How market organisation, information and inventory risk turn orders into observable prices, fills and trading costs.
Market microstructure
Market microstructure is the study of how trading arrangements and participants’ behaviour turn buy and sell intentions into transactions, prices and trading costs. For a quant, its practical consequence is that a forecast about an asset is only part of a trading strategy: the rules, available liquidity and information revealed during execution affect whether the forecast can be monetised.
The subject includes exchange order books, dealer markets, auctions, order routing and the information in trades and quotes. It asks both how a market is organised and what that organisation does to observed prices. A particular spread formula, matching algorithm or market-making model addresses part of that subject; none defines the whole field.
Start with a price you cannot automatically trade
Suppose a fictional share has a best bid of USD 99 and best ask of USD 101. The midprice is their arithmetic average:
m=299+101=100 USD per share.
The bid-ask spread is USD 2 per share. The ask is USD 1 above the midpoint.
Assume the quote stays fixed, enough shares are available at the ask, and an aggressive buy for 100 shares executes there in full. Ignore fees and funding. The buyer spends USD 10,100, while the acquired shares marked at the same midpoint are worth USD 10,000. The difference is USD 100, or USD 1 per share.
Synthetic quote: bid USD 99, midpoint 100 and ask 101 per share. Buying 100 shares at the ask costs USD 10,100; their midpoint mark is USD 10,000, a USD 100 benchmark difference before fees.
Original synthetic frozen-quote ledger, USD per share and USD cash amounts. Enough quantity is assumed at the ask. No fees, quote changes or additional book levels are included.
This is a one-way execution concession against a stated benchmark. It is not a forecast of a subsequent loss, a round-trip cost or the seller’s net profit. Nor does the midpoint establish the share’s fundamental value. It is an observable quote-based reference; an order at that price may fail to execute.
Already, a seemingly simple question—“What is the price?”—has several answers: a bid, an ask, a midpoint and a transaction price. Microstructure explains why these can differ and which is relevant to a particular decision.
The mechanism matters
In a limit order book, resting orders specify prices and quantities. An incoming order can consume that liquidity or wait for a counterparty. More size may require worse prices; displayed depth may change before the order arrives. In a dealer market, a customer may instead request a quote from an intermediary trading for its own account. An auction concentrates matching at a specified event. These arrangements create different execution opportunities and information sets.
Allocation rules matter even when the price is the same. Under price-time priority, earlier eligible orders at a price have priority over later ones. Seeing a trade at your limit price therefore does not prove your resting order filled. The venue’s actual rules must determine the allocation. Nasdaq’s OUCH overview describes price-time matching, while NYSE’s parity explanation describes a different allocation approach at the same price. Neither description should be applied to every venue or every order type.
The Order Books lesson develops the consequences: average prices for orders that consume several levels, allocation under stated priority rules, and the limits of aggregate snapshots. The public page is a preview; the full lesson requires membership.
Why supplying liquidity carries risk
A participant willing to quote both sides does more than collect a spread. Two central risks help explain its decisions:
Information risk. A counterparty may be trading because it knows something the quote supplier has not yet incorporated. A passive buy can fill just before the market falls; a passive sell can fill just before it rises. This is adverse selection. Fills are selected by other traders’ decisions, so the returns on filled orders can differ from returns measured at arbitrary times.
Inventory risk. Buying from customers accumulates a position whose value can change before it is sold or hedged. A participant engaged in market making may adjust quotes, quantities or hedges to manage that exposure. A quoted spread is therefore not automatically a realised profit margin.
These mechanisms can interact with processing costs, competition and trading rules. Their importance depends on the market and circumstances; a wide spread does not uniquely identify its cause. Lee and Martin’s New York Fed discussion illustrates how asymmetric information and inventory or liquidity costs interact in a stylised dealer-market model. Its results depend on that model’s assumptions.
What a quant must preserve in the data
A transaction records a match that happened. A quote records terms being offered at an observation time. Neither alone records the price available to a different trader arriving later with a different size.
This distinction affects both research and execution. Alternating trades at the bid and ask can create short-horizon price variation even when the midpoint is unchanged. A backtest that assumes every buy and sell occurs at the midpoint can omit a trading concession. A passive-fill model that awards a fill whenever a trade touches its price can ignore queue priority. And an observed price move after a trade does not, by itself, establish the trade’s causal market impact: the trade and the price move may respond to the same information.
Useful microstructure analysis therefore names its observations, timing, venue rules, order size and benchmark. Those choices connect a statistical signal to an executable transaction—and make the resulting transaction-cost estimate interpretable.
Sources
Rishi K. Narang, Inside the Black Box: A Simple Guide to Quantitative and High-Frequency Trading, 2nd edition, Wiley, 2013, Chapter 7, printed pp. 120–124, especially “Aggressive versus Passive” — execution choices, priority, adverse selection and quote-based benchmarks. The numerical ledger and figure are original.
