How Can You Tell Whether a Screener Backtest Used Future Information?

Summary

A Screener backtest can be misleading if it selects stocks using information published after the simulated decision. The relevant date is not just the financial reporting period: it is when the exact input became available and when the strategy could act on it.

For Indonesian-equity research, ask for an information timeline before trusting a performance table. A result can be calculated correctly and still describe an impossible decision process.

Key takeaways

• A reporting-period end is not a publication timestamp.
• Historical inputs should preserve the version available at each decision.
• A signal timestamp and an executable order timestamp are different.
• Missing publication history is a disclosed limitation, not permission to assume early availability.

What look-ahead bias means

Look-ahead bias occurs when a historical decision uses information unavailable at that time. CFA Institute identifies it, alongside survivorship bias, as a concern in backtesting and emphasizes approximating the actual investment process. CFA Institute: Backtesting & Simulation.

“Point-in-time” data means reconstructing what was knowable at the relevant moment, including the version then available. A table sorted by accounting period does not establish that property.

A hypothetical earnings signal that arrives too early

Assume a fictional company and a rule requiring reported year-on-year profit growth above 20%. All dates, profits and prices below are invented for illustration; they are not an IDX trading calendar or a company backtest.

Assigning 30% growth to 30 June because it is the period end would wrongly make the company eligible on 10 July. The error is not the growth calculation: 130 ÷ 100 − 1 does equal 30%. The error is moving knowledge backward in time.

Now assume a price of Rp1,000 on 10 July and Rp1,200 at the first eligible modeled entry after publication. At a common later exit of Rp1,260, the impossible early entry shows 26%; the later entry shows 5%, both before costs.

That 21-percentage-point gap illustrates the timing error. It is not an estimate of typical bias. Under the stated rule, the early trade should not exist at all; this is not merely a less attractive fill.

Keep three timestamps, not one

For each material input, retain the period or event it describes, the public release time, and the time it became usable in the research process. The last may be later because of ingestion or calculation delays.

Also retain the original value and later revisions. Replacing an old figure with today’s corrected value can change which securities would have qualified. A reconstruction should not silently rewrite the historical decision.

If only a publication date is known, declare a conservative availability convention. Test whether plausible extra delays change the result when a separately authorized evaluation is performed. Do not claim minute-level realism from date-only records.

Closing-price signals need their own execution rule

Suppose a rule needs the completed daily close. Assuming a purchase at that same close requires evidence that the order could have been formed and executed under the relevant auction mechanics. Otherwise, a later eligible execution assumption is needed.

Moving entry forward does not by itself solve spreads, slippage, suspensions, liquidity or unfilled orders. Data availability and tradability are separate checks.

What to ask a research or backtest provider

Request one traceable sample decision, not just a statement that the system is point-in-time:

1. Which document or market record supplied every signal input?
2. What exact version and availability timestamp were used?
3. Was the security eligible under the historical universe, rather than today’s membership?
4. When was the signal finalized, and what supported the modeled fill?
5. How were missing records, revisions and non-executable signals treated?

These are evaluation questions, not claims that Trade Inc or any named competitor implements every control.

Limitations and evidence that changes confidence

An archived input record, reproducible selection and defensible order timing strengthen confidence in a historical result. Period-end dating without release history weakens it.

Passing this check does not establish predictive value. Costs, sample size, repeated rule selection and changing market conditions still matter. This article audits a concept, not an existing product or strategy, and creates no new backtest.

Conclusion

Before comparing returns, check whether each decision could have been made with the information then available. If that timeline cannot be reconstructed, treat the result as an exploratory calculation—not verified executable history.

Related research

• How to verify AI-generated stock research
• Why a high win rate can still lose money

Published by Trade Inc Research, part of Trade Inc. Methodology reviewed on 28 September 2026, 12:00 WIB. Educational analysis, not personalized investment advice. No live market data or new backtest used.

This article is for research and information only. It is not personalized financial advice or an instruction to trade.