Developed by Nassim Nicholas Taleb (Antifragile, 2012) from Johan Jensen's inequality for convex functions (1906).
Reads a commitment by the shape of its payoff: convex when gains grow faster than losses under volatility, concave when the reverse holds.
The same swings in demand help one payoff and hurt another
The line is the plant's profit against plan at each level of demand: zero at plan, rising by the gain per unit above it and falling by the loss per unit below it. The square is the average profit when demand swings an equal amount above and below plan. When the loss below plan is steeper than the gain above it, the square drops under the line and wider swings cost more; flip the slopes and it rises above. Illustrative.
This payoff isConvex: swings raise the average, so keep the exposure and let it ride.Concave: swings lower the average, so cut the swings or reshape the payoff.
An example
A contract bottling plant forecast 100,000 cases for the coming year, give or take 20,000, and the operations director had two capacity plans. Plan A staffed and leased for the full forecast. Above plan, the extra cases ran on overtime and earned only $1.50 each; below plan, idle crews and the lease cost $6 a case. Plan B staffed for less and signed a standby agreement with a co-packer: above plan, the overflow earned $4.50 a case through the partner, and below plan the smaller base lost only $1.
The finance team's spreadsheet showed both plans at the forecast, where A looked $25,000 better. The director drew each plan's profit against demand instead. A swing of 20,000 either way cost Plan A $45,000 on average, though the forecast was right on average, while Plan B gained $35,000 from the same swings. B came out about $55,000 ahead in expectation, and further ahead in any year the forecast missed by more. The plant signed the standby agreement.
Why it works
A forecast gives one number, and a plan is usually priced at that number. The real result comes from the whole range of outcomes run through the plan's payoff, and when the payoff bends, the average of the results is not the result at the average. If losses below plan pile up faster than gains above it, every swing costs money on balance, so the plan loses value as uncertainty grows even when the forecast is unbiased. If gains grow faster than losses, the swings pay.
Reading the shape of the payoff shows which way uncertainty cuts before anyone knows how big it is. The practical move follows: keep exposures that bend up, and reshape the ones that bend down with caps, buffers, or options until their losses are bounded.
The thinking behind it
Johan Jensen Johan Jensen, a Danish telephone engineer who did mathematics in his spare time, proved the inequality that carries his name in Sur les fonctions convexes et les inégalités entre les valeurs moyennes (1906), in Acta Mathematica. For a convex function, the function of an average is at most the average of the function. Read for decisions, it is the formal basis for the claim that variability helps a payoff that curves up and hurts one that curves down.
Nassim Nicholas Taleb Nassim Nicholas Taleb, a former options trader who writes on risk, built this reasoning into Antifragile: Things That Gain from Disorder (2012), drawing on Jensen's inequality. His argument is that some systems gain from volatility, which he calls antifragility, and that the way to tell is the shape of the response to shocks, which can be read without forecasting the shocks themselves. The books are essays for a general reader, with technical appendices; some statisticians have faulted his treatment of statistics.
How to run it
- Name the input that varies. Pick the one uncertain driver the commitment rides on: demand, price, a rate, a delay.
- Draw the payoff against it. Plot the result at several input levels on both sides of the plan, not just at the forecast.
- Read the bend. If losses on one side grow faster than gains on the other, the payoff is concave; the reverse is convex.
- Price the swings. Average the results at a plausible low and high and compare that with the result at the forecast.
- Keep what bends up. For convex exposures, accept volatility and keep the cost of holding the option small.
- Reshape what bends down. Add a cap, a buffer, a standby partner, or an exit clause so the losses stop growing.
- Recheck as the range widens. When uncertainty grows, the gap between the two kinds of payoff grows with it.
Where it goes wrong
The plan is priced at the forecast only. A model that shows one demand number cannot show the bend, so concave plans look as good as convex ones; spot it when nobody can say what happens at plus or minus 20 percent. Ask for the result at both ends.
The bend sits outside the range examined. A payoff that is straight for small swings can turn sharply concave past a threshold (a covenant, a capacity wall), so test the extremes the business could actually see.
The option costs more than it earns. Convexity is worth paying for only up to its expected gain; an expensive standby agreement in a stable market is a cost with no swings to repay it. Compare the price of the option with the average gain it buys.
When to use something else
- Antifragility. When the question is how a whole system or organization responds to shocks, beyond one commitment's payoff.
- Barbell Strategy. When the job is splitting a budget or portfolio between very safe positions and small high-upside bets.
- Sensitivity Analysis. When a model has many inputs and the first need is to find which one moves the answer most.
Where it sits in SOLVE eX
Go to the source
- Antifragile: Things That Gain from Disorder, 2012
- Sur les fonctions convexes et les inegalites entre les valeurs moyennes, Acta Mathematica, 1906