Stop Chasing Tails: Some Long‐Only Opening Lines in Portfolio Theory It appears that Kyle Bass doesn’t know how to be long stocks. Whether this is a personal problem or not, it brings up the central problem of a permabear.
Being a permabear is painful like being long volatility during the summer of 2009. There is a time to be a bear, just there is time to be long volatility. The “perma” part is what sucks. It is cool to be a huge fan of long dated treasuries. The facts give indication that the United States—the global economy—is stuck in an irreversible cycle of deleveraging for some years to come. But that doesn’t mean there is no value in some equity exposure. At the very least, utilities generate reasonably reliable inflation‐adjusted yield. Two things at work: Investing return requires taking risk. Simple humility to realize that you can’t predict the future means that you must hedge both risking and derisking portfolios.
Center Bets versus Fat Tails: The Topology of Extremistan Everybody knows what “fat tails” if only vaguely. They refer to a higher probability of extreme events embodied in thicker tails of a distribution, compared to the normal distribution. Fat tails imply the relevance of the extreme events in life. Specifying the exact distribution is unimportant, because the distribution evolves over time. What matters is recognizing 1) the grave error in making based assumptions purely on linear cause and effect constructs and 2) most predictions underestimate the risks and rewards. Statistical models like regression are particularly prone to this defect, because they assume normality or near‐normality that does not factor in extremes… it puts weight on the body of a distribution and assumes the shape of the distribution doesn’t evolve over time. Even well‐worn tools like correlation are inadequate in capturing a non‐linear reality. That said, you hedge linear events: you can’t hedge non‐linear event very effectively. So these tools do have a useful place: they are ways to categorize thought, and they work well when the tail risks are small. Just like anything that maps human systems, they don’t work well when tail risks are large. There it is. The issue is not about a proper choice of probability distribution. It is not about human inability to find the “right” distribution because groping around for it is too hard. It is about reality constantly reinventing itself. All probability distributions have the potential of constant change at any given time and state, because the world undergoes phase transitions. It is not an issue of finding the unknown. It is an issue of accepting the unknowable. This is why all predictions fail and why everyone is eventually wrong. Not even the simplest processes can escape this graveyard. Consider the simple example of a coin flip off your thumb. One asks: “How
can the probability of a head or tail ever be appreciably different from 50%?” Well, if you flip a coin a million times off your thumb, that thumb will be rubbed to the bone and eventually fall off. You will be forced to flip the coin in a different way that ultimately can affect the distribution of heads to tails. The world does not resemble anything as simple as a coin flip anyway: predictions are exponentially harder. All one can do is base decisions on available data, use models that accommodate the widest range of potential states of the world, recognize that as the time‐frame gets longer, the forecast error grows exponentially. Who knows? You may get lucky. Forecast Error and Human Limits “My daughter was telling me about an Adonis who lives in a fraternity house opposite her school. ‘He’s so handsome, Muddy,’ she sighed. ‘So handsome! Why his face is simply carved with dueling!’” ‐“Time Mellowed Germany”, National Geographic, 1934
The chick that wrote the article I quoted form above is a classic case of the divergent errors of estimates compounding exponentially not arithmetically over time. The dude with the carved up face in 1934 may well have driving tank in Kursk in 1943. Probability is just a rule‐based way to synthesize what we observe. Since probability distributions are in general unobservable, all one can do is start with some prior distribution based on intuition and then modify it based on evidence. So if an event is repeated often enough the evidence will give “shape” to a distribution and most people will agree on what it should be. Only people with very different starting intuitions will disagree significantly. Even though their prior is very different, given enough time even diametric views will converge (more on this later). Due to measurement error and other practical limitations, nobody can ever prove something is 100% correct or 100% wrong. A probability distribution is a belief, with supporting evidence, and not a fact. The degree of successful prediction then is due to the starting intuition, the hunch that is your prior. In a sense, alpha is the pure luck of your assumptions being better than others at the start. See my picture. Distributions evolve, starting off flat like the uniform. Over time, sampling makes them look normal. Finally, the way‐out‐there extremes make the tails fat. Over time, the distribution fluctuates between flat and extreme as the underlying dynamical system parameters change.
