Cap the Downside, Keep the Option

The Core Insight

Societe Generale sold close to 70 billion dollars of stock on January 21, 2008, into a thin holiday market. World markets dropped close to 10 percent and the firm lost close to 6 billion dollars. A sale one tenth the size, 7 billion dollars, produces no loss at all. Size did the damage.

Nassim Nicholas Taleb published Antifragile in 2012 as the main volume of his Incerto. The argument runs to one sentence. Fragility is measurable and predictable where risk is not, because everything nonlinear either gains or loses from volatility.

The word did not exist. Taleb searched Latin, Greek, Romance, Slavic, Germanic, Semitic and Turkic families and found no term for the opposite of fragile. Mail glasses to Siberia and you stamp the box fragile. The logical opposite of that stamp is a box marked please mishandle.

Most people who manage risk work on the forecast, extending the data and tightening the confidence interval. Taleb argues that the shape of the exposure decides the outcome, and the shape is visible today. A glass on a table loses from volatility whatever the weather does next.

The Framework

Exposure comes in three states, and most accounting stops at two.

• The fragile loses from disorder and is unharmed at best. The emblem is Damocles under a sword hung on a horse hair.
• The middle state returns to its starting condition, unharmed at best and at worst. The emblem is the Phoenix.
• The antifragile gains from disorder. The emblem is the Hydra, which grows two heads for every one cut off.

Written as arithmetic, antifragility is fragility with a negative sign in front of it. Fragility means more downside than upside. Antifragility means more upside than downside. Seneca ran the second one at scale, holding three hundred million denarii while traveling with what a shipwreck leaves behind.

The one operational instruction is the barbell. Put 90 percent in cash or another store of value, and 10 percent in maximally risky securities. The maximum loss is known and capped at 10 percent, and the upside stays open. Medium risk carries the largest measurement error, so a portfolio held entirely there carries a real chance of ruin.

The shape repeats outside finance. Taleb's personal version bans smoking, sugar, motorcycles, city bicycles and any plane without a professional pilot, and permits every other risk. Georges Simenon wrote sixty days a year, did nothing for three hundred, and published more than two hundred novels.

Hugo Schulz described hormesis in 1888 after small doses of poison stimulated yeast growth while larger doses harmed it. Taleb inverts the textbook reading, making hormesis the normal case and its absence the injury.

Key Ideas

Optionality Is a Substitute for Prediction

Thales put a down payment on the seasonal use of every olive press near Miletus and Chios. The harvest came in large, and he released them on his own terms. Aristotle credited superior astronomy, and Taleb corrects him. Thales bought the right to use the presses without the obligation, at a small known price.

An option is asymmetry plus rationality, and the rationality only has to recognize the good outcome after it arrives. Thales needed no theory of weather. Nature runs the same filter, and about half of all embryos end in spontaneous abortion.

Research payoffs have unbounded upside and a floor under the loss, so the return scales with the count of trials. His funding rule is 1 over N: with n options, put equal amounts into all of them. The chapter closes with four rules.

• Rank every opportunity by how much optionality it carries.
• Prefer open-ended payoffs to closed-ended ones.
• Invest in people who change direction six or seven times across a career, and skip the business plan.
• Keep the barbell underneath every position.

Convexity Pays More Than the Average

Take a six-sided die that pays its face value. Its expected payoff is 3.5, and the square of that is 12.25. The average of the squares runs 1, 4, 9, 16, 25 and 36 over six, which is 15.17. The gap is a 24 percent edge, and it comes from the convexity alone.

The reverse runs the same way. The square root of 3.5 is 1.87, and the average of the square roots is 1.80. A linear payoff needs to be right more than half the time, and a convex payoff needs much less. You can guess worse than random and still come out ahead.

Size and Speed Manufacture Fragility

Harm from a shock rises faster than the shock. A king swore to crush his son with a large stone, so the sage had it cut into pebbles. Pacing the living room does no damage, and thirty feet is roughly the cutoff for death from a fall.

Add 10 percent more cars and travel time jumps 50 percent. Two hours at an average of seventy degrees Fahrenheit sounds ideal. The first hour runs at zero and the second at one hundred and forty, and you get a funeral.

A government projects 9 percent unemployment and a two hundred billion dollar deficit. At 8 percent the deficit runs 75 billion, at 9 percent 200 billion, and at 10 percent 550 billion. The average of the function is 312 billion, so the single-point forecast understates the hole by 112 billion.

Efficiency removes the slack that absorbs error. Wheat prices more than tripled between 2004 and 2007 on a 1 percent rise in net demand. Taleb built a detector with Raphael Douady that measures the acceleration of harm as the shock grows. Applied to Fannie Mae in 2003, a move up produced large losses and a move down small profits, both accelerating. It works with a wrong model, the way a defective scale still shows whether you are gaining weight.

Removal Is the Safer Intervention

Charlatans give positive advice, and only positive advice. Chess grandmasters win by not losing, and people get rich by not going bust. Negative knowledge holds up better under error, because one black swan disproves every white one.

