Wednesday, May 8, 2013

Dynamic Retirement Strategies - Inflight Brainstorming

There is an important distinction between static and dynamic retirement strategies. Static strategies belong to a "set it and forget it" group. A basic example would be the 4% rule, which suggests withdrawing 4% of one's assets at the retirement date, keeping track of how much income this represents, and then adjusting this income amount by inflation in subsequent years to determine future withdrawals.  What I investigated in my recent article on the efficient frontier for retirement income was also static in nature. I assumed that the retiree made all of their decisions at the retirement date (how to divide financial assets between stocks, bonds, and different types of annuities) and stuck with this decision for their remainder of their retirement.

In contrast, dynamic strategies allow for changing plans throughout retirement. I have a bit of trouble defining what a dynamic strategy is. For example, a constant percentage strategy, in which one decides at retirement to withdraw 4% of their remaining portfolio balance in each year of retirement, could be considered a dynamic strategy since future inflation-adjusted withdrawal amounts fluctuate dynamically in response to the market returns and past withdrawals taken from the portfolio. At the same time, to me, deciding at retirement to withdraw 4% of remaining assets each year does sound like a "set it and forget it" type approach as well. Nonetheless, it is dynamic in the sense that the ongoing realized experience of retirement will impact decisions in a way that doesn't happen for a constant inflation-adjusted spending amount strategy.

But there are other types of strategies, which while still representing behaviors which respond to a set of decision rules developed at the time of retirement, could really be considered more dynamic in nature since there is no telling what will happen. Blanchett, Kowara, and Chen (2012) developed the most advanced version of this type of decision rule framework in relation to taking withdrawals from their portfolio with their "mortality-updating constant probability of failure" model. With this sophisticated model, withdrawals in each year of retirement are determined by remaining life expectancy and an updated portfolio value, with the idea of choosing a withdrawal amount each year that will maintain a constant probability of failure looking forward from that point of time. 

Another important area where dynamic strategies can play an important role relates to the decision of if and when to annuitize, with an effort to combine systematic withdrawals and annuities into a more complete retirement income framework. This is an area which has been explored quite extensively in highly mathematical academic studies, but the findings of such studies have not yet received detailed scrutiny from the wider investing public about the applicability of the assumptions found in the models. Nonetheless, without getting into specifics about assumptions and so forth, a general finding of such studies is that the annuity decision should indeed be dynamic, with chunks of annuitization taking place throughout the retirement period in response to the updated situation faced by retirees. These studies also tend to find that for any one-time annuitization decisions, it is often better to wait until one is in their 70s.

When thinking about dynamic retirement strategies, there are two different ways to try to simulate strategies to determine which approaches may be superior. The less sophisticated approach would be to test different sets of decision rules in a somewhat ad hoc fashion using Monte Carlo simulations to see what comes out best based on the outcome measures chosen by the researcher. A more sophisticated approach involved using "dynamic programming," in which one works backward through retirement. Determine what the optimal decisions are at the end of retirement, and then work backward to determine what is optimal earlier on based on the probabilities of different subsequent outcomes and how one would optimize in those outcomes. In this approach, you may act differently now when you specifically incorporate how you might respond in the future to the realized outcomes of your retirement. 

But without getting into greater details about the modeling approach, it is somewhat clear that dynamic retirement strategies, both with regard to the amount to withdraw from the portfolio and with regard to decisions relating to when and how much to annuitize, will be optimal to strategies in which future decisions are unrelated to the ongoing experience of the retiree. Dynamic decisions about annuitization must also incorporate the irreversibility of the decision and the real option value of waiting, and so it is a matter of figuring out the types of conditions that should trigger annuitization. The conditions experienced at the start of retirement, such as the extremely low bond yields facing today's retirees, could also impact the optimal decision rules for a dynamic retirement income strategy.

Now, in an effort to brainstorm, what all should be incorporated in a modeling approach which seeks to determine an optimal dynamic retirement income strategy? Here is an effort to start a list.

First, the decision rules to guide a dynamic retirement income strategy should allow for a way to make updated decisions each year about the appropriate asset allocation,  how much of remaining financial assets to annuitize (and which of the many different types of available annuity products to use), and also how much insurance to have for long-term care, health, and life.

