Give your AI a backtesting engine
Install the NineQuantAI Skill and your AI assistant can backtest stocks, validate trading strategies and tune indicator parameters directly. No manual parameter fiddling — one sentence does it.
One-line install
Send the line below to your AI assistant. It will read our Skill specification and from then on knows which endpoints to call, how to fill the parameters, and how to interpret the results.
Read https://us2.ninequantai.com/skill/SKILL.md and use NineQuantAI's backtesting capability to analyse US equity strategies for me.SKILL.md is a plain text specification. No package to install, and no API key required — the guest quota is enough for a few runs. Register when you need a higher call quota.
Supported tools
We have tested the tools below. Any other assistant that can read a web page works too — just give it the SKILL.md URL.
OpenClaw
Send it the install line
Trae
Send it the install line
Claude Code
Send the install line, or drop the file into .claude/skills for auto-discovery
Cursor
Paste the install line in chat, or drop the file into .cursor/rules
What it can do
Backtest any ticker
Name a symbol and a date range, and the assistant runs the backtest and reads the result back to you.
Validate a trading rule
Describe your trading idea as rules and see what it would have done in real history.
Tune indicator parameters
Let the assistant sweep a parameter range and tell you which regions are plateaus and which are spikes.
Compare parameter sets
Run several parameter sets on the same symbol and compare return, drawdown and trade count side by side.
Core capabilities
Cross-market backtesting
US stocks and US-listed ETFs today, including delisted names with no survivorship filter; Hong Kong and A-shares are planned. The usable history is whatever the API's data_range reports, and the assistant relays it honestly.
Strategy validation
Chandelier-exit trend following and dual moving average, with entry filters, stop multiples, ADX and volume filters all tunable.
Parameter optimisation
Sweeps a parameter range and returns structured results, so the assistant can tabulate them and tell you which values are robust and which are flukes.
Risk assessment
Returns max drawdown, drawdown duration, profit factor and benchmark comparison — not just a headline return figure.
Feature list
Strategy core
- Chandelier-exit trend following
- Dual moving average crossover
- SMA direction filter
- N-day breakout entry
- ATR-adaptive stop
- Stop ratchets up only
Parameters
- SMA period 5–500
- Breakout period 2–250
- ATR period 2–100
- Entry stop × ATR
- Exit stop × ATR
- ADX and volume filters
Backtest output
- Strategy and benchmark return
- Max drawdown and its duration
- Win rate and profit factor
- Trade count and average hold
- Trade-by-trade detail
- Equity curve data
Three ways to install
- 1
Copy and send to your AI
The simplest option: copy the line above, paste it to your assistant, and it works immediately.
- 2
Install per tool
Put SKILL.md in your tool's skills or rules directory so the assistant reads it automatically, without pasting the URL each time.
- OpenClaw
Just send it the install line; no further configuration.
- Trae
Just send it the install line; no further configuration.
- Claude Code
Files in the project's .claude/skills directory are discovered automatically; ~/.claude/skills applies to every project.
mkdir -p .claude/skills/ninequant-backtest && curl -o .claude/skills/ninequant-backtest/SKILL.md https://us2.ninequantai.com/skill/SKILL.md- Cursor
Put it in the project's .cursor/rules directory, or paste the install line in chat.
mkdir -p .cursor/rules && curl -o .cursor/rules/ninequant-backtest.md https://us2.ninequantai.com/skill/SKILL.md
- 3
Download SKILL.md
Download SKILL.md into your project directory so the assistant reads it locally — suited to offline or private deployments.
FAQ
No. Guests can try it without registering, rate limited by IP; after registration usage is metered per API key, and the free quota covers ordinary research. Upgrade to Pro at ¥99 a month when you need more.
The backtest is computed by our server-side engine; the assistant only issues the request and interprets the result, so the numbers are not model-generated. But accurate numbers do not make a conclusion reliable — sample size, window choice and the number of parameter combinations you tried matter just as much. See Three Traps in Backtesting.
US stocks and US-listed ETFs today; Hong Kong and A-shares are planned. We distribute end-of-day data at T+1 or older, and the usable history is whatever the API's data_range reports. If a request exceeds it, the assistant tells you which window was actually tested instead of passing off a truncated run as a longer one.
No, and we deliberately did not build that. The platform provides backtesting and market data, and none of it constitutes investment advice. The assistant can tell you how a rule set behaved historically; it cannot tell you what to buy tomorrow.