2.4.6 Advanced Extensions

Basic Prerequisites: 04 · Queue-Reactive Models and 05 · Failure Modes & Practice.

---

1. Intuition & Practical Objective

The queue view scales up. This page is the launchpad for the extensions that turn a single-queue fill model into a live trading system:

1. The Cont–Kukanov fill function - the exact expected-fill formula for a random outflow, and how fill ratio collapses with queue position.
2. Multi-venue overbooking - why posting more than you need across several venues is optimal, and how diversification kills non-execution risk.
3. Queue-reactive intensities - making $\lambda (q),\mu (q)$ state-dependent (the Huang–Lehalle–Rosenbaum model).
4. Order-flow-imbalance (OFI) price impact - the linear law $\Delta P=\beta \mathrm{OFI}/\mathrm{depth}$ that links fills to price moves.

Each is one honest step past the tractable core. Everything farther (Hawkes self-exciting arrivals, queue-reactive market making, cross-venue routing under fees) is linked from here.

---

2. Mathematical Ground Truth & Derivations

2.1 Cont–Kukanov fill function and the value of being early

For queue position $x$ ahead, order size $L$, and random outflow $\xi \sim F$,

$$\mathbb{E}[\text{filled}]=\mathbb{E}[(\xi -Q{)}^{+}-(\xi -Q-L{)}^{+}].$$

The fill ratio $\mathbb{E}[\text{filled}]/L$ is maximised by small $Q$ (be at the front). For an exponential(mean $m$) outflow there is a closed form: with $\xi \sim \text{Exp}(1/m)$,

$$\mathbb{E}[(\xi -Q{)}^{+}]=m{e}^{-Q/m},\mathbb{E}[\text{filled}]=m(e^{-Q/m}-e^{-(Q+L)/m}).$$

2.2 Optimal multi-venue overbooking

With $K$ venues, queue positions $x_k$ ahead, and the same limit price, total bought is

$$A(X,\xi )=M+\sum _{k=1}^K[({\xi }_k-x_k{)}^{+}-({\xi }_k-x_k-L_k{)}^{+}],$$

where $M$ is the market-order catch-up at the horizon. Because the ${\xi }_k$ are imperfectly correlated, posting $L_k$ on each of several venues (“overbooking”) and cancelling the unused rest drives non-execution risk down faster than it raises impact - the fills are a diversified portfolio. This is the optimal-placement insight of Cont–Kukanov (2017): size is determined jointly by queue position and the correlation of outflows across venues.

2.3 Queue-reactive intensities

Replace constants by functions of the current size: market-order intensity $\mu (q)$ and limit-order intensity $\lambda (q)$, both empirically increasing and concave (saturating) in $q$. The generator becomes state-dependent and the ergodic distribution is solvable numerically. The concave shape is the quantitative statement of “liquidity begets liquidity”: thin books attract refills, deep books attract aggressive flow.

2.4 OFI price impact (Cont–Kukanov–Stoikov 2014)

Define the order-flow imbalance at the best quotes over an interval as the signed change in best-queue sizes:

$${\mathrm{OFI}}_k=\sum _{\text{events in }k}[\text{signed change in bid size}-\text{signed change in ask size}].$$

Over short intervals,

$$\boxed{\Delta P_k=\beta \frac{{\mathrm{OFI}}_k}{{\text{depth}}_k}+{\epsilon }_k}$$

a linear relation whose slope is inversely proportional to market depth, robust across time scales and stocks. Combined with a scaling argument it implies the empirical square-root relation between price change and traded volume. OFI is the single variable that ties queue dynamics to price - the empirical bridge between this folder and the market-impact literature.

---

3. Computational Implementation - fill ratio, overbooking, OFI

Stdlib + numpy. We compute the CK fill ratio from Monte Carlo, show multi-venue overbooking slashing non-execution risk, tabulate a concave reactive intensity, and fit the OFI price-impact slope.

Three lessons in the numbers. (i) Queue position is a fill-ratio tax: moving from the front ($Q=0$, ratio $0.850$) to behind 3,000 shares ($Q=3000$, ratio $0.312$) cuts the fill ratio by nearly two-thirds. (ii) Overbooking buys execution insurance cheaply: two venues take non-execution from $28\%$ to $7.4\%$, four venues to $0.47\%$ - the diversification of imperfectly correlated outflows. (iii) The OFI regression recovers the theoretical slope to five digits with an honest $R^2=0.896$ - queue/flow imbalance is the linear driver of short-horizon price change.

---

4. Failure Modes & First-Principles Breakdowns

1. Ignoring outflow correlation across venues. Overbooking only diversifies when ${\xi }_k$ are imperfectly correlated. On venues with the same taker base (or during a market-wide sweep) the fills co-move and the diversification benefit vanishes exactly when you need it.
2. OFI is an aggregate, not a queue. $\Delta P=\beta \mathrm{OFI}/\mathrm{depth}$ prices average impact; it is blind to your queue position. Using a single $\beta$ to size a passive edge double-counts price impact that your position may never experience (or misses impact your fills do incur).
3. Reactive intensity mis-specification. The concave $\lambda (q)$ is an empirical shape, not a law; on venues with different tick sizes / fee structures the curvature differs. Calibrating on the wrong venue’s shape biases the fill forecast.
4. Static overbooking. The optimal $L_k$ depends on the live queue positions $x_k$ and the remaining time - a static split is stale within seconds. Production systems re-optimise continuously.
5. Fees and rebates dominate at the margin. $r_k$ can flip sign across venues (maker-taker vs inverted); the queue-optimal placement is wrong if the fee term is dropped. This is a first-order cost, not a rounding error.

---

5. References

• Cont, Rama; Kukanov, Arseniy - “Optimal order placement in limit order markets,” Quantitative Finance 17(4), 553–571 (2017)
• Cont, Rama; Kukanov, Arseniy; Stoikov, Sasha - “The price impact of order book events,” J. Financial Markets 17, 47–88 (2014)
• Huang, Lehalle & Rosenbaum - “Simulating and analyzing order book data: the queue-reactive model,” JASA 110(509), 107–122 (2015)
• Rosu (2009)
• Cartea, Jaimungal & Penalva - Algorithmic and High-Frequency Trading (2015)
• Gueant - The Financial Mathematics of Market Liquidity (2016)

---

6. Connected Graph Bridges

• Back: 05 · Failure Modes & Practice · Index Hub
• Sibling in-pillar: Optimal Execution & Almgren–Chriss · Market Microstructure & Order Types
• Impact & market-making: Market Impact & Depth · Avellaneda–Stoikov & Optimal Quoting
• Routing/fees: Execution Algorithms: VWAP, TWAP, POV