Dark Pools and Midpoint Execution
A dark pool matches orders without displaying pre-trade size or price to the market. For liquid equities, the common interaction is a midpoint execution: a buyer and seller trade at (bid + ask) / 2, splitting the quoted spread. That apparent half-spread saving is valuable only when the fill does not reveal information and the price does not move against the residual order.
What a midpoint fill means
If the NBBO is 100.00 bid and 100.04 offer, a midpoint buy fills at 100.02. Relative to lifting the offer, the immediate mark-out looks favorable by 2 cents. But quotes can be stale, venues can use different reference rules, and the counterparty may be selling because it predicts a short-term decline. Execution quality is therefore a distribution of post-fill mark-outs, not a fill rate.
| Metric | Definition | Interpretation |
|---|---|---|
| Fill rate | Filled / routed shares | Access to liquidity |
| Fill size | Shares per match | Information leakage proxy |
| 1s/5m mark-out | Mid after fill minus fill | Adverse selection |
| Opportunity cost | Unfilled residual movement | Cost of waiting |
Conditional routing
A smart router should classify dark venues by symbol, time of day, side, urgency, and recent mark-outs. A high fill-rate venue with consistently negative buy-side mark-outs is not cheap liquidity. It may be systematically interacting with informed flow. Read market microstructure and adverse selection as the economic foundation for this measurement.
def midpoint_markout(fill_px, mid_after, side):
# positive is favorable for the aggressor
return side * (mid_after - fill_px)
# Estimate E[markout | venue, symbol bucket, side, volatility, urgency]
Use a randomized small flow allocation to continue estimating venue quality; routing all flow to last month's winner creates selection bias. Cap participation and child-order size so a dark order cannot become an easily inferred parent order. Minimum-execution quantity can reduce information leakage but also sacrifices useful small matches.
Reference price and regulatory details
Midpoint eligibility depends on the protected quote, tick size, and venue rules. Odd-lot quotes, locked/crossed markets, auctions, and trading halts require explicit handling. A midpoint order may also be pegged or repriced, which creates order-state and timestamp complexity. Preserve the NBBO snapshot, venue acknowledgment, execution timestamp, and subsequent quote path for every fill; without them, TCA cannot distinguish price improvement from a measurement artifact.
Dark pools do not eliminate market impact. Unfilled dark exposure delays displayed liquidity seeking, while a sequence of dark fills may leak enough information to worsen later lit execution. Optimize the combined schedule: dark allocation, displayed participation, and urgency for completing the residual. This is a routing problem, not a binary dark-versus-lit decision; see smart order routing.
Implementation discipline
The model is only one layer of the trade. Store input timestamps, vendor identifiers, contract specifications, FX conversion conventions, and the exact version of each risk model alongside every signal. A result that cannot be reproduced from raw inputs is not a research result; it is an anecdote. Use point-in-time constituent data, distinguish an indicative quote from an executable quote, and make the decision clock explicit. For an execution-sensitive strategy, a close price or composite midpoint is usually an unattainable benchmark rather than a fill.
Evaluate the full distribution, not just a headline Sharpe: drawdown, tail loss, turnover, capacity, exposure concentration, and degradation during the stressed periods that motivated the idea. Split design, parameter choice, and final evaluation across separate samples. If alternatives were explored, record them and account for the search using the Deflated Sharpe Ratio. Finally, run a paper or shadow phase with the same order, reconciliation, and limit logic planned for production. A backtest should establish a conditional expectation, not a promise of a tradable return.
Key takeaways
- Midpoint executions save half the quoted spread only before adverse selection and delay costs.
- Measure conditional post-fill mark-outs, not raw fill rate, by venue and market state.
- Preserve quote and execution timestamps to make TCA defensible.
- Control minimum size, exposure duration, and residual urgency to limit information leakage.
- Optimize dark and lit liquidity together at the parent-order level.