Fragment On Monte Carlo

David Cantor recently sent me a decent piece on Monte Carlo analysis by Sandidge, Tharp and Powell in the Investments and Wealth Monitor that covered the limitations of Monte Carlo analysis in retirement situations. It's pretty good and takes a proper posture of caution about overthinking MC. I myself once imputed magic to MC after my divorce. I did not gain a proper skepticism on this until I had built quite a few simulators of different types and became intimate with the black box: its coding, its errors, its flaws, its assumptions, its biases, its inability to say much, really, about the future.  

Dirk Cotton wrote on this a bit as I have as well. Kitces covered it at least once. Most analysts, if honest, will point out the flaws. The real problem, I think, is mostly between naive retail investors and naive advisors. And it is there that the link above carries its own weight and confronts the problems directly. You can read it for yourself. I can't remember the article in detail but I'll highlight a few things that came to mind as I read it.

1. Meaning. 

Telling a freshly minted retiree that their chance of success is 80%, and leaving it at that, is a bit of dead end and empty of meaning (even allowing that this is a wee bit of a strawman). It's an epistemological sink hole. Saying of the state of one's retirement that there is an 80% chance of success is like saying of the state of one's drive to grandma's house that it is "50 miles per hour."  The dashboard (either one) is useful but only in terms of getting us from point a to point b in a timely, safe manner while we remind ourselves that the drive, like retirement, is an unfolding process within which much adjustment and adaptation -- dynamism in the article -- will obviously happen.  We know this. 

The better approach I think, pointed out in the article and by me in no small number of posts, is one that is achieved by way of "triangulation." And here they use the mental model of hurricane forecasting (I did a great post on this once, btw...). Otoh, I don't think they take this metaphor far enough. The "spaghetti" models one sees in hurricane forecasts are generally all a form of MC projection: many paths but all the same or similar method (I think. maybe. idk). So, I propose that the Hurricane model metaphor misses a couple things: 1) dynamism ie in the sense of the continuous updates over time that I mentioned in my hurricane post not just the spaghetti models themselves at some arbitrary point, and 2) other frameworks, by which I mean in a Ret fin context not just many MC sims or many MC sims updated over time but also other evaluative frameworks that may or may not be within the MC fold. Here I think of maybe "perfect withdrawal rates" or "consumption utility" or deterministic formulas or whatever... but at least different perspectives brought to bear on the same question in order to give a general sense of what is coming at some point in time...and then continuously updated. I'm not sure the Hurricane models do that (except over time), but they might. And I live in FL so I do pay attention. 

The other thing here, mentioned in the original article, is that fail rates are a bit of a bogus metric at the retail investor level. What does 20% fail expectation even mean? What is the threshold for good or bad (hint, doesn't exist)? That estimate is a snapshot, not a movie, right? Why did we pick 30 years? Why isn't it 100% success?  But there are better questions to ask. For me I think this whole framework of inquiry is better denominated in time questions. This was made manifest in Milevsky's Seven Equations book. The proper positioning of our insight is more like this: How long will it last? How long will I live? Those would not be point estimates, by the way. Statistically they are distributions but conversationally, they are part of an ongoing dialogue as long as one lives. This questioning, this practice of ret-fin epistemology, never really ends.  

2. People Don't Really "Mathematically Ruin."

This is a point that my friend Dirk used to make often. In a coded simulation, the years tick by like a metronome -- tick tick tick -- and our trajectory, telemetry wise, is informing us that we are tick tick headed towards zero. Yet the sim does nothing: tick tick. This is the inevitable sorta linear mathematical ruin thing.  Who does this or allows this to happen? Uh, no one. If one sees ruin coming in a tick tick fashion way in advance one becomes afraid and averts...or should try. Spending is reduced, jobs are taken, family is engaged, allocations are changed, annuity ripcords are pulled. Something, anything. Something changes. This is the dynamism that the link speaks to. (More rational to fear a slow-then-fast bankruptcy cascade as Dirk would have said)

Yes, of course, sophisticated programmers can gin up rules or paths or adjustments or pseudo-dynamism in MC code in R or C++ or Python... but they, the rules, are always: a) generally pretty much arbitrary, and b) have no "stomach" i.e., fear. Who knows how one really reacts in situ. As an aside: a tidbit I once picked up from Gordon Irlam was that in an extreme or troubled situation, the un-wealthy will (or should or might) counterintuitively take on immense risk. Though they will most certainly fail anyway and by taking on risk they will most likely enhance the chance of that fail, the new risk is in some sense the only way out. It is a type of lottery ticket and quite rational if you think about it. Rarely sim-ed by the way. 

So, "fail rates" in an MC sim are in some sense mooted or at least corrupt from the very beginning, sophisticated programming notwithstanding. Prove me wrong. In this I agree with the linked article. 

3. Judgement

I have spouted quant drivel here on this blog ad nauseum for close to a decade. Yes numbers are important. Yes, triangulation among many methods is better than single point numbers. Yes, process control and managing and monitoring regimes trump the f out of single event optimization schemes. But often the truth is in the seat of the pants. I have a bad habit -- or at least bad as my kids frame it -- of chatting up random strangers on random topics. I have done that reasonably often -- covid season notwithstanding -- over the years and have on more than one occasion done so on retirement finance where older folks have chuckled at (if not outright mocked) my anal perseveration on numbers and models. Generally speaking they say they know what is coming and when times are good they relax and when times are bad or "worser" they more or less retrench. This is why I have been teased on all the quant models. Rightfully so. On the other hand, I once had a fellow quant roll his eyes when I said, after a decade of doing this stuff, that all I really do is keep a balance sheet and an income statement and keep a finger on the pulse of things. What?! no Kolmogorov? No Yaari!? Well, OK maybe Yaari but I have a pretty loose grip on stuff these days, more like our grandparents did than an academic or well informed advisor would do now.  I mean, I have been retired since ~2009 so I have not really been tested by dark things yet; biggest bull market in ages. So far so good. I might rue my dalliance with quant-disdain sometime soon. TBD.