The best intel on a deal is with another trader
4 sections
Before a trader sends an offer, there are a few things they would like to know. What did this buyer pay last week? Does this buyer pay on time? Did the last cargo of this grade pass inspection at that port?
No screen shows these. But the answers exist. For almost every deal, somebody did the same deal a short time ago and knows. That person is usually another trader.
This post looks at what a screen does not show, why the person who knows is another trader, why that trader rarely tells, and what AI agents change.
What no screen shows
For most physical trades, the facts that decide the margin sit in private deals.
Many products have no screen price. Copper and wheat trade on an exchange, and anyone can see the price. Many products do not. Take minor metals such as cobalt, gallium and molybdenum. A few have exchange contracts, but these are rarely traded. The prices people use come from price reporting agencies. Their reporters “collect prices from consumers, producers and traders” and publish a price every week or two.1 Reuters wrote that this process “relies on human judgement”.1 So even the published price is built from what traders tell a reporter.
A published price is an average. A deal is one case. A price for a standard grade at a main port does not tell a trader what one buyer will pay for one blend this week. The facts that move the margin are specific.
- Price. What this buyer, or buyers like it, paid on the last deal.
- Payment. Whether this buyer pays on time, and what it does when the market falls after the contract is signed.
- Quality. Which cargoes were rejected or downgraded at this port, and why.
- Rules in practice. How the port and customs handle this product today, which can differ from the written rule.
The trader who did the deal knows
The person who knows is a trader who has settled a deal with that buyer or that supplier.
Take a trader who is about to sell 100 tonnes of blended milk powder to a buyer in Manila at $2,480 a tonne. Another trader sold to the same group of buyers last week at $2,540. That one fact is worth $6,000 on this deal.2
Often the answer is in pieces. One trader supplies these buyers and knows what they pay. A second had a cargo downgraded there last month. A third got a better bid in another market. None of them has the whole answer, and none of it is written down where the first trader can look.
We think another trader is the best source, for three reasons.
They were paid, or they were not. What they know comes from a deal that was settled. It is a fact, not a view of the market.
It is recent. A deal from last week says more than a report from last quarter.
Their own money was on it. A trader who carried the cargo checked the buyer with care, because a mistake was their own loss.
Traders know this already. Most have a few people they call before a deal.
Why they rarely tell
Traders do share with each other, but in a small circle, and only what is safe to give.
This is not special to trading. In the 1980s Eric von Hippel of MIT studied eleven American steel minimills. At ten of them, people routinely gave process know-how to other firms, sometimes to rivals.3 They followed two rules. An engineer helped someone who could help back later. And if the know-how was “vital to his own firm’s competitive position”, he did not give it.3
We think intel moves between traders in the same way, and four things keep it small.
Asking takes time. To find who sold to those buyers last month, a trader may have to make ten calls. For one container it is not worth it, so the trader does not ask.
A favour has to be returned. A trader asks people they can help in return. That keeps the circle small, and it is smallest for a new firm.
The answer can cost the person who gives it. Telling a rival what a buyer paid can lose the next deal with that buyer. A trader in another market loses nothing by telling. But the trader who asks rarely knows who that is.
Nobody is paid. The trader who knows gets a favour in return, and no money.
So today a trader asks only when a lot of money depends on the answer, and mostly asks people they already know. A large firm has more of this inside the firm: a trader can ask a colleague on another desk or in another office. A firm of three people has its own circle and nothing more.
What AI agents change
An AI agent makes asking cheap. Trust and pay still have to be built.
Asking becomes cheap. An AI agent that works on a deal knows what the trader needs to find out. It can work out who is likely to know and send them a request, at any hour, for a deal of any size. The ten calls become no work for the trader, so a small request is worth sending too.
That is not enough. Intel does not move because it has become easy to ask for. We think three more things are needed.
- A record. The trader who asks has to know whether the last answers from this person were right. That needs a record of the answers and of how the deals ended.
- Control. The trader who answers has to choose what to share and with whom: for example, never with a firm that sells to the same buyer. Some intel should not be shared at all, such as anything a trader received under a promise to keep it private.
- Pay. A favour owed does not work between strangers. We think the fair price of an answer is a part of the extra margin it brought, paid after the deal has made money.
In our first post we wrote that AI agents let a small firm do the work of a team. Intel is the other thing a big firm gives its traders. We think small firms can have that too, and they can get it from each other.
Footnotes
-
Pratima Desai, Reuters, Inside the opaque world of minor metals pricing, July 2023. ↩ ↩2
-
The numbers are an example, not data from one deal. ↩
-
Eric von Hippel, The Sources of Innovation, chapter 6, Oxford University Press, 1988. ↩ ↩2
Flowpact is by invitation. Request an invitation
How AI agents will change commodity trading