Satish Vutukuru

← All writing

Markets as information systems

· 1 min read

The most useful thing I have ever heard about markets came from an engineer, not an economist. He said: a market is a distributed system whose only consensus protocol is the price.

This is not the whole truth. But it is the part that engineers tend to miss when they reason about markets, and it is the part that economists tend to take for granted.

The protocol is lossy

Prices compress an enormous amount of private information into a single number. The compression is lossy by design. The order book does not tell you why someone is selling, only that they are. This is a feature when the private reasons are noise. It is a bug when the private reasons are correlated — when everyone is selling for the same reason, the price arrives at the new equilibrium with everyone already in it.

The model is not the market

Quantitative finance is full of beautifully calibrated models of markets that have never traded a share. The interesting question is not whether the model is correct in the abstract. The interesting question is what the market will look like in the world in which everyone is using a model of this kind.

Sometimes the answer is: like the model. More often: not quite, and the gap is where the returns are.

A useful working assumption

Markets aggregate information. They do not aggregate wisdom. The two are related, but they are not the same, and when they diverge it is usually expensive to be on the wrong side of the difference.

Related

When the second copy stopped being free

For forty years, the thing that made software the best business in the world was a quirk: serving one more user cost almost nothing. AI inference puts a meter back on every answer, and that quietly unwinds the economics that scale, freemium, and 80% margins were all built on.

The verification economy

Agents made production cheap, so the cost of work is moving to the checking step. Checking is too big for human attention, which means it gets delegated too: my agent verifies your agent's work. That recursion has to bottom out somewhere, and it bottoms out the way commerce always has.

The fungibility gradient

Enterprise AI strategy leans on the idea that models are interchangeable: put a gateway in front and swap freely. That is true at the transport layer, false at the behavior layer, and the gap between the two grows with every step your application takes.

← All writing