Tuesday, December 1, 2015

100% Certain That We’re Not Sure

In my previous post, Uncertainty Denial, I promised to suggest a way to accept and plan for the uncertainty of future market returns rather than to deny it. Future market returns, future expenses and how long a healthy person will live are critical to your retirement plan and all are uncertain. The uncertainty is inescapable. Relying solely on averages or likelihoods in a retirement plan is risky.

In this post, I’ll talk about the uncertainty of market returns and the impact that has on our ability to optimize the financial parameters of a retirement plan, like spending rates, equity allocations and how much we need to fund a safe floor of income.

Let’s say you decide to base your plan on the assumption that your portfolio will return 5% annually in real dollars with a standard deviation of 11% over the remainder of your lifetime. You should have very little confidence that this prediction is in any way accurate, but let’s stick a pin in that for now and pretend that we’re pretty sure of our prediction.

That 11% standard deviation quantifies the uncertainty of our estimate. It means that roughly two-thirds of the annual returns will fall between a 6% loss and a 16% gain, a pretty broad range. About 95% of those returns will fall between a 17% loss and a 27% gain.

The following chart shows terminal portfolio values (wealth at death) from a Monte Carlo simulation for a retiree spending $40,000 a year from a $1M initial portfolio balance (the 4% rule) and living exactly 30 years in retirement. The red dots are scenarios that ended in ruin before 30 years. The lighter blue dots represent scenarios that survived 30 years but ended up with less than the $1M the scenarios began with. The dark blue dots represent scenarios that funded thirty years and ended up with more than the portfolio’s initial value. The probability of ruin is 6.7% for all scenarios.

(I fudged a bit on the TPV of the red dots by continuing spending after the portfolio was depleted for the sake of visual clarity. The values of these failed portfolios should all be precisely zero, but that places all the red dots on the x-axis and it is impossible to see how many there are.)

Figure 1 above shows a lot of uncertainty. About 14% of the scenarios had annual returns less than 2.5%, or half the expected return. About 11% had returns greater than 7.5%, or 50% greater than the expected 5%. Figure 2 below shows the same simulation as above, but adds life expectancy modeled from mortality tables for a 65-year old couple.

There are several points of interest regarding Figure 2. First, the points are more scattered than the first graph, showing the degree of uncertainty that life expectancy adds. That means real retirement outcomes are less predictable than studies assuming a 30-year retirement. Second, there are fewer red dots in this chart than in Figure 1 because many retirees who would have run out of money had they lived 30 years would die before going broke in real life. (Only about 6% of 65-year old males will live another 30 years or more.)

The probability of ruin drops from 6.7% to 3% for all scenarios when we include life expectancy in the model. This is consistent with a study by Stout and Mitchell that found that considering life expectancy in the sustainable withdrawals rate (SWR) model lowers the probability of ruin by about half. That’s another way of saying that fixed-length retirement models overstate the probability of ruin.

While 3% of the 10,000 scenarios ended with failed portfolios before death, 82% of those failures (1,400/1,700) occurred when returns were worst, 18% occurred with below average returns, and 0.01% occurred when returns were greater than the expected 5%.

Shorter lifetimes than 30 years means less time to grow those big terminal portfolios found in the upper right quadrant of both figures, and when joint lifetimes for a couple are considered instead of assuming they always retire for 30 years, the median terminal portfolio value falls 22% in this example.

Those huge terminal portfolio values from SWR studies that leave us starry-eyed are possible, but they require all four of these pieces of good luck: excellent market returns, a long life, a fortunate sequence of those returns, and no spending shocks. On the other hand, any one of poor market returns, an unfortunate sequence of returns, or spending shocks combined with a long life can be financially catastrophic.

Third, notice the dots (scenario outcomes) in Figure 2 near the 5% annual return line. The outcomes range from a couple of failed scenarios (red dots) to one leaving about $3M to heirs. Outcomes range from portfolio failure to a huge success even though they all experienced a 5% average annual return on investments. Or look at the outcomes with an annual average 10% return that range from $1M to $8M. These examples illustrate sequence of returns risk.

