpriced.

Does anybody know if the Journal of Financial Planning articles are per reviewed? Perhaps that is what he is getting now.

Getting into the weeds now!

They model the returns using what is know as an autoregressive model. These kind of models assume that the values from the current year can be used to predict the next year, and that the next year value is a linear function of the current value. They derive the parameters for the model (I assume) by fitting it to historical data.

They start with a "seed" CAPE (cyclically adjusted PE ratio), and then assume that the CAPE changes each year using the following recurrence relation:

CAPE(t) = 2.11 + .87*CAPE(t-1) + e(CAPE)

where e(CAPE) is normalized white noise with a mean of 0 and standard deviation of 4.0 They then estimate the total stock return using this equation:

r(s) = .24 - .0083*CAPE(t) + e(S)

where e(S) is more normalized noise with a mean of 0 and std dev of 0.2.
 
I don't think that is a useful way to look at it. Averages be damned, you either live to 95+ or you don't. I want to be prepared in case I do.

Me, too. I'm not advocating that people shouldn't prepare. I'm saying a few things though that I think get overlooked.

1. First, is that everyone does draw the line of preparation somewhere. That is, while you may want to be prepared in case you live to 95, do you also demand the same degree of certainty to prepare to live to be, say, 105? Or 110? Some people do live that long. So we all have to draw a line somewhere. Maybe you draw it at 95 and someone else might draw it at 90 and someone else at 100.

2. My point is that drawing the line is not just something where you are either have a 100% chance of success or no chance of success. That is, if you say want a 100% chance of success for 95, that doesn't mean that you have 0% chance of success for 96 or even 100. That is, many might say that while they want 100% for 90, they would be fine with 95% for 95, and 90% for 100. This is because the chances of living to 100 is less than the chance of living to 90. So I want a different level of preparation for 90 than for 100 (of course, where one draws that line will differ from person to person but most everyone will draw that line somewhere).
 
Getting into the weeds now!

They model the returns using what is know as an autoregressive model. These kind of models assume that the values from the current year can be used to predict the next year, and that the next year value is a linear function of the current value. They derive the parameters for the model (I assume) by fitting it to historical data.

They start with a "seed" CAPE (cyclically adjusted PE ratio), and then assume that the CAPE changes each year using the following recurrence relation:

CAPE(t) = 2.11 + .87*CAPE(t-1) + e(CAPE)

where e(CAPE) is normalized white noise with a mean of 0 and standard deviation of 4.0 They then estimate the total stock return using this equation:

r(s) = .24 - .0083*CAPE(t) + e(S)

where e(S) is more normalized noise with a mean of 0 and std dev of 0.2.

Fred, it's great that you have the math and programming skills to analyze these data. IIRC you are in the financial services industry. I also thought something looked odd about the risks associated with different asset allocation. It is not clear to me from the link that the paper has been independently published or peer reviewed. It wouldn't be the first time that false results have been published, whether due to deliberate data manipulation or error. If something doesn't look right to informed readers, it should be questioned. Therefore I think this is one instance in which it would be quite appropriate to contact the principal author to ask for the raw data so that we (I mean you) can look it over and validate it (or not). Would you be willing to volunteer on behalf of the forum?
 
I see you are using the formula in the paper. I follow the logic of the formula although I'll admit things like "normalized noise" are terms I'm not familiar with. So what I think you are saying is even using their formula you have trouble reproducing their results.

What I am saying is I find their reduction of the expected stock market returns from 12% to 10% (nominal) arbitrary and unduly pessimistic. Meaning my formula would be
r(s) = .26 (i.e. +2%) - .0083*CAPE(t) + e(S).

The reason being. In the 20th century the US enjoyed conditions conducive to economic growth, the rule of law, modestly regulated capitalism, political stability, and our infrastructure was not ravaged by wars or revolutions. The world as whole in the 21st century is lot more like the US back in the 20th century. Capitalism is widespread, democracy is catching on,and while there is still plenty of wars, the prospect of world wars has diminished. I therefore expect the rest of the worlds capital markets to catch up with the US rather than US markets to slip backwards.
 
