So, what would it look like if we projected some constant spend rates starting in 1969 and compared them to some benchmark? I realize some of this has been done before, notably by Bengen in 1994 and the Trinity Study in 1998(?) and countless others thereafter. But I will try again because, like a tongue that keeps seeking the problem with a flawed tooth I can't help myself. I need to touch it again.
This time I want to run three spend rates (3, 3.5, and 4%) against the market data of those times. I will constant spend (set the spend at year zero and inflate at hist. rates thereafter). All of this will be done from the perspective of a 58 year old, partly because this is personal and partly because 58 stands in for a mid-range estimate for an earlier-than-average retirement. Since this is a fast back-of-the-envelope kind of thing given the amateur-hack approach I've used in the past there will be methodological flaws that would bother even me if I had more time to do it better. But let's proceed anyway. The other thing I want to do is to focus less on math and rationality and more on psychology and behaviors. How would it have felt? What would I have done? The goal is to see what happens to the "effective" spend rates if we constant spend and to consider how we might have reacted to them. I just finished Andrew Lo's book Adaptive Markets and he makes the case that finance (and markets and people that interact with them) are more about biology than physics and the biology is driven more by combinations of ideas like fear (emotion), selection, reproduction, and adaptation than it is by immutable rules as in physics. I'm thinking that is mostly true and this way of looking at things will have a good place in future retirement finance developments.
Let's do this:
1. Get some data: I took market data (SnP and short and long term treasuries) from Aswath Damodaran at Stern and from inflationdata.com. This was annual data not monthly and a quick and dirty approach to running the data.[1]
2. Spend: For a given 2Asset allocation, I ran a constant inflated spend against market returns while inflating spending at the historically recorded rate. Spend for better or worse is drawn in proportion to the current, end of prior period, allocation (probably worse?).
3. Chart: We'll chart the effective spend rate over time. This is the spend rate if we did not change the inflated spend and divided the current year spend "budget" by the end of last year's wealth. This is unrealistic because I am ignoring taxes and fees for now and certainly not going dynamic with spending...which we should.
4. Create a benchmark. Since we need to compare to something and since the comparative spending needs to reflect expectations like returns, risk-aversion, inflation, and adjusted longevity expectations, we'll plot some simple results for age-adjusted, risk-aversion-adjusted, and continuously recalculated spend rates using two simple tools. First I'll use Blanchett's "simple formula" for withdrawal since he's a recognized authority and I trust his judgement. Second I'll use the RH40 formula because, well because it's mine. I could do others but again this is a fast back of the napkin thing. I showed in the past that these two approaches are "ok" for very conservative spending at various ages...search rh40 on this site. There is no connection to periodic levels of wealth here, this is just a generic age-based suggestion as if we were starting over in that year.
5. Add a late age RMD-style benchmark. Since even Blanchett recommends that with his own equation an RMD style calculation is probably better with 10-15 years left in a retirement horizon, and since Gordon Irlam at AAcalc.com likes a very similar RMD-style approach that I believe is more related to Merton math than it is to simulation, we will also, late in the retirement sketch (15 years to go when bench-marked against age 95) throw up some RMD style calcs for reference. The trick here is that the basic idea is the RMD-style withdrawal rate is something like W/e where W is wealth and e is longevity. Forget that W is tricky because it includes the PV of future income as well as assets, focus on the idea that "e" is hard. How many years? That depends on how one looks at it. In this case I will use longevity estimates from AA calc based on three assumption sets: 1) average health, 80th percentile expectation, SS2013 data, 2) average health 95th percentile expectation, SS 2013 data, and 3) above average health, 95th percentile, Society of Actuaries Individual Annuity Mortality table data.
6. ...Then we'll "look and feel." Forget ruin rates and probabilities of success and years solvent and rational left-brain stuff for now. Let's just look at the chart and guess how we might have felt as an early retiree.
Given all that, this is what it might look like if I'm even close to getting everything right:
What do we have here then?
- Green lines: these are the constant inflated spend rates for three levels (3, 3.5 and 4% at the outset) but displayed at what the effective rate would be in each year if the inflated spend is divided into any year's current wealth level (end of prior year).
