Cross Validation Of Commission Schedules Over Time

When to Use

Invoke this skill when running historical backtests spanning multiple years (e.g. 2016–2024) on US cash equities. Retail commission rates moved repeatedly over short windows — one major US broker's standard online equity rate went $8.95 → $6.95 → $4.95 → $0.00 between February 2017 and October 2019 — so a single flat commission figure is wrong for most of a multi-year sample. Applying today's zero-commission structure retroactively inflates backtested P&L for any strategy that trades frequently. This skill models date-effective fee schedules, adds the US regulatory pass-through fees that survived the move to zero commission, and quantifies the P&L difference against the naive assumption.

When NOT to Use

• Non-US venues. Stamp duty, STT, and exchange transaction levies are not modelled.
• Volume-tiered schedules where the rate depends on trailing volume — use exchange-fee-tier-and-rebate-structure-analysis.
• Maker/taker rebate economics — use post-only-and-maker-taker-fee-optimization.
• Slippage, spread, and market impact. This skill models explicit fees only; implicit costs belong to transaction-cost-analysis-tca-integration.

Prerequisites

• Trade execution log with timestamps, symbol, positive share quantity, price, and side (BUY/SELL).
• Your broker's published historical commission schedule with effective dates. The shipped DEFAULT_SCHWAB_RETAIL_SCHEDULE is a worked reference example, not a substitute.
• For US regulatory fees: SEC Section 31 rates and FINRA TAF rates covering the backtest window (references/standards.md lists the primary sources).

Workflow

1. Construct the Time-Varying Fee Schedule: One CommissionTier per period during which the published rate did not change, with inclusive effective dates. Match the broker's actual structure — flat ticket, per-share with floor/cap, or percent-of-value. Do not blend a ticket fee with a per-share fee unless the broker charges both. Overlapping tiers are rejected at construction; gaps are logged and raise at lookup.
2. Resolve the Tier for Each Trade Date: Parse the timestamp to a calendar date and select the covering tier. If the date is unparseable or uncovered, fail loudly — never fall back to the latest tier, because the latest tier is $0.00 and that silently reintroduces the exact bias being corrected. Convert timestamps to the schedule's timezone first; a UTC timestamp near a session boundary can land on the wrong calendar day.
3. Compute the Trade Commission: max(raw, min_trade_fee) then, when the broker publishes one, min(..., trade_value * max_pct_of_value) — the cap is applied after the floor and dominates it. Pass share quantity as a positive number with an explicit side; a signed quantity would silently reduce a per-share fee.
4. Add Regulatory Pass-Through Fees: SEC Section 31 and FINRA TAF apply to sales only and both changed rates repeatedly. Supply a RegulatoryFeeTier list; when omitted, results carry regulatory_fees_modeled=False so the report states the cost was excluded rather than measured as zero.
5. Audit Fee Schedule Impact: Price every trade under the historical schedule and under a flat modern_baseline, and report the delta in dollars and as a percentage of starting capital. That delta is the P&L a naive backtest would have fabricated.

Full procedure: see references/workflows.md. Standards reference: see references/standards.md. Printable pre-flight checklist: see assets/checklist.md.

Common Pitfalls

• Retroactive Zero-Commission: Charging $0.00 back to 2015. US retail zero commission began in October 2019 — and on different dates per broker (TD Ameritrade Oct 3, Schwab and E*TRADE Oct 7, 2019). Using one broker's cutover date for another mis-costs the intervening days.
• Treating "Zero Commission" as Zero Cost: The SEC Section 31 fee and the FINRA Trading Activity Fee still apply to every sale and are passed through to the seller. A backtest that charges nothing to a post-2019 exit under-costs every exit in the sample.
• Silent Fallback on an Unresolvable Date: An unparseable timestamp, a date before the schedule starts, or a schedule gap must raise. If the modeler quietly applies its most recent tier instead, a whole pre-2019 backtest can be priced at $0.00 and still look like it modelled commissions.
• Modelling a Structure the Broker Never Charged: A flat-ticket broker charges a ticket fee and no per-share fee. Adding a per-share component "for realism" invents costs; conversely, dropping a per-share broker's floor or cap mis-prices both ends of the size distribution.
• Ignoring Minimum Ticket Charges: Applying $0.005/share without the $1.00 per-order minimum under-costs a 10-share trade by 20×.
• Ignoring the Percent-of-Value Cap: 10,000 shares at $0.10 is $1,000 of notional; an uncapped $0.005/share fee charges $50 — 5% of the trade — where the broker's 1% cap charges $10.

Verification

• Price the same trade on 2019-10-06 and 2019-10-07 against the reference schedule and confirm $4.95 vs $0.00 — an off-by-one at the cutover is the most common schedule bug.
• Submit a trade dated before the schedule's coverage and confirm it raises rather than returning $0.00.
• Price a 1,000-share $200 sell under a Section 31 + TAF schedule and confirm the total cost is non-zero even in the zero-commission era.
• Run python -m unittest discover -s skills/cross-validation-of-commission-schedules-over-time/scripts and confirm a 100% pass rate.

Related Skills

• transaction-cost-analysis-tca-integration
• post-only-and-maker-taker-fee-optimization
• multi-year-regime-coverage-requirement
• exchange-fee-tier-and-rebate-structure-analysis