Universa Decennial Letter 2018
Universa Decennial Letter 2018
Universa Decennial Letter 2018
DECENNIAL LETTER
MARCH 2008 - FEBRUARY 2018
This month marks the ten-year anniversary of Universa’s tail hedging program (the “Black Swan Protection
Protocol”). As we ring the bells and reflect on how far we’ve come, I am reminded of an old Russian proverb that
warns, “Dwell on the past, lose an eye. Forget the past, lose both eyes.”
What a diverse decade it has been, spanning a great bust and boom, some very high volatility and some very low.
It was a superb test for us, and a little retrospection is in order. So let’s review how we performed as a risk
mitigation strategy for you, including in comparison with other strategies that also presumed to serve such a
function. Risk mitigation performance must of course be measured by its “portfolio effect”—specifically, the
impact it has on the compound annual growth rate (CAGR) of the entire portfolio whose risk it is trying to
mitigate. As I will discuss later in this letter, this is all that really matters in risk mitigation, and has always been
our focus. It is where the rubber meets the road.
Below is Universa’s ten-year life-to-date legacy of risk mitigation performance. We paired our actual net
performance (monthly administrator-provided net returns, using yours from your start date, expressed as returns
on a standardized capital investment) with an SPX position (a realistic proxy for the systematic risk being
mitigated) to create a hypothetical “risk-mitigated portfolio.” That portfolio’s net performance is summarized
below, along with the similarly-constructed portfolio performance of five other standard-bearers in risk
mitigation. Consider this a risk mitigation scorecard for the past decade.
iShares 20Y+ Treasury (25%) + SPX (75%) 12.8 12.0 9.7 -15.1
CBOE Eurekahedge Long Volatility (25%) +
10.9 10.4 9.1 -13.8
100 SPX (75%)
Gold (25%) + SPX (75%) 14.2 10.6 8.5 -25.4
Hedge Fund Index (25%) + SPX (75%) 14.6 12.2 8.2 -27.5
50
2008 2010 2012 2014 2016 2018 CTA Index (25%) + SPX (75%) 12.4 11.2 7.9 -21.5
The equity curve shows each risk-mitigated portfolio’s growth of capital over the last ten years. The table shows
each portfolio’s latest one-year trailing 12-month return (labeled “1Y”), last five-year and ten-year CAGR (or
“geometric mean” annual return, labeled “5Y” and “10Y”, respectively), and lowest annual return over the last ten
years (labeled “10Y min”). Green cells indicate the highest two returns among the six portfolios in any given
column, and red cells indicate the lowest two.
In our ten-year life-to-date, a 3.33% portfolio allocation of capital to Universa’s tail hedge has added 2.6% to the
CAGR of an SPX portfolio (the SPX total CAGR over that period was 9.7%). To put this in perspective, this is
the mathematical equivalent of that same 3.33% allocated to a ten-year annuity yielding about 76% per year. In
contrast, each of the other risk mitigation strategies actually subtracted value over the same period, regardless of
their allocation sizes. (Other risk mitigation strategies omitted from this comparison basically produced results
within the ranges of these five standard-bearers, or worse; they are shown in the Appendix.)
Moreover, during the last one-year and five-year periods, when the SPX experienced very positive returns,
Universa’s risk mitigation strategy also outperformed these alternative risk mitigation strategies. It is common for
people to simplistically view tail hedging, the way we do it here at Universa, as a “drag” on their portfolio in the
absence of a market crash. However, when framed correctly, as we have tried to do here with our risk mitigation
scorecard, the picture changes. It becomes apparent what a real portfolio drag most other risk mitigation strategies
have actually been. (This is like the joke about the camper who only needs to outrun his friend, rather than the
bear that’s chasing them. Over the past decade, Universa’s risk mitigation strategy needed to outrun only these
other strategies—but we also outran the SPX.) Despite our firepower in a market crash (and my scorn for
monetary-bubbles), we have always been truly agnostic as to the direction of the stock market.
The 3.33% portfolio allocation size to Universa was chosen because it is (and has always been) the approximate
effective allocation size recommended in practice at Universa (relative to a client’s total equity exposure). The
25% portfolio allocation size to the other risk mitigation strategies was chosen to be meaningful and realistic for
an average investor (relative to their total equity exposure). That turned out to be insufficient for any of those
strategies to provide a level of downside protection anywhere close to the level Universa provided. This is clearly
evidenced in the “10Y min” column, a good proxy for the systematic risk remaining in each portfolio (thanks to
the 2008 data point in our time series). The HFRI allocation, for instance, actually resembled adding more SPX-
like risk to the portfolio.