Last edited:
Also, don't forget William Bernstein's advice in the "Retirement Calculator from Hell, Part III" about the futility of low failure rates on long term withdrawals from a portfolio. Although I think he is too pessimistic in the article, it is general advice well worth noting:

The Retirement Calculator from Hell, Part III
Unfortunately you clipped off his closing remarks right after your excerpt. His point was that factors other than market performance may override your calculated probability of success, not that higher probability rates are "futile." Higher probability of success is beneficial unless something geopolitical or otherwise changes everything. He certainly did not mean that any financial plan over 80% success rate was pointless.
Retirement Calc from Hell said:
Mind you, this is not a call for wild abandon. The above table constrains the retiree desiring a theoretical 97% success rate (of portfolio survival) from spending more than 3% per year of the initial real amount of his nest egg. Taking the accident propensity of the species into account would allow him to spend about 4%. But if you believe that we’re about to encounter a bad returns sequence or simply wish to leave a few baubles to your heirs, you’re right back to 3% again.

So live a little, and enjoy your money, for tomorrow we may be consumed by the ghosts of Hitler, Lenin, and Attila the Hun. And at withdrawals of 3% to 4% of your nest egg, don’t spend it all in one place.
 
Me, too. I'm not advocating that people shouldn't prepare. I'm saying a few things though that I think get overlooked.

1. First, is that everyone does draw the line of preparation somewhere. That is, while you may want to be prepared in case you live to 95, do you also demand the same degree of certainty to prepare to live to be, say, 105? Or 110? ...

OK, but what I find is if you plan for a 40-50 year retirement, your WR starts to approach a 'forever' portfolio anyhow. So if I'm alive and my brain is functioning, I'll re-evaluate as I go.

I see you are using the formula in the paper. I follow the logic of the formula although I'll admit things like "normalized noise" are terms I'm not familiar with. ...

They need to thrown in a term to adjust for the Boltzmann Constant ;)

-ERD50
 
1. First, is that everyone does draw the line of preparation somewhere. That is, while you may want to be prepared in case you live to 95, do you also demand the same degree of certainty to prepare to live to be, say, 105? Or 110? Some people do live that long. So we all have to draw a line somewhere.

Actually, retirees don't have to have a line. Retirees with enough savings, multiple income sources and/or low enough expenses may never run out of money. Some might actually save money in retirement for charities, trust funds, grandchildren's college funds or simply force of habit.

Some retiree households do find they can live well on SS alone, especially if they retire in a low cost of living location. Then they can just invest their pensions (if they have one or more) and reinvest their investment income.
 
Last edited:
Unfortunately you clipped off his closing remarks right after your excerpt. His point was that factors other than market performance may override your calculated probability of success, not that higher probability rates are "futile." Higher probability of success is beneficial unless something geopolitical or otherwise changes everything. He certainly did not mean that any financial plan over 80% success rate was pointless.

Agreed, I get tired of that being taken out of context (or just not passing common sense).

Sure, if a catastrophe occurs, all bets are off, a 'portfolio' isn't going to be of any comfort. But if it doesn't, a 4% WR has a higher likelihood of failing before you do than a 3.5% WR, and a 3.3% WR, or a 3.0% WR, etc. There's no way around that.

-ERD50
 
I just think they spend all their time coming up with parameters and models that will make things look even worse than their last paper!

Follow the money (the annuity industry).

Great discussion!

I'm sticking with my 3.33% withdrawal of remaining portfolio, and even plan to increase it if I get old enough.
 
Agreed, I get tired of that being taken out of context (or just not passing common sense).

Sure, if a catastrophe occurs, all bets are off, a 'portfolio' isn't going to be of any comfort. But if it doesn't, a 4% WR has a higher likelihood of failing before you do than a 3.5% WR, and a 3.3% WR, or a 3.0% WR, etc. There's no way around that.

-ERD50

Geez, I did NOT quote that article out of context, not in the least. I can't cut and paste the entire article into a discussion like this.

No one said that a 3.5% WR is not safer or more prudent than a 4% WR, etc. That is not what we are getting at.

The main point of the article is that it is NOT possible to have, for example, a 97% ex-ante financial success possibility, or even close. No matter how much money you have or planning that you do. Because matters other than investment returns take precedence at that point and saving more yields greatly diminishing returns against other higher outside probabilities that you cannot control.

And I do agree that one should be conservative like he mentioned if you expect lower returns in the future (something I agree with 100%, I don't use FIREcalc for projections because I think ex-ante returns are lower now). That is different than the main point of the article (but put there so people would not spend with wild abandon for the future) but an excellent point indeed.

Bernstein himself was recently quoted as saying the stock and bond markets are offering lower ex-ante returns now than anytime in history. THAT is something worth saving more for.
 