In 1930s New York, doctors examined 389 children and recommended surgery for 174 of them. A second set saw the remaining 215 and sent 99 to surgery. A third set saw the remaining 116 and recommended 52. Morbidity ran at 2 to 4 percent, with a death in about every 15,000 operations, and nobody looked for the break-even point.

Harm from the treatment stays roughly constant while the benefit tracks severity. In hypertension, mild cases have a 5.6 percent chance of benefiting, high cases 26 percent, and severe cases 72 percent. The iatrogenics live in the patient rather than the treatment.

If yearly data is half signal, daily data is 95 percent noise and hourly data is 99.5 percent noise. The rule is to look only at very large changes. Taleb claims a 1 percent modification cuts fragility by about 99 percent.

Skin in the Game Prices the Advice

Hammurabi's code is about 3,800 years old and settles it in one line. If a builder builds a house that collapses and kills the owner, the builder is put to death. The rule buys information rather than revenge. The builder knows more than any inspector about what sits in the foundations.

Modern finance broke the rule twice over. Robert Rubin collected about 120 million dollars in Citibank bonuses over a decade, Citibank collapsed, and taxpayers covered the loss. Joseph Stiglitz assessed Fannie Mae with Peter and Jonathan Orszag and put the risk of a default at effectively zero. Fannie Mae went bust at a cost of hundreds of billions, and Stiglitz published a book in 2010 claiming to have predicted the crisis.

As Taleb writes it, the US stock market cost retirees more than three trillion dollars over a dozen years against government money market funds. Managers at those same companies got richer by close to four hundred billion dollars through options. A market that rises 50 percent and returns to flat pays the managers.

The cure fits one instruction. Never ask anyone for an opinion, a forecast or a recommendation, and ask instead what they hold in their portfolio. The chief ethical rule sets the boundary. Never buy your own antifragility with someone else's fragility.

Time Already Ran the Test

Split the world into the perishable and the nonperishable. A 40-year-old man has about 44 years to go, and at 41 he has a little more than 43. A book in print forty years can be expected to stay in print another forty, and one that survives another decade gets fifty more.

Richard Gott ranked the Broadway shows running in mid-1993 by age, and the ranking predicted survival with 95 percent accuracy. As a child he compared the Great Pyramid, fifty-seven hundred years old, to the Berlin Wall at twelve, and guessed right.

Forecasting by addition is the disease Taleb calls neomania. Subtractive prophecy removes what is fragile and keeps what already paid its dues. Taverns date back twenty-five centuries, wine six millennia, and glass drinking vessels at least twenty-nine hundred years. Read as little as feasible from the last twenty years.

Practical Applications

Find the barbell in your own balance sheet first. Name the cash floor you never risk and the fraction you can lose whole. Move every medium-risk position to one side or the other.

Fund trials rather than plans. Take a fixed budget, split it equally across every candidate bet, and keep each stake small enough that the loss is survivable. The payoff scales with the count of trials.

Run the removal pass before the addition pass. List what you are adding this quarter, then price what taking one item out does to the worst case. Taleb's filter is one line: if you have more than one reason to do something, do not do it.

Change what you ask advisors. Stop asking for the recommendation and ask what they hold, what they are exposed to, and what they lose if they are wrong.

Apply the age filter to what you adopt. A technology twenty-five years old has a fair chance of running another twenty-five, and one released this quarter carries no record at all.

Who This Is For

Founders and operators carrying one concentrated exposure get the most from this book. Anyone whose downside is open and whose upside is capped holds the wrong shape. People running a portfolio of small bets get the optionality chapters.

Skip it if you want a forecasting method. The book supplies exposure design, and it is openly hostile to prediction as a discipline.

The book is long, repetitive and busy settling scores. Taleb recycles the turkey, the wheeled suitcase and the lecturing-birds image many times with the same worked example. The prologue, conclusion and glossary restate the core several times over. Roughly two thirds of the text is ornament. That covers fiction about a character named Nero Tulip, wine lists, etymological digression, and named attacks on people who annoyed the author.

Several claims arrive asserted rather than demonstrated, including hormesis as the normal case and a 1 percent modification cutting fragility by 99 percent.

The barbell allocation is the sharpest case. Taleb prints 90 percent and 10 percent with no backtest, no drawdown series and no account of what the cash side earns. That is the first evidence any portfolio manager asks for. He then concedes that the form is not essential, and that anything removing the risk of ruin qualifies. The concession is honest, and it drains the number of its authority.

The mechanism survives all of it. Asymmetry of payoff is arithmetic, and Jensen's inequality is a theorem. A capped loss with an open upside beats a symmetric bet of the same average, and the die example prices the gap.

The Decision

The test is one page and it takes an afternoon. Name your single largest exposure: one customer, one platform, one funding source, one employer. Write what a small shock costs you, then what twice that shock costs. A cost that more than doubles marks a concave position, and concavity means the size is the problem.

Cut the size first. Cap the loss at a number you name in advance, and put what is left into bets that pay unbounded. Do that before you spend another hour on the forecast.