Four-day ETF example: execution timing matters

Synthetic teaching inputs: synthetic-4d-v1. ETF_A and ETF_B are fictional. Values use SYNTH, an artificial accounting unit. This tests timing and accounting, not real-market effectiveness.

Complete input and rules

Start at 10,000, comparing 60/40 fixed weights with 50/50 buy and hold. The readable table below represents the long CSV date,asset,total_return_index.

Date	ETF_A	ETF_B
2026-10-26	100	100
2026-10-27	110	100
2026-10-28	121	100
2026-10-29	108.9	100

The manifest declares reinvested distributions, calendar, version and hashes. Do not add dividends again. The full synthetic calendar defines October 30 as month-end; the last supplied row is not month-end.

Signal after October 26 close, execute at October 27 valuation close. Holdings earn returns only after execution. Assumed friction is 10 basis points per side, charged on risk-asset turnover without counting cash again. Cash return is zero; no leverage or real share quantities. Initial turnover is 1, costing 10 units for each portfolio. This is a model approximation, not a broker tariff.

Reproducible result

Date	Strategy 60/40	Benchmark 50/50	Event
10-26	10,000.00	10,000.00	Signal, cash
10-27	9,990.00	9,990.00	Cost and initial allocation
10-28	10,589.40	10,489.50	Existing holdings earn returns
10-29	9,930.06	9,940.05	No month-end rebalance
Cumulative return	-0.70%	-0.60%	Versus initial value
Maximum drawdown	-6.23%	-5.24%	Initial value included

October 28 strategy value is 9990×(0.6×121/110+0.4); October 29 is 9990×(0.6×108.9/110+0.4). Drawdown compares each value with its running peak. Neither portfolio earns the earlier 100→110 gain.

Counterexample and next check

The constructed 60/40 allocation loses slightly more: a valid finding. Executing at the October 26 close would improperly capture pre-execution returns. Do not annualize four observations or invent a Sharpe ratio.

Keep confirmed rules, input manifest, daily equity, allocations and events. Remove one ETF_B row and validation should identify the gap and stop. A 63-interval momentum request needs 64 complete observations; this sample must fail. A final-day signal without a next day remains unexecuted.

Limits and source

These are self-authored inputs with hand-checkable arithmetic. Investor.gov’s ETF overview supplies background, not performance evidence. The model omits real fills, taxes, lots and liquidity. Host verification is separate. Methods are free; model and data costs may differ.

Use strategy specification before ETF verification.