Perhaps I screwed up something in my program, but, at least at first glance, there appear to be problems with both the model and the model calculations written up in the paper.


You should double check with the authors (and ask for their code). In my opinion, errors in papers (and code) are quite common (and reviewers never double check at that level of detail).


One of the things that I find most suspicious about the paper is this. The found the optimum AA was 40% equities and 60% bonds. At time when bonds are at historic lows (when the paper was written) such an AA makes no sense even if you believe stocks are overpriced.

I think this could be plausible since the problem with equities is having to draw from your portfolio when there has been a bad sequence of returns.

BUT what makes me suspicious about their paper is that none of their forecast models have references and they don't talk much about how they set their parameters. They also make seemingly arbitrary adjustments (-2% for US), their inflation model has 5 parameters fit from maybe 100 data points, they have to arbitrarily cap PE10 at 5 and 45, and they don't do any sensitivity analysis or blind tests of their forecasts (on data held out from parameter estimation).
 
Geez, I did NOT quote that article out of context, not in the least. I can't cut and paste the entire article into a discussion like this. ...

Sorry, I didn't mean to point that at you specifically.

Over the years, a number of posters have said something to the effect that anything past 80% in any calculator is meaningless, and I don't think that is what was intended, and I don't think it makes sense. My comment was more towards that past history.

-ERD50
 
Sorry, I didn't mean to point that at you specifically.

Over the years, a number of posters have said something to the effect that anything past 80% in any calculator is meaningless, and I don't think that is what was intended, and I don't think it makes sense. My comment was more towards that past history.

-ERD50
OK, thanks ERD50 :) Yes, I agree that 80% is too low and in my original post stated that Bernstein was just too negative in his calculations. But I think the underlying idea is sound and worth thinking about . . .
 
You should double check with the authors (and ask for their code). In my opinion, errors in papers (and code) are quite common (and reviewers never double check at that level of detail).




I think this could be plausible since the problem with equities is having to draw from your portfolio when there has been a bad sequence of returns.

BUT what makes me suspicious about their paper is that none of their forecast models have references and they don't talk much about how they set their parameters. They also make seemingly arbitrary adjustments (-2% for US), their inflation model has 5 parameters fit from maybe 100 data points, they have to arbitrarily cap PE10 at 5 and 45, and they don't do any sensitivity analysis or blind tests of their forecasts (on data held out from parameter estimation).

Good points from all, especially the 2% adjustment. If I plug this in, I get a success rate of about 95%. And it's very hard to justify this adjustment, given that they use CAPE to predict stock returns, and the CAPE is apparently based on US stocks, while the adjustment is supposed to represent an international portfolio. Another problem is that the arbitrary PE10 5<->45 limits arbitrarily constrains the stock return to a range between about -14% to 20%. And the variance of the stock return distribution is much smaller than the S&P 500 variance.

I've convinced myself that the model has so many flaws that it isn't worth spending any more time on it (although independent verification would be welcome). And makes me desire a decent open source retirement calculator.
 
And makes me desire a decent open source retirement calculator.

I've long thought this should be something a group of us should try for. It's not that hard, and there are surely enough people here who have the skills.
 
I think this could be plausible since the problem with equities is having to draw from your portfolio when there has been a bad sequence of returns.

BUT what makes me suspicious about their paper is that none of their forecast models have references and they don't talk much about how they set their parameters. They also make seemingly arbitrary adjustments (-2% for US), their inflation model has 5 parameters fit from maybe 100 data points, they have to arbitrarily cap PE10 at 5 and 45, and they don't do any sensitivity analysis or blind tests of their forecasts (on data held out from parameter estimation).

Yes it is certainly true that if you have a bad sequences at the beginning a heavy equity concentration makes retirement challenging. But I haven't seen a FIRECalc run or any other research that shows that anything under 60% equities is optimum. On the other hand if you restrict equity returns to below historically levels then this make sense. I'm just saying that logically if bond yields are historically low it make sense to hold less bonds not more to have the maximum SWR.

All very good points.
 
Your CDC chart may be misleading, because it blends sexes and races. Men clearly live shorter lives on average and being a single male I only have to worry about me. The obits I read every day show a trend, very few 90+ old men.

And the few 90+ men that you do see are usually still leaving a surviving wife behind. Widowers don't tend to live a long time once they are alone. They tend to either remarry or die fairly quickly (with some exceptions, obviously).
 
Back
Top Bottom