Measuring Liquidity and Transaction Costs in Practice (Chapter 21)
You cannot manage what you cannot measure. Harris opens Chapter 21 with this principle and then spends the rest of the chapter explaining just how hard it is to measure transaction costs properly. The basic idea is simple: compare what you paid to some benchmark price. But the choice of benchmark determines everything, and every benchmark has flaws.
This is one of the most practical chapters in the entire book. If you work anywhere near institutional trading, this is required reading.
Three Types of Transaction Costs
Harris breaks costs into three categories.
Explicit costs are the easy ones. Commissions, exchange fees, taxes, the cost of running a trading desk. A cost accountant can add these up.
Implicit costs are harder. These are the costs that arise because your trading moves prices. You buy at the ask, you sell at the bid. That spread is a cost. If your large order pushes prices against you, that market impact is a cost too. But measuring these requires a benchmark, some reference price that tells you what you would have paid if your trade had no impact. And that hypothetical price does not exist in the real world.
Missed trade opportunity costs are the hardest. These are the profits you lost because you did not trade or did not trade enough. Harris gives a concrete example: a speculator wants to buy 100 cotton futures at 65 cents per pound, submits a limit order at 64.95, and the price rises to 68 cents without filling. The missed opportunity cost is enormous, 137,500 dollars for 100 contracts. But most traders never track this number.
Here is a subtle but important point: transaction costs for one side are revenues for the other. Your cost of buying at the ask is the dealer’s profit from selling at the ask. When you add up transaction costs for all buyers and sellers in a trade, they sum to zero. This means sell-side institutions actually benefit from higher transaction costs, even though they market their services as providing low-cost execution. They do it because they compete with each other for buy-side business.
The Benchmark Problem
Every method for measuring implicit transaction costs compares your trade price against some benchmark. The formula is simple: trade price minus benchmark price, with the sign adjusted for whether you are buying or selling. But the choice of benchmark is where all the complexity lives.
Effective Spread
The effective spread compares your trade price against the quotation midpoint (the average of bid and ask prices) at the time of your trade. It is intuitive and easy to compute. If you buy at the ask and the midpoint is in the middle of the bid-ask spread, your cost is half the spread.
Retail traders use this method instinctively. They look at the bid and ask, execute, and compare. But the effective spread has a serious limitation: it tells you nothing about whether your broker timed the trade well. If the midpoint moves 50 cents against you while your broker waits, the effective spread does not capture that cost.
Realized Spread
The realized spread uses a quotation midpoint observed some time after the trade, typically 5, 15, or 60 minutes later. This interests dealers because their actual profits depend on where prices go after they take a position. Realized spreads tend to be smaller than effective spreads because prices often move against the dealer (toward the informed trader’s direction) after the trade.
The difference between effective and realized spreads measures what dealers lose to informed traders. Good information for understanding dealer economics, but not great for measuring your own transaction costs.
Implementation Shortfall
This is the one Harris clearly prefers. Implementation shortfall (popularized by Andre Perold) compares your actual portfolio to a paper portfolio that traded at the quotation midpoint at the moment you decided to trade. The difference between the two measures everything: the cost of the trade itself, the market impact, the timing delay, and the cost of any shares you failed to buy.
The key advantage: the benchmark price is determined before your order has any impact on the market. This solves a major problem that plagues other methods.
VWAP
The volume-weighted average price compares your trade price to the average price of all trades that day, weighted by volume. It tells you whether you traded better or worse than the average market participant.
VWAP is popular with institutional investment sponsors because it does not require intraday data. You just need daily summaries. But it has real problems.
Why Every Benchmark Has Flaws
Harris systematically evaluates each benchmark against several criteria. This is where the chapter gets really valuable.
Accuracy. The further the benchmark price is from the trade in time, the noisier the estimate. Random events unrelated to your trade can make a good execution look terrible or a bad execution look great. A trader who gets an excellent price in the morning looks awful if the stock tanks in the afternoon and the analyst uses the closing price as the benchmark. Effective spreads are the least noisy because the benchmark is contemporaneous. Everything else adds noise proportional to the time gap.
