CFA AI Investment Challenge 2026 · Team 18

AI-Powered
Optimal Trade Execution

Minimising market impact on large institutional orders through regime-aware reinforcement learning, classical benchmarks, and transparent AI governance.

The Challenge

Why Optimal Execution Matters

When institutions trade large orders, poor execution can cost millions. Every basis point matters.

Market Impact

Large sell orders push prices down against the trader. The faster you trade, the more you move the market — but waiting carries its own risks.

Timing Risk

Markets shift between calm and volatile states. A strategy that works in calm conditions can be catastrophic during stressed periods.

Accountability

Regulators and compliance demand transparent, explainable decisions. "The AI decided" is not sufficient justification for a trading desk.

Our Solution

Four Pillars of Intelligent Execution

We combine statistical market analysis, adaptive AI, proven benchmarks, and human-readable governance.

Regime Detection

Hidden Markov Model identifies whether the market is in a calm or volatile state using realised volatility and order-book signals.

Reinforcement Learning

A PPO agent learns to optimally pace trade execution — balancing urgency against market impact in each regime through trial and reward.

Classical Benchmarks

We compare every trade against TWAP, VWAP, Almgren–Chriss, and immediate execution to prove real improvement.

LLM Governance

Every execution decision is explained in plain English by Claude, providing compliance-ready audit trail for portfolio managers.

Architecture

End-to-End Pipeline

From raw market data to governed execution in five stages.

Market Data

Price, volume, BBO

Regime Detection

HMM classification

RL Agent

PPO policy

Execution

Benchmark comparison

Governance

LLM explanation

Performance

Key Results

100%

Order Completion Rate

The RL agent completes full order liquidation within the execution horizon across all test scenarios.

Regime-Aware

Adaptive Strategy

HMM detects calm vs. volatile markets. The agent adjusts pace accordingly — slower when calm, faster under stress.

Explainable

Every Decision Governed

Claude generates compliance-ready explanations for each execution decision, covering regime context, pacing rationale, and risk.

See It in Action

Walk through a complete execution scenario in our Case Study, or configure and run your own in the Execution Lab.