Those decisions should relate to updated information available about:

-the remaining portfolio value of financial assets, which is influenced by realized market returns, asset allocation, and past withdrawal decisions

-beliefs about appropriate future assumptions for market returns, which may be influenced by past realized market returns, and which also will affect assumptions about the future evolution of annuity prices

-an updated budget for spending needs over the remainder of the retiree's life, which could be impacted by unexpected surprises in one's past budget (such as new unexpected expenses), updated realizations about how future spending needs may change, and some sort of hedonic adjustment which gets at the idea that people may get used to whatever their current budget is and have a harder time making future reductions (which would imply greater caution about increasing one's budget in response to good market outcomes)

-updates about health status and remaining life expectancy both for oneself and for society (which affects future annuity prices)

-some sort of reflection about one's relative position in society. For instance, if there is a big market drop, then many people may tighten their belt at once and so the relative impact is less severe. Does one have a strategy which will mimick societal changes or which is contrarian in nature? What I mean, for example, is that complete annuitization may cause one to fall behind in a relative sense when markets go up, but would also cause one to experienced increased relative wealth when others' portfolios drop

-a retiree's cognitive skills are important, and any dynamic strategy which moves away from a "set it and forget it" approach must incorporate the notion that retirees will experience increased difficulty to implement complex dynamic strategies as they continue to age

-the tax basis for one's investments as well as one's tax situation will continue to evolve throughout retirement

-risk aversion could change throughout retirement as well, and is not necessarily constant. Finke, Pfau, and Williams (2012) talks about risk aversion in retirement as it relates to "spending flexibility."  That also relates to the idea of risk capacity, which gets at how one is able to ride out drops in their wealth. When dynamically updating a retirement income strategy, it is important no monitor changes in this risk aversion. Some of the factors relevant here include how a retiree feels about the idea of outliving their financial assets (longevity risk aversion) and what income would still be available if that happened, any changes in the volatility of unknown future spending needs (such as the need to care for other family members, to cover out-of-pocket health expenses, in some sense the potential losses which could still be experienced by one's investment portfolio, long-term care needs, and potential for any voluntary job loss if one is still relying on part-time labor income).  Also relevant here is a psychological matter of how adaptable one feels about the idea of reducing living expenses if necessary.

It's a tall order, and only some of these matters may be incorporated in the financial planning software available to help today's retirees, but if researchers can keep plugging away at it, we will hopefully get to the point where more dynamic retirement income strategies can be vetted and compared on a basis that fulfills the needs of real retirees and their unknown future experiences. 


  1. Hi Wade,

    An excellent summary! I think though that the idea of coming up with a model that will remove decisions and predict the future is a Holy Grail. Tough choices are a part of life even in retirement. Predictions about the future are also cyclic and tend to lag what the past has thrown, i.e., there are no predictions that would last a retiree’s entire remaining lifetime. We humans are also biased by what we perceive to be true when in truth some things haven’t changed in the retirement arena. Here’s an example of perceptions on pensions:

    This said about predictability, the observation does not invalidate research or the models. Many tend to ignore or discount the conclusions and observations of research because the software doesn’t exist on a wider basis for all to use. This throws the baby out with the bath water. The dynamic model in Frank, Mitchell and Blanchett ( series of papers in an example of a model that replicates a series of annual decisions a retiree may make over the remaining years of their lifetimes. The software that replicates these decisions is not available, however the impact of earlier decisions, observations, conclusions and practical applications are still applicable. Because you see, the retiree goes slowly year by year through what the model replicates in an instant.

    Advantages a dynamic approach has over the static approach are 1) primes the retiree to realize spending and allocations decisions will still need to be made in the future, just as they were in the past during their working years when income, economy and markets changed; 2) problems are spotted earlier than later and adjustments tend to be smaller compared to those needed when a decision is delayed; 3) both 1 and 2 above help if markets and the economy don’t recover as fast as assumed based on past history (future tends to be different in unexpected ways).

