AlgoQuant Agent

Status: in development. AlgoQuant Agent is a new Numerical Method product that has not been released yet. This page describes what it is being built to do. There is no download, price or release date yet; to follow its progress or to discuss early access, contact us.

The problem

Turning a trading idea into a tested strategy takes days of routine work: writing the strategy code, wiring up data, running backtests and sensitivity analyses, and writing up the results. Large language models can now do much of that routine work. Left unchecked, they also make it trivially easy to try thousands of variants and keep the one that happened to look best, which is the fastest known way to produce a strategy that fails out of sample.

The approach

AlgoQuant Agent is an AI research assistant built around AlgoQuant. The researcher describes an idea in plain language. The agent writes the strategy against AlgoQuant’s interfaces, runs the backtest and the robustness checks, and produces a report. Three design rules constrain it:

• Everything is recorded. Every variant the agent tries is logged, so the number of trials is known and can be used to adjust reported performance for multiple testing.
• Validation is out of sample by construction. The agent cannot see the data reserved for final evaluation.
• A human decides. The agent proposes and documents; nothing it produces is deployed without explicit review by the researcher.

What it is being built to do

• Translate a written idea into AlgoQuant strategy code that a researcher can read and edit.
• Run backtests, Monte Carlo and bootstrap analyses, and sensitivity analysis with AlgoQuant’s existing tools.
• Write a validation report that states assumptions, the number of variants tried and the out-of-sample result.
• Keep a reproducible record of every experiment.

How it fits

The agent uses the models in Quantitative Trading Strategies as reference implementations and the training controls planned in Tempo for machine-learning strategies.