Model Risk

You run a hedge fund. You think there’s a major profit opportunity in ruble-denominated caviar futures.

Before investing your fund’s total net worth in the plan, you hire a bunch of Harvard PhDs to put together a risk model. It’s a mathematical equation that quantifies your risk exposure.

You put in the data. It spits out an answer: you couldn’t possibly lose more than 20% of your investment. Confident that even the worst-case scenario isn’t that bad, you put the futures position in place.

The next day, Vladimir Putin announces that he’s de-valuing the ruble...and, in an unrelated decision, he's also outlawing caviar exports.

Your investment tanks. You lose everything. You bring in the Harvard PhDs to yell at them. They just shrug. “That wasn’t in our model,” they meekly point out.

Model risk. The risk that the intricate, game theory-informed, A.I.-optimized model you used as the underpinning of your decision wasn’t actually very good. It's the worry that the process you're using to figure out risk is, in itself, not perfect.