6.8.3 Liquidity as a Priced Risk Factor

Basic Prerequisites: Econometrics & Time Series and 02 · Illiquidity Measures.

---

1. Intuition & Practical Objective

The level of illiquidity being priced (page 01) is not the whole story. The deeper, harder-to-hedge risk is that liquidity itself moves over time, and that some stocks are more exposed to those moves than others. A stock that behaves like a normal, liquid stock most of the time but becomes expensive to trade exactly when the whole market becomes expensive to trade is riskier than its average illiquidity suggests. Investors demand compensation for bearing that exposure.

This page builds the standard machinery: an aggregate liquidity factor, its innovations (surprises), and each stock’s liquidity beta - the sensitivity of its return to those innovations. The empirical punchline of Pástor & Stambaugh (2003): sorting stocks on liquidity beta produces a long-short spread of about 7.5% per year in abnormal returns - high-liquidity-beta stocks earn more.

The two-premium recap. Page 01 priced the level of illiquidity ($E[c^i]$). This page prices the covariance of the return with aggregate liquidity - the “PS channel” that Acharya–Pedersen formalize as their ${\beta }_3$ (the return-vs-market-illiquidity term; AP number the four betas ${\beta }_1..{\beta }_4$, so our ${\beta }_1,{\beta }_2,{\beta }_3$ are their ${\beta }_2,{\beta }_3,{\beta }_4$). They are different risks and both are compensated.

---

2. Mathematical Ground Truth & Derivations

Step 1 - per-stock liquidity measure. For each stock, each month, run a time-series regression of the stock’s excess return on the signed dollar volume (a price-impact slope). The slope coefficient ${\hat{\gamma }}_{i,t}$ is the stock’s liquidity: a large negative $\hat{\gamma }$ (returns fall sharply when volume is heavy) means illiquid.

Step 2 - aggregate liquidity. Equal-weight the per-stock measures into a market-wide series:

$${\hat{\gamma }}_t=\frac{1}{N_t}\sum _{i=1}^{N_t}{\hat{\gamma }}_{i,t},\Delta {\hat{\gamma }}_t={\hat{\gamma }}_t-{\hat{\gamma }}_{t-1}.$$

Step 3 - the innovation. Because the level of liquidity may be forecastable (and expected changes would contaminate risk measures), Pástor–Stambaugh extract innovations by regressing the change on its own lag and the lagged deviation from trend:

$$\Delta {\hat{\gamma }}_t=a+b\Delta {\hat{\gamma }}_{t-1}+c\left(\frac{m_{t-1}}{m_1}\right){\hat{\gamma }}_{t-1}+u_t,$$

then define the liquidity innovation $L_t=\frac{1}{100}{\hat{u}}_t$ (the $\frac{1}{100}$ is just scaling).

Step 4 - the liquidity beta. For stock $i$, regress its return on the innovation:

$$r_t^i={\alpha }^i+{\beta }_L^i{L}_t+{\epsilon }_t^i,{\beta }_L^i=\frac{\mathrm{cov}(r_t^i,L_t)}{\mathrm{var}(L_t)}.$$

A high (more positive) ${\beta }_L^i$ means the stock’s return tends to rise when aggregate liquidity improves and fall when it dries up - it is exposed to liquidity risk.

Step 5 - the price. Cross-sectionally, expected return increases in ${\beta }_L^i$:

$$E[r^i]=\alpha +\lambda {\beta }_L^i+\cdots$$

with $\lambda >0$: bearing liquidity risk is compensated. Pástor–Stambaugh measure the spread between high- and low-liquidity-beta deciles at roughly 7.5% annually after controlling for market, size, value and momentum.

---

3. Computational Implementation - a liquidity factor and sorted portfolios

Simulate a market with a true aggregate liquidity-innovation factor $L_t$, give stocks different true liquidity betas, then recover those betas and show that a long-short portfolio sorted on estimated beta earns a positive spread. Stdlib only.

Even in this small synthetic sample, sorting on the estimated liquidity beta delivers a monotone spread: the high-beta portfolio earns more than the low-beta portfolio. That spread is the liquidity-risk premium. (Real-world estimates are noisier, and Pástor–Stambaugh reach 7.5%/yr by first predicting betas from firm characteristics rather than using raw historical betas - see §5.)

---

4. Failure Modes & First-Principles Breakdowns

• Innovations, not levels. Using the level of liquidity instead of its surprise contaminates the beta (expected liquidity changes predict returns). The residual-based innovation $L_t={\hat{u}}_t/100$ exists precisely to purge this.
• Beta estimation noise. Historical liquidity betas are noisy, so decile sorts on raw betas understate the true spread; the Pástor–Stambaugh fix is to sort on predicted betas (from market cap, return, and other predictors).
• Small-stock tilt. Liquidity betas are highest for small, low-price, low-volume stocks - so a liquidity-factor strategy is easily confused with a size strategy unless you control for size (they do).
• Factor measurability. The aggregate liquidity series has occasional large downward spikes (Oct 1987, Sep 1998/LTCM, COVID) - the factor is fat-tailed, so OLS betas are sensitive to a few extreme months.

---

5. References

• Pástor & Stambaugh (2003). Liquidity risk and expected stock returns. JPE 111(3), 642–685. The construction (eqs. 6–8), the innovations, the predicted-beta sorts, and the 7.5%/yr spread.
• Acharya & Pedersen (2005). Asset pricing with liquidity risk. JFE 77(2)
• Chordia, Roll & Subrahmanyam (2000). Commonality in liquidity. Journal of Financial Economics 56
• Hasbrouck. Market Microstructure: Foundations

---

6. Connected Graph Bridges

• Next: 04 · Liquidity Risk & Crises · 05 · Failure Modes
• Back: 02 · Illiquidity Measures · Index Hub
• Sibling: Toxic Order Flow & VPIN (information risk as a second priced microstructure risk, Easley et al. 2002)