Identifier: C5T5L3 · Revision: 2026-10-26 · Classification: domestic

Decision and objectives

The practical task is to detect look-ahead and selection bias in historical comparisons. Make the decision reproducible: name the exposure, time horizon, cash requirement and comparison baseline before selecting a product.

Objectives: Detect look-ahead and selection bias in historical comparisons; Recalculate a portfolio including failed or excluded products; Add turnover and funding costs on a transparent basis.

Prerequisite: C5T5L2; completed Classes 1–4 and the Stock Market School foundation or Forex School foundation. Every example is hypothetical. Prices, dates, rates, fees, contract factors, policies and tax schedules are supplied teaching inputs, not current offers or legal parameters. No live account is needed.

Mechanics

A backtest must use information available at each decision time. Look-ahead occurs when later holdings, revised data or future rankings influence an earlier choice. Survivorship bias omits products that failed, merged or disappeared. Selection bias can arise when only a favorable market window or the best strategy among many trials is reported.

A correct arithmetic average cannot repair a biased sample. Record the selection universe as it existed, publication lags, delisting treatment and rebalancing instructions. Keep a genuine out-of-sample period or scenario challenge separate from the period used to choose the rule.

Price and return drivers

Equal-weight simple one-period return is the arithmetic mean of the supplied constituent total returns when weights start equal and no rebalancing occurs. Turnover cost should use actual traded notional and directions, not a flat fee guessed after observing performance. The constant-notional illustration below deliberately excludes compounding; label that assumption.

Worked case

Hypothetical initial equal allocation among funds A, B and C. Their period total returns are +10%, +5% and −30%; C later closes. Full-universe return = (10 + 5 − 30)/3 = −5%. A website retaining only A and B reports (10 + 5)/2 = +7.5%, overstating this comparison by 12.5 percentage points. Separately, a hypothetical strategy on fixed TRY 100,000 notional earns TRY 20,000 gross, executes ten round trips costing 0.30% each and pays TRY 4,000 funding. Trading cost = 100,000 × 0.003 × 10 = TRY 3,000; net gain TRY 13,000, or 13%, before tax. Gross 20% is not the implementable net return.

Adverse case

A fund holdings report dated 30 April but published 10 May cannot justify a trade assumed on 1 May using those holdings. A revised economic series downloaded now may also differ from the release available then. If the strategy is chosen after comparing 100 rules and only its best in-sample outcome is shown, the winner may reflect selection noise; a favorable historical result supplies no guaranteed forward edge.

Practical use

Use official dated disclosures, original release timestamps and a reproducible trade/cost ledger. Include delisted or terminated products where they belonged in the original universe. Capture bid/ask, financing, taxes, redemption constraints and capacity when comparing routes. Product existence and return history do not establish current investor eligibility or live executable depth.

Use permitted domestic access routes and the actual instrument documents. Verify intermediary authority, investor eligibility, order/redemption terms, settlement, custody and executable size before implementation. Reconcile spread and commissions, funding, collateral, custody, product expenses, conversion and exit charges; an amount already embedded in a fill or fund unit price is not deducted again. A published mark or product description does not promise an executable price or liquidity.

Tax is excluded unless a numerical schedule is expressly supplied as hypothetical. For actual rules record income year, residence and individual/corporate status, instrument and acquisition/contract dates, taxable event and rule effective date. The GIB investment-income guide, checked 2026-10-26, covers 2025 individual income; its 2026 publication date alone cannot settle a 2026 liability.

Decision boundaries

Reject a claimed advantage that disappears when the full universe, publication timing or realistic costs are included. A simpler baseline with fewer trades can be preferable. Report uncertainty rather than tune assumptions until the historical strategy wins.

Practice and references

Exercise: hypothetical equal-start holdings return +8%, +2%, −16%. Solution: full return −2%; dropping the loser reports +5%, a seven-percentage-point distortion. On a separate fixed TRY 50,000 strategy with gross gain TRY 6,000, costs TRY 1,000 and funding TRY 500, net is TRY 4,500, or 9%, before tax. Do not compound this constant-notional example retroactively.

Takeaway: Add turnover and funding costs on a transparent basis with the cash dates, costs and remaining risks stated.

Product card: portfolio-decision-record. Formula: Fixed-notional historical net gain. Glossary: Look-ahead bias, Survivorship bias. Next lesson: C5T5L4. Topic test.

Sources checked 2026-10-26:

Shared reference guide