The ChatGPT method, honestly
The workflow is: describe your strategy to ChatGPT, ask it to write a Python backtest, then copy that code into an environment (like a notebook), supply price data, fix whatever doesn't run, and read the output.
It works, and it's a great way to learn. But notice how many manual steps there are - and how each one is a chance for a subtle error to creep in.
Step by step
If you go this route, here's the realistic sequence:
- Describe the strategy precisely - asset, entry, exit, costs - or ChatGPT will guess the gaps.
- Ask for a complete, runnable Python backtest using a known library.
- Get historical data yourself (an exchange export or a data API) and feed it in.
- Run it, then debug - missing imports, data-format mismatches, and edge cases are common.
- Scrutinise for lookahead bias and check costs are modelled, because the code will run even if it's subtly wrong.
Where it goes wrong
The biggest risk is silent error. ChatGPT-written code can run perfectly and still be wrong - using a future price, omitting fees, or mishandling the data - producing a confident, beautiful, fake result.
You also own all the plumbing: data sourcing, environment setup, debugging. For a quick idea, that's a lot of overhead, and the parts most likely to introduce bias are exactly the parts you're handling manually.
The faster alternative
Premiss is essentially the ChatGPT method, productised and made safe: you type the idea in your own words - no code - and it writes the real backtest, sources the real data, runs it correctly, and shows you the code, the trades, and a verified result - without the copy-paste-and-debug loop. Same describe-the-idea input; none of the manual plumbing or the silent-error risk.