Next, notice that there is one lonely failed scenario in the Above Average Returns range and none in the Best Returns range. Had I increased the number of simulations from say, 10,000 to 100,000, more failures would show up in all ranges and a few would appear in the Best Returns range. These additional failures are improbable, but not impossible. The best range of returns is not immune to portfolio failure; failures there are just less likely.

Lastly – and perhaps most importantly, if you retired 10,000 times then your history of terminal portfolio balances might look a lot like Figure 2. But unless you're Dr. Who, you only get one retirement and one of those dots.

(If you are Dr. Who, please call me. I need to know if Clara is permanently dead. The suspense is killing me.)

Table 1 below shows the statistics for the simulation that includes joint life expectancy.

Table 1. Portfolio Return Scenarios Including Joint Life Expectancy
Outcome Worst Returns Below Average Returns Above Average Returns Best Returns
Range of Returns <= 2.5% 2.5% to 5% 5% to 7.5% > 7.5%
% of Scenarios 14% 33% 34% 18%
Median Successful TPV ($) 441,160 905,200 1,883,960 3,314,862
% Failed Scenarios 14% 3% 0% 0%

Next, let’s add rows to Table 2 for the 95th-percentile sustainable spending rate using Milevsky’s formula and the recommended asset allocation using the AACalc website. Those last two rows of Table 2 show the spending rate and asset allocation we would have chosen had we known we would end up with those returns instead of the 5% average return that we assumed.

Table 2. SWR and Equity Allocation Recommendations
Outcome Worst Returns Below Average Returns Above Average Returns Best Returns
Range of Returns <= 2.5% 2.5% to 5% 5% to 7.5% > 7.5%
% of Scenarios 14% 33% 34% 18%
Mean Annual Return 1.0% 3.9% 6.1% 9.3%
% Failed Scenarios 14% 3% 0% 0%
Sustainable Spending Rate 1.1% 2.6% 3.9% 5.9%
Equity Allocation 15% 55% 100% 95%

And herein lies the planning predicament. Although 5% returns with an 11% standard deviation sounded precise, it actually provides a 1-in-7 possibility of a terrible outcome and nearly a 1-in-5 possibility of a really great outcome. For which scenario do we plan? Worst Returns with 1.1% annual spending and a 15% equity allocation? Best Returns with a 5.9% spending rate and 95% equity allocation? Somewhere in the middle?

A lot of plans will focus on the middle, but a good plan will consider them all.

"Looks like we’re 100% certain that we’re not sure." Jerry Horn, Twin Peaks.
[Tweet this]

Here’s my thinking. The first order of business is to eliminate catastrophes – those red dots. We can’t avoid red dots when we fund retirement with an equity portfolio because even good returns on average can fall victim to an unfortunate sequence of returns. We can, however, make red dots non-catastrophic by building a floor of safe income with TIPS bond ladders, Social Security benefits, fixed annuities and the like to insure that we can meet our non-discretionary spending needs. Then, even if our equity portfolio becomes a red dot, we don’t lose our standard of living.

Do this step first because most American households will find that once they allocate their assets to the floor and set aside some cash for liquidity, there won’t be much left to invest for upside. Floors are expensive, especially in the current low-interest rate environment. Most households won’t even be able to build an adequate floor. For those who can, the cost of the floor is likely to significantly change the size of the upside equity portfolio.

If you end up in Above Average Returns or Best Returns, you won’t have many money problems. I would hope for those outcomes, but not plan on them. We should plan for somewhere between Worst Returns and Below Average Returns, but this is still a broad range of equity allocations and spending rates. By doing so, we reduce the upside potential of better returns, but that is the cost of safety.

Many retirement income plans appear to make assumptions about these parameters that are far more certain than we can actually calculate. If the right answer is somewhere between 15% and 55%, tweaking an allocation by 5% seems hard to defend.

Whether I would choose closer to the Worst Returns or to the Below Average Returns parameters would depend largely on my capacity to build a floor. With a larger floor, I can take more equity risk and lean toward Below Average Returns. I also consider short-term volatility in setting my equity allocation, as I explained in The Role of Xanax in Asset Allocations. That might overrule these recommended allocations.

