“When you know a thing, to hold that you know it; and when you do not know a thing, to allow that you do not know it — this is knowledge.” — Confucius
This is my series of thoughts on relevant investment topics. In our ongoing research of our own dilemmas, we recognize that these same issues are likely at the heart of struggles that we all face. With equal measure of frustration and excitement, we hope to contribute to the marketplace of ideas and discussion. And with the goal of interaction and feedback, please reach out with responses or topics of interest.
Seth Klarman: Investors Can No Longer Rely On Mean Reversion
"For most of the last century," Seth Klarman noted in his second-quarter letter to Baupost's investors, "a reasonable approach to assessing a company's future prospects was to expect mean reversion." He went on to explain that fluctuations in business performance were largely cyclical, and investors could profit from this buying low and selling high. Also Read More
Last week Los Angeles experienced what us Southern Californians call crazy weather, with one heavy rainy day after another. For all of you who live in parts of the country (or world) where rain is part of life, this feels like just another day, but to us Angelino’s, it’s madness. In fact, driving anywhere last week was akin to being in a once in hundred year snowstorm on the east coast, with traffic grinding to a snail’s pace as arm’s length puddles scatter SoCal highways. But the worst of it is how it hits home -- literally. While heavy weather regions ensure buildings go up with pounding water in mind, not here!
With just a few days of heavy rains, I was faced with water dripping, more so coagulating, in the strangest of places around my house. It was seeping out of one of the walls in my living room, gliding down the surface of my office ceiling, and sitting -- dead center -- on the floor next at the bottom of the basement stairway, as if it jumped out of nowhere to create a puddle island. It is frustrating, to say the least, to deal with these leaks, or more specifically, their sources. Finding the source of water infiltrating the home can be an outrageous task. Water meanders, left and right, down the path of least resistance, and then exits to reveal itself, sometimes wholly disconnected from any nearby opening we can lay our eyes upon.
The real kicker, though, is that no matter how difficult the sourcing journey is, the water continues, relentlessly, to cause damage, both visible and hidden. The puddle on the floor, or discoloration in the ceiling, is readily apparent, but trouble also grows inside the walls, or even worse, gets absorbed, forming black mold which will eventually affect our breathing. In my last post, I discussed confounding variables. Confounders are a tricky bunch in that they remain in hiding until such time that they decide to appear. Much like sneaky leaks, the water may only show up inside with heavy rains, such that all prior drizzles are absorbed by materials inside the walls, creating damage unbeknownst to us until the accumulated internal decay begets disaster.
So the big question becomes: how do we identify the leak, and its original source, before the damage compounds? Can we simply embark on a search and destroy of all the confounders that may be creeping about and distorting our correlations? Is this even possible?
In a 2001 white paper on Conditional Skewness in Asset Pricing, Harvey and Siddique (professors at Duke/Georgetown, respectively) determined that “ignoring negative skewness in evaluating alpha leads to overestimation of alpha and the appearance of delivering alpha when the risk is merely being transferred.” They hit the nail on the head, negative skew is not just inflated alpha, it reveals hidden risk. But where their hammer missed the nail (to be fair, they were not trying to build an entire house!) was with regards to enlightening us as to where that negative skewness actually comes from, e.g. what is actually responsible for that fatter left tail? From a risk perspective, the visible excess density in the left tail is a result of events that have already happened, but concurrently, imply a tendency to larger magnitude negatives from yet-to-happen events. Thus, by accepting fat tails, we leave ourselves, effectively, short volatility, or analogously, with more water damage to deal with when the next storm hits.
Moreover, given that fat tails have undefined variance, we have no precise way to estimate the magnitude of the yet-to-happen event. (While undefined variance is a mathematical abstraction versus the real possibility of an “infinite” magnitude move, in reality, there obviously cannot be an infinite drawdown. The pragmatic basis of undefined is that the tail does not converge, and therefore provides no way for us to estimate the “worst” possible event). And as if that danger wasn’t enough, to boot, we are empirically bankrupt, having too few prior events from which to extrapolate future ones. We are literally doomed, unless -- and this is the big unless -- we can either identify the culprit, or fully protect from it without even having to identify it; which brings us back, full swing, to the task at hand, deal with the confounders!
To that end, I am going leap straight over the notion of finding the source, dubbing this a near impossibility and futile endeavor. The very nature of a confounder is that it lurks, both creepy and covert, with little to no indication of its very existence until its ready to rear its head. In fact, the only time we learn of it is historical presence is upon assessing the damage it unleashed after having surfaced. The source, for all intents and purposes, is lost in obscurity. The unfortunate reality is that, most often, we cannot know from whence the negative skew will come, nor when.
