Friday, April 10, 2015

Time to Retire the Probability of Ruin?

The following post originally appeared on Advisor Perspectives, a blog for financial planners, in April 2015.

Perhaps no other retirement finance concept has gotten more ink than the "probability of ruin", which is interesting because few other economic concepts model the real world as poorly. The gist is that a retiree who spends a constant dollar amount throughout retirement is exposed to the risk of outliving her savings. The problem with the economic model is that no one would actually do that.

A constant-dollar spending assumption served us well when William Bengen formalized the notion of path dependence, but it seems that even he didn't anticipate that planners might implement the strategy by rote. In Conserving Client Portfolios During Retirement, Bengen states, ". . . the adviser should examine the projected current withdrawal rate through the entire time horizon of the clients, not just the first year of retirement." He even put it in italics.

Probability of ruin is largely an artifact of those constant-dollar SWR studies.

(Note: You can hover over yellow terms.)

Once constant-dollar SWR escaped the lab, though, the mainstream press and popular retirement newsletters latched onto the belief that constant-dollar spending was OK. These same publications softened their stances quite a bit after retirements were destroyed in the Great Recession, but I recently received a Kiplinger retirement newsletter still espousing the constant-dollar strategy.

Nearly all retirement income strategies avoid the probability of ruin issue in theory by spreading spending thinner and thinner as savings decline – a rational rationing strategy. Annual Recalculated Virtual Annuity (ARVA download PDF), RMD,  Milevsky's formula for sustainable withdrawals without simulation, and the 3D dynamic updating strategy (download PDF) of Frank, Mitchell and Blanchett all spread spending over remaining life expectancy. Constant-percentage SWR doesn't, but won't deplete a portfolio in theory, either.

Unfortunately, these strategies reduce the probability of ruin by reducing spending, our "standard of living."

Life annuities are different because they augment their payouts with a mortality premium and both guarantee a standard of living and provide longevity protection.

I was rereading a study recently about evaluating retirement strategies and noticed the author suggested that retiree's want an income strategy that guarantees they will not outlive their savings. While this is no doubt true, what we really want is an income strategy that guarantees we won't lose our standard of living. That's a bigger ask than avoiding ruin. A strategy that ignores standard of living but guarantees the retiree won't go broke is easy to develop, though perhaps not very useful.

Probability of ruin analyses are not good models of rational human behavior. Michael Kitces has often argued that no one actually implements a constant-dollar SWR strategy. If a retiree's portfolio grows significantly after retirement, any rational person would begin to spend a bit more and surely the reverse is true. What rational person would lose a large portion of their savings and believe they wouldn't need to spend less?

Wade Pfau says it a little differently: constant-dollar SWR is a research technique, not a retirement strategy.

Following this line of reasoning to its logical end, I have calculated the true probability of ruin resulting from consistent overspending with sequence of returns risk for nearly all retirees under nearly all economic conditions.


That's the probability I estimate that an actual retiree will just keep spending the same amount every year as she faces clear prospects of ruin. She will, instead, reduce spending and face the prospects of a diminished standard of living. This is a more rational model and the risk that we should be discussing. The probability of a reduced standard of living is more difficult to quantify than probability of ruin with constant-dollar spending, which is why we use the latter in the first place, but ease of calculation doesn't make it a useful metric. 

I have given this question a great deal of thought relative to my own retirement finances and I cannot imagine the impact of outliving my savings, nor can I assign a probability of doing so that I would consider "acceptable." I can say that my number is far closer to zero than to 5%. 

A small probability of a catastrophic outcome is not something that most of us can internalize. We only know that we want to avoid it.

Probability of ruin is a useful research technique that can allow us to compare the relevant benefit of two strategies, but translating that to real-life benefit isn't straightforward. The concept can rarely be communicated with a retiree, unless that retiree has an unusually strong understanding of probabilities, and even then an acceptable probability for losing one's savings late in life is just too big an ask.

I wonder if it's time to retire the probability of ruin, or at least drive it back into the lab. It's not clear to me that we are helping retirees by focusing on it.

Michael Kitces has posted two outstanding columns (more, actually) on Social Security benefits at his blog, Nerd's Eye View. The first, Valuing Social Security Benefits As An Asset On The Household Balance Sheet, was posted April 8, 2015. This led me to a 2014 post, How Delaying Social Security Can Be The Best Long-Term Investment Or Annuity Money Can Buy.  I highly recommend them, and Nerd's Eye View, in general.