If we were to try to calibrate ex post the allocation sizes for each of the risk mitigation strategies in order to
maximize their ten-year portfolio CAGRs, the optimal size for each strategy (with the exception of Universa’s and
the Treasury strategy) would actually be 0%. That means for best results, you shouldn’t have added any of those
strategies to your SPX portfolio at all. Despite our efforts, the risk mitigation scorecard remains something of an
apples-to-oranges comparison.
The source of Universa’s risk mitigation outperformance is no secret: It was driven by our “convexity”—the
degree of portfolio loss-protection provided for a given capital allocation, or the “bang-for-the-buck”—that is, of
course, our particular modus operandi at Universa. This translates into a lower capital allocation of only 3.33%
needed to produce a meaningfully larger protective crash profit. And, that 3.33% (even when experiencing losses)
poses less “drag” relative to the far greater amount of capital invested in the SPX the rest of the time.
While the past can be an imperfect gauge of future risk mitigation value, it can nonetheless provide insight that
prevents us from “flying blind,” which seems to describe many of these more orthodox risk mitigation efforts
today. A strategy that worked in the past naturally isn’t guaranteed to work again in the future. Yet, if a strategy
didn’t work in the past, isn’t there something inherently unscientific about expecting that it will in the future? In
other words, Popper’s falsificationism applies: Though we cannot necessarily accept a strategy as always
effective, we recognize when we must reject it.
Heeding the warnings of the old Russian proverb, we at Universa don’t like the thought of losing even one eye!
So, of course, we don’t rely just on the past in demonstrating our risk mitigation. As Universa investors, you have
had the benefit of knowing first-hand (either from the risk reports we provide regularly or your own stress tests of
your transparent portfolio) that we have delivered consistent and robust crash protection throughout your tenure as
clients. We have employed no forecasting, no timing, no finger to the wind—none of which ultimately adds risk
mitigation value, by definition. Effective risk mitigation requires consistency and robustness. It shouldn’t be a
black box or a mere statistical regularity. Our risk mitigation approach plays too significant a role in a portfolio
for that, more than the mere incremental “alpha” of a typical allocation.
So much of what passes for risk mitigation strategies simply is not that; rather, they result in what Peter Lynch
referred to as “diworsification.” They may moderately lower portfolio risk, but more importantly they also lower
its CAGR. When done right, and as we have seen first-hand and delivered these past ten years, effective risk
mitigation does more than just lower risk. It transforms the entire portfolio, adding unique value that no other type
of investment can. It is the tail that wags the dog.
Why do we use the CAGR, or geometric mean annual return, as our metric to evaluate the effectiveness of a risk
mitigation strategy?
A vexing conundrum known to most investors is losing, for instance, 50% one period and then making 100% the
next; you’ve experienced an impressive arithmetic mean (or “ensemble average”) return of 25%, yet you just
barely made it back to even with a geometric mean (or “time average”) of 0%. As a single wager, a coin toss of
either a +100% or -50% return looks smart, but as an ongoing and compounding parlay, it’s a big waste of time.
Your long-run performance (of just breaking even) would be highly “non-ergodic”.
The arithmetic mean return of an investment just doesn’t convey all that much about its risk mitigation value, and
needn’t translate into a portfolio’s CAGR. As it turned out, over the past ten years the arithmetic mean annual
return on invested capital in the standalone Universa tail hedge was quite high at 52.7%, while the arithmetic
mean annual returns of the other five standalone strategies ranged from 1% to 6.3%. But this doesn’t really tell the
story of the relative risk mitigation performance of all these strategies.
As I have previously written and demonstrated to you elsewhere, even if we hypothetically adjusted Universa’s
standalone arithmetic mean annual return to exactly zero over the past ten years (by, say, hypothetically lowering
our annual return in 2008 accordingly), the CAGR of that combined hypothetical Universa tail hedge and SPX
risk-mitigated portfolio still would have outperformed the SPX alone by more than 0.6%—and that would still be
greater than any of the other alternative strategies (all with positive standalone arithmetic mean annual returns).
(For the Universa allocation, this would equate to a similar-size allocation to a ten-year annuity yielding about
24% per year.) The extreme crash convexity of our return profile is far more important than our arithmetic mean
return in adding risk mitigation value to your portfolio.
