6.4.6 Advanced Extensions

Basic Prerequisites: Econometrics & Time Series and 03 · The GM Model.

---

1. Intuition & Practical Objective

The GM model reveals why the spread exists; this page launches the extensions that make it operational. GM gives belief-driven inference (informed traders arrive, the maker learns). Its three main descendants, all bridged here, attack different practical gaps:

1. Kyle (1985) - strategic size, not just direction. The informed trader chooses how much to trade, optimally hiding in noise volume. Out comes Kyle’s lambda $\lambda$ - price impact per unit of order flow - and the celebrated result that exactly half of the private information is impounded in one round: $Var[v\mid y]={\Sigma }_0/2$.
2. PIN (Easley–Kiefer–O’Hara 1997) - estimating the informed fraction $\pi$. $\pi$ is a parameter; PIN estimates it from daily buy/sell counts via a Poisson mixture, giving the empirical number the whole GM machinery needs.
3. Event uncertainty (Easley–O’Hara 1992) - time is information. When information events are themselves random, the arrival pattern (lulls, bursts) is informative - the bridge to modern toxicity and order-arrival models.

The objective: compute a Kyle equilibrium, recover its price-impact and impounding facts, and get the PIN formula in hand - all runnable.

---

2. Mathematical Ground Truth & Derivations

2.1 Kyle (1985) one-shot equilibrium (Hasbrouck Ch 7)

$v\sim N(p_0,{\Sigma }_0)$; informed demand $x=\frac{v-p_0}{2\lambda }=\beta (v-p_0)$; noise $u\sim N(0,{\sigma }_u^2)$; market maker sees total flow $y=x+u$ and posts

$$p=\lambda y+\mu ,\lambda =\frac{1}{2}\sqrt{\frac{{\Sigma }_0}{{\sigma }_u^2}},\beta =\sqrt{{\sigma }_u^2/{\Sigma }_0},\mu =p_0.$$

Kyle’s lambda $\lambda$ is the price impact per unit order flow; depth is $1/\lambda$. Magnitudes: $Var[v\mid p]=Var[v\mid y]={\Sigma }_0/2$ - half the private information is impounded, independent of noise intensity - and informed profit $\mathbb{E}\pi =\frac{(v-p_0{)}^2}{2}\sqrt{{\sigma }_u^2/{\Sigma }_0}$ rises with the squared value divergence and with noise-trading variance (camouflage). Multi-period: the insider “slices and dices,” $\Delta x_n={\beta }_n(v-p_{n-1})\Delta t$, each round pricing in a further sliver (Hasbrouck Ch 7 §7.2). Price impact model in practice: $\Delta P=\lambda Q$ - the market-impact workhorse.

2.2 PIN - the empirical informed probability (Hasbrouck Ch 6; EKOH 1997)

On each of many days an information event occurs with probability $\alpha$; when it does, informed traders arrive at intensity $\mu$ and always trade on the right side; uninformed arrive on both sides at intensity $ϵ$. Daily buys $b\sim Poisson$, sells $s\sim Poisson$, so over $D$ days the likelihood is a Poisson-mixture of textbooks (eq 6.3) whose model parameters $\alpha ,\mu ,ϵ$ are estimated by MLE, giving

$$\mathrm{PIN}=\frac{\alpha \mu }{\alpha \mu +2ϵ}\in [0,1].$$

The product $\alpha \mu$ is identified (so PIN is stable/estimable even though $\alpha ,\mu$ separately are imprecise - Hasbrouck Ch 6). PIN is the empirical estimate of the GM informed-fraction and of order-flow toxicity.

2.3 Event uncertainty & the timing of information (Easley–O’Hara 1992)

When the arrival of information itself is uncertain, the absence of trades is informative: no-trade intervals are low-information states, bursts are high-information. Trade timing becomes a signal - the step from “does order direction tell me anything?” to “does the pattern of order arrivals tell me anything?” That is the theoretical backbone of modern order-flow-imbalance and toxic-flow detection.

---

3. Computational Implementation - Kyle lambda, impounding, and PIN (stdlib only)

The run confirms the two invariant facts of the informed-trading family: Kyle’s price immediately impounds the informed demand (residual $p-v\approx 0$), and exactly half the private information is incorporated in a single round ($Var[v\mid y]=2.0={\Sigma }_0/2$). And PIN translates the model into an operational toxicity number: $0.43\to 0.69$ as informed intensity rises. These are the numbers a maker or a researcher actually uses to set toxicity hedges and measure price impact.

---

4. Failure Modes & First-Principles Breakdowns

1. Kyle’s “lambda” is not a constant. The one-shot $\lambda =\frac{1}{2}\sqrt{{\Sigma }_0/{\sigma }_u^2}$ is a model value; empirically, price impact is concave in size, time-varying, and inventory/vol sensitive. Treating $\lambda$ as a static per-share parameter misprices large orders (see Avellaneda–Stoikov for the inventory-driven correction).
2. PIN’s parameter bundle. Only $\alpha \mu$ is well identified; $\alpha$ and $\mu$ separately are imprecise (Hasbrouck Ch 6). Don’t over-interpret individual PIN pieces; the product is the stable quantity.
3. The martingale property is about the maker’s info, not your model. “Prices are martingales” in GM refers to the maker’s information set. If your model under-uses the right conditioning variables, the residual predictability is yours to miss - the semi-strong-form claim does not protect a poorly-specified model.
4. One-shot Kyle hides the multi-period slicing. The single-round result (half impounded) is the kernel of a sequence in which the insider spreads a finite information advantage over many rounds to raise total profit. Reading the one-shot as “all information enters instantly” mistakes the kernel for the process.

---

5. References

• Kyle (1985), Continuous auctions and insider trading, Econometrica 53(6), 1315–1335
• Easley, Kiefer & O’Hara (1997), The information content of the trading process, JFE 44(1), 159–186
• Easley & O’Hara (1992), Time and the process of security price adjustment, J. Finance 47(2), 577–605
• Hasbrouck (2007), Empirical Market Microstructure
• Glosten (1994), Is the electronic open limit order book inevitable?, J. Finance 49(4), 1127–1161

---

6. Connected Graph Bridges

• Back: 05 · Failure Modes · Index Hub
• Forward/in-pillar: Toxic Order Flow & VPIN (PIN in production) · inventory) · Spread Decomposition & the Roll Model (measuring λ empirically)
• Base: Econometrics & Time Series · Probability & Measure Theory