Regardless, retirees should revisit this decision annually because some uncertainty is reduced with time. As we age, for example, our remaining lifetime becomes more certain and the range of possible TPV's becomes more clear.

Now, I have to return to my earlier statement above that these calculations are based on the assumption that we are pretty confident in our prediction of 5% future returns with an 11% standard deviation – and that we have no reason to be.

To quote a recent study by Pastor and Stambaugh entitled, “Are Stocks Really Less Volatile in the Long Run?” (they aren’t, by the way):
Expected return is notoriously hard to estimate. . . “estimation risk” . . .  reflects the fact that, after observing the available data, an investor remains uncertain about the parameters of the joint process generating returns, expected returns, and the observed predictors. That parameter uncertainty adds to the overall variance of returns [risk] assessed by an investor.
In other words, we are adding to the uncertainty quantified by an 11% standard deviation for our 5% return assumption the uncertainty that 5% and 11% were the right guesses in the first place. My calculations above, for example, only show the uncertainty assuming we were certain about our return assumption. Were 4% a better estimate of future returns, for example, sustainable spending rates would have ranged from 0.7% to 5.2% and equity allocation recommendations from 0% to 95%, a sizable difference from the 5% guesstimate.

The process I suggest may seem pointless given the amount of uncertainty involved, but a plan that results from our best efforts to quantify that uncertainty is better than no plan at all. If we can’t identify a plan that will work with certainty, we can at least rule out plans that probably won’t work and identify plans that are more likely to work. But, we have to recognize that uncertainty, even in the latter.

I suspect that most planners put way more faith in expected outcomes than is prudent. The unexpected outcomes are the ones to worry about.

The best strategy is to "buy insurance" to avoid catastrophic outcomes and invest what’s left. We buy portfolio failure insurance by building a floor of Social Security and pension benefits, TIPS bond ladders, and fixed annuities and we maximize that floor by delaying Social Security benefits as long as we can.

If we’re lucky, our wealth will grow. If we’re unlucky, we’ll still have a roof over our heads.

Don’t confuse precise probability calculations with certainty. As Jerry Horne said to his evil brother, Ben, after torching their lumber mill for insurance fraud in the second season of Twin Peaks, “Looks like we’re 100% certain that we’re not sure.”

Please check out my next post in this series, Positive Feedback Loops: The Other Roads to Ruin.


  1. "Build a floor" is fine advice for those entering retirement. Is there any similarly fruitful and cautious step one can take in (say) the decade before?

    1. You bet.

      A lot of households who had planned to retire in 2007 or 2008 saw their portfolios decline precipitously as the market fell more than 50%. They were forced to postpone retirement and probably lower their expectations even then. I think most of them would agree their equity allocation had been too high. I retired in 2005, but a 40% equity allocation saved me from most of the carnage.

      So, reduce your equity allocation in the decade prior to retiring. 40% to 50% equity is probably a good ballpark, but it will vary significantly depending on your complete financial situation.

      You can also start building that floor so the income begins in your first year of retirement.

      I like the idea of paying off your mortgage before retirement if that is within reach and certainly paying off any consumer debt. This debt leverages risk and relatively small decreases in annual spending can show a significant improvement in your cash flow and balance sheet.

      Thanks for writing!

  2. Thank you for the clear analysis, insights and their implications. Nicely done! . I really enjoyed this post and it affirms a lot of my prior conclusions -- as we always like affirmation of what we thought, don't we ;)
    I wonder, though, if things are a bit more uncertain than even your analysis would suggest. From easy observation, projected income from "safe" investments and income streams like Social Security and Pensions are not certain. Social Security has been increasingly taxed (and Federal and State levels) and even Pensions that were "promised" have been cut mid-stream.

    1. Correct!

      And insurance companies that provide annuities could fail. If one or two fail, state insurance regulators should be able to step in, but if there is a systematic failure like we saw on Wall Street at the end of 2007, the states would be overwhelmed.

