The Square-Root Market Impact Law
The square-root impact law is the empirical observation that the expected price impact of a metaorder grows roughly with the square root of its participation in daily volume, rather than linearly with shares traded. A common form is
\[ I(Q) = Y\,\sigma\,\left(\frac{|Q|}{V}\right)^\delta, \qquad \delta \approx \tfrac12, \]
where \(Q\) is signed parent quantity, \(V\) is relevant market volume, \(\sigma\) is a volatility scale, and \(Y\) is a dimensionless coefficient. This formula is central to capacity planning because it says that doubling assets does not merely double cost.
It is a conditional empirical regularity, not a law of mechanics. Its usefulness comes from expressing impact in normalized units across instruments and regimes; its danger comes from treating a noisy average as a guaranteed price path.
Why concavity matters
If impact were linear, two equal halves of a parent order would incur the same total expected impact as one whole order under fixed conditions. With square-root impact, the marginal cost declines with completed quantity:
\[ \frac{dI}{dQ} \propto Q^{-1/2}. \]
That does not mean traders should rush the first shares. Volatility, information leakage, temporary liquidity, and benchmark risk change during execution. The formula describes aggregate displacement associated with a metaorder, not a complete optimal schedule.
| Quantity fraction \(Q/V\) | \(\sqrt{Q/V}\) | Relative impact if \(Y\sigma=1\) |
|---|---|---|
| 0.01% | 0.010 | 1.0% |
| 0.10% | 0.032 | 3.2% |
| 1.00% | 0.100 | 10.0% |
| 4.00% | 0.200 | 20.0% |
The table shows why capacity can fail quietly. A strategy scaled tenfold from 0.01% to 0.10% ADV faces roughly 3.2 times, not ten times, the normalized impact—yet dollar turnover also grows, so total P&L erosion can be severe.
Define the object before calibrating it
A trade print is not a metaorder. The parent order has a signed quantity, start and end time, execution style, and often an unobserved decision price. Public data generally cannot recover these labels perfectly. If parent orders are reconstructed by a sign-based clustering rule, report that rule and test sensitivity.
Choose a target price response: end-of-execution displacement, close-to-decision movement, or post-trade peak and decay. These targets mix different economics.
import numpy as np
def square_root_impact(q, daily_volume, volatility, y=0.5):
participation = np.maximum(np.abs(q) / daily_volume, 1e-12)
return np.sign(q) * y * volatility * np.sqrt(participation)
def expected_cost_bps(q, daily_volume, volatility, y=0.5):
return 1e4 * np.abs(square_root_impact(q, daily_volume, volatility, y))
Use point-in-time forecast volume and volatility in production. Completed daily volume is available only after the fact; using it in a morning pre-trade forecast leaks the very liquidity that makes the order cheap.
Calibrate conditional distributions, not a single exponent
Fit log impact against log participation with robust methods, but do not discard negative or low-signal observations solely to make the fit clean. A practical model often estimates \(Y\) by asset, side, venue, volatility regime, spread regime, and duration. Report prediction intervals. The residual variation around the average curve is often larger than the estimated mean at modest participation.
| Input | Why it shifts impact |
|---|---|
| Volatility | Repricing and inventory risk rise |
| Spread/depth | Immediate liquidity is scarcer |
| Participation rate | Faster trading consumes replenishment |
| Duration | More time allows recovery but adds risk |
| Information state | Correlated alpha flow can persist |
The formula's \(\sigma\sqrt{Q/V}\) scaling is often more portable than raw share impact, but coefficients are not universal. ETFs, small caps, futures, auctions, and crypto venues differ in matching, fragmentation, and arbitrage. Out-of-sample cost prediction is the standard, not an appealing cross-sectional regression.
Impact versus implementation shortfall
Market impact is only one component of implementation shortfall. Spread, fees, delay, opportunity cost, and price drift can dominate small orders. Conversely, a buy order with positive alpha may experience a favorable price move while still causing impact; using arrival-price P&L alone can conflate forecast return and trading cost. The measurement framework in transaction cost analysis separates these pieces.
The law also does not determine permanent impact. Prices can revert after a mechanical liquidity shock or remain elevated because the order revealed information. Estimate response paths and condition them on execution duration and market regime.
Turning the law into capacity limits
For each proposed rebalance, forecast cost under a conservative coefficient range and compare it with gross expected alpha. Aggregate names can interact: simultaneous buys in correlated names may compete for the same liquidity providers. Capacity is therefore a portfolio-level stress test, not a per-name ADV filter.
Use participation caps, minimum expected net alpha, and escalation rules for thin names. The broader portfolio implications are covered in strategy capacity and market impact; schedule design is covered by Almgren-Chriss optimal execution.
Key takeaways
- Square-root impact normalizes expected metaorder displacement by volatility and volume participation.
- It is concave and useful for capacity planning, but it is not a complete execution-cost model.
- Define parent-order labels, response horizon, and point-in-time volume forecasts before fitting.
- Calibrate conditional distributions and uncertainty by regime; a universal coefficient is rarely defensible.
- Combine impact with spread, fees, opportunity cost, and correlated portfolio demand before sizing capital.