Extremistan: Outliers Fatten Tails and Central Events Have More Weight
Mediocristan: Central Limits and Normality
Insufficient Data Leads to Uniformity of Outcome Likelihoods
Sources of Unpredictability So given this reduction of probability to degrees of belief, the essential issue would seem to be one of convergence. It is not. The real issue is that the unobserved probability distribution is unstable. History is a dynamical system that produces bad events when it “breaks down” and good events when it starts working “right” again. In terms of statistics, this reduces to drawing observations from a rapidly changing distribution—what is observed comes from a distribution uniquely defined at all points in time (the extreme case). This back to the coin flipping example: flipping a coin produces very stable outcomes and probabilities over a long time period, but ultimately your thumbnail will rub down to the quick, and you’ll flip the coin in an outcome altering way. But note something about the fat tail distribution: the center is even more concentrated than under the assumption of normality. So if you believe in fat tails, then it is more important to find the precise center than anything else. Worry about the tails after you focus the center. You may risk mange those tails terrific, but if you manage reward awful, you can lose big. Often when market sentiment is divided, the center bet is bi‐modal. Don’t lose sight of the importance of adjusting risk AND return. See picture. Missing the Center for the Tails
If your portfolio doesn’t cover the center adequately, the tails don’t matter at all. That is the pain of the permabear. Everything is concentrated on one tail. Some Tail Measures Wide differences in inflation breakevens across maturities Low Volatility implies perception of thin tails
Fat Tails, Risk Aversion and Leverage: Some Long‐Only Opening Lines in Portfolio Theory I’ve thought about portfolio building under alternative assumptions, and I use the Mediocristan‐ Extremistan shorthand employed in Taleb’s book The Black Swan, the Talmud, and hypermodern chess theory. No shorts involved. Mediocristan Mediocristan is the same as assuming normality, where the tails are not fat. You don’t need to worry about extreme events. One portfolio option finds alpha, and gets exposure to a wide center. Banks stocks are a strong center bet, but their common stock is more exposed to negative tail shocks. Their senior debt is more robust against shocks, in the extreme because of government guarantees.
Technology stocks are more exposed to positive shocks because they innovate. Ownership of these companies is desired, being a creditor is less desirable. So you pick solid companies across sectors and hold varying positions across their entire capital structure. You won’t insure against extremes like high inflation or financial systems collapses, but when you assume normality, there’s no need.
The Talmud Rule states that one should diversify. The portfolio below holds more senior debt than many, but it’s my chart porn. The point is indexation across the capital structure. Hold a diversity of debt from HG to HY. Preferred shares for max yield. Less diversified large caps because they have less room to grow, mid‐ and small cap to get exposure to growth and general positive shocks.
Extremistan Someone who believes in fat tails believes that there will either be a very small move or a very large move, and not much in between. I’m a believer: no matter how long we collect observations, something will come along that will surprise us at some point if we look over an ever‐longer timeframe. So we need both insurance and low‐probability/high‐reward bets far from the center to protect some part of our plans if things go wrong. Still, you must play the center, because money will concentrate on assets with more likely outcomes. There is more stability in investments when the vast majority of investors are more prepared for center best, especially if you beat them to it. You will lose loads if you bet entirely on the fat tails. It remains to buy tail exposure at the right price. Crashes are moments of clarity for risk, where the market correctly emphasizes tail outcomes. Melt‐ups are where to pick up cheap insurance. The center is where you accumulate safe yield. In Nasim Taleb’s actually very good book he advocated 90% allocation in treasuries (tinkering with maturities?) and 10% in whatever you hit on the dartboard. He understandably has modified his definition of safety to include more assets like gold or whatever. I frankly think it is ironic that the safest possible assets are difficult to identify. The rest should be invested in black swan hedges and dragon king positions, objectively based on some random selection. Good and bad shocks can’t be predicted, so smart people have no advantage. There is no alpha in Extremistan. Theory lies. Picture.
Diversification in Extremistan
Commodities (5%) Small Cap Common Stock (15%)
Utility Common Stock (10%) Large Cap Common Stock (10%) Preferred Shares/Convertibles (10%)
Senior Corporate Debt (20%)
Government Securities (30%)
Degree of Diversification
Some Tail Measures There is some profit in playing with tails, because tail risk is often underpriced and overpriced. Positive tails are more often than not overpriced. Come to think of it, so is buying puts. Buy underpriced tail‐ risk, and sell overpriced tail‐risk. Some indications of this are: Wide differences in inflation breakevens across maturities Low Volatility implies perception of thin tails
Maximize your exposure to positive tails, and minimize your exposure to negative tails. This may be obvious to the trader who buys low and sells high, but tails are hard to price.