2.3.6 Advanced Extensions

Basic Prerequisites: 03 - Implementation Shortfall and 05 - Failure Modes & Practice.

---

1. Intuition & Practical Objective

The three failure modes of page 05 each have a structured response, and together they define how a real desk’s execution stack is built. This page is the launchpad - it shows the four extensions that turn a static VWAP/TWAP into something a modern execution desk ships:

1. Benchmark-aware scheduling - the choice between a day benchmark (VWAP) and a decision benchmark (arrival price / IS) is not cosmetic: on a trending day they demand opposite schedules.
2. Adaptive / arrival-price (IS) execution - solve the schedule against a live impact + risk model and re-solve as price and volume arrive (the bridge to the Almgren–Chriss frontier).
3. POV / participation-based execution - a real-time-volume-following schedule that is structurally immune to profile misestimation.
4. Hybrids and smart order routing - the modern engine blends these: start on arrival-price, switch to POV on volume, place child orders as marketables vs. passive across venues.

Why this order: benchmarks frame the objective; arrival-price gives you an optimal curve; POV gives you robustness; routing gives you venue-grade (how each child is actually placed). Together they close the loop opened in page 01.

The one-sentence essence. “A static schedule is a point solution; a production engine is a feedback system that re-aims at its benchmark as price and volume arrive - the benchmark determines the target, the adaptive curve the path, POV the robustness, and routing the venue-level tactics.”

---

2. Mathematical Ground Truth & Derivations

2.1 Benchmark choice is a scheduling decision. On a day with upward drift and ${\varphi }_t$ the volume profile, a VWAP engine executes near $\text{VWAP}={\sum }_t{\varphi }_t{p}_t$, an IS/arrival engine trying to minimize $\mathbb{E}[(\bar{p}-m_0)]$ should front-load (buy early, before the drift). Formally, Perold’s IS with a full fill is $\bar{p}-m_0$; its variance-minimizing trajectory under drift points decisively early, while the VWAP objective is clock-agnostic to drift. Same order, opposite direction.

2.2 Arrival-price = the Almgren–Chriss curve (the bridge). The IS-optimal child schedule is exactly the AC trajectory

$$x_t=X\frac{\sinh (\kappa (T-t))}{\sinh (\kappa T)},\kappa =\sqrt{\frac{\lambda {\sigma }^2}{\eta }},$$

from 03 - The Almgren–Chriss Model - an exponentially-decaying, front-loaded curve (heavy at the open, light at the close), which is precisely the arrival-price flavor. VWAP/TWAP are the limit cases where you instruct “match volume (VWAP)” or “match clock (TWAP)” instead of “follow the impact-risk map (IS).”

2.3 POV = the profile-immune schedule. With live volume $v_t^{\text{live}}$ arriving and participation rate $\rho$,

$$q_t=\rho v_t^{\text{live}},\Rightarrow w_t^{\text{exec}}={\varphi }_t^{\text{realized}}\Rightarrow \text{TE}=0.$$

POV does not predict the profile; it follows it, so on a news day (page 05 failure 2) POV stays matched to realized volume by construction - at the cost of fully accepting whatever adverse selection that volume carries.

---

3. Computational Implementation - benchmark-aware and POV-robust

On a rising day, compare a VWAP engine vs a front-loaded (arrival-price) schedule, then show POV self-pacing on a volume spike. Stdlib only; reproduced exactly.

Read the numbers. On the rising day, the VWAP engine executes at the VWAP ($+84.1$ bps of IS vs the decision price) while the front-loaded arrival-price engine buys early and lands at $+65.67$ bps - $18.4$ bps cheaper precisely because it ignored the day benchmark and followed the drift instead. That $18.4$ bps is the value of picking the right benchmark (IS on a trend) and the whole point of benchmark-aware scheduling. Second, on a late volume spike the fixed VWAP engine sizes its child from the forecast profile, which allots $8.9\%$ of the parent to a bucket that then realizes $22.8\%$ of the day - under-participating the spike and carrying residual size into the thin close - while POV sizes its child to the actual arriving volume share ($\sim 8,000$ scrip scaled to live volume) and never over-commits.

---

4. Failure Modes & First-Principles Breakdowns

1. Benchmark mismatch is the silent alpha leak. Run a VWAP algo on a trending day and every broker report says “great, +0.0 bps vs VWAP” while IS says you left $\sim$18 bps against the decision. The report is honest to the wrong benchmark.
2. POV robust ≠ POV optimal. Following realized volume eliminates profile risk but buys full exposure to whatever volume arrives - including every informed move. POV trades control of when for the removal of what-if.
3. Adaptive re-solving adds its own risk. Re-solver feedback can chase noise, overtrade into micro-spikes, and (in a worst case) become its own detectable pattern. Production systems damp the control and randomize the updates.
4. Hybrids inherit both families’ failure modes. A blend of IS-start → POV-volume → smart-routing must be tuned, not bolted: the IS component is drift-sensitive, the POV component trend-adverse, the routing component venue-selective. Every layer has a failure; the stack is defense-in-depth (page 05).

---

5. References

• Almgren & Chriss - “Optimal execution of portfolio transactions,” Journal of Risk 3(2), 5-40 (2000). The arrival-price / IS-optimal schedule that this page’s advanced engine implements.
• Bertsimas & Lo - “Optimal control of execution costs,” J. Financial Markets 1(1), 1-50 (1998). The dynamic-programming seed of adaptive execution.
• Kissell - The Science of Algorithmic Trading (2014); Kissell, Glantz & Malamut - Optimal Trading Strategies (2003). IS, POV and benchmark-aware execution in desk practice.
• Johnson - Algorithmic Trading & DMA (2010). POV / participation algos and smart routing of child orders.
• Cartea, Jaimungal & Penalva - Algorithmic and High-Frequency Trading (2015). A stochastic-control formalization of adaptive execution.

---

6. Connected Graph Bridges

• Handoff: Optimal Execution & Almgren–Chriss · AC Advanced Extensions
• Microstructure base: Market Impact & Depth · Adverse Selection & Glosten–Milgrom
• Routing/venue: Market Microstructure & Order Types
• Transaction costs: Constraints & Transaction Costs