If one were to assume the risk free rate was around 2% -- which is a sketchy assumption since it was maybe more like 0 over that time frame -- then an optimal portfolio allocation would have been that tangency thing somewhere around 40% allocated to SPY and 60% to AGG before we get into talking about being levered or not. So let's call it 40% to a risk asset. Is that an optimal portfolio going forward? Who knows. Probably not. We'll never know until we get there and can look back. [that green dot is a trading strategy, btw, and not part of this post]
But that way of thinking is the thinking of the very young or institutional investors. It is optimization for people who have long or infinite time-frames and/or no consumption requirement against the portfolio. It is for people that are convinced that risk is the same as variance and those who do not have a palpable fear of lifestyle degradation or running out of money in their one "whack at the cat." It is, in other words, not the way a retiree thinks.
I noticed this shift a few years ago when I realized at some unremembered moment that I was not managing a portfolio for abstractions like efficiency I was managing it for a purpose. Or, as Michael Zwecher puts it "Retirement requires outcomes, not just expectations...an exclusive focus on risk-return implies consumption foregone rather than consumption deferred...[it] ignores why we create portfolios: to make the most of our deferred consumption." So, the optimization game had changed on me while I wasn't watching. Now, if not risk-return what was I optimizing for? I suppose the following is the starter list:
- Not running out of money before death
- Limiting the magnitude of failing in terms of time spent underwater
- Avoiding moderate to severe degradation of lifestyle even if I don't run out
- Sustainability of lifestyle but also with some degree of positive confidence about it
- Keeping an option open for gains in prosperity
- Keeping an option open for small to medium bequests
- Keeping reserves available for big unexpected expenses that affect the prior points
- There may be others...
That's a nice list but short of going back to paid employment (still an option but not forever and fading fast) or harsh spending cuts and a grinding vigilance (a retiree's lot at some point, perhaps) how does one get to where one needs to go when thinking only in terms of managing the portfolio? Me? After five years of reading and studying this stuff I had all sorts of competing concepts and narratives running around my head that are tough to reconcile. There are no small number of articles and papers that claim high fixed equity allocations are optimal if not necessary. Others disagree. Some have glide paths up, others down. If simulating a retirement and optimizing the sim for success rates the allocation percent can be in a high range (50-80%, say) while optimizing for a minimum in the number of years in a "fail" state can say "low" (maybe 20%). Retirement income professionals advocate floor-and-upside portfolios where the allocation depends on how you look at the problem: if 70% (or 80 or 100) of your portfolio is committed to a floor then perhaps 100% of the excess is in risk assets (or it's really a 30/70 risk position, right?). Economic theory by way of techniques such as backward induction, with or without utility functions, says "it all depends" (it would say that wouldn't it) and the allocation will vary based on year of plan, spending expectations, and size of portfolio when optimized working backwards from the endgame. My own courageous approach, in a post-2008 cringe or crouch or whatever you want to call it was to retreat from a 70-30 allocation to something more like a 40-60 or 50-50 risk position (ignoring some weird alt stuff that skews it a little) and to avoid the problem altogether. That retreat, btw, was actually mostly done in the first quarter of 2007 (avoiding almost all of the fall for at least some of my assets) in one of the few feats of timing genius I can ever truly claim.
So now what? MVO, for those that use it, seems (seemed) to say 40% risk-assets, at least when we look backwards. My current position on allocation, through nothing other than indolence and a still lingering recession-induced financial PTSD, is 40% give or take a few points. Everyone else says, well, they say just about everything else. Let's think this through, then, with some, but not all, of the tools I have -- using a retiree frame of mind -- and see what we find. (there are no doubt an infinite number of ways to go about this, I'm just picking this particular path to see where it goes)
First, even though the RIIA might tell me to create a floor and annuitize a whole bunch (maybe all) of my assets, as an early retiree I'm not quite ready for that so we can take that off the table for at least a couple years. That leaves standard portfolio management and asset allocation (and work and spending control for that matter...). Now, just for fun let's use simulation as one of our main tools in going down the path I am taking. There are all sorts of reasons to dislike or disdain simulation for retirement planning but we will leave that to the side for now. I went back to my simulator and tightened up as many assumptions as I could to match my situation as well as I could. Most of the assumptions I'll keep private but I had things in there like "to age 95" rather than random longevity, return suppression by at least a couple percent for the next 10 years, spending that steps down in three stages based on the plan, spending variance that matches my experience, a provision for SS, etc. Then when I do a trial-and-error on allocations, keeping the other assumptions constant, this is what I come up with:
The x axis is the % allocated to a risk asset. Y1 is the fail rate percent, Y2 is the median value for fail duration in years. The black line is fail rate at different allocation settings and has a wide flat range between 40-80% with a technical optimum at ~50%. The blue line is the median fail duration for different allocations. That line has an optimal point at about 10-20%, call it 15 but otherwise rises with additional risk allocation. You see my point now about reconciling? At least the fail rate optimum is closer to what I actually do in practice. A few months back with looser assumptions the result was hovering in the 60-80% range.
