How to Count the Cost of Waiting
For a retail limit order, the price on the filled shares is only half of the economic story. This chapter focuses on the measurement problem created by non-execution: how to combine the price paid for shares that filled with the opportunity cost of shares that did not. The goal is to read implementation shortfall, fill rate, and effective spread together, then interpret the study’s lower measured cost for limit orders without turning a strong sample result into a universal trading rule.
The distinction follows directly from the choice examined in the previous chapter. A marketable order is designed to obtain liquidity now. A nonmarketable limit order accepts uncertainty about completion in exchange for greater price control. If the limit order is judged only by the shares that execute, it can look unusually cheap. If it is judged only by the shares that fail to execute, it can look like a mistake. A useful measure must keep both parts in view.
Why a favorable fill can still be an expensive trade
Start with the executed portion. An effective-spread measure compares the execution price with a benchmark, in this study the midpoint of the NBBO when the broker receives the order. For a buy, execution below that midpoint is favorable relative to the benchmark; for a sell, execution above it is favorable. The measure asks a focused question: how much did the filled trade cost relative to the market available at the order’s arrival?
That question matters, but it cannot answer what happened to the unfilled shares. Suppose a customer submits a buy limit order and only half the shares execute. If the market then rises, the remaining shares are no longer available at the original limit price. The investor may cancel and buy later at a higher price, or may decide not to complete the position. The first outcome adds price drift and a new spread to the economic cost. The second preserves cash but gives up the intended exposure. Either way, the favorable price on the first half does not describe the whole decision.
The study addresses this problem with implementation shortfall, or IS. Conceptually, the measure is a weighted combination:
- the effective spread for the fraction of shares that filled; and
- an opportunity cost for the fraction that did not, based on an imputed price at the end of the order’s life.
For an order that expires at the close, the study uses the closing price on the submission day. For an order ending through cancellation or execution, it uses the opposite-side quote at the final event: the ask for an unfilled buy and the bid for an unfilled sell. This is a demanding benchmark. It includes the price drift during the order’s life and the spread that might be paid if the investor crosses the market after canceling.
The measure also makes an explicit assumption: the investor was committed to completing the full top order. That assumption is necessary for a like-for-like comparison between an immediately executable order and a limit order that leaves shares unfilled. But it is not a statement about every investor’s intention. An investor who deliberately accepts missing a discretionary purchase has a different personal objective from an investor who must obtain a complete hedge or position. IS makes the cost of non-execution visible under the completion objective; it does not decide whether that objective was appropriate.
A simple illustration shows why the weighting matters. Imagine a 100-share buy limit order in which 60 shares execute at a price 20 basis points favorable to the arrival midpoint, while the remaining 40 shares are assigned a 30-basis-point opportunity cost because the market moved away. The executed component contributes 60% × negative 20 basis points, or negative 12 basis points, and the unfilled component contributes 40% × positive 30 basis points, or positive 12 basis points. The combined IS is approximately zero. If the same favorable price occurred but 80 shares filled, the favorable component would dominate; if only 20 filled, the opportunity-cost component would dominate. The illustration is not a new estimate from the study. It shows why fill rate is part of the cost mechanism rather than a separate descriptive footnote.
The same logic explains why two investors can receive different personal results from orders in the same statistical category. One order may fill almost completely before the price moves, while another may remain unfilled through a sharp move. The category average combines those paths. It is useful for judging the population’s typical trade-off, but it cannot tell an individual investor which branch their next order will take. The more a conclusion depends on the unfilled portion, the more important it becomes to report the benchmark used for that portion and the investor’s assumed commitment to completion.
The endpoint also matters. Using the opposite quote at the final event charges the unfilled shares for both price drift and the spread that might be paid to complete the trade. Using a closing price for an order that expires at the close gives a different but stated comparison. Neither imputation recreates the investor’s actual future transaction in every case. The point is to make a hidden consequence measurable and comparable, not to pretend that an unexecuted order has one observable market price after the fact.
The sign of IS also needs care. In the study’s signed basis-point convention, a negative average IS for limit orders means that the combined execution and imputed non-execution outcome was favorable relative to the chosen benchmark. It does not mean the broker paid the customer, that the investor earned a portfolio return, or that the result is a guarantee for a future order. It is a benchmark-relative trading-cost measure built from a particular lifecycle assumption.
This is why effective spread and IS can point in different directions without contradicting each other. Effective spread can show a favorable result for filled limit-order shares while the opportunity cost of unfilled shares is positive. IS combines the two. If the fill rate is high enough and the filled shares receive a sufficiently favorable benchmark comparison, the all-in measure can remain favorable even after charging the unfilled portion for price drift. If the order fills rarely or the market moves sharply away, the opportunity-cost term can dominate.
The study reports precisely that tension. Marketable orders in the sample have an average fill rate close to 100%, while nonmarketable limit orders average about 65%. That lower limit-order completion rate is the cost of seeking price control. Yet the limit orders’ average effective spread is about negative 20.2 basis points, while marketable orders average 1.04 basis points. The limit orders also carry an opportunity cost of approximately 16 basis points for unexecuted shares. After combining the two components, limit orders have an average IS of about negative 8.2 basis points, compared with roughly 1.1 basis points for marketable orders.
