Implementation Shortfall

Formalized by Perold in 1988, and the reason he titled the paper "Paper versus reality."

You look at the screen, see a price, and decide to sell. By the time you've actually finished selling, you've received a different — worse — average price. Implementation shortfall is that gap, measured in basis points (1 bps = 0.01%).

$$\text{IS} = \frac{S_0 Q - \sum_t v_t P_t}{S_0 Q} \times 10{,}000$$

where $S_0$ is the mid-price when you decided, $Q$ is the total size, and $v_t$ units filled at price $P_t$ at each step. Positive means you did worse than the decision price. Lower is better.

Why it's the right metric

It captures both halves of Market Impact in one number — the immediate cost of crossing the spread and the slower price drift your own trading caused — without needing to attribute the damage to one or the other. That makes it directly comparable across strategies: a TWAP schedule, an Almgren–Chriss schedule, and a learned RL policy can all be scored on the same axis.

The part people get wrong

Reporting the mean implementation shortfall of a strategy tells you very little on its own. Two strategies with identical means can have wildly different distributions, and it's the tail that bankrupts you. Any serious evaluation reports the spread — standard deviation, the 95th and 99th percentiles, and conditional value-at-risk — not just the average.

Getting that distinction right is most of what makes the execution research I do hard.