“You cannot manage outcomes, you can only manage risk.” - Peter Bernstein
I therefore jump, head on, into the only practical, and very real, solution. For this, I am reminded of the well known and powerful point, often quoted, from Peter Bernstein, who says “you cannot manage outcomes, you can only manage risk.” Adapted to confounding variables, I’ll rephrase: “you cannot find confounders, you can only protect their destructive path.” In other words, the only means forward is to proactively prevent. Which begs the ultimate question, how do we protect the path if we don’t know the path? While we may not be able to find the source of the leak, we can thoroughly assess every potential corner, so that, across vast locations, we can rigorously apply layer upon layer of sealant. Further, we can look to the boat industry, and monsoon cities, for how they protect from the smallest of droplets seeping their way into structural elements. We must ensure that the water, from whichever way it comes, is trapped, and left out in the cold to evaporate.
In the world of protecting all possible paths, we run into several problems. As brought up in previous posts, we first face the issue of ergodicity, where the probabilities we deal with in our portfolios are time and/or path dependent; our experienced “average” will not equal the ensemble average. Next, we face the issue of asymmetry, wherein most financial assets move up at different rates and magnitudes than they move down. An old mentor of mine from my early days on a proprietary trading desk used to say “stocks climb stairs and go down elevators”. And finally, we face the problem of regimes, or more specifically, highly distinguished phases or market cycles, during which the entire “character” of the market can change. We can thank Fama and French for identifying the main influential factors, but their mere identification does very little to help us work with timing their dislocations, such as when value rises and growth cracks, or yield leaps to investor preference over quality.
So where then do we begin? Common approaches to confounder control includes things like: diversification, orthogonality across portfolio components, or volatility normalization (aka risk parity). But these all have massive flaws. Diversification, while partially beneficial, fails in exposure reduction. Going long 50 names instead of 25 names is certainly “twice” as diversified with regards to idiosyncratic risk, but has equal directional exposure, and arguably, greater systemic exposure in getting more proximate to the index; “diversification works” is empty without context.
When is low correlation be too high?
Seeking low correlation is equally suspect. First off, if there is no deeply substantiated rationale for why what has historically behaved in opposition should continue to do so, there is no reliability. Correlations, in and of themselves, are highly undependable. Moreover, “low” is too light for a hedging objective: “low” is still something. Protection should be either negative, or more negative, to the risk generator. More importantly, and as addressed above, the greater risk lies in the lurking confounder. A perfect example is the typical market neutral portfolio, which has “low” correlation to many other strategies, but doesn’t necessarily hedge others, or worse, can have its own concurrent trauma. Look at the grave nature of a theoretical market neutral portfolio, when the purported neutrality can be disrupted by a confounder that sends shorts “up” and longs “down” when value rallies and growth falls, thereby generating losses on both sides of the portfolio. The hedge becomes the risk.
Finally, risk parity, as a tool for hedging, is fundamentally upside down to the objective. Allocating by volatility expects the “lower risk” (aka, lower vol) assets to outperform the “higher risk” (aka, higher vol) ones during stress. But ironically, this is implicitly short volatility, and will result in the exact opposite outcome when stressed. The implicit short vol nature arises from transferring exposure from realized to unrealized vol, or said differently, a higher vol asset that “lives” at that vol level may actually be safer than a lower vol asset that can shift to a higher vol regime. To state the obvious, low volatility can’t get much lower, but can definitely get much higher! Technically speaking, Vol of Vol is more right skewed for lower vol. As such, this back-door short vol is likely what goes wrong when everything else does. The goal of a hedge is get long vol, not short.
Elsewhere, I have seen promotion around “systematic trend following” as a hedge to equity stress. Sure, they can be a great hedge, but only for trending down markets (e.g. 2008), not for whipsaws (or high volatility of volatility), like Feb and Oct of 2018. Going back to the problem of ergodicity, they are path dependent to autocorrelation, aka trend, and so will not always be there for you, or even more concerning, can add to the pain during the stress events that do not trend.
In grand summation, the ultimate hedge, must be built to protect the heaviest storms and most clandestine confounders, must satisfy all conditions; ergodic, long volatility, and long asymmetry. When your car crashes, it doesn’t matter how or why, you are covered. Fortunately, there is a beautiful solution to achieving this, which begins with utilizing hedging components that are perfectly associated with investment components, and ends with making them an alpha source. In simple terms, the best alpha is hedging alpha, which I shall talk more about in my next post.
Chief Investment Officer
Logica Captial Advisers, LLC