  1. Enjoyed this article Dirk. As usual thought provoking (always have to read a couple times), and useful. I agree probability of ruin is not what I am looking for, but instead the probability of reducing my standard of living. One thing I have wondered about is whether most of the income strategies above that you mention (ARVA, RMD, MIlevsky etc.) take into account declining income needs as one ages. It seems that most I have seen assume constant dollar spending in real dollars when I believe that most retirees income needs will peak probably early in retirement then proceed down as you mentioned by about 3% per year. I believe there were several ways to increase the odds of not reducing one's standard of living in retirement vs the 4% rule such as adding small cap stocks, intl stocks etc.,and cutting back on spending. It would seem that the cutting back on spending may occur naturally as one ages, and I just wonder whether the models mentioned take this into account? Thanks, Brad.

    1. Brad, this is a good question and one that has come up a couple of times.

      Nearly all retirement income strategies calculate the maximum amount of your savings that you can safely (sustainably) spend in the current year assuming that you will spend that same amount every remaining year of your retirement. (RMD doesn't because the IRS' goal is simply to ensure that you spend most of your IRA and get taxed before you die.) None are really concerned with your expenses.

      Your desired expenses may be more or less than the maximum amount that you can safely spend from savings in any given year. You have to work that out with a budget, not a retirement income strategy.

      If you believe your expenses will be lower in the future, then of course you can spend a little more now. Two techniques that can consider future expenses are Monte Carlo simulation and consumption-smoothing.

      Retirement income strategies are principles upon which you can build a retirement plan and not a complete plan in themselves.

  2. Thanks Dirk, OK, I blew that. Can't mix expenses with retirement income techniques. Two separate things. I had to pull ESPlanner sheets, and review. I see where there are limited expense items included (housing, med part B, funeral), and then available discretionary spending to come up with total spending. I guess as long as my non-listed expenses are no more than discretionary spending amount then I am OK. Thanks again.. Brad

    1. That's how E$Planner works, you give it a few non-discretionary expenses and it figure out how much you can spend (discretionary) on top of that. E$Planner also has options to reduce spending consistently throughout retirement, or increase it consistently, or make it irregular.

  3. How right you are Dirk about Probability of Ruin. Kitces wrote a blog ( that reframes what you're talking about into something people may get their arms around ... the Possibility of Adjusting spending as well as the Possibility of Excess.

    Monte Carlo simulations look at what percentage of simulations fail to reach the end of the selected time period. It is a measurement of relative spending ability - and NOT something anyone should rely on, since the whole regimen of the simulations method is to discern a prudent point solution today based on a range of potential future outcomes - yet nobody knows which of those outcomes (or even an outcome not simulated) may result into the far future.

    However, repeating the exercise each year gives signals at to when spending may be getting dangerously high. Hint: the withdrawal rate has gone up - but how high is too high? The answer to the question was looked at in Frank, Mitchell & Blanchett's JFP paper Nov 2011 ("Probability-of-Failure-Based Decision Rules to Manage Sequence Risk in Retirement"). Each failure rate has it's own withdrawal rate, and vice versa. Given the same portfolio allocation characteristics, when the withdrawal rate goes up, so does the failure rate. But the differences between withdrawal rates narrows as failure rates go up - thus, there's a point when rising failure rates signals spending problems. Percentage of failing simulations is a valuable tool that provides more insight that relying solely on withdrawal rates alone.

    The use of failure rates from simulations is not quite the same as a probability of failure, although the terms are used interchangeably.

    Another point to be considered is the assumption that future economy's and/or markets are able to support a given standard of living. One assumes that would be true - or even better to have rising standards of living. However, it would be quite unlikely for workers to experience stagnant or falling standard of living, and retirees to expect anything different themselves. One can not get more out of the economy/markets than what they can provide in the aggregate. Thus, standard of living should also be evaluated in conjunction with events going on around - and adjustments made accordingly. Which gets to the point I also agree with, human behavior makes adjustments to everything - and thus, re-evaluating every year with signals established during the year, a.k.a., annual reviews, is important.

    A great post again Dirk!

    1. Thanks, Larry. And the economics point in you last paragraph is a good one to keep in mind.

  4. hi Mr Cotton, i was wondering if you have read this article by Mr Wade Pfau and perhaps blog a little of your thoughts about it >

    1. Yes, Kyith, I have read the paper. I suppose my initial thoughts are that it's a great start at a methodology to compare variable-spending strategies. I share Joe Tomlinson's concern that asking "if XYZ" ends up being too complicated a question for retirees. It is also important to note that some variable spending strategies show larger spending amounts in bad times because they take on more risk. But like all of Wade's work, it is a great analysis.