Ok, so simulation is what I'll call the weak form of retirement allocation analysis. Backward induction, on the other hand, is supposed to be the strong form, an assertion I'll just roll with for now. That approach, which I tried to tackle in a previous post, comes to some difficult conclusions. This was the allocation map for a 40k inflated spend which is not my spending but will serve as a proxy for now to make the point:
My map, by the way, looks the same it's just for different portfolio values. Both maps say the same thing: maybe a 0% stock allocation is appropriate today (for the starting portfolios in question) and that all sorts of allocation values might apply in the future depending on what year and on what value the portfolio lands. Then, there is a better than even chance of ending up with a 100% allocation if you get to the end of the plan and are still in good shape. That's a lot to digest, though, and there is a lot of possibility for a lot of changes and the implication for today seems like it is really aggressively conservative. For the "possibility of changes," note that here is the freq distribution of possible allocation to the risk asset across all years and wealth levels (ex 100% which has a lot of occurrences):
But here I had a mini-micro epiphany. I went in and looked at the data and more often than not the risk allocation for the current year wasn't really zero it was more like 9-10%. That small difference reminded me that in the article that started me down the induction path the conclusion was that the implication of the analysis can be that "bond heavy" (not necessarily 100%!) portfolios are called for early and/or maybe in situations where wealth is declining into a caution zone. (That caution zone and the direction from which one approaches it is what can help one avoid the need for annuities in some cases -- see aacalc.com -- and why I put the RIIA suggestions to the side for now.) That "bond-heavy" comment, then, made me realize that I didn't really have to go totally to 0% risk now or, necessarily for that matter, in the future, I could just "lean" more or less on bonds depending on the year and situation. So, first I let the simulator run with the "pure" backward-induction map and it's many allocation choices. The result was a 19.7% fail rate and a median fail-years duration of 5, so a loss on fail rate (goes higher) and a gain (goes lower) on fail duration (vs. 7.4%/8 for the trial and error simulation with fixed allocations), something I can't chart. Then, with my mini-epiphany, I changed the BI allocation map so that it was not infinitely granular like the freq distribution or map above but was forced into "bands." By that I mean that I made all BI-map allocation recommendations between 0 and 50% (risk-asset) force themselves into a "band 1" (say a 30% allocation to risk), recommendations between 50-85% were forced to 70% and everything else to 100%[1]. With that simplification in hand I re-ran the simulator with different values for the value used for "force to band1" (i.e., band 1 was varied from 0% to 70% allocation to the risk asset[2]). That allowed me to do what the article had implied: go "bond heavy" in the situations that the map called for but without the extreme version of going to 100% bonds. That, in turn, allowed me to stay marginally true to the spirit of "strong-form," economically-correct theory implied by the backward induction approach and its dynamic allocations, an assertion that is probably as sketchy as it sounds now when I re-read it, and chart it (in practice this comes across as nothing more than dressed up simulation). Here is what the simulated fail rate and fail duration look like when using the "banded" backward induction map and then we vary band 1 from 0-70% allocation to the risk asset, results that I find a bit underwhelming but maybe helpful:
The black dotted line is the fail rate estimates using different values for band 1. The optimal point, which is hard to see in the chart, is at a 40% band-1 allocation with a 7.4% fail rate but it's still pretty flat between 40 and 70 and does not differ much at all from the regular simulation. The blue dotted line is the median fail duration using the banded backward induction allocations. This is where the benefit to backward induction resides and, contra the solid blue line, there are optima (re fail duration) all the way from 0% to 40% band-1 allocations, not just at 15% in the original blue line. What that means in this solidly unscientific approach using massively idiosyncratic assumptions that mean nothing to anyone but me, I can get a momentary convergence, at least this year, for these assumptions, at about a 40% allocation to a risk asset.
Miracle of miracles that happens to be where I already am! That means that if I use my standard deep analytic technique of asking myself "should I do something or should I do nothing at all" I get a pretty convincing answer of "do nothing" ... at least this year, anyway. Not only is that where I started at the beginning of this journey but there is probably a small line of people that know me that would be willing to step to the podium to confirm that doing nothing at all is something that I am particularly good at anyway. Any way you cut it, it's optimal enough for me.
Postscript 2/7/17
On a whim I decided to run it with the same assumptions but adding a 3% probability of a 3 x initial-spend shock in any given sim year. The results were harder to make a solid judgment about but it gave me some interesting insight I hadn't thought about above. Here is what it looks like on the same scale:
The surface differences are that the median duration of fail minimum moves right a bit if we band at a dynamic 40/70/100. The minimum fail rate is 11.9% when we are using a binary backward induction table that allocates 100% to equity or backs down to 70% equity if wealth levels and year match what is in the table. What that means is that if 70/100 was our only choice of a binary combo, and it is in this example, and if a median fail duration of 8 were acceptable, and it probably would be because its better than the fixed allocation minimum of 10yrs when the fixed allocation is 60% and a little higher than the absolute minimum of 6, then the binary 70/100 would more or less dominate every other allocation choice, fixed or banded-dynamic. You would never choose any other scheme. That means I should probably someday explore various choices of binary combos for dynamic allocation. My guess is that retreating from 100% to say 50% (rather than 70 % in this example) might be interesting. But that is for another day.
-------------------------------------
[1] these are arbitrary choices and all of this is suspiciously close to the corrupt practice of forcing data to produce the results one wants but since I am not an academic or a practitioner, just some guy in his home office, perhaps I can get away with it... Maybe we can just call it some kind of "heuristic."
[2] so the forced-to-a-band levels, as they were varied, were like this in terms of allocation to a risk asset: 0/70/100, 10/70/100, 20/70/100, 30/70/100, etc up to 70/70/100 where it is really just two bands, 70 and 100.
The simplified map would look like this when forced into three bands:
No comments:
Post a Comment