AI agents for quantitative research

From market hypothesis to tested strategy.

SignalThesis agents explore your authorized market data, generate analysis code, coordinate backtests, and write up what they found, including what the results leave out.

research-notebook / run 01 Illustrative example

H₁ HYPOTHESISShort-term momentum persists after earnings announcements in liquid equities.

Data explored
Code generated
Backtest run
Report drafted
STRATEGY vs BENCHMARK (SHAPE ONLY)

After modeled transaction costs. The gap from the gross line is shown, not hidden.

LIMITATIONS LOGGED

  • Cost model assumes fixed spreads; results are sensitive to this.
  • Dataset coverage gaps flagged for two periods.
  • Performance concentrated in one market regime.
Awaiting researcher review Not for live use until reviewed
§ 01Method

A research process, written down as it runs.

Agents handle the repetitive parts of quantitative research. Every step leaves a record your team can inspect, rerun, and challenge.

01

Explore

Agents survey authorized market datasets: coverage, structure, gaps.

02

Hypothesize

Research questions are framed as testable hypotheses before any result is seen.

03

Code

Analysis and strategy code is generated and kept with the run.

04

Backtest

Backtests are coordinated with explicit settings, costs, and assumptions.

05

Document

Findings are written up, limitations included, for your team to review.

Reproducible by design: data, code, and configuration travel together, so any result can be rerun.

§ 02Core workflows

Five workflows. One research record.

From the first look at a dataset to the report your committee reads.

2.1

Market data exploration

Understand what a dataset covers, where it is thin, and what it can support.

Data profile
2.2

Signal research

Turn an idea into a defined signal and test whether it holds up.

Signal study
2.3

Strategy backtesting

Run backtests with transaction costs and assumptions stated up front.

Reproducible backtest
2.4

Portfolio scenario analysis

See how a strategy behaves across scenarios and alongside existing positions.

Scenario set
2.5

Research reporting

Document results, methods, and limitations in a consistent format.

Research report
§ 03The first agent

Give it a hypothesis. Get back a backtest you can rerun and a report you can read.

IN

Your trading hypothesis

Stated in plain language, the way your researchers already write them.

A

Data explored

The agent checks what the authorized data can and cannot support.

B

Code generated

Analysis and backtest code is written and stored with the run.

C

Backtest coordinated

Costs, assumptions, and settings are recorded alongside the results.

OUT

Backtest + research report

Results and limitations, ready for your team's review.

RESEARCH REPORT / OUTLINEStatus: draft for review

What every report contains

A consistent structure, so findings can be compared and limitations are never an afterthought.

  1. 1
    HypothesisThe question, as stated before testing
    Defined
  2. 2
    DataSources, coverage, and known gaps
    Profiled
  3. 3
    MethodSignal definition and backtest settings
    Recorded
  4. 4
    ResultsGross and after modeled costs
    Both shown
  5. 5
    Risk considerationsWhere and how the strategy can fail
    Flagged
  6. 6
    LimitationsAssumptions and data constraints, stated plainly
    Flagged
  7. 7
    ReproducibilityCode and configuration to rerun the test
    Attached
Outline shown for illustration. Structure can follow your team's own template.
§ 04Rigor before live use

Every backtest leaves something out. SignalThesis writes it down before a strategy is considered for live use.

[a]

Assumptions

Every modeling choice is recorded, so reviewers see what the result depends on.

[b]

Data limitations

Coverage gaps and dataset constraints are flagged, not smoothed over.

[c]

Transaction costs

Results are shown after modeled costs, next to the gross numbers.

[d]

Risk considerations

Concentration, regime dependence, and failure points are part of the report.

Authorized data only

Agents work with the market datasets your team is authorized to use.

Reproducible workflows

Any finding can be traced back to its data, code, and configuration, and run again.

Researchers decide

Agents accelerate the research. Your team evaluates the evidence and decides what moves forward.

§ 05Who it's for

Built for teams that test before they trade.

5.1

Quantitative research teams

5.2

Investment funds

5.3

Proprietary trading firms

5.4

Financial research companies

Get in touch

Discuss your research process.

Tell us how your team researches and tests ideas today. We will follow up to talk through where agents could fit.

  • 01How hypotheses move from idea to backtest
  • 02The datasets and tools you already use
  • 03How results are reviewed before live use

We only use these details to respond to your enquiry.

Thank you.

We have your details and will be in touch to set up a conversation.