Tick Size Pilot Program Impact Assessment

When to Use

Use this skill when a minimum pricing increment has changed — or is about to — and the desk needs a measured answer to "what did that do to our execution costs and our queue?" rather than an inference from the tick ratio. Concretely:

• A US retick. The amended Rule 612 introduces a $0.005 quoting increment for symbols the listing exchange designates tick-constrained (Time Weighted Average Quoted Spread <= $0.015). It is adopted but not yet operative: SEC exemptive relief of 2026-06-11 (Release 34-105656) deferred compliance to the first business day of November 2027. When it lands it is a narrowing for the affected symbols — the mirror image of the Pilot.
• An EU band reclassification. An RTS 11 liquidity band changes with the annual ADNT calculation, moving an instrument's tick without any venue rule changing.
• A venue retick on any exchange that revises its price-step schedule by circular.
• Retrospective study of the SEC Tick Size Pilot Program itself, whose 20 TB of collected data remains the largest controlled tick-size experiment on record.

The engine decomposes spreads into quoted, effective and realized (5-minute) components on the Rule 605 formulas, computes top-of-book depth, order-to-trade ratio, share fill rate and adverse selection in basis points, compares two regimes, and maps the measured deltas onto passive market-making, TWAP/VWAP slicing, momentum-taking and stat-arb parameters.

When NOT to Use

• To produce or check a Rule 605 report. The formulas match, but two things do not: Rule 605 measures effective spread against the NBBO midpoint at the time of order receipt (17 CFR 242.600(b)(8)), whereas this engine measures against the midpoint of the snapshot carrying the trade. Rule 605 also requires realized spread at five horizons (50 ms, 1 s, 15 s, 1 min, 5 min), not one. Numbers from this engine are research and TCA inputs, not a filing.
• To decide whether an order price is legal. This engine measures the consequences of a tick regime; it does not validate a price against one. For price alignment and tick compliance, use exchange-tick-size-regime-tracking.
• To forecast the effect of a proposed retick. The published Pilot outcomes range from a -17% spread change to +203% depending purely on the security's pre-change spread class. There is no transferable multiplier, and this engine deliberately hard-codes no effect size.
• On a sample that does not isolate the tick change. A pre/post comparison with no control group attributes every concurrent market-wide move — a volatility regime, an index reconstitution — to the tick. The Pilot used a randomly stratified control group and difference-in-differences for exactly this reason.
• On quote-only data, to reason about execution costs. Without trades the engine returns None for effective spread, realized spread and adverse selection. That is the correct answer, not a gap to fill.

Prerequisites

• Python 3.10+ (standard library only).
• L1 quote snapshots with bid/ask price and size, timestamped to at least millisecond precision.
• Trade records tagged with aggressor side, from exchange execution records where available, or a signing rule (Lee-Ready, tick test) where not — signing error propagates directly into effective and realized spread.
• Trade size on every trade, or the engine falls back to equal weighting and says so. Rule 605 spread averages are share-weighted (17 CFR 242.600(b)(8), (12), (13)); an equal-weighted average over-counts odd lots.
• The consolidated midpoint 5 minutes after each execution, already honouring the end-of-session proviso in 17 CFR 242.600(b)(13): where fewer than 5 minutes of regular trading hours remain, the midpoint of the final NBBO of the session is the correct input. Where it cannot be observed, pass None.
• Matched baseline and test samples for the same symbol, ideally with a control group of untouched symbols measured over the identical window.

Workflow

1. Assemble TickSnapshot series for the baseline and test regimes. One snapshot per quote; attach last_trade_price, last_trade_size, last_trade_is_buy and future_mid_price_5m to the snapshot the trade printed against.

   • Decision point — can you observe the 5-minute midpoint? If the horizon runs past the close, apply the Rule 605 proviso and use the session's final NBBO midpoint. If you cannot, pass None; the trade is then excluded from the realized-spread sample rather than imputed.

2. Evaluate each regime — evaluate_microstructure_metrics(symbol, regime, snapshots, total_messages=, total_fills=, total_shares_ordered=, total_shares_executed=).

   • Decision point — how dirty is the feed? The default InvalidSnapshotPolicy.SKIP excludes crossed quotes, non-finite values and non-positive prices, counts them in excluded_snapshot_count, and continues. Use RAISE only on a feed you expect to be clean. Always read excluded_snapshot_count before trusting the result: a large exclusion count is a data problem, not a market finding.
   • Decision point — check weighting before comparing. SHARE_WEIGHTED means every trade carried a size. EQUAL_WEIGHTED means at least one did not, and the spread averages are no longer comparable to a Rule 605 figure or to a share-weighted baseline.
   • Order-to-trade ratio is total_messages / total_fills. Share fill rate is total_shares_executed / total_shares_ordered — the Pilot's definition. These are different measurements from different denominators; supplying only the message counters leaves the fill rate None.

3. Compare — compare_regimes(baseline, test).

   • Percentage-change fields are None when the baseline metric is absent, zero or negative, and are named in undefined_metrics. A zero baseline effective spread is a real outcome — every print at the midpoint — not a data error, so it is reported as undefined rather than as an infinite or sign-flipped percentage.
   • fill_rate_change_pp is a difference in percentage points. The others are percentage changes. Do not mix them in a report.
   • Decision point — did quoted and effective diverge? They usually do, and the gap is the finding. Under the Pilot, quoted spreads widened 14–24% while share-weighted effective spreads rose 54–59% in cents per share; the two are reported against different denominators and neither is a proxy for the other.

