How income may compare between Dynamic and Safe approaches

I recently attended an academic research conference for the Academy of Financial Services (AFS). I had the great pleasure of visiting with Wade Pfau, David Blanchett, Michael Kitces and Joe Tomlinson as well as the attending academics from the various universities across the country researching many aspects of personal financial planning. This article will discuss a few observations I made from discussions we had as well as from other paper presentations made at this, and other, conferences. Notes at the end provide further insights for the advanced reader. First some general comments.

• Of course it is unlikely for a couple to remain alive as a couple into their 100’s (but possible). At some point, only one of them will be alive and the cash flow values would adjust up because of the now shorter horizon based on just one age, not two ages. This would increase withdrawal rates, thus increasing cash flow/income (a point that most software doesn’t factor in for survival income replacement planning).
• Here’s a detailed blog discussing the Dynamic Updating approach shown in the figures (all values other than the Safe Withdrawal Rate (SWR), which in general starts at 4% and then increases the income each year by inflation, typically 3% each year.*
• Real dollars are depicted; thus, the cash flow from the SWR would remain constant in today’s dollars as well.
• I introduce a term, “ceiling”  as a notional concept representing potential upside spending relative to the floor, where spending could be higher at any given point in time – however, as you’ll see later, the trade off is less spending later because of the lower resulting balances later (even though withdrawal rates go up as one ages due to shorter time periods remaining).
  • Part of the Dynamic Updating process adjusts for the time period measured from the current age going forward, while all other simulations or calculations don’t adjust for this effect (except Ken Steiner’s work using a deterministic approach).
• As you age, you have less time left. Thus, you don’t need as much “principal” for the remaining years as you did for younger years. Therefore, it’s okay to spend down some principal incrementally over time.
• As you read along, please keep in mind that one is not actually tracking along a specific line, but within the bounds depicted by the various lines. The lines themselves depict later consequences of earlier spending choices and decisions. Life and global events are a lot more fluid than any calculations about the future may suggest.
• “Dynamic Updating” and “Safe” approaches are really just different points on the same spectrum – in other words bookends, or two sides of the same retirement coin. The main difference is that the safe method has ill defined adjustments with general decision rules as guidelines; while the dynamic method is purposefully built on adjustments for BOTH market AND longevity changes over time as the retiree ages.
  • Quoting Michael from our conversation on this, “if you don’t define the upside adjustments at all, you just end out with huge “inefficient” surpluses (a la the Sharpe critique of SWR), but I’ve never viewed that as a real criticism; the upside adjustments are ill-defined, but for planners who meet regularly with clients, at some point I expect they WILL adjust upwards. We just need to do a better job (under the “SWR” approach) of defining those parameters to make “safe” upwards adjustments …”
• Kitces discusses the Floor with Upside approaches in his blog. The below is a floor with upside discussion from the Dynamic Updating point of view … the biggest difference here though is that the Dynamic Updating method may actually measure how far up the upside may go, as well as see the consequences for spending more earlier (e.g., Go-Go line below). The model supports intuition that says if you spend it now, you don’t have it to spend later. The SWR approach doesn’t have a good way of defining or measuring much of anything once started (as Kitces eludes to in the point above).

Figure 1

The ceiling line, in Figure 1, labeled 50th Percentile refers to the middle range of simulated cash flows. The 75th Percentile looks at what may be possible IF returns are consistently poor. Higher income may be possible IF returns are consistently better than 50th percentile, but are not depicted here. In actual life, one could choose how much income they would want between any of the depicted values, including taking more as shown in the 10 Year Go-Go years – which clearly shows less income later because of reduced portfolio balances.

Two equity allocations are also depicted, a balanced 50% stock/50% bond and a conservative 40% stock/60% bond.^ It may be tempting to interpret Figure 1 ending points as suggesting that the balanced allocation (higher stock exposure) is better than the conservative allocation because it shows more income later. However, if you look at Figure 2, it becomes clear the reason why is that the withdrawal rate for a more volatile allocation is less!