The mathematical intuition behind how this works was first laid out in a 1738 paper by the great Swiss
mathematician and physicist Daniel Bernoulli (the same Bernoulli who taught us how airplanes fly). Bernoulli put
a stake in the ground for using the geometric mean to appraise the value and expected performance of an
investment or wager, calling the geometric mean expectation a “moral expectation.” (Unfortunately, he
confusingly expressed this as a logarithmic utility function, merely an intuitive detail for him—but a detail that
would become a profound distraction for economists for centuries to come.)
Bernoulli explained how a logarithmic mapping of winnings and losses, expressed as a return on one’s total
wealth, solved the famous St. Petersburg paradox (named after his Russian residence at the time, though
formulated by his cousin Nicolaus—the Swiss family Bernoulli were most prodigious indeed). His idea was that
wagers should be evaluated by the probability-weighted logarithms of their outcomes, mathematically equivalent
to the geometric mean of their probability-weighted outcomes. (Thus, whether we interpret his point as log utility
maximization or geometric mean maximization doesn’t matter.) The paradox was no longer a paradox. When you
maximize the expected geometric mean of a wager, you maximize your end-point wealth. (For a good, not-too-
technical read on this, see the 2005 book Fortune’s Formula by William Poundstone—my tattered copy has been
sitting on my desk all these past ten years.)
A plot of the logarithmic function (an example is shown in the picture, over Bernoulli’s left shoulder) shows how
this all works. Consider that the geometric mean return is mathematically (the exponential of) the average of the
logarithms (the y-axis) of the arithmetic returns (the x-axis). Because the logarithm is a concave function that
curves downward, it increasingly penalizes negative arithmetic returns the more negative they are. Thus, the more
negative they are (as we roll down the curve), the more they increasingly lower the geometric mean return (and
the end-point wealth). That’s not esoteric quant finance; that’s precisely the way compounding works in the real,
physical world (and regardless of your return distribution assumptions).
The destructive impact this curvature of the log function has on geometric mean returns and wealth is what I call
the “volatility tax.” (It is why our smart coin tosses ended up being a waste of time.)
What’s more, think of the geometric mean return as a superior measure of risk. This is because, unlike the more
traditionally used standard deviation (or volatility), for instance, the geometric mean more accurately reveals the
consequence of risk. Just as standard deviation is the square root of the average squared return deviation, the
geometric mean is the exponential of the average log-return. The former over-weights larger deviations, positive
or negative. The latter over-weights larger negative deviations; so, it measures the incidence of extreme loss more
accurately, and it actually has real economic meaning.
Bernoulli even went on to apply this geometric mean criterion to valuing insurance—something near and dear to
our hearts here at Universa. He demonstrated how a shipper’s logarithmic losses relative to his wealth (which,
again, penalize larger losses) are what drives his economic value in insuring against those losses. A large relative
loss disproportionately lowers his geometric mean return as a shipper, because it leaves him with much less to
reinvest and compound on his next shipment. So that log-loss shows the true (higher) value-added for him of
insuring against it. Almost three hundred years ago, this Swiss mathematician understood the intuition of what
really drives risk mitigation value better than most quants do today. (Leave it to the Swiss.)
Bernoulli saw this disproportionate cost from compounding losses as “nature’s admonition to avoid the dice
altogether,” which “follows from the concavity of curve” in the logarithmic function. He could just as easily have
said “nature’s admonition to mitigate risk with extreme convexity.” He is Universa’s Patron Saint.
Being exposed to this treacherous “concavity of curve” is a lot like being extremely “short gamma” far below the
market (which as options traders know is another way of saying your incremental losses get bigger as the market
moves down). This makes hedging that nonlinear cost via extreme “convexity of curve”—or extreme “long
gamma” far below the market—seem like a natural fit. That’s good intuition behind why Universa’s specific type
of extremely convex, insurance-like payoff has been so effective at raising a portfolio’s geometric mean return—
as well as why mere negatively-correlated payoffs just can’t really move the risk mitigation needle.
Mathematically, it is the rare big loss, not the frequent small losses, that matters most to long-run compounding.
Over time, Bernoulli’s geometric mean maximization criterion eventually did put up a good fight for attention
among risk practitioners, most notably through Williams in 1933, Kelly (whose “Kelly criterion” is simply a
“geometric mean criterion”) in 1956, Latané in 1959, and Breiman in 1961 (who showed that a geometric mean
maximizing strategy both minimized the time to reach a target level of wealth and maximized the level of wealth
reached after a given amount of time—and who wouldn’t sign up for that?).
