I woke up to retfinophobia -- defined as the fear of or unease about one's financial future in retirement...although now that I made that word up I suppose it could also mean fear of retirement finance content...like this blog -- in 2010 when I started to ask myself the deceptively simple but devilishly difficult (impossible) to answer question: "yikes! am I going to run out of money?" The attempt to answer that question, which is destined to fail since it attempts to predict the future, involves some complex thinking and math while it also leans on a melange of many things all stewed together to come up with some assessment of "how much risk do I have" or "how long will the money last" or "how much do I need now." This is where I started in 2010, a start that was initially inter-mediated by others but then slowly was taken onto my own shoulders as I started to learn more over time. This was also, by the way, the genesis for this blog. I realize now, however, that some of that early time might have been better spent understanding the individual components of the problem, each of which are independent of the others, rather than confusing myself with the opaque and naively-understood blend of all of them. The blend would have made a lot more sense -- and is quite useful once known -- if I had learned the basics of the independent processes first.
By "blend" I mean stuff like ruin risk assessments or Monte Carlo simulations or even the risk capacity judgments coming from the present value analysis of the household balance sheet although that last is much closer to what I'll be talking about here. There may be others. All of them seem to synthesize retirement risk into something approaching unitary answers based on the modeling and integration of complex and interrelated but mostly independent underlying real world processes. The independent processes I am talking about, the ones that can coalesce into the net retirement judgments, are at least three: 1) the longevity "process," 2) the return generation process, and 3) spending. Let's take at least a superficial look at these one at at time.
1. The longevity "process." This should be self evident but an awful lot of ret-fin lit seems to give this process short shrift. In order to simplify the planning and the math it is often seductive to simply say "the planning horizon is 30 years." But this ignores both the individual situation where the plan could be anything from 1 to 60 years (with varying degrees of probability that are really not all accurate for one person in one life) as well as the mass-population level where the aggregate longevity process has been studied well and is quite predictable (at given points in time but is a moving target otherwise). I think that in retrospect I should have spent more time understanding human longevity as a process and a probability. What I mean by this is that there are mortality tables and probability distributions and math to know about but here one must also first note that: a) there are different cohorts that can be measured and different measures can be taken from the same table, b) the math and estimation are usually a moving target, and c) there are different degrees of usefulness to this kind of knowledge.
a) Cohorts and Measures. As an example of cohort differences I'll note that there are average-average tables like the Social Security Life tables that measure people across a wide swath and then there are other cohort tables like the Society of Actuaries table that measures annuitant mortality. SOA tables are based on people that choose to annuitize so since they may have inside information on their personal or family health and therefore may be biased towards purchasing annuities they may live longer than average so the statistics are different than the SS table. i.e., the statistics generally trend later/older. But even within these tables what longevity expectation do we use? The average (mean)? Maybe. That can be a useful summary statistic but these are skewed distributions so other measures may be important too like the median or the mode or the 95th percentile or even the max. What one might use would be a function of something like conservatism, risk aversion (that may be the same thing), age, insight into personal health, access to "floor" income, influence of advisors, etc. The essential point here though is that having an advisor plug in 30 years does not really cut it. How old are you? How healthy? How risk averse? What is changing in the estimates? One must know how this works and then "own" the knowledge. The thing to remember is that longevity is a "process" and the probabilities mean the duration of your retirement is not a number it is a distribution with a (kinda) known shape and terminal scope. It's not a "30," it's a "maybe" and that maybe has a shape.
b) Moving Target. Even if your advisor punts and switches you from a 30 year planning horizon to an average or median or even the 95th percentile for your age you are not off the hook yet. One needs to know that the shape of the longevity distribution is changing over time. And this is more complex even than it sounds. While there is no evidence I can report here that the terminal age for humans is advancing past around 120, it has been true for a while that the average/mode/median has been shifting out a bit with changes to med-tech and lifestyle changes (that's your kicker by the way, lifestyle changes are the lever that has the biggest control on this). So all else equal, for a 60 year old the mean expectation used to be x but now it is y. This, by the way, is the basis for adjustments to the tables like the SOA does for its IAM table. But wait! In the US this movement in the mean is now in retrograde motion. The mean/mode is actually moving back down. But if one is not personally in the grips of an opioid-crisis-demographic maybe one does not really have to think about this. Or maybe yes. Who knows? Something to think about though.