The apparent contradiction is the result to understand, not a reason to discard one metric. Limit orders give up some completion certainty, but the shares that do execute can receive a favorable price comparison. The study’s high fill rate for retail limit orders means that the favorable executed portion is large enough to outweigh the imputed cost of unfilled shares on average. The cost advantage is therefore not “free price improvement.” It is the measured result of a particular balance between price, fill probability, and opportunity cost.
What the retail comparison establishes—and what it does not
The empirical setting is unusually broad for a retail-order study. The researchers use OATS data from May 2020, covering more than 27 million top orders from individual accounts handled by 19 active U.S. retail brokers and 300 stocks stratified by size. The data contain order receipt, routing, execution, cancellation, and price information. That lifecycle view is what makes an opportunity-cost calculation possible; a database containing only completed trades could not observe the order that waited and then disappeared.
Within that defined sample, the difference remains after the researchers address several obvious comparability problems. Their regressions compare marketable and limit orders with stock-day fixed effects, so the comparison is not driven simply by one order type appearing in one group of stocks or on different days. They control for order size and the quoted spread at arrival. Additional specifications add broker effects and match orders within the same stock, five-minute interval, and buy-or-sell direction. The marketable-order coefficient remains positive: marketable orders have an IS roughly 9 to 10 basis points higher than limit orders in these comparisons.
Each control closes a different interpretive loophole. Stock-day effects reduce the chance that the result is just a difference between large, liquid names and smaller, wider-spread names. Arrival spread helps account for the market’s condition when the order entered, while order size recognizes that larger orders can be harder to execute. Broker effects absorb some persistent differences in clientele, interfaces, and handling practices. Matching within a short stock-time-direction window asks a particularly concrete question: when buy and sell orders of both types are submitted under nearly the same observable conditions, does the measured gap remain? The answer in the study is yes.
The value of those controls is not that they create a randomized experiment. They narrow the set of plausible explanations. If limit orders were cheaper only because they were used in different stocks, at different times, or with different spreads, the gap should shrink substantially after those factors were aligned. It does not. The persistence supports the conclusion that patience and liquidity provision are economically relevant in the sample.
But “supports” is the right verb. Investors choose their order types. Those who use limit orders may differ in urgency, information, experience, monitoring behavior, or willingness to miss a trade. The study tests several ways of comparing orders under similar observable conditions, but it cannot observe every motive behind each instruction. The result is a strong conditional association, not proof that a randomly selected investor could lower costs simply by replacing every marketable order with a limit order.
The size pattern reinforces the conditional reading. The cost difference between limit and marketable orders is about eight basis points in large stocks, 18 basis points in medium stocks, and 21 basis points in small stocks. Smaller stocks generally have wider quoted spreads, so the reward for supplying liquidity can be larger. At the same time, a wider spread can reflect less dependable liquidity and a higher risk that an order will not complete at the desired price. The result describes an economic trade-off, not a ranking in which small stocks are automatically better places to use limit orders.
The measure also has an applicability boundary created by the sample. The authors identify ordinary held orders from individual accounts, filter out certain modified, oversized, merged, incomplete, or anomalous lifecycles, and analyze the 19 brokers together rather than publishing broker-specific results. May 2020 is a historical window, and OATS was later replaced by the Consolidated Audit Trail. The paper therefore offers strong evidence about the order population and market conditions it studied, not a live performance table for every current broker.
That boundary matters particularly when the headline is shortened to “limit orders are cheaper.” Cheaper in what sense? The study’s answer is lower implementation shortfall under an assumption that the investor intended to complete the full order. It is not a claim that limit orders always produce higher portfolio returns, that their lower measured IS is independent of investor selection, or that a customer should tolerate non-execution regardless of purpose.
A reader can use the evidence more carefully by keeping four questions together:
- Completion: What proportion of the submitted quantity filled, and was that proportion acceptable for the objective?
- Price: How did the filled shares compare with the arrival midpoint or another stated benchmark?
- Non-execution: What price or event was used to represent the unfilled shares, and how sensitive is the conclusion to that imputation?
- Inference: Does the comparison align orders under similar stocks, times, sizes, spreads, brokers, and directions, and what unobserved differences remain?
Those questions turn a statistical result into a judgment about evidence quality. They also prevent a common mistake: using a favorable average to erase the individual order’s failure mode. A trader who needed all shares immediately may experience a limit order very differently from the sample’s average, even if the order belongs to the same category. Conversely, a patient trader who values price discipline may regard a non-fill as an acceptable outcome rather than an economic loss.
The main conclusion is consequently two-sided. In this large, defined retail sample, nonmarketable limit orders have lower measured implementation shortfall than marketable orders, and the difference survives controls for several observable conditions. That is meaningful evidence that retail investors can benefit from supplying liquidity rather than demanding immediacy. It is not a universal instruction to use limit orders, because the result depends on the completion assumption, the market environment, the order population, and the investor’s purpose.
The next question is how retail orders achieve the combination of price control and relatively high fill rates observed here. Many are placed behind the best quotes and remain open for much longer than fleeting market-wide limit orders. The following chapter traces how price aggressiveness and time create that outcome—and where patience can stop helping.