Ironically, even Markowitz had become a proponent of the geometric mean criterion by 1959 (and very much so
by 1976). But it was too late; his 1952 mean-variance (“Modern Portfolio Theory”) framework was already on its
way inexplicably to becoming the dominant framework used even today for portfolio construction. Markowitz’s is
a distinctly one-period framework, in which time is irrelevant, as opposed to the multi-period, multiplicative,
cumulative, reality-based framework of Bernoulli. Like our non-ergodic coin toss, in a one-period world the
arithmetic mean matters, and in a multi-period world it’s the geometric mean. In both the one-period and the
multi-period world, volatility is bad. In the former, it is because volatility means “risk”; in the latter, it is because
volatility means higher losses that lower the geometric mean return and the end-point wealth.
Nassim Nicholas Taleb, my long-time colleague and Universa’s Distinguished Scientific Advisor, in his recently
published book Skin in the Game, very insightfully took up the point of the non-ergodicity of the one-period
ensemble (arithmetic) average versus the multi-period time (geometric) average, whereby the occurrence of
-100% merely lowers the former, but brings ruin to the latter. As Nassim wrote, “more than two decades ago,
practitioners such as Mark Spitznagel and myself built our entire business careers around…the effect of the
difference between ensemble and time.” That pretty much sums it up.
Bernoulli’s call to map returns through the logarithmic function was a normative one, not a descriptive one. We
experience the markets’ returns in arithmetic space; a return over any given time interval is arithmetic, by
definition. If we had only one bet to make in our life, maximizing this arithmetic return might be appropriate. But
from a long-term investment standpoint, when we have many multiplicative bets to make, or many bets whose
results compound over time, we need to map present arithmetic returns into future geometric returns in order to
maximize our end point wealth. To do so, we need to experience the markets’ returns through the lens of the
logarithmic function. We need to be Bernoullian logarithmic risk-takers. (This distinction is the best instance I
know of what I have called the “direct” versus the “roundabout” approach, the latter being the most productive for
the long term.) While that may sound unnaturally Spock-like, it’s actually the most natural thing we could do as
successful investors.
Raising a portfolio’s long-term geometric return is the very point of any effective risk mitigation strategy. All risk
mitigation strategies aim to do it, but, as we saw over the past decade, most fall short. Effective risk mitigation
happens to be a really hard thing to do.
Our legacy to-date speaks for itself. I hope that this retrospective portfolio-level view of what we have delivered
over the past ten years gives you a helpful, more holistic perspective on the benefit that Universa’s tail hedge has
provided to your portfolio. It should also attest to the destructiveness that the volatility tax—or Bernoulli’s
treacherous “concavity of curve”—has had on equity portfolios, and the benefit from effectively mitigating it.
At Universa, we are engaged in what can only be seen as something of a new paradigm in portfolio construction
and risk mitigation. We put Markowitz’s Modern Portfolio Theory on its head. While what we started ten years
ago (and personally even ten years before that) has since grown into its own “tail hedging asset class,” such
nomenclature is more style than substance. Despite our growth and maturation over these ten years, our approach
remains very much an unorthodox niche strategy and way of looking at investing, and I don’t see that changing
anytime soon.
We should expect the next decade to be at least as diverse as the last, with another great bust and boom and ever
wilder swings in volatility. After all, the very root of all of that last time was the economic and market distortions
of monetary interventionism, which is now certainly greater than ever before. But whatever turbulence lies ahead,
thanks to effective risk mitigation, together we shall remain agnostic again. We shall, in the words of Bernoulli,
“avoid the dice.”
Make no mistake, effective risk mitigation will remain our absolute focus here at Universa, just as it was from the
start, and we intend to continue providing all of our investors with a similar level of risk mitigation value-added in
our second decade as we have in our first. I look forward to showing that to you again in 2028, when I write you
our next decennial letter.
Cordially,
Mark Spitznagel
is an unmanaged, capitalization-weighted index of the common stocks of 500 large U.S. companies designed to measure the performance of
the broad U.S. economy. In contrast, the BSPP strategy invests in options, futures (including options thereon) and other instruments as well
as short sales, and includes a component designed to profit during months in which the SPX experiences significant declines. The SPX’s
performance reflects the reinvestment of interest, dividends and other earnings.
No Duty to Update. Neither Universa nor any of its affiliates assumes any duty to update or correct any information in this
document for subsequent changes of any kind.