4. Recalibrate — recommend_strategy_tuning(algo_type, comparison) for PASSIVE_MARKET_MAKING, TWAP_VWAP_SLICING, MOMENTUM_TAKER or STAT_ARB.

   • Output is advisory screening only. Every branch is gated on the module-level screening constants (SPREAD_FINDING_THRESHOLD_PCT, DEPTH_FINDING_THRESHOLD_PCT, ADVERSE_SELECTION_FINDING_BPS, MARKET_MAKING_ADVERSE_SELECTION_BPS, SLICING_SPREAD_THRESHOLD_PCT), which are this engine's reporting triggers — not regulatory limits and not empirical constants. Tune them to the desk's tolerances.
   • An undefined input metric produces an explicit "Cannot assess…" line rather than silence, so an empty recommendation list means "measured, nothing triggered" and never "could not measure".

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

Common Pitfalls

• Reading the tick ratio as the cost ratio. A $0.01 → $0.05 retick is a 5x wider tick, and the Pilot's quoted spreads rose 14–24% while share-weighted effective spreads rose ~54–59%. Neither is 400%, because midpoint executions and price improvement absorb much of a widened quote. Setting a TWAP price cap from the tick ratio prices in a cost the book never charged.
• Equal-weighting the spread averages. Rule 605 share-weights every spread statistic. An equal-weighted mean gives a 100-share odd lot the same influence as a 10,000-share block and will not reconcile against any published execution-quality figure.
• Substituting one metric for another when data is missing. Quoted spread is not a stand-in for effective spread, and half the effective spread is not a stand-in for realized spread. A fabricated component silently becomes a fabricated adverse-selection number and then a fabricated recalibration.
• Ignoring the end-of-session proviso. A trade at 15:58 has no midpoint 5 minutes later. Rule 605 requires the session's final NBBO midpoint; carrying the next session's open instead prices an overnight gap into the realized spread and reports it as adverse selection.
• Letting one crossed quote kill the batch. Crossed and locked NBBOs are transient but real in consolidated data. A locked quote's spread genuinely is zero; a crossed quote is not a measurement and must be excluded and counted, not averaged in and not fatal.
• Comparing a test period to a bare pre-period. Without a control group the estimate absorbs every market-wide change in the window. The Pilot's own headline numbers are difference-in-differences against a randomly stratified control group, and several of them were not statistically significant even so.
• Assuming a widened tick means a worse fill rate. It did not in the Pilot: shares executed rose from 1.1% to 2.2% in Test Group 3, against 1.2% → 1.5% for the control. Deeper queues and higher fill rates coexisted because the trade-at prohibition pushed volume onto displayed quotes.
• Confusing fill rate with the reciprocal of the order-to-trade ratio. Fills per message and executed shares per ordered share answer different questions; reporting one under the other's name understates or overstates passive performance by whatever the average order size is.
• Treating the amended Rule 612 half-penny tier as live. It is deferred to November 2027 and is a per-symbol assignment by the listing exchange, not a function of price. A baseline built on the assumption that tick-constrained symbols already quote in half-pennies is measuring a regime that does not yet exist.
• Citing RTS 28 for EU tick sizes. The MiFID II tick regime is RTS 11 (Commission Delegated Regulation (EU) 2017/588). RTS 28 was the top-five-execution-venue report, and that obligation was removed in the MiFID II/MiFIR review (ESMA deprioritised supervision from 2024-02-13).

Verification

• Reproduce the Pilot Assessment's published worked example (footnote 11): selling at 10.00 into a 10.00 x 10.05 quote gives an effective spread of 0.05; selling at 10.01 into the same quote gives 0.03; a print at 10.025 gives exactly 0.0.
• Confirm share weighting: 100 shares at an effective spread of 0.05 and 900 at 0.02 must average 0.023, not 0.035.
• Confirm a quote-only sample returns None for avg_effective_spread, avg_realized_spread_5m and adverse_selection_bps — never the quoted spread and never half of it.
• Confirm calculate_quoted_spread(10.00, 10.00) returns 0.0 (locked) and calculate_quoted_spread(10.05, 10.00) raises (crossed).
• Confirm a crossed snapshot inside a batch is skipped and counted in excluded_snapshot_count under the default policy, and raises under InvalidSnapshotPolicy.RAISE.
• Confirm compare_regimes returns None — not ZeroDivisionError, not a sign-flipped percentage — for a zero or negative baseline effective spread, and names the metric in undefined_metrics.
• Confirm a +400% quoted spread with a nearly flat effective spread does not trigger a TWAP passive re-weighting.
• Run the suite and confirm a 100% pass rate:

python -m unittest discover -s skills/tick-size-pilot-program-impact-assessment/scripts

Related Skills

• exchange-tick-size-regime-tracking
• order-book-depth-processing-l2-l3
• adverse-selection-measurement-for-passive-orders
• queue-position-modeling-for-passive-orders
• peg-order-types-for-passive-execution
• execution-slippage-attribution-timing-vs-sizing