There is a volatility drag because more simulations would fail due to higher volatility, so the simulations determine a lower withdrawal rate when you seek the same failure rate between both allocations (failure rates are how you compare apples to apples between withdrawals). The effect therefore, it that less is spent due to higher volatility and more remains in the portfolio to continue to grow. Of course, the higher expected returns of the higher allocation then also serve to build that balance more, so even higher income is potentially available in those later years. What happens under the SWR regime, is that balances under most scenarios grow even more and become a bequest because spending is constrained a bit more under the method. Under the dynamic approach, bequests do happen as well, because there is always a portfolio balance to support income into the older expected ages. They would simply be smaller bequests, unless designed to be larger by less spending along the way.

Let’s look at an example: From the slide share numbers # discussed in this blog – lower left corner in slide –most of the time monthly income may be as high as $3867/mo. for the upcoming year (it would adjust in subsequent years based on facts known then). However, should markets decline during the current year (illustrated in yellow, orange and red bands), then the withdrawal rate in percentage terms (WR%) would rise with a rise in the signal from the related Possibility of Adjustment (POA)** (one can’t tell when the WR% becomes a problem – the POA is a clear signal related directly with WR%).@ If one foreshadows spending retrenchment and pre-calculates what the floor may need to be, in this example $3477/mo (far right of red line values in box) … then this is a floor value for  the expenses target, and the amount between the floor and the upper ceiling value ($390 – or 11% of floor).

In other words, when markets go down, ALL the expenses don’t need to be cut – only the fraction that is above the floor. The further one goes above the floor, the greater should be the expectation that some expenses may need to be reduced when markets decline. In this example, 89% of the expenses are still “safe,” at least to the extent that the markets and economy don’t have a greater shock than 2008; in which case spending wouldn’t be “safe” in either paradigm! You can then see then that 89% of potential income  are really 100% of the floor expenses and these won’t need to be messed with (the blog demonstrates how this value relates to the SWR mathematically) typically because these values relate very closely with the SWR value. Thus, 11% of the expenses should be, by definition above the floor, and may be spent or not spent depending on what has happened during that particular year.

So as a practitioner, and a deeper insight into measuring and monitoring boundaries, it then becomes important to understand what the client’s situation is so that one who under-saved is not inadvertently put into the position of having to cut spending when they’re already pushing their spending limit constrained by the lower spending floor value. Those who have saved sufficiently, or even over-saved, typically don’t face issues (gets to Meir Statman’s “Precarious Middle” demographic) and may not need to spend all they could anyway (see Different strokes paragraph below).

Figure 2

What is the effect of spending more early?

Figure 3 expands on the early retirement years. After the 10 year period, the withdrawals revert back to the conservative allocation, chosen because the conservative allocation, counter intuitively, has higher withdrawal rates leading to more cash flow (see Figure 2). However, if the retiree desires to switch to lesser spending to preserve balances, that is an option too. This gets to the point that spending is not set in concrete and may vary within the ranges depicted in the figures.

Notice in Figure 3 that the starting cash flows are higher for the conservative (40% stock) portfolio – one would think the higher stock allocation in the balanced portfolio would provide more income early on. Volatility drag mentioned above affects the withdrawal rate as shown in Figure 2. Implied portfolio balances for the Dynamic values may be computed by simply taking the annual cash flow any given year and dividing by the withdrawal rate (Figure 4 for specific numbers). Here it becomes clearer that the reason an 85 year old couple may have a higher annual spending potential with the balanced allocation ($73,253 / 6.69% = $1,094,962 balance) relative to the more conservative allocation ($65,353 / 6.86% = $952,667). Again, this is because the higher allocation forces a lower withdrawal each year (Figure 2) due to volatility drag.

Figure 3.

Different stokes for different folks.

Circling back to the AFS conference, one of the research paper presentations showed a propensity for those who saved before retirement to continue to save during retirement. This manifested itself by them NOT spending all that they COULD spend. Those who did not save before retirement suffered a consumption gap where they were short of money. {A link to this paper will be put here once it becomes available in AFS Proceedings}. This finding suggests then that all retirees are not created equal in need, where a deeper understanding of spending throughout a lifetime is useful. A balanced allocation would result in a higher balance later that may help offset future income gaps … however, in the near term, because of the lower withdrawal rate relative to a more conservative allocation (Figure 2), such a strategy would require some relative belt tightening in the earlier retirement years. In other words, more equity exposure results in higher income potential later on (Figure 1), but less relative income earlier (Figure 2).*** Under a Go-Go scenario where they could get more income early on, possibly applied for example to, partially of fully, fill the income gap until Social Security maximizes better for the couple, they’d have less relative income later (Figure 1).

