Skip to main content
Status of this page: P0 draft. Impact coefficients are market- and period-specific; the numerical example is illustrative with an assumed prefactor.

1. Impact models

  • Kyle (1985) linear model: price change = lambda x signed net order flow. lambda falls as noise trading rises. It is an equilibrium coefficient of an information model, not automatically the causal impact of one trade.
  • Square-root law: metaorder impact is roughly Y x sigma x sqrt(Q / V), where sigma is daily volatility, Q is the order size, V is daily volume, and Y is a constant of order one. The shape is widely reported across markets; Y must be calibrated.
  • Temporary vs permanent: temporary impact decays after execution; permanent impact is the information content. Almgren and Chriss (2001) use linear permanent and temporary components.
Illustration. sigma = 2% daily, participation Q / V = 1%, Y = 1 (assumed): impact = 0.02 x sqrt(0.01) = 0.002 = 20 bps. Halving the order size reduces impact by a factor of sqrt(0.5), i.e. about 29%, not 50%.

2. Execution schedules

Almgren-Chriss. Minimise expected cost plus lambda x variance. In the linear model the efficient frontier of strategies (lowest expected cost for a given risk) can be constructed explicitly, and the risk-neutral solution is a uniform (TWAP-like) schedule; more risk aversion front-loads. The same framework leads to liquidity-adjusted VaR.

3. TCA metrics

  • Implementation shortfall = (execution price - decision price) x side, plus fees, and an opportunity cost term for unfilled shares.
  • Arrival slippage vs mid at order arrival.
  • VWAP slippage vs interval VWAP (can be gamed by trading with the benchmark).
  • Reversion: mid move after completion; high reversion indicates temporary impact, low indicates information.
  • Spread capture for passive fills, measured with post-fill mid (queue position and fills).

4. Building a cost model

  1. Collect parent order data: size, start/end, participation, symbol, venue, side.
  2. Fit cost = a x spread/2 + b x sigma x (Q/V)^c on parent orders; test c near 0.5.
  3. Check out of sample and by volatility regime.
  4. Use the fitted function as the cost term in optimisers and as the fee in backtests.

5. Caveats

  • Own-flow impact is confounded with signal: orders placed when alpha is strong move with the market regardless of impact.
  • HK-specific: closing auction participation and the tick table both change the cost curve (auction mechanisms, tick size rules).

Points to verify

  • Calibrated value of Y for HK cash equities and ETFs; the example uses an assumed 1.
  • Exact statement of the Almgren-Chriss solution for the risk-averse case (hyperbolic functions in the original).

Sources

  • Almgren and Chriss (2001), Optimal execution of portfolio transactions, Journal of Risk 3(2). Abstract
  • Kyle (1985), Continuous Auctions and Insider Trading. Link
  • Maglaras, Moallemi and Zheng (2015) for a microstructure impact model. PDF