AI Trading Systems

How LLMs and AI agents could support market decision-making — built from first principles, grounded in real code, and honest about the risks that make control non-negotiable.

This section develops the LLM and agent half of my proposed PhD: how large language models and AI agents might support market decision-making, and — just as important — the controls that make that credible rather than reckless. The other half of the site (the Equation and Machine Learning libraries, the Quant Lab) is quantitative and verified against real data; this section is its qualitative, engineering counterpart, built to the same standard of honesty. The emphasis throughout is transparency and control: an agent should be a testable, auditable component of a system, never an unsupervised black box placing trades.

Where the quant sections compute and check every number, these entries ground every claim in real, runnable code — the actual Anthropic Messages API structure, correct model IDs, and the schema-constrained, guarded patterns my MarketLens AI app uses — so the design principles are shown working, not just asserted.

Published entries

Added one at a time, to the same standard as the rest of the site. The roadmap below is the plan.

AI Risk Controls

safety

You cannot make a language model trustworthy; you make the system safe around an untrustworthy component. This entry synthesises the whole section into one control stack — input, invocation, output, action and monitoring — governed by two principles: defence in depth, and fail safe. The layers, what each defends against, and the honest limit that this buys bounded, auditable failure, not none.

7 min

Agent Memory & Planning

agents
safety

Tools give an agent hands; memory gives it continuity and planning gives it reach. Both are what make an agent capable of more than a single call — and both are autonomy dials that a system near capital turns down, not up. Short-term versus long-term memory, the planning patterns, why long-horizon planning is unreliable, and the point-in-time, bounded, gated discipline that keeps them safe.

7 min

Hallucination Risk When LLMs Meet Financial Data

safety

A language model produces plausible text, not true text; the two usually coincide and occasionally, fluently, do not. In a chatbot that is an annoyance. In a system that trades on the output it is a liability. The taxonomy of financial hallucination, the honest limits of every fix, and the one rule that matters most: use the model for text, never for numbers you can look up.

9 min

LLMs versus Conventional ML

foundations

The instinct is to stage a duel between LLMs and ‘traditional’ machine learning and crown a winner. That framing is already out of date: by 2026 there are three distinct families, and the useful question is not which is best but which belongs where — a discrimination a market system depends on, and one the efficient market puts a hard ceiling on.

7 min

Multi-Agent Coordination

agents
safety

The section began by defining an agent and ends by wiring several together. The fashionable image of multi-agent AI — models debating in natural language until they agree — is exactly what a trading system must not build. Real coordination is a structured, typed shared state, a fixed decision flow, a deterministic arbiter with the final say, and honest uncertainty carried through. The engineering of coordination, and how it meets the game theory the ML tier already formalised.

8 min

Retrieval-Augmented Generation (RAG)

news agent
safety

The schema stops a model emitting a trade or a malformed field, but not a fluent misreading of a fact. RAG is the layer that grounds the model in retrieved, current, authoritative source text so it reasons over facts you provide rather than facts it half-remembers. How it works, the real code, why every retrieval must be point-in-time, and what it does and does not fix.

8 min

Schema-Constrained News Extraction

news agent
safety

The News & Narrative Agent points a fluent, hallucinating language model at market-moving text and asks it to produce inputs a trading system will act on. The single decision that makes that safe rather than reckless is to constrain the output to a bounded schema — typed features, never instructions. The real Anthropic API code that enforces it, the guardrails around it, and why the schema is the safety mechanism.

10 min

Sentiment & Narrative Interpretation

news agent

The seductive pitch for news in trading is ‘positive headline, buy.’ The efficient-market evidence on this site says that is mostly noise — obvious news is priced in milliseconds. So this entry goes past the sentiment score to what text can genuinely offer a market system: novelty, narrative, and risk context — and is honest that its home is the Regime and Risk agents, not a buy signal.

7 min

Tool-Using Agents

agents
safety

Tools are what turn a language model from a thing that answers into a thing that acts, and they are the fix for its worst weakness — recalling facts it should look up. But every tool is also an attack surface, so the tool set is where control lives. The mechanics in real code, the design rules for a market system, and why the tools you withhold matter as much as the ones you grant.

8 min

What Is an AI Agent?

foundations
agents

The word ‘agent’ does a lot of work in this proposal — six of them. Before building on the idea it is worth being precise: the difference between a language model and an agent is the difference between a system that answers and one that acts, and in finance that difference is everything. The agent loop, the real Anthropic API code behind it, and why the proposal keeps its agents deliberately un-autonomous.

9 min

Why an LLM Must Never Directly Place Trades

safety

The single non-negotiable rule of the whole design: no language model is an execution authority. The case for it — hallucination, non-determinism, miscalibration, and above all prompt injection through the untrusted text the News Agent reads — and the deterministic risk gate that enforces it, so that a useful but untrustworthy model can inform every decision and make none.

8 min
No matching items

The roadmap

Foundations

What is an AI agent? · LLMs versus conventional machine-learning models.

The News & Narrative Agent

Schema-constrained news extraction · Retrieval-Augmented Generation (RAG) · sentiment and narrative interpretation.

Agent capabilities

Tool-using agents · agent memory and planning · multi-agent coordination.

Safety and control

AI risk controls · hallucination risk when LLMs meet financial data · why an LLM must never directly place trades.

Complete — eleven entries across four tiers, each grounded in real code and honest about its limits: the qualitative counterpart to the proposal’s quantitative evidence.

The proposed multi-agent architecture

A concrete system I want to study — deliberately decomposed so each component can be tested and evaluated on its own, rather than trusting a single model end to end:

  • Market Regime Agent — identifies the current market state and estimates regime probabilities (HMMs, mixtures, change-point detection — foundations already built).
  • News & Narrative Agent — converts unstructured text into bounded, schema-constrained, timestamp-safe market features; never free-form trading instructions.
  • RL Trading Agent — selects exposure, timing, and position-management actions from the combined state, trained on risk-adjusted outcomes.
  • Risk Management Agent — approves, scales, or rejects any proposed action against hard, pre-defined limits (exposure, drawdown, Expected Shortfall).
  • Review & Learning Agent — evaluates, after the fact, whether each decision was justified given the information available at the time, and attributes outcomes per agent.
  • Behaviour Detection Agent (exploratory) — flags microstructure patterns consistent with elevated systematic activity.

The Review & Learning Agent matters most: without a rigorous, honest evaluation loop, a system like this is just automation of overconfidence. The PhD Research page presents the full formal proposal built around this architecture — objectives, hypotheses, methodology, and the evidence already in place across this site.