Execution without measurement is folklore. Transaction Cost Analysis (TCA) is the discipline that closes the loop: define a benchmark, measure slippage against it, attribute the causes, and feed the findings back into strategy choice and parameters. This part covers the benchmark landscape, the arithmetic, and the pitfalls that make naive TCA worse than none.

The Benchmark Landscape

A benchmark is the reference price you compare fills against. The choice is not cosmetic - it defines what “good” means and therefore what behavior it incentivizes. Benchmarks fall into three families by when their information exists:

Pre-Trade Benchmarks

• Decision price: the price when the PM made the call. Measures the full cost of implementation, including any delay before trading started.

• Arrival price: the mid-quote when the order reached the desk or algorithm - the anchor of the implementation-shortfall framework (Part 10) and the standard for alpha-bearing flow.

• Prior close / open: convenient references for overnight decision processes and for baskets benchmarked to a print.

During-Trade Benchmarks

• Interval VWAP: the market’s volume-weighted average over your window - beat it and you traded better than the average participant (Part 5).

• PWP (Participation-Weighted Price): the price a hypothetical X% POV execution would have achieved - a fairer yardstick for participation-style mandates (Part 6).

Post-Trade Benchmarks

• Closing price: the standard for index-tracking flows, whose true objective is matching the close.

• Reversion marks (mid at T+5min, T+30min, next open): not targets, but diagnostics - covered below.

The Arithmetic, Standardized

Slippage is quoted in basis points, signed so positive = cost:

cost (bps) = side × (execution price − benchmark) / benchmark × 10,000, with side = +1 for buys, −1 for sells.

A buy filled at 101 versus an arrival of 100 costs 100 bps; a sell filled there earns 100 bps. On top of per-fill arithmetic, the shortfall decomposition splits total cost into delay, trading, and opportunity components - attributing cost to the stage that caused it, and pricing unfilled quantity honestly rather than ignoring it (Part 3).

Reversion: Separating Impact from Alpha

The most informative post-trade diagnostic is what the price does after you finish:

• Price reverts toward pre-trade levels → you paid temporary impact: your own pressure moved the price, and it relaxed once you stopped. Persistent reversion says trade slower, display less, or use more midpoint liquidity.

• Price keeps going in your direction → the order carried real alpha (or permanent information), and the drift you paid was the market catching up. Persistent continuation says trade faster - your patience is donating alpha to the market.

This single diagnostic, aggregated across many orders, is the main evidence for tuning urgency - and the empirical input to the impact models of Part 13.

Pitfalls That Invalidate TCA

• Noise dominates single orders. One order’s slippage is mostly market randomness. Statistically meaningful conclusions need aggregation across many orders, with outliers examined separately rather than silently averaged in.

• Selection bias. Comparing algorithms on the orders each was chosen for confounds strategy with difficulty: the desk sends hard orders to the careful algorithm and easy ones to the cheap one, then concludes the cheap one is better. Fair comparisons need matched samples or randomized routing experiments.

• Benchmark gaming. Every benchmark can be traded to: a VWAP-benchmarked desk can hug the curve and never save a basis point; a close-benchmarked desk can dump everything into the auction. The metric improves while the fund’s actual outcome doesn’t.

• Sloppy timestamps and marks. Arrival measured at order creation versus market arrival can differ by seconds - enough to reassign whole basis points between “delay” and “trading.” In crypto, the benchmark must also name its venue: there is no consolidated tape, and “the price” differs across exchanges (Part 14).

• Ignoring unfilled quantity. A 60%-filled order with beautiful per-fill slippage may still be a bad execution once opportunity cost on the remainder is charged.

TCA as a Loop, Not a Report

Mature desks run measurement as a cycle:

1. Pre-trade: estimate expected cost and risk (via the models of Part 13), choose strategy and horizon accordingly, and record the estimate.

2. In-flight: monitor slippage versus the estimate in real time, with alerts when an order breaches its expected envelope - catching problems while they can still be corrected.

3. Post-trade: aggregate, attribute (spread vs. impact vs. timing; venue-level fill quality), compare against the pre-trade estimate, and adjust parameters - urgency, participation caps, venue tiers - for the next cycle.

The pre-trade estimate deserves its own machinery: how do you predict the cost of an order before trading a share of it? That is Part 13: Market Impact Models.