      TIPS bonds are safest, but may not provide adequate longevity insurance.

      William Bernstein wrote a great piece years ago, at efficientfrontier.com if memory serves, saying that, given all the external geopolitical and global financial variables, the best probability of a successfully fund retirement is probably no better than about 80%.

      This post is only about the uncertainty of meeting retirement spending needs with a volatile portfolio of stocks and bonds. Beyond that, my guess is things are a lot more uncertain than even my analysis suggests.

      But we deal with a lot of uncertainty in our everyday lives and generally get by just fine. My point is that we need to recognize that uncertainty and deal with it instead of pretending we can accurately predict the future.

      Great point. Thanks for writing!

  3. Thanks for this column. Fantastic stuff. Unless I am missing something, your analysis seems to discount for adjustments to the plan over time. It seems that the plan is made and then left unattended for the duration. In reality there would be adjustments to both investment allocation and spending in response to sequence of returns. Is it possible to model those variables into the Monte Carlo simulation?

  4. Just out of curiosity, what software was used?

    1. Will, I'm a computer geek. – I write my own. This was written in R.

      AACalc is free server-based software available at the link provided in the post. I wrote an R script to implement Milevsky's formula and provided a public Excel version in an earlier post.

    2. This comment has been removed by the author.

  5. Good question, Paul!

    First, I wouldn't call this a plan, it's just a modeled example of the first year of a plan.

    As I say above, "Regardless, retirees should revisit this decision annually because some uncertainty is reduced with time." There is really no justification for a "set-and-forget" retirement plan. I have written about this several times, including three posts that began with "Think Like a Bayesian Pig." Those posts link to some excellent papers written by Larry Frank and others on the topic of dynamic updating.

    It would be irrational, as you point out, to not change these parameters as your financial situation changes over time, assuming you can. There may be other factors, like spending shocks, that preclude less spending and that's the real risk to standard of living. I'll be writing about that soon.

    Certainly it is possible to simulate future adjustments by creating rules (policies) that a retiree might follow – the Frank papers do, for example – but that doesn't solve this uncertainty problem. It simply exposes another level of uncertainty. If we are uncertain of next year's return and variance, we are far less certain of returns 10 or 20 years into the future.

    I guess my bottom line would be that, yes, we can build a better model than this, but it will still contain tremendous uncertainty. Retirement income models are best for understanding the underlying processes. They aren't very good at predicting your future.

    Thanks for writing!

  6. Great post once again Dirk!

    Indeed, return uncertainty is as unknown next year as it is for any future year. Charts that show the range of returns over various periods of time trick our brains into interpreting the future as being more certain than is really is (here’s an example http://awmfinancial.com/uncategorized/a-chart-summarizing-market-volatility/ – not picking on the company because all of them show the chart this way … counting the numbers UP 1, 3, 5, 10, etc.) and lead to this belief that stocks are less risky in the long run. I have always maintained that such charts are illustrated incorrectly. Such charts should be reinterpreted in two ways – both from flipping the chart over:

    1. The numbers should be arranged to count DOWN because it is the PAST 20 years, PAST 15 years, etc. where those numbers come from. Thus, the graph when flipped over from its current form, would properly show us that the next year is as likely to be as volatile statistically as the past one.

    2. Once the graph is flipped over, one can then move it along the table slowly, to also tell the mind that as time goes on, that graph moves along to show that the past years and the upcoming years (unknown returns and deviations) would produce a similar set of graphs over time – simply with changing numbers as to average returns and deviations.

    It would be interesting Dirk to see your results using your return and STD values as you’ve done with a slight modification to the portfolio values from which the income comes from. This gets to the Ocean Analogy we’ve talked about ( http://blog.betterfinancialeducation.com/sustainable-retirement/is-all-of-your-portfolio-at-risk-of-loss/ ) where spending is set on the statistical (say 1 STD) portfolio value of the “trough” instead of the “peak” portfolio value – that peak value keeps changing from peak, to trough, to peak, to trough, as the portfolio value goes up and down with the markets. The real portfolio value is still there that generates the TPV; however spending is taken from a lower “floor” value. The purpose of course is to reduce the number and extent of spending reductions in real life. And more would be reserved for TPV in your single simulation model (30 year period).