Split orders. Large traders break their orders into pieces. The first piece moves the market, making subsequent pieces more expensive. The effective spread underestimates total costs because the benchmark midpoint keeps adjusting to reflect the impact of earlier fills. The VWAP does even worse because the trader’s own trades influence the average. Implementation shortfall gets this right because the benchmark is set before any piece of the order has hit the market.
Harris demonstrates this with a clear example. A trader splits a 4,000-share buy into two 2,000-share pieces. The effective spread shows the same cost for both pieces (5 cents each). But the second piece actually cost more because the first piece moved the market up. Implementation shortfall correctly shows 5 cents for the first trade and 15 cents for the second.
Momentum and contrarian bias. If you buy after prices rise (momentum trading) and use the opening price as your benchmark, you will always look expensive because the opening price is low relative to when you traded. Contrarian traders who buy after prices fall will look cheap. Neither result reflects execution quality. It just reflects when you decided to trade relative to when the benchmark was set.
Informed trader bias. Well-informed traders who buy before prices rise will show low transaction costs when measured against later benchmarks. But this is not because they traded well. It is because they were right about the direction. The closing price and VWAP benchmarks underestimate costs for informed traders because those benchmarks already reflect the information the trader was trading on.
Gaming. This is the most insidious problem. If brokers know you will evaluate them using a specific benchmark, they can manipulate their behavior to look good without actually serving you well. A broker evaluated on effective spread can always make their costs look negative by exclusively using limit orders, buying at the bid and selling at the ask. But this means they never take liquidity, which means they may fail to fill urgent orders while chasing the market around.
Brokers evaluated on VWAP can game by spreading their trades evenly throughout the day so their average price matches the market average. Brokers evaluated on closing prices can hold orders until the close. In all cases, the broker optimizes for the metric instead of optimizing for the client.
Implementation shortfall cannot be gamed because the benchmark is set before the broker receives the order. Harris is clear: when data costs are not a concern, implementation shortfall is the best method.
The Plexus Iceberg
Harris describes the Plexus Group’s breakdown of implementation shortfall into components. They call it the Iceberg of Transaction Costs because the visible costs (commissions and market impact) are just the tip. The invisible costs (timing delays and missed trades) are the massive bulk underwater.
Their estimates: 12 basis points in commissions, 20 basis points in market impact, 53 basis points in timing costs as prices move away from the manager, and 16 basis points in orders that never fill. The hidden costs are more than twice the visible costs.
The critical insight: costs are easily shifted between categories. If a trader reduces visible market impact by waiting longer, they may just be converting market impact into timing cost. The total implementation shortfall captures everything, which is why Plexus measures all components.
Predicting Future Costs
Measuring past transaction costs is only useful if you can use the information to predict future costs. Traders build regression models that explain past transaction costs using variables like order size, price placement, bid/ask spreads, recent volume, price momentum, and market capitalization.
The most important order variable is size. Large orders cost more. The most important market variables are average volume and volatility. High-volume, low-volatility markets are cheap to trade in. Low-volume, high-volatility markets are expensive.
Portfolio strategists need these predictions before adopting a trading strategy. A strategy that looks profitable on paper might be unprofitable once you account for the cost of actually implementing it in real markets. This is especially true for large traders in illiquid markets.
The Cooperation Imperative
Harris ends with an important organizational point. Portfolio strategists and the traders who implement their decisions need to cooperate, not compete. Strategists need to tell traders why they are trading so traders can calibrate how aggressively to execute. Traders need to tell strategists what is feasible so strategists can form realistic orders.
Incentive systems that separately reward strategists and traders create adversaries. Strategists blame traders for bad execution. Traders blame strategists for impossible orders. Joint incentive contracts that reward the combined effort are better. You cannot disentangle contributions when people must cooperate to produce the best result.
For most active investment managers, reducing transaction costs through better execution is actually easier and more reliable than improving portfolio selection. The marginal dollar spent on better trading infrastructure often generates more value than the marginal dollar spent on better research.
This post is part of a series on Trading and Exchanges: Market Microstructure for Practitioners by Larry Harris (Oxford University Press, 2003).