Smart Order Router

While TSX has the most liquidity, other venues may offer better prices, lower fees, or price improvement. Smart routing across venues often achieves better overall execution than routing to a single venue.

Getting a better price than the displayed quote. If NBBO is $50.00 bid / $50.02 ask and you buy at $50.01, you got $0.01 price improvement. Dark pools and midpoint orders often provide price improvement.

Market orders: when speed matters and stock is liquid. Limit orders: when price matters and you can wait. For larger orders or wider spreads, limits protect you from poor fills. Many use 'marketable limits' (limit at current price) as a hybrid.

Dark pools are venues where orders aren't displayed. Benefits: anonymity, potential price improvement, reduced market impact. Drawbacks: no fill guarantee. Good for larger orders; small retail orders usually get routed by broker automatically.

Check execution quality reports (most brokers provide them). Compare fill prices to NBBO at order time. Look at fill rates and any price improvement. In Canada, brokers must seek 'best execution' under IIROC rules.

TWAP: even execution over time; good when volume pattern doesn't matter. VWAP: matches historical volume pattern; good when you want to match the benchmark VWAP. VWAP is more common for institutional orders.

Conservative: 5-10% (minimal impact, slower). Moderate: 10-20% (balanced). Aggressive: 20-30%+ (faster but more impact). Consider: order urgency, stock liquidity, and acceptable market impact.

Maker-taker venues charge takers and rebate makers. Inverted venues (like Alpha) rebate takers. For passive orders, route to maker-rebate venues. For aggressive orders crossing spread, inverted venues may be cheaper.

Use icebergs for large orders where you don't want to reveal true size. Display 10-20% of total size. Benefits: reduced market impact, hide intentions. Available on most lit venues.

Compare to benchmarks: arrival price (implementation shortfall), VWAP, TWAP. Calculate slippage, fill rate, market impact. Use TCA over time to identify patterns and improve routing strategies.

Components: 1) Market data aggregation from all venues, 2) Venue scoring algorithm (price, size, fill probability, cost, latency), 3) Route optimizer, 4) Execution engine with venue connectors, 5) Monitoring and TCA. Start with rule-based, evolve to ML-based.

Gradient boosting (XGBoost, LightGBM) works well for tabular fill prediction. Features: spread, depth, volume, queue position, historical fills. For sequence data (order book dynamics), RNN/LSTM can capture temporal patterns.

Common models: Square-root model (impact ~ sqrt(size/ADV)), Linear model, Almgren-Chriss (temporary + permanent impact). Calibrate to your own execution data. Consider temporary (dissipates) vs permanent (persists) impact separately.

Monitor execution in real-time: fill rate vs target, price drift, volume. Adjust: speed up if behind or price moving away, slow down if ahead or price favorable. Use feedback loops to continuously optimize execution throughout the order.

RL can learn optimal execution policies without explicit programming. State: market microstructure features. Action: routing, sizing, timing decisions. Reward: execution quality. Advantage: adapts to changing markets. Challenge: sample efficiency, simulation quality.