In a nutshell:

The authors compare five retirement withdrawal strategies (under the constraint that the retiree considers bequests to be of zero importance) and define a measure of Withdrawal Efficiency Rate as a means to compare them. They conclude that the commonly mentioned/used “Constant Dollar” (first year 4% of original assets, then adjusted annually for inflation) is the least efficient of the five considered, while one targeting a constant probability of failure with withdrawals continuously adjusted for remaining longevity and factoring in the effect of return history is the best. This should not come as a surprise to readers, since as a practical matter it should be obvious that “spending regimes that dynamically adjust for changes in both market and mortality uncertainties outperform the more traditional approaches”.

The details

For those of you who, like I, are junkies for papers on the subject of “withdrawal strategies” in retirement, Blanchett, Kowara and Chen have published an excellent paper entitled  “Optimal withdrawal strategy for retirement income portfolios”.

 The authors’ contributions are in a number of important areas:

-defining a “Withdrawal Efficiency Rate” (WER) “which measures the relative efficiency of various withdrawal strategies. The Withdrawal Efficiency Rate compares the withdrawals received by the retiree by following a specific strategy to what could have been obtained had the retiree had “perfect information” at the beginning of retirement.” Although past results do not necessarily reflective of future outcomes, i.e. determining which approach would have performed the best with return distributions based on past returns, there are still significant lessons to be learned about the (uncertain) future

-explicitly addressing the more often than not ignored  practicalities like the required adaptability over a 20-40 year retirement; they “…show that spending regimes that dynamically adjust for changes in both market and mortality uncertainties outperform the more traditional approaches.” It should not be a surprise to anyone that adaptability over a 20-40 year retirement will lead to superior results. The benchmark for the WER is based on that “If retirees knew the future return and the years they will live, i.e., if the retiree had “perfect information,” he or she (or they for a couple) would be able to determine the precise amount of income that could be generated from the portfolio for life, eliminating any uncertainty about a shortfall (running out of money before death) or surplus (not spending all the money during the lifetime).”

“For the analysis, a Monte Carlo simulation is created where life expectancies and returns are randomized.” The authors choose a conservative approach for historical return statistics (by reducing expected returns 50 bp and increasing standard deviation by 200 bp). The authors compare outcomes for five strategies and four equity allocations (0%, 20%, 40% and 60%). “Life expectancies… (are) based on the Annuity 2000 Mortality Table. The primary simulation will be based on the joint life expectancy of a couple…”

The 5 approaches simulated are:

1. Constant (real) Dollar Amount (“Constant dollar”): Withdrawal amount is determined by some percent (typically 4% suggested by Bengen’s insightful original work) of initial assets of the retirees which leads to “a fixed initial amount, increased annually by inflation”
2. Constant Percentage: (“Endowment Approach”), sometimes also referred to as  “Proportional Approach”) in which withdrawal amount is a “fixed percentage of portfolio value” in each year
3. Changing Percentage probability of failure fixed retirement Period (“Constant Failure Percentage”) in which withdrawal amountis based on maintaining a constant probability of failure over the expected fixed retirement period
4. Changing Percentage in which withdrawal amount is “1 divided by remaining retirement duration (life expectancy…based on survivorship experience)” called Life Expectancy Withdrawal Approach (“RMD Method”)
5.  Changing Percentage: probability of failure mortality updating (“Mortality Updating Failure Percentage”) in which withdrawal amountbased on “maintaining a constant probability of failure over the estimated remaining retirement duration…updated based on survivorship experience”

Conclusion

If my understanding of the paper is correct, under the condition of no bequest preference, the authors conclude that the 5 approaches approximately rank (lowest to highest) as follows (the table also shows whether each approach uses fixed or adaptable longevity and whether the approach adapts to market uncertainties):

The authors’ conclusion is that:  “Among the five withdrawal strategies considered and for each of the four different portfolios (equity allocations), the “Mortality Updating Failure Percentage” strategy, was the optimal withdrawal strategy; while the “Constant Failure Percentage” approach, was the second best in three out of four equity allocations considered. For three out of four equity allocations, the “Constant Dollar” strategy was the worst. Also, interestingly, the “Endowment Approach”,increases in relative efficiency for higher equity allocations, while the “RMD Method” declines in relative efficiency. ..The results suggest that the primary method employed by many practitioners, where a constant real dollar amount is withdrawn from the portfolio until it “fails” (called the “Constant Dollar” approach in this study) is often (actually it is only better at zero equity allocation, and only marginally so, than the “Endowment Approach”) the least-efficient approach to maximizing lifetime income for a retiree. The optimal withdrawal strategy points to approaches that incorporate mortality probability where the projected distribution period is updated based on the mortality experience of the retiree (or retirees) and the withdrawal percentage is determined based on maintaining constant probability of failure. This approach best replicates how a financial planner would (or at least should) determine the available income from a portfolio for each year during retirement. As a practical matter, (note that:) for retirees who can’t replicate the results presented here or don’t have access to them, the RMD method emerges as a reasonable alternative to the more common constant dollar and constant percentage of assets withdrawal strategies.”

Using the optimal strategy, Strategy 5, “the initial withdrawal percentages as a percentage of the portfolio balance for the 40% equity portfolios” for couples at ages 60, 65, 70, 75 and 80 start at 4.7%, 4.9%, 5.3%, 6.4% and 7.2% respectively.

The bottom line, whether you fully agree with the somewhat complex measure of efficiency or might choose a different measure of efficiency, is that the results are intuitively sensible (assuming you give zero value/importance to leaving an estate) and the more responsive the approach of withdrawal rate to changes in returns and remaining longevity, the more “efficient” the withdrawal strategy. Of course, depending on the importance of leaving a bequest the efficiency (and ranking) of various withdrawal approached might change, e.g. with a higher than zero importance weighting to some level of estate, the constant withdrawal rate (the Endowment Approach or as some refer to it as the proportional withdrawal approach) might become the most efficient approach

Subject to having potentially misunderstood some of the finer points of the authors, I do have some minor quibbles with some of the approaches/perspectives in the paper, but overall it least makes explicit and clarifies the pros and cons of approaches that some retirees and their advisors have been using implicitly. My first quibble revolves around the fact that the authors implicitly assume (and explicitly acknowledge as such) that the retirees assign zero importance/value to any residual estate as it is neither included in planning the strategy nor in valuing the residual assets; and while it is true that the bequest motive might be secondary for most retirees, I’d argue that the subjective weight assigned to bequests is more nuanced for most retirees. Second, if/when the authors assume that assets will be drawn down to zero (no bequest motive) would it not make sense to include consideration of some annuitization strategy, up front or as assets are drawn down, for comparative purposes? Third, one might also quibble the definition of the Withdrawal Efficiency Rate used compare withdrawal strategies based on the past return statistics, which as we know might not be predictive of the future. I would also have been interested in seeing how much improvement we would have seen with the Endowment Strategy when “floor-and-ceiling” constraints were added as described in my Vanguard GLWB vs. other decumulation strategies blog. (No doubt some others will also argue with the Monte Carlo method in general or with the distributions used in particular.) Whatever quibbles one might have, the paper should assign to its final resting place, the advisor practices based on the “Constant Dollar” approach!