But even if we keep the shape of the distribution stable for any given age we are not really done. There is something called conditional survival probability. This is the phenomenon that says that if one survives to an incremental year the odds of the survival beyond that year change (go up). If my median survival at 50 is 83 or whatever, if I survive to 60 it might now be 87 (depending on the cohort again). While one probably can't buck the age-120 "wall" the conditional probabilities do change for each year survived. This just means that retirement planning needs to be a continuous process. One needs to re-estimate this stuff each year because among all the other factors, longevity estimates change, whether by age or the progress of science. While it might not be totally straightforward to do these calculations oneself it is also not rocket science. I don't lay it out here but it is not all that hard to figure out.
The other factor that is often ignored is that retirements are getting longer not just because of longevity science but also because retirement start dates are getting earlier whether by choice or some involuntary factor like layoffs or caregiving for others. That means the start date, like the end date, is a random variable too. And we have not even discussed the stresses created by a world shifting from defined benefit mode to defined contribution mode for pensions. What's a "pension" a millenial might ask? Good question. You would probably wish you had one if you knew anything about longevity, wouldn't you?
c) Usefulness. Once longevity math is known, the shift from "30 years fixed" to "probability based thinking" can be quite useful. First it can add to the conservatism of a plan which in the absence of DB pensions or full annuitization is probably a good thing. Then, if one can do the math, it also adds to one's ability to analyse different kinds of risk and/or financial instruments. All sorts of retirement math seems to depend on knowing something about the vector of conditional survival probabilities for a given age. And here I am thinking about things like annuity price estimation or maybe risk-of-ruin calculations but there are perhaps others like estimating the present value of spending plans. Maybe there are others I have not thought about yet. So, yes, useful.
This kind of thinking is important because longevity is a process independent of other retirement processes though it is, in the end, very intimately related to outcomes. Change your portfolio and longevity is not affected. Change spending and longevity is not affected. I mean, maybe health care spending can change it but you know what I mean. The reverse is not necessarily true I guess so understanding longevity in retirement planning (at least at an aggregate US level; considered individually it is always a roll of the dice) is quite useful if not absolutely necessary. If you have an advisor that blows over this subject or is incoherent on it maybe you need to think twice about the advisor or perhaps brush up on it yourself. I did.
2. The return-generation process. Note first that I chose my words carefully here. I did not frame this as a portfolio choice or portfolio management thing. While it is true that the size of (or absence of or sudden changes in) the portfolio can trigger either behavioral effects in spending or maybe related pre-packaged execution of financial-planning spending rules, it is not true that either spending or longevity have anything to do with the underlying return generation process in and of itself. It is entirely independent. Ok, so that may be self evident but lets look at it a couple ways -- single period view and multiple period view -- first before we move on.
a) Single period view. I think it is relatively well known and appreciated by retail retirees that asset allocation can have a positive effect on a return generating process. A blend of stocks and bonds and maybe other semi-correlated classes can have a better risk return profile than each in isolation. This is basic modern portfolio theory and most advisors and retirees will have this conversation if the retiree is not already on board, as they should be. In other words, my portfolio, diversified, will have a more efficient return for a given level of risk than if not diversified; I don't get paid for risk I can get rid of by way of diversification. This is pretty basic stuff but again the point here is that understanding this is important to retirement outcomes and the emphasis in this post is still that the return process -- especially here in single period mode -- is independent of the other two processes mentioned above. Point 2a should probably receive a "duh" for its obviousness.
b) Multi period view. This is where I think retail retirees, myself included, are often under-aware or under-trained. Take some of the different portfolio returns and volatility one might have from 2a and now insert them into a multi-period context (say like retirement...and this is before we even get to spending). This is the domain of geometric mean return analysis where it can be less self evident to some that long term returns will usually be worse than one's initial (arithmetic) expectations and that high return high vol portfolios (return generating processes I should say) might have worse outcomes over time than lower return and moderate to lower vol strategies. It is also less self evident that that analysis can produce different answers over short time frames than over very long time frames. We can also say it is also not always self evident, even to otherwise savvy advisors, that the multi-period geometric return process can influence the asset allocation decision since it is theoretically possible that a high return portfolio will be less efficient than a lower return one in terms of the multi-period interaction with the single period efficient frontier.
And, to be clear, as I inferred in the first sentence of this section, this is before we even get to spending/portfolio considerations. Ex-spending there is no "ruin" as such in the return generation process, just a limit of zero. That is vaguely interesting. But add spending to the mix and "to and through zero within a lifetime" (i.e., ruin) becomes a fun retirement game to play and about which the return generation process has a lot to say and the results of which are of keen interest to a lot of retirees, myself included. But that is only when we combine portfolios and spending. I touch on that a tiny bit below rather than here.
