AI in trading
Agents and algorithms can take real work off a trading desk. Before they write to the books, the controls have to be in place. We build both.
Where agents and algorithms work
Confirmations and settlement exceptions. An agent checks each confirmation against the booked trade. It chases the breaks and drafts the fix for someone to approve.
Reconciliation. Positions, invoices and exchange statements compared every day. Each difference explained, not only listed.
Surveillance triage. Alerts sorted before an analyst opens them. Each one arrives with the orders and trades behind it.
Bidding and dispatch. Day-ahead and intraday bids for a battery or a portfolio. Dispatch that follows the market within limits you set.
The controls that come first
Under REMIT II, software that sets price, volume or timing is algorithmic trading. That holds whatever the vendor calls it. In plain words, the rules ask for four things.
Notification. Your regulator and ACER are told that you trade by algorithm.
Limits that stop erroneous orders. Thresholds stop a bad order before it is sent. A plan covers the day the system fails.
Testing and monitoring. The algorithm is fully tested before it goes live, and watched while it runs.
Records the regulator can check. The regulator can also ask for your strategies, limits and test results.
Most firms meet these duties with a way to stop the algorithm at once.
Agents that never place an order still write to the books. A wrong confirmation costs the same whoever made it. So the rules of the book go in before the agent does.
That is the job of Morpholog, our open-source compliance core. This page is a summary, not legal advice.
An example: confirmations and settlement
Most confirmations match the booked trade. The rest wait in a queue for someone in operations. Settlement breaks pile up behind them.
An agent reads each confirmation and compares it with the trade. Where they differ, it proposes a correction or flags the counterparty's error.
Every proposal goes through Morpholog. A correction that would breach a limit or reopen a settled period is refused. The agent gets the reason in business words. It can fix the proposal, or hand it to a person.
Each accepted change is logged with the reason it was allowed. When an auditor asks about a trade, the answer is a query, not an investigation.
The analysts keep the exceptions that need judgement. This is an illustration of the work, not a client case.
How we work
With your team. We work alongside your AI, operations and trading technology people. The controls go in first, then the first agent workflow. Your team owns both when we leave.
Through your consultancy. A consultancy delivering an AI programme holds the client. We bring the trading knowledge and the compliance core.
A first piece of work is usually one agent workflow, with its controls and its records.