    Now, none of this removes what you say … you are spot on … I just wanted to point out that the application of research conclusions is still possible using more widely available software that replicates just one year at a time. However, without knowledge about how present decisions affect what is possible in the future, the use of software that looks only at the present year has been difficult. Frank, Mitchell and Blanchett (March 2012; link above) and your reference to Blanchett, Kowara, and Chen (also in 2012) (Blanchett the common thread; and your description of models applies to both papers; differences are model construction and practical applications).

    The difference between the two approaches is that Frank, Mitchell and Blanchett approach can be replicated one year at a time using current Monte Carlo software with the number of years determined by reference to Period Life Tables (either the 2000 Annuity Table, now 2012) or Social Security Table). Knowing that a high simulation percentage of failure rate is an early signal for spending too much at the moment, preserves the portfolio value with a small spending adjustment so that future cash flows may be higher than they would be if higher spending now were to be continued. This is logical … however before the relationship wasn’t as clear when models only used single simulations because the relationship between cash flow and portfolio balances can’t be explored in a single simulation model (static), rather than stringing together a series of simulations in a model to investigate outcomes in the future based on decisions today (dynamic).

    An excellent post Wade that discusses, and continues the dialogue, about how to use observations and conclusions from ongoing research, even though software may not have caught up for use by the wider practitioner and public groups. Best wishes settling in out East at American College!

  2. As a proponent of what you refer to as a dynamic withdrawal strategy, I encourage you to pursue research on these types of strategies. I also agree with Mr. Frank's comments that there is no perfect approach and tough choices remain during retirement. The approach I advocate in my website produces a budget (not a required withdrawal amount)that a retiree can choose to ignore (hopefully by making an informed decision to do so). I also believe that it is important for the dynamic strategy to include some sort of reasonable algorythm to smooth the impact on next year's spending budget of the many changes in assumptions or circumstances that may take place since the prior year's evaluation.

  3. Wade, I would hope to see some analysis and discussion of different levels of spending at different stages of retirement. One of the concerns I have with the approaches normally discussed on this blog is that they will lead to underspending in the earlier part of retirement when one is more active as opposed to later (i.e., in your 80s) when spending levels are likely to be lower. How does one take this into account in formulating a dynamic retirement strategy? Thanks

    1. Thanks for the comment.

      I have taken a look at these sorts of issues. First, about the idea that spending may decrease with age as people slow down, please see this column:

      Secondly, about the issue that people may choose to spend more early and less later, please see this one:

    2. Hi Wade, if I may respectively add Thirdly ... a paper on how to use Monte Carlo simulations and the period life tables together to measure either goal of "Consumption Orientation" (spend more money early on)or "Inheritance Orientation" (bequest/or spend more later) .

      Gets to the fundamental question "How does an adviser/person know which one they're actually doing?" Different people have different spending goals when retired.

  4. Sorry to go off topic here, but I don't see any other way to ask this. Have you looked at any of the work done by Stacy Schaus at PIMCO on the risk capacity of pre-retirees and the glide path of most target date funds vs PIMCO's recommended glide path and asset allocation? If so, any comments on PIMCO's recommendations that pre-retirees would have 65% to eventually 85% of their portfolio in fixed income investments (albeit diversified over 8 subclasses)? It seems to fit in well with the retirement portfolio of immediate annuities and stocks once in retirement your work suggests, but not so well with the traditional recommendations of at least 50% in stocks to support the traditional 4% withdrawal. Looking forward to your next round of studies. Thank you for sharing your work.

    1. Hi Kathy,

      I haven't read the specifics of PIMCO's recommendations, but a few years ago I did some research on glidepaths during the accumulation phase. What you've written is in line with what I was coming up with for conservative investors if you are talking about retirement date allocations, though closely to 65% than 85% fixed income. After retirement begins, everything I've seen and done suggests there is no need to keep reducing the stock allocation.

  5. Thanks for replying, Wade. I was focusing more on the pre-retirees that I work with - people who expect to retire within 3-5 years. We've been working to figure out what their retirement income might be based on their current portfolios. If they have to retire, we'll all feel pretty badly if we have to cut their retirement income by 10-20% because the market decides to crash just before they retire.