    This reduces the likelihood of spending adjustments, except in those statistical occasions where markets forces actual portfolio values declines below the trough value (in your simulator, it is then that spending would be stressing results since you are doing single period, i.e., in this case a 30 year period, rather than multiple period simulations as in our work).

    This is simply recognizing that most of an investment portfolio, when properly diversified, doesn’t have wide swings in value, that portfolio value below the trough, and thus this too may represent a floor for spending over and above Social Security (and a pension if there is one).

    Thought experiments suggest better outcomes to TPV … of which in real life are actually reserving for the statistical time periods (read older ages) that some may age to (and others may not). Simply a method to spend a slightly more by setting the time period to a statistical age one is more likely to have (i.e., 24 years for example at 65 rather than 30 where “about 6% of 65-year old males will live another 30 years or more “ ) and thus spend the money when one is more likely to be around to spend it. Yet – hedge the bet by reserving and calculating what should be in reserve for those potential statistical years in the future for those who actually do demonstrate they’re the ones outliving their cohorts.

    Your programming skills are par none!

    Clara’s status has me on seat’s edge too!

    1. Clara Oswald dead? Say it isn't so!

      I just referred another commenter to your work in this area and should have included this link to BetterFinancialEducation.

      We could tweak the model, as you suggest, but constant-spending models are only valuable to study underlying processes. They are too simple and irrational to provide real predictive value. In real life, no one is going to go broke because they kept spending the same dollar amount every year . . . but some are going to lose their standard of living because they stopped spending too late. Constant-spending models do, I think, work well to explain the uncertainty issues. Your model is much better for planning.

      While I agree with you that "most of an investment portfolio, when properly diversified, doesn’t have wide swings in value" and thus can act as a "soft floor" under normal circumstances, this fact may bring little comfort if there is a severe spending shock.

      I'll write about that soon and look forward to your comments!

      Someone else asked about my software. It's publicly available at GITHUB for anyone familiar with coding in R. If you don't know R, I wouldn't call my scripts "user friendly."

      Great comments, Larry. Thanks for your continued contributions to the discussions here.

    2. Indeed ... "a severe spending shock" would likely torpedo any portfolio value remaining. However, in a "fixed" income world, the money wouldn't be their either (unless the retiree traded future income for a lump sum - but then they get the same result - reduced spending in the future). In other words, spending has a stochastic element to it just as markets do ... they're separate issues.

      The challenge is to match them both up over one's lifetime. Even while working, that doesn't work out well for some due to life's curve balls we are each thrown, or not. So there are actually three sets of stochastic values: Markets, longevity, and spending. Modeling is looking at the first two. Some try to incorporate the third and in my experience require an even larger sum of money to deal with the differences between forecasts and actual life as we live it because nothing is completely predictable. My phone bill, as an example, looks quite different today than it did just 10 years ago; as do most of my "floor" expenses.

      Great conversation and stimulating thoughts Dirk!

      PS. A great point between modeling (read simulations) and planning too!

    3. Some of the floor's income stream would still be there. If my portfolio disappeared entirely, I would still have income from Social Security, annuities and pensions, although your "soft floor" would be gone. While this is unlikely to result from market volatility, as you correctly point out, market volatility isn't the only risk to that wealth. Spending is volatile, too, as you also point out.

      I don't think they're separate issues at all. They're separate variables, true. But you can't build a retirement income model without considering the spending variable.

      I do agree that if your retirement plan collapses, it probably won't be due to market volatility of a well-diversifed portfoliio.

  7. Dirk, this goes in my Retirement Cafe Hall of Fame articles. Great article. I like the summary paragraph.

    The best strategy is to "buy insurance" to avoid catastrophic outcomes and invest what’s left. We buy portfolio failure insurance by building a floor of Social Security and pension benefits, TIPS bond ladders, and fixed annuities and we maximize that floor by delaying Social Security benefits as long as we can.

