MARKETS

The cost is the finding

5 min readMonezys

There is a particular kind of disappointment that arrives about six weeks into building anything that trades, lends, routes or settles. The idea works. You can see the edge in the data, it is stable across periods, it survives the obvious robustness checks. Then somebody adds up what it costs to actually do, and the edge is gone — not reduced, gone, and sometimes considerably past gone.

This is not bad luck and it is not a failure of the idea. It is the normal result, and the only unusual thing about it is how late most teams find out.

Gross is a number about the world. Net is a number about you.

A gross result tells you something true: that a pattern exists, that a price moves a certain way after a certain signal, that a certain kind of borrower repays. It is a statement about the market. It is also not the number that determines whether the company works.

The net result is a statement about you — your spreads, your fees, your settlement failures, your slippage, the person who has to check something by hand, the support ticket, the chargeback, the reconciliation break that takes an afternoon. Those come out of exactly the same number the upside goes into, and they are specific to how you are built rather than to how the market is.

A result that has not been measured net of what it costs to run is not a result. It is a reason to go and find the result.

Why it gets found late

Three reasons, usually, and none of them is carelessness.

  • The costs are spread across functions. The spread is a trading number, the failed settlement is an operations number, and the manual check is a headcount number. Nobody owns the sum.
  • Early testing happens at a size where costs round to nothing. The first hundred transactions are free in every sense that matters. The ten thousandth is not.
  • The gross number is more interesting to look at, and it arrives first.

What to do instead

Build the cost model before the signal, not after it. It does not have to be good — a crude, pessimistic, explicitly-labelled guess is worth far more at week one than a precise one at month six, because its job is to tell you how large the edge has to be before the thing is worth building at all.

That number becomes a threshold. If the gross edge is not comfortably above it, you have not found a business, and you now know that in week one. If it is, you have something, and you have also learnt exactly which cost to attack first — which is usually a more valuable finding than the signal was.

The cost often is the opportunity

The useful inversion: if a cost is large enough to kill a good idea, it is large enough to be a business on its own. Somebody is paying it. The question of who pays it, how often, and what they would pay to stop, is frequently a better question than the one you started with.

Illustrative example: a strategy earning roughly three basis points per transaction gross, paying roughly ten to cross the spread on the way out, is not a marginal business — it is a reliable way to lose money at scale. Finding that in week one costs a fortnight. Finding it in year two costs the company.