Figure 4 (data for figures 1-3).

Notice in Figure 4 that the initial expected retirement time period at age 60 is 33 years (couple of the same age using Annuity 2000) – yet when the retirees reaches age 93, there’s still a possibility of 11 more years! The longevity table at the bottom of Figure 4 shows the percent of cohorts expected to outlive (i.e., percentile) time periods at those ages. For example, 50% of cohorts may expect to outlive 33 years at age 60, while once at age 93 20% of those cohorts may expect to outlive another 11 years (as couples). Again, life tables for individuals are different and would apply when one of the couple dies, or would be used immediately if the person were single. The time periods would be shorter, so withdrawal rate and cash flow would be higher relative to couples. Males would compare “better” to females for the same reason (do you want more money, or more life?).

The “Go-Go” years are developed by looking at a time period where 70% of cohorts at age 60 (65% at age 65). Since more are expected to survive, this means the derived time period is shorter and the resulting withdrawal rate is higher = higher cash flows for those years. Of course, the expectation is to outlive those years! Thus, cash flow needs to be reduced eventually or the money is also outlived! The dynamic model illustrates this in Figures 1-3.##

Longevity is dynamic as well. All data, and life itself, is stochastic!

Moral of the Story: The Dynamic Updating approach illustrates the choices that may be desired throughout retirement and what choices are possible once the pros and cons of decisions are more deeply evaluated year by year. Intuition is often not the same as what may actually turn out and, as can be seen, many aspects of withdrawals for income are counterintuitive.

Safe withdrawal rate approach and the Dynamic Updating approach are both sides of the same retirement income coin.

Notes.

*Other early work on the dynamic withdrawal retirement income approach was done by my co-researcher Dr John Mitchell: Dynamic Retirement Withdrawal Planning Academic journal article By Stout, R. Gene; Mitchell, John B.Financial Services Review , Vol. 15, No. 2  , Summer 2006.

^A spectrum of allocations is possible and our research looked at the full three dimensions (see link in @note below), from all stock to all bonds/cash. For easier interpretation of the Figures here, only two are depicted.

#Note that the ages, allocations, income amounts, etc in the slide share do not correspond to Figures 1 – 4 numbers. The objective here is to understand concepts that must be applied to each individual situation to arrive at your numbers.

**Possibility of Adjustment (POA) is a Kitces term I like to use with clients because it conveys the concept of adjusting spending better than how simulations determine this though measure the percent of simulations that fail to reach the end of the period simulated (POF).

@The main point is that WR% alone is a POOR method for comparison of data. Why? From our [David and my] JFP June 2010 paper: “To represent all the variables incorporated into a withdrawal percentage value, a more complete withdrawal rate expression would be WR% (N, I, i, s , μ, t, Δt) (values in ( )  to left should be in subscript, however the blog software limits use of subscripts)   where, for the portfolio, N = number of distribution years of the portfolio simulation, I = inflation, i = rate of return,  s= standard deviation, μ = probability of failure, t = time set at a fixed target end date, for example age 95, and Δt = change in time (years) to represent aging of the person. There are two “time” functions to represent the number of years for the portfolio, and the number of years of the person.” In other words, we [Mitchell, Blanchett and I (4th, 5th and 6th papers)] found that POF (μ  component) is a decent, and sensitive, signal for better withdrawals decision making.

***This is the effect Pfau and Kitces see in their rising equity glidepaths, and also why T-Bills improve results (the volatility drag is reduced using T-bills versus other bonds – the longer term the bond, the more equity like the volatility). Under a Dynamic Updating approach when one is above the floor with spending, the rising equity glidepath is actually muted back to a lower allocation because the shorter time periods remaining results in lower volatility requirements when measuring POF again (this is also in Pfau/Kitces data for shorter periods – but not reported in their papers because they didn’t factor in the effect of aging into those short periods).

##The process, each year, used is 1) Determine the length of the time period from life tables, 2) set the desired POA (POF) threshold, 3) retrieve the resulting withdrawal rate and apply it to the current portfolio value that year.

Our work is described in this series of postings for those interested in gaining a deeper understanding.