    I like the idea of transferring risk if possible to help eliminate worst case scenarios. This is why I like, for us, long term care (LTC) insurance (since it is affordable) since the range of possible LTC outcomes is so large (like stock/bond portfolio returns).

    I seem to recall that transferring risk and building a floor via a single premium immediate annuity (SPIA) instead of bonds may allow a higher higher legacy (and maybe higher SWR) as well. The below article is interesting in that regard.


    This is somewhat related. I am also interested in trying to optimize my income in retirement by determining what is the best mix between systematic withdrawals, annuities etc. I contacted the company at the link below, but so far have not been able to find a financial adviser near me that uses their software.


    Great article Dirk. Thanks, Brad.

    1. Thanks, Brad.

      I am familiar with Income Discovery and know its founder, Manish Malhotra. Great guy, extremely knowledgeable, very impressive. I haven't dealt with his company, but I'm sure it reflects his competence. How valuable it is to try to optimize these things given the inherent uncertainty I describe is another question, but even if their answer turns out to be sub-optimal it might still be an improvement. As I mention in the post, "a plan that results from our best efforts to quantify that uncertainty is better than no plan at all." At some point, however, continued tweaking just gives you unwarranted confidence. If you can't predict future returns and volatility with any precision, how do you optimize parameters based on them?

      (By the way, does the adviser have to be physically close by?)

      Regarding Wade's post, you probably know that he is my hero (OK, second to Bernstein) and a friend (wish Bernstein was, too), but I don't agree that there is no place for bond funds in a retirement portfolio. First, bond funds may be ideal for securing liquidity and Wade quickly agreed with this exception when I challenged him.

      Using annuities instead of bond funds in a retirement plan means that you have a floor of annuities and an upside portfolio of all equities. That's fine unless you would be concerned about the volatility of an all-stock portfolio even if you have a secure floor. The mini-survey I did recently showed that the vast majority of respondents would still want some bonds to dampen portfolio volatility, and I certainly would.

      So, I believe there is a second use for bond funds in a retirement plan, dampening short-term portfolio volatility, and one that Wade actually addresses in an example in the piece of a couple that wouldn't completely buy into the all-annuity approach.

      Having said that, if I had to choose between my retirement advice and Wade's, I'd probably go with his. I think in this case, though, he may be over-claiming.

  8. My graduate adviser put it like this "It is possible to precisely wrong".

  9. Thanks Dirk,

    Yes, even if I had an annuity as a part of my secure floor I would never put all the rest of my assets in stocks. I value the dampening effect of intermediate bonds as well as the cash reserves of short term bonds (to ride out any pronounced and prolonged drops in the market).

    A representative from Income Discovery gave me the name Rick Fine of Sensible Financial in Waltham Mass as a contact. Ann, and I will probably be ready to work on the details of the distribution plan in a couple years so I have not called him. Maybe by then they will have someone closer to me.

    One last question and comment please. I think you may have said the optimal time to look at an annuity/tips ladder is around age 70. Am I correct about this?

    Also, I would throw all three of you (Bernstein, Pfau and Cotton) into my pantheon of retirement thinkers/heroes Thanks, Brad.

    1. That's so kind of you to say, Brad, but in all honesty I feel privileged that the two of them even let me read their work!

      I believe the optimal age to annuitize is around 70, but there is a lot of conflicting opinion on this. By 70, mortality credits have begun to accrue and by 80 or so many insurers won't (or are not allowed to) sell them.

      Another option to consider is an Advanced Life Deferred Annuity (ALDA), or a deferred annuity that only pays off at an advanced age, typically 80 to 85, but is cheaper than an immediate annuity. Wade Pfau published a paper on their advantages as did David Blanchett.

      I would suggest that you read this paper by Moshe Milevsky. He concedes that the optimal age is over 60 and under 90 but adds, "Nevertheless, I remain agnostic on the precise age at which this product allocation should take place."

      For what it's worth, I plan to partially annuitize at some point and unless something changes my belief before then, I plan to do so at about age 70.

