How to Backtest a Strategy From a Research Paper or PDF
You backtest a research paper by pulling the exact rules out of the PDF, rewriting them as precise entry, exit, and risk conditions, and running them on historical data to see if the claimed edge shows up for you. Academic and blog papers often describe a strategy in words and math but leave out details like exact timing, position sizing, or how they handled costs. The job is to translate the paper into rules a computer can follow, then test it honestly on data the paper didn't cherry-pick. An AI agent speeds this up. You can hand it the paper's idea, it drafts the testable rules, runs the backtest, and shows where the claims hold up and where they fall apart once real trading costs and different time periods are included.
Academic and vendor PDFs describe edges in prose, not code. How traders extract the rules from a research paper and turn them into a backtest instead of taking the claim on faith.
Key points
- Research papers describe strategies in words and formulas, so the first step is translating them into exact entry, exit, and risk rules.
- Papers often leave out practical details like trading costs, slippage, and position sizing, and those gaps can flip a winning result into a losing one.
- Test the idea on a different time period than the paper used, since results tuned to one stretch of history often don't repeat.
- Watch for cherry-picked date ranges and hindsight, which can make an average strategy look far better than it really is.
- An AI agent can read the idea, draft the testable rules, and run the backtest for you, so you're not stuck hand-coding formulas.
- A paper's edge is worth studying, but replicating it yourself is the only way to see if it survives real costs and fresh data.
Frequently asked questions
Can I really backtest a strategy from a PDF I found online?
Yes, as long as the paper spells out the rules clearly enough to reproduce. You extract the entry, exit, and risk logic, fill in any missing details like costs and sizing with sensible assumptions, and run it on historical data. If a paper is too vague to turn into exact rules, that vagueness is itself a warning sign.
What details do research papers usually leave out?
Common gaps are trading costs and slippage, exact entry and exit timing, how positions were sized, what happened during the worst stretches, and whether the test avoided stocks that were later delisted. Filling these in realistically often shrinks the reported edge.
Why do published strategies sometimes stop working after the paper comes out?
Once an edge is public, more traders act on it, which can crowd it out and erode the advantage. On top of that, some published results were partly luck or fit too closely to one date range. That's exactly why you re-test on fresh data instead of trusting the paper's numbers.
How do I know if the paper just got lucky with its dates?
Run the same rules on a different time window than the one in the paper, ideally including a rough patch for that market. If the edge only appears in the exact period the authors chose, it probably fit that stretch rather than found something durable.
What's the easiest way to test a paper's idea myself?
In Agenticks, describe the paper's rules to AlgoAgent and it drafts the testable logic, runs the backtest on real historical data, and reports where the claims hold and where they break, so you can replicate the study without writing code.
Related on Agenticks
This content is for educational purposes only and does not constitute financial advice. Trading involves risk, including possible loss of capital.