Policy Risk Is Hedgeable. Almost Nobody Hedges It.

Treasury has an instrument for everything except the risks that now move earnings most.

Currency, rates, and commodity inputs each have a deep market, a desk that quotes them, and a hedge ratio a treasurer can defend. The exposures that reprice earnings hardest no longer sit there. A tariff line changes overnight, a compliance regime lands as direct cost, a bill rewrites sourcing economics, a shipping lane closes. The impact is direct and measurable. There is no instrument on the shelf, so the exposure is carried in full and explained afterwards as volatility.

A tariff notice is not a loss until it meets your exposure.

The same policy change means very different things to two companies in the same industry. What decides the number is the lanes they run, the inputs they buy, where their revenue sits, and whether their contracts let them pass the cost through. Signal data alone cannot produce a loss estimate: it has to be joined to an item master, a supplier list, and a set of customs entries. Most companies cannot produce that figure for themselves, which is why the risk goes uncovered even when the event is obvious to everyone watching.

Commodity futures worked because farmers showed up. Event markets are still waiting for theirs.

Regulated prediction markets already list event contracts on the exact questions that decide these exposures: whether a tariff takes effect, whether a rule is finalised, whether a threshold is crossed by a date. The venue exists. What is missing is the other side of the trade. Almost every participant is there on an opinion rather than an exposure, so the books stay thin and a corporate-sized order would move the price against itself before it filled. Commodity futures did not begin as speculation. They began because producers had a harvest to protect and someone was willing to take the other side. Event markets have the speculators already. The hedgers are the missing half, and they are the half that makes a market deep enough to be worth using.

A binary hedge is honest arithmetic: cover a loss and you pay about what the market thinks that loss is worth.

A contract that pays one dollar on the event costs its market price. Covering a loss therefore takes L / (1 - p) contracts and L x p / (1 - p) in cash, paid up front and in full: no margin, no leverage. That arithmetic has a hard edge. At fifty cents the hedge costs as much as the loss it covers, and above that it is worse than absorbing it. Protection is worth buying in the band where an event is probable enough to hurt and still cheap enough to insure. Past it, the market is not telling a company to hedge, it is telling them to change the supply chain. A system that sizes a hedge without saying which side of that line you are on has not done the work.

Trust is built not by making bold claims, but by showing the model.

Language models are good at reading a docket, classifying what changed, and routing it to the exposure it touches. They are the wrong tool for pricing it. The market already publishes a probability, continuously and at no cost. The work worth doing is measuring the exposure that probability would land on, and judging whether the quoted price is cheap or dear against it. Model above market and protection is on sale. Model below market and the honest recommendation is to retain the risk or fix it operationally. The bar is not accuracy alone, it is a skeptical treasurer being able to walk backward from the recommendation to the filing and the model run that produced it.