      The precise time to annuitize is another example of part of a retirement income plan that we should consider uncertain.

  10. Makes sense. I first ran across this floor and extra concept from Wm Bernstein's have 20-25 years of "drawdown" (my word) in "safe" products (my words) and invest the rest anyway you like. I think, when possible the "floor" + other approach makes more sense than the focus on overall equity vs fixed income allocation. Nothing is set it and forget it but this is a sleep well approach if you have enough.

  11. Thanks Dirk. I will take a look at the articles. I thought you believed more in TIPS ladders for guaranteed income vs annuities (due in part to default risk)? Brad

  12. Dirk
    Thank you for this article, I will need a lot of time to absorb this much data :-).
    It would be useful to apply Wade's Retirement Dashboard assumptions with your analysis.

    1. I asked Wade about the assumptions that went into his Dashboard. The model is somewhat complex, but he uses Shiller's data that shows a long-term CGR for stocks of 8.8% with a standard deviation of 17.9%.

      A 50% equity portfolio with an 8.8% return for stocks and 2.25% for bonds, as he suggests for year one, would return 5.375% nominal. IFA.com calculates an 11% standard deviation long term for a 50/50 portfolio.

      Wade suggests 1.2% inflation going forward, so we have a 50/50 portfolio with a real return of about 4.2% with a standard deviation of about 11%, compared to my example of 5.5% real returns with standard deviation of 11%, so the chances of lower returns with his estimates are greater.

      Plugging Wade's numbers into my model, about 1 in 6 scenarios returned a CGR of 2% or less, 1 in 6 returned 7% or more and 2 in 3 returned 2% to 7%.

      Hope that helps, but you will see a broad range of possible outcomes no matter whose projections you use and keep in mind that future returns are notoriously difficult to forecast. My point is that even if Wade's projection is precisely correct, and I'm sure he doesn't believe that it is, there is still a great deal of uncertainty in a 4.2% estimate with an 11% standard deviation.

  13. Your blog post makes me more convinced that the approach of allocating funds into buckets -- with TIPS or other 'safe' investments funding the "floor' bucket -- may be the way for me and the DW to go. So that leads to a question about your thoughts on dealing with the uncertainty surrounding life expectancy. Say, for instance, that both members of a couple are now 65 years of age, would you have thoughts on how they should think about (and or calculate) how many years of expected cash outflow to fund in their “floor” bucket? Up to just their joint life expectancy? Or less to start? Or more based on their guess about living past the average life expediencies?

    1. Excellent question.

      I prefer to plan with dynamic updating of several variables periodically, including life expectancy, expected spending needs, and expected market returns. Your life expectancy will decrease non-linearly throughout retirement and your plan should evolve to incorporate that.

      The floor portfolio is different, however, in that its purpose is to protect you from a catastrophic failure, such as loss of your savings portfolio or an exceptionally long life. I think it should protect you in the worst case, so you should assume a very long life for that floor allocation and probably include a healthy dollop of annuitization with fixed annuities and Social Security benefits (preferably delayed).

      I honestly don't find any value in a healthy person guessing how long they might live.

      Thanks for asking!

  14. Excellent blog post, Dirk in explaining the additional uncertainty about returns, and great comments too. Re: best age to annuitize, I like the idea of spreading SPIA purchases over 10 years or so, say every couple of years from 65 to 75. That may reduce financial risk somewhat, but it mainly reduces "regret risk" if there is a surprise interest rate jump after purchase. As for stock/bond mix in retirement, if one builds a reasonably safe floor, then the stock/bond mix relates to funds for discretionary spending. I like dynamic spending strategies that adjust based on investment experience by calculating withdrawals as a percent of the current portfolio (RMDs or slightly more aggressive percentages are examples). With a dynamic strategy the stock/bond mix choice reflects whether one wants high average discretionary spending that bounces around a lot from year to year (high stock allocation) or lower average but more stable spending (low stock allocation). It's good to see more folks coming around to the safety-first, floor/upside, life-cycle approach, although there still seem to be a majority of planners recommending probability-based approaches.