1
Look-ahead bias
The test uses a value that became known after the simulated decision.
Lesson · Backtesting and Trading Systems
Signal Timing, Look-Ahead Bias, and Data Leakage
Prevent the strategy from using future prices, completed-bar values, revised data, or preprocessing information that was unavailable at decision time.
Core concepts
What to understand
Keep each definition operational so it can be implemented and tested.
2
Data leakage
Training or preprocessing exposes information from validation or future periods.
3
Publication lag
Fundamental and macro data must enter the test when they were released, not when they describe.
Interactive lab
Signal Timing and Leakage Lab
Switch between valid next-period execution and impossible same-period fills. Add future leakage and observe the inflated result.
Signal Timing and Leakage Lab
Deterministic synthetic educational data · no live market connection
BACKTEST MODEL
Process
Build a causal event sequence
For every field, record when it became knowable.
Observe
Complete the bar or receive the event.
Calculate
Update indicators and evaluate conditions.
Execute
Submit an eligible order at a later executable price.
Knowledge check
Check your understanding
Select one answer for each question. Feedback appears immediately.
1. A signal calculated from today’s close can normally execute:
Execution must occur after the signal becomes available.
2. Using a full-sample mean to normalize all periods can create:
Future observations affect past transformed values.
3. Fundamental data should enter a test on:
Publication and revision timing determine when the information was knowable.
4. A breakout is confirmed by bar 40's close. The earliest bar the rule can be filled in is:
Signal at the close of bar 40, order at the next opportunity, fill at bar 41's open — the lab's Invalid assumptions counter is the number of fills that broke this order.
Practical rules
- Timestamp signals, orders, and fills separately.
- Lag data by actual availability, not reporting period.
- Fit transformations on training data only.
Common mistakes
- Buying at the close that generated the signal.
- Using revised economic data as originally released data.
- Centering the full dataset before splitting it.
References
Official reading
The labs are simplified simulations. These sources cover historical performance, data, costs, or execution limitations.
FINRA · Smart Beta—What You Need to Know
Explains that historical backtesting is hypothetical and does not predict future performance.
FINRA Rule 2210 · Communications with the Public
Includes restrictions on performance projections and misleading performance claims.