Thirty years ago I began in the open outcry pits on the LIFFE floor in London. Almost everything about how a trade is executed has changed since then. Almost nothing about what happens after it has.
The thing people remember about the floor is the noise. What mattered was what happened when the noise stopped.
A price was agreed by two people who could see each other. That part was fast, and in its own crude way it was honest: the market was a room, and the room knew what it had just done. Then the trade left the room. It went onto a card, into a clerk's hands, along a chain of institutions none of whom had been standing there, and into a settlement process that would tell you some days later whether the thing you thought you owned had actually arrived. The agreement took two seconds. The administration took a week.
The market had two systems then. It has two systems now. There is the system that finds the price, and there is the system that moves the money and the asset. The first was theatre. The second was invisible. Over thirty years the first has been dismantled and rebuilt from the ground up. The second has been wrapped in better software and left broadly where it was found.
I am not sentimental about the pit. Anyone who tells you the floor was fairer was not standing at the back of it. Height helped. Volume helped. Knowing which broker's clerk owed you a favour helped. Screens took a great deal of that away, and they took spreads with it. Prices are tighter, access is wider, timestamps are auditable, and nobody loses a hundred thousand because two men misheard each other. Electronification of execution was a genuine public good and I would not undo a minute of it.
What I want to account for is the other half of the trade, because I spent three decades watching it not move.
I traded bonds, swaps, foreign exchange and options, and I ran desks in London, Madrid, New York and Geneva. The instruments changed with every desk. The chain behind them did not. Whatever I had just done on the screen ended up in the same place: an instruction sent to a correspondent, an account held at a bank in another country, a cut-off time set by somebody else's clock, and a window in the middle of the day during which I had delivered and had not yet received. We knew that window existed. We funded against it, we set limits around it, we watched it close every afternoon. What we never did was ask why it existed at all.
The costs of a trade behaved very differently on either side of that line. On the execution side, the line went almost vertically down over thirty years. On the settlement side, it did not follow, because that cost was never really a technology cost. It was the price of a chain of institutions, each of whom had a reason to be in it.
The commission convention survived electronification completely intact. It was renamed. It was tiered. It was bundled into a platform fee, then unbundled again, then rebadged as connectivity, market data, clearing, and access. It was never abolished, because it is not a cost of the system. It is the revenue of the system, and no institution has ever funded the removal of its own revenue line.
That is the whole argument, and it is not a conspiracy. It is capital allocation working exactly as it should. Making execution faster increased volumes, and per ticket revenue is a function of volume, so the money went there and did very well. Rebuilding settlement shortens the chain, and every party in the chain earns from being in it, so the money did not go there. Technology was aimed, over thirty years and with great skill, at the parts of the market that did not threaten the fee. That sentence explains more about the architecture of wholesale markets than any amount of writing about latency.
The moment I understood this was not dramatic. I was looking at a settlement instruction, and I could not tell from its content whether it had been produced that morning or fifteen years earlier. The wrapper had changed. The message had not. The same kinds of parties were named in the same order. The same buffer of days sat between the agreement and the money. Somebody in operations was going to spend the afternoon on the phone about it, exactly as somebody had spent the afternoon on the phone about it when I was standing in a ring.
A reconciliation exists because two records of the same event exist and nobody fully trusts either of them. Every reconciliation team on every desk in the world is evidence that the plumbing produces two versions of the truth and then employs people to argue about which one is real. Over thirty years we did not remove that condition. We automated the argument, and we called it progress. It was progress, of a sort, in the same way that a faster fax machine is progress.
Here is the part that is inconvenient for me, and for what I am now building.
First, I did not discover this problem from outside. I priced it into my own profit and loss for years and did not think of it as a problem, because it sat in a column marked cost of doing business, and everyone I competed with carried the same column. That is how structural costs survive. They are not hidden. They are shared, and shared costs stop looking like costs.
Second, the old chain has one enormous advantage over anything intended to replace it: it has been through crises. It has failed, and been repaired, and the market has learned in detail how it behaves when everything else is going wrong. Netting through a correspondent is not only rent extraction. It is also a very efficient way to economise on liquidity, and a fully prefunded alternative is not automatically cheaper in every respect. Anyone who tells you otherwise is selling. A new rail earns that knowledge the hard way, over years, and it has not earned it yet.
Third, IDBX is pre revenue. It is not authorised, and nothing here should be read as suggesting otherwise. The hardest parts of what we are attempting are not software problems: they are permissions granted by other people, on their timetable, and consortium formation among institutions with every reason to move slowly. I can describe the design. I cannot promise the outcome, and I will not put dates on approvals that are not mine to give.
What I can say is that the argument comes from inside the model. I helped run the model. I was paid by the model. I am not describing a market I read about.
The pit taught me one thing worth carrying out of it. A market is a place where people agree a price. Everything after that is administration. We spent thirty years making the agreement instantaneous, and we left the administration approximately where we found it in 1996, with better screens on the front of it.
That administration is the work that is left. It is not glamorous, it does not demo well, and it is where nearly all of the remaining cost lives.