6.3.6 Advanced Extensions

Basic Prerequisites: 04 · Quote Skewing and 05 · Failure Modes.

---

1. Intuition & Practical Objective

Every defect of the baseline inventory model motivates an extension. This page is the launchpad for the upgrades that matter in production: optimal liquidation (how to unwind a large inventory when you must), combining the skew with adverse selection (position control is not information control), and multi-asset / portfolio inventory (correlated positions share one risk budget). Each is linked to its dedicated in-pillar page.

Why these first? Optimal liquidation is what a maker actually does when the skew fails (the cap hit of 05 triggers it). Adverse selection is the other killer - the one inventory control cannot touch. Multi-asset is the scaling reality: real makers quote thousands of correlated instruments and cannot treat inventory one stock at a time.

---

2. Mathematical Ground Truth & Derivations

2.1 Optimal liquidation (Almgren–Chriss)

When a large position must be unwound over horizon $T$, trading it all at once pays full market impact, and dribbling it out leaves price risk. Almgren & Chriss (2000) minimize

$$\mathbb{E}[\text{cost}]+\lambda \mathrm{Var}[\text{cost}]$$

with temporary impact $\eta$ and permanent impact ${\gamma }_{\text{perm}}$. The optimal remaining-inventory trajectory is

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

starting at $x_0=X$ and decaying to $x_T=0$ with characteristic time $1/\kappa$ - this is the controlled mean reversion of 02, and the bridge to the execution pillar. A higher risk aversion $\lambda$ or variance ${\sigma }^2$ ⇒ faster liquidation (larger $\kappa$, shorter half-life); higher impact $\eta$ ⇒ slower (protect the price).

2.2 Combining the skew with adverse selection

Quote skewing prices inventory risk over uninformed flow. Add a fraction $p_{\text{tox}}$ of informed fills that move the mid against you by $J$ after you fill. Each fill then carries an adverse cost $p_{\text{tox}}J$ that the skew’s spread does not cover. The break-even condition for one side is

$$\underset{\text{half-spread}}{\underbrace{a^{\ast }}}=c+\frac{1}{k}\ge p_{\text{tox}}J.$$

Once $p_{\text{tox}}J$ exceeds the half-spread, the maker loses on informed flow no matter how well inventory is controlled. The fix is an information term in the spread (Glosten–Milgrom logic, toxicity-gated quoting) - not a bigger skew. This is the direct link to Toxic Order Flow & VPIN and Adverse Selection & Glosten–Milgrom.

2.3 Multi-asset / portfolio inventory

A real maker quotes $N$ correlated instruments sharing capital and margin. The inventory becomes a vector $\mathbf{q}$ with covariance $\Sigma$, and the reservation price of asset $i$ generalizes to

$$r_i={\bar{S}}_i-\gamma {\mathbf{e}}_i^{\top }\Sigma \mathbf{q}\tau ,$$

so a long in a positively-correlated neighbour counts as (partial) excess inventory in asset $i$, and vice versa. The skew is no longer per-asset but portfolio-wide: the risk budget is shared. Ignoring $\Sigma$ double-counts risk and quotes $N$ times the intended position.

2.4 Guéant–Lehalle–Fernandez-Tapia: inventory constraints as a well-posed problem

Guéant, Lehalle & Fernandez-Tapia (2013) re-solve the market-making problem with the inventory cap $|q|\le Q$. Under exponential intensities, a change of variables turns the HJB into a system of linear ODEs for $v_q(t)$:

$${\dot{v}}_q(t)=\alpha q^2{v}_q(t)-\eta (v_{q-1}(t)+v_{q+1}(t)),\alpha =\frac{k}{2}\gamma {\sigma }^2,\eta =A(1+\frac{\gamma }{k}{)}^{-(1+k/\gamma )},$$

on the $2Q+1$ inventory states, with closed-form asymptotics for the quotes that reduce to the A–S spread plus an inventory-dependent correction. This supplies the verification theorem the raw A–S quotes lacked - the formal statement that “the skew encourages flattening, the cap guarantees it” (05).

---

3. Computational Implementation - optimal liquidation + adverse selection

(a) We compute the Almgren–Chriss liquidation path and verify it starts at $X$ and decays to 0, with the half-life set by $\kappa$. (b) We add informed flow ($p_{\text{tox}}$) to a skewing maker and show the spread’s half-markup $a^{\ast }=c+1/k$ stops covering it.

(a) The liquidation path decays from $1000$ shares to $0$ with a half-life $=\ln 2/\kappa =1.16$ yr, front-loading the unwind (speed $1114$ shares/yr at $t=0$, still $943$ at $t=T$). This is the optimal rate - faster and you pay impact, slower and you carry price risk. (b) Adverse selection is the failure inventory control cannot fix: with $p_{\text{tox}}=0.3$ mean P&L falls from $53.5$ to $28.6$; at $p_{\text{tox}}=0.6$ it collapses to $3.6$ while inventory std stays flat - the skew keeps the position under control but has no defence against informed flow. The fix is a wider, information-priced spread, not a bigger skew.

---

4. Failure Modes & First-Principles Breakdowns

1. Optimal liquidation ≠ instant liquidation. Dumping the whole position pays full impact; the Almgren–Chriss path trades impact against price risk. Getting the impact coefficient $\eta$ wrong mis-sizes the whole unwind.
2. Adverse selection survives every inventory fix. Capping inventory and skewing quotes do nothing about informed flow; the fix must add an information term or an external toxicity gate (VPIN). Do not mistake position control for safety against being picked off.
3. Multi-asset coupling is ignored at your peril. Treating $N$ correlated instruments independently double-counts risk and quotes $N$ times the intended inventory; the covariance matrix must enter the reservation price.
4. The closed-form asymptotics have validity ranges. The Guéant approximation is excellent for small $|q|$ and degrades near the cap; use the exact linear-ODE solution when the limit is tight.
5. Estimated parameters feed every extension. $\eta$, $\lambda$, $p_{\text{tox}}$, $J$, $\Sigma$ are all estimated; a spurious value (a noisy toxicity rate, a wrong correlation) shifts quotes and turns a neutral maker into a directional bet.

---

5. References

• Almgren & Chriss (2000), Optimal execution of portfolio transactions, Journal of Risk 3(2)
• Guéant, Lehalle & Fernandez-Tapia (2013), Dealing with the inventory risk, Math. & Financial Econ. 7(4)
• Cartea, Jaimungal & Penalva (2015), Algorithmic and High-Frequency Trading, Cambridge UP
• Cartea & Jaimungal (2015), Risk metrics and fine tuning of high-frequency trading strategies, Mathematical Finance 25(3), 576–611
• Easley, López de Prado & O’Hara (2012), Flow toxicity and liquidity in a high-frequency world, RFS 25(5)

---

6. Connected Graph Bridges

• Back: 05 · Failure Modes · Index Hub
• In-pillar forward: Adverse Selection & Glosten–Milgrom · Toxic Order Flow & VPIN · A–S: Advanced Extensions
• Cross-pillar: Optimal Execution & Almgren–Chriss · POV · VaR & Expected Shortfall