Backtesting concepts: look-ahead bias and overfitting
Education only — not a recommendation to buy, sell, or hold anything.
This page explains common ways historical tests fool people. It is not a backtest of any strategy, not performance reporting, and not a recommendation to trade. Hypothetical results are illustrative only. Past ≠ future. Fake or simulated results are not live Voltron performance.
What a backtest is (concept)
A backtest applies a set of rules to historical data to see what would have happened if those rules had been followed. It is a teaching and research tool — not proof that the same rules will work next.
---
Look-ahead bias
Look-ahead bias means the test used information that would not have been available at the decision time.
Examples (educational):
- Using the day's close to decide a trade that supposedly happened at the open.
- Restating index membership with today's list for past years.
- "Knowing" an earnings number before the release timestamp in the dataset.
- Chart pattern labels drawn with future bars already visible.
If the simulator knows the future, the grade is fake.
---
Overfitting
Overfitting means the rules were tuned so hard to one history that they describe noise instead of a durable process.
Warning signs in education:
- Dozens of free parameters for a short sample
- "Works" on one ticker/era and collapses everywhere else
- Rules rewritten after peeking at the equity curve
- Ignoring costs, gaps, and liquidity (fees, liquidity)
A beautiful curve can still be a costume for randomness (edge vs randomness).
---
Other biases worth naming
| Bias | One-line idea |
|---|---|
| Survivorship | Only studying names that still exist |
| Data-snooping | Trying many ideas and keeping the lucky one |
| Regime blindness | One market climate ≠ all climates |
---
Healthier research habits (still not advice)
- Write rules before running the history.
- Include costs and realistic fills.
- Hold out a period you do not tune on.
- Prefer simple processes you can journal (trading journal).
- Treat every backtest claim on the internet as marketing until methodology is clear.
Voltron will not publish performance claims as live results unless Ledger-verified and Mad King + Sentry approved (see site DISCLAIMERS).
---
Tie-in to paper practice
Use /mock/ to practice process, not to invent a curve to brag about. Coach tips grade habits from simulated journals — not"your edge is proven."
---
Related
---
Sentry CLEAR · Education only. No strategy performance claims.