The point here is that the return generation process by itself is worthy of consideration deeper than "I might make x%" for several reasons: it is more complex than it looks on the surface, it is a process that is independent of the others, the perspective changes in odd ways in multi-period mode, the interactions with the other two processes can be intense when combined, and the consequences of the choices to be made are significant. We haven't even talked about taxation yet.
3. Spending. Much retirement analysis joins spending and portfolios at the hip. Discussion of retirement "withdrawals" is usually framed by scaling the retirement process in terms of spending with respect to the portfolio...or vice versa. It's a rate or a percent or, inverted, a wealth unit. And don't get me wrong, this is an important and proper way to do it. In fact one of my significant aha moments in 2017 was realizing that ruin risk is the mathematical intersection of two independent probability distributions (processes): a) human longevity (as above using conditional survival probability), and b) portfolio longevity in years, a single probability distribution that is created by combining return generation processes (and portfolios) with spending. But maybe you see here that "portfolio longevity in years" is itself made up of two things: an underlying return generation process and a spending process. We just said that the return generation process is solitary, independent, and unaffected by either longevity or spending. That means that spending must then be independent itself and worthy of much deeper consideration all on its own or at least deeper than I have seen in the ret-fin literature to-date. Maybe that elision is because many of the academics and advanced practitioners, while whip-smart, professional, and focused, are not yet in the throws of retfinophobia themselves.
To tease this out let's take a look at the assumption used and abused by a few too many ret-fin papers over the last 50 years, the constant inflation-adjusted spend. Why constant? I don't really know but I have to presume that it is either analytically tractable (easy, lazy?) or based on the idea that retirees fetishize the utility of constant lifestyle to the exclusion of all other considerations (say risk?). But even if we stipulate to the probably wrong and over-simplified utility argument and even if we give a passing nod to the idea of the probably true behavioral adaptation to changes in ambient risk and/or the size of the portfolio, this "constant" assumption forecloses on any further nuanced introspection into the true nature of spending. Consider this: spending varies above and beyond changes due to inflation or a constant budget. Sometimes there is just plain random variance. I can count on one finger the number of papers I've seen that deal with this. Supposedly there are also secular trends observed in real spending: spending goes down over time with changes in mobility and/or the desire to consume and then goes up again with late age health care spending. Blanchett, among others, postulates (and convincingly shows) this U shaped trend pattern to spending into late age. Then there are spend shocks. Dirk Cotton once wrote a good cover on the chaos-math nature of shocks and bankruptcy that does not conform to standard probability modeling or thinking about retirement in general and spending in particular. All this tells me, as a retiree, that constant spend modeling is more likely than not a false model. It is also, as I pointed out in a past post, a very active and aggressive risk taking position as a portfolio declines but that is another subject altogether.
But even if we add something like "spending rules" to our financial modeling I don't feel like it does much for us. First, the rules are still often framed in terms of a response to a stimulus that is more often than not portfolio driven, though that is not always necessarily the case. That means it disrespects spending as an independent process which I am claiming that it is. Second, the rules, to me anyway, smell of abstractions and unreliability and undeserved certitude. This, as with the constant spend assumption above, forecloses on a deeper understanding of real world spending.
All this means that I think spending needs a much more sophisticated and isolated approach than I typically see out there. The closest I have come to this recently was an idea borrowed from pension liability analysis (think retirement spending): stochastic present value. There, return generation processes are taken off the table (sort of..wait a moment and I'll contradict myself) and we work in present time with current asset values. But that means we then need to deal with the present value of spending. That, in turn, means we focus very very closely on the future nature and shape of spending in isolation, draw it into the present as a risky distribution, and then compare it to assets currently available to serve the liability. That's pretty good stuff but the problem here though is that the SPV technique uses stochastic processes for generating discount rates (that is a return generation process in disguise, by the way) and, the way I do it, longevity assumptions (the other process we wanted to isolate from). That means even SPV will rob me of a clear eyed, deep, and nuanced view of spending as an isolated and independent process, a view I think it deserves. At least there is a hyper-focus on the internal processes of spending over time with SPV. That's a step forward.
Hindsight Conclusion
So, at the end here what is my hindsight "proposition?" It is, in simple terms, that I wish -- rather than starting out with a complex and integrated analysis of retirement finance risk and then later coming to a much deeper understanding of and appreciation for the underlying independent real world processes involved -- that I had done the reverse: come to a deep understanding of and appreciation for the independent processes involved first and then integrated them into a judgement about future retirement risk, a judgement that by its nature will always be flawed, which is to say wrong.
------------------------
update 2/6/18: there are other independent processes like inflation, taxes, etc. Some of that can be embedded in the return generation process as real returns. Maybe another post...
No comments:
Post a Comment