- Black Lines: these are the results of running Blanchett's simple rule but recalculated each year. The allocation assumption is 60/40 alloc, fees are put to zero here, the upper line reflects a success expectation of 80%, the lower line 95%. The terminal age is 95 except that at age 85 I start moving that out a bit to a later age (say 105 by 105) to mimic actuarial expectations a bit. Not scientifically, just a bit. Without that it would just curve up a little bit more sharply near the end so it probably doesn't matter much.
- Dotted Line: My RH40 calc: withdrawal = Age / (40 - Age/3), recalc-ed each year. Just for fun.
- Blue lines: These are RMD-style spend calcs when retirement duration longevity expectation is less than 15 years where the spendrate = W (just net retirement assets here, no PV calcs of SS or anything) divided by "e" (longevity expectation). The top blue line is based on the x-axis age and uses SS2013 data, average health, and an 80th percentile longevity expectation. The middle blue line is the same as the top but has a 95th percentile longevity expectation. The lowest line uses SOA Individual Annuity Mortality table data that reflect better health expectations and a 95th percentile longevity expectation.
What can we make of this?
1. Let's start with the 3% constant spend. This is a safe spend rate and we more or less know from common sense and the chart that it will likely work over the long haul (not necessarily in our real future, mind you, just "likely"). The thing I notice here, though, is that it is well above a (lower) benchmark for something like 17 years. That's a long time. My guess is that would have been very uncomfortable and would have evinced some bad behavior at times like panic selling. The peak effective rate was 6.1%. Knowing what I know now (max spend rates doing the math are typically lower than 6% in mid-term retirement years I recall from somewhere, though not every article I've ever read agrees with that) that would have made me pretty nervous. I also notice that by the time I get old I am in a better and very conservative pattern and can probably loosen up a bit though maybe its a little late to still have as much fun as I would have in my 60s.
2. 3.5% spend: Ok, so I am above a safe benchmark for something like 28 years here. The peak rate in the first 24 years is over 8%. Of course it works but really, 28 years? 8%? If I had been aware of that in this hypothetical world there is no way I would have let that be status quo the whole way. Something would have changed (most likely spending). I didn't model it but note that panic selling and going to bond-heavy would not have helped (quite the opposite). On the other hand, at late age, it again looks like I can again loosen up a bit.
3. Now let's look at 4% spend. Sure it ends up "surviving" around thirty years (in my flawed run and in Bengen '94 and Trinity '98 which I have not exactly finished reading) but who, among those who are younger than 65 when they retire, especially in the current world, assumes thirty years anymore? And who want's to be in "maybe I'll survive" mode that whole time?[2] This is playing with fire both economically and emotionally. Look at the chart: the effective spend rate is well above a "safe benchmark" for the entirety of retirement and pulls away fast. How do you suppose that would have felt? How would you have behaved? This is sustainable on paper for 30 years but I doubt anyone would have stayed the course on this; I would have bitten my nails the whole way and lifestyle changes would have come at a steeper and steeper price as the years passed. This is not a course for the faint of heart, certainly not for me.
4. I notice that when we kick into the final stretch at age ~81 we are not even then really home free. The Blanchett suggested RMD-style spend (blue), given average health expectations and 95th percentile longevity expectations (middle blue), comports quite well with the other benchmarks. But minor changes in longevity/health expectations have large effects on the RMD-style spend rate. If you are sick, spend like crazy, I guess, otherwise hoard like everyone else might.
5. The last thing I see is that for long/early retirements it is clear that they are different than short/late retirements. That early (over)spend can get away from you fast and is solved mainly by three things short of lottery wins and going back to work: an earlier than expected death, spending control, or (not shown) a higher equity allocation that can maybe dig you out of a hole...or just as easily leave you there or even worse off. At late ages, there appears to be a little more latitude. Unfortunately it may not matter much then due to things like death, ill health, or lack of mobility. As the chart and its benchmarks make clear, a better game might be to play conservative early, test the wind each year, triangulate using whatever information is currently available and then maybe loosen up as the distance to the end shortens...even though it'll be less fun than it would have been 10 years before.
------------------------
[1] I have a little self doubt on my sources, exact type, and periodicity of data as well as my methods of use. Others can gin up longer runs than 30 years and others can sustain higher spend rates. My heart is pure though. If uncomfortable I guess we can just view this some kind of funky one-run simulation using customized data in order to illustrate a point.
[2] I'm also wondering who exactly plans and executes static vs. dynamic plans these days. Rational behavior, without any attempt at attribution on my part, has always been dynamic.
No comments:
Post a Comment