Advanced Specialisations

The matching engine is the only place a trade actually happens. It is a deterministic, single-threaded loop that takes messages in a fixed sequence, applies each one to an in-memory book, and emits the fills and quote updates the whole market then sees.

A passive order is a race between two clocks, the queue in front of you draining, and the price walking away. Estimating who wins is what decides whether you post or cross.

The mid-price sits halfway between the bid and the ask no matter how lopsided the book is. The microprice weights it by queue sizes, giving a fair value that leans toward the thin side, which is the side about to be eaten.

Retail orders are small, round-lot, and routed through a predictable handful of wholesalers, which makes them identifiable from the tape even though no trade is tagged "retail." Being able to tell retail from institutional flow is worth real money to whoever can do it.

Every market maker's core problem is that the next order to hit their quote might come from someone who knows something they don't. Distinguishing informed from uninformed flow, even imperfectly, is what separates a profitable market maker from one that gets picked off.

A large institutional order sliced into hundreds of small child orders is designed to look like ordinary flow. Detecting it anyway, from the tape alone, is a core skill for both predatory traders looking to front-run it and desks trying to measure their own footprint.

Order flow imbalance predicts the next few seconds to minutes of price better than almost anything else available in real time, which is why it sits at the core of high-frequency trading models even though its predictive power decays almost as fast as it appears.

Not every counterparty is priced the same by a market maker. Toxicity scoring turns a counterparty's trading history into a number that predicts how much they'll cost you the next time you fill their order.

A price drop makes market makers more cautious, so they quote less; less liquidity makes the same order move price further; that bigger move makes market makers even more cautious. This feedback loop is how an ordinary selloff becomes a liquidity crisis.

Individual stocks' liquidity doesn't move independently, spreads and depth across a whole market tend to widen and thin together, driven by the same funding and risk-appetite shocks that hit every dealer at once.

A stock trading across sixteen venues looks less liquid on any single book than it did on one exchange, but add up the depth across venues and often nothing has actually been lost. Fragmentation moves liquidity around more than it destroys it.

After a trade eats through the order book, how fast does depth come back? The liquidity half-life turns that recovery speed into a single number you can compare across stocks and regimes.

Liquidity is not a fixed pool sitting in the order book, it is supplied by market makers who can walk away, and demanded by traders who need to fill now. Prices move when the two sides get out of balance.

An efficient price reacts to new information once, correctly, and then holds still, measuring how far real prices fall short of that ideal is how researchers put a number on market quality.

When an exchange changes a rule, a new tick size, a new order type, a new fee schedule, proving what the change actually did requires comparing the affected market to one that didn't change, not just watching a before-and-after chart.

A "good" market isn't just one with a tight spread, it lets traders transact cheaply, in size, without the price bouncing around for no reason, and it stays that way even under stress.

At the open and close, exchanges stop trading continuously and instead collect orders for a few minutes, then clear them all at a single price. That single step fixes problems continuous trading can't.

The exchange has a book full of buy and sell limit orders at different prices and needs one number. The algorithm picks the price that trades the most shares, then breaks any tie by minimizing leftover imbalance and finally by distance from a reference price.

A stock's first trade ever is not a continuous market finding its footing, it is a single call auction that has to set a fair opening price with no prior trading history to anchor it.

When an index provider changes its constituents or weights, every fund that tracks that index must trade at the same moment, the closing auction, creating some of the largest, most predictable liquidity events in markets.

Liquidity is not one number. Kyle's three dimensions separate what it costs to turn around right now (tightness), how much size the book absorbs before the price moves (depth), and how fast it heals afterwards (resiliency).

Every observed tick is the efficient price plus a small trading-mechanics error. That error is harmless at daily frequency and catastrophic at tick frequency, where it makes realised volatility blow up and returns look artificially mean-reverting.

An FPGA is a chip you wire into the exact circuit your strategy needs, instead of running instructions one at a time on a general-purpose CPU, trading a little flexibility for a latency advantage measured in tens of microseconds.

When the same asset trades on two venues and news reaches one microseconds before the other, a faster firm can trade on the stale, not-yet-updated price before it catches up, a race that requires no forecasting skill, only speed.

A backtest that reacts to a price the instant it appears in the data is trading with information it wouldn't actually have had yet, realistic execution latency has to be inserted between "signal fires" and "order arrives," or the backtest is quietly cheating.

Tick-to-trade latency is the clock from a market data update landing on your wire to your resulting order leaving the building, every microsecond in that path is a microsecond someone faster can trade on the same information first.

Regulation NMS is the 2005 US rule set that ties every stock exchange together into one "National Market System", it's the reason your broker is legally required to get you the best displayed price, no matter which exchange is quoting it.

The National Best Bid and Offer is the single best price across every US exchange, stitched together in real time from a dozen-plus separate venues, it's the reference price nearly every US equity rule is written against.

Europe's answer to Reg NMS isn't a hard rule tied to displayed prices, it's a broader legal duty requiring brokers to take "all sufficient steps" to get clients the best overall outcome, weighing price, cost, speed and likelihood of execution together.

Many retail brokers don't send your order to an exchange at all, they sell it to a wholesale market maker who fills it directly, and pays the broker for the privilege, which is why your "commission-free" trade still generates revenue for someone.

The single print that becomes "today's closing price" isn't the last continuous trade of the day, it's a separate call auction, run after a dedicated order-collection period, that now handles a huge and growing share of total daily volume.

Amending a resting order isn't a small edit made in place, under the hood it's usually a cancel of the old order and the birth of a brand new one, which is why a "small" price change can silently cost you your place in the queue.

The order entry gateway is the piece of exchange infrastructure that sits between a trading firm's systems and the matching engine, validating, sequencing, and forwarding every order before it's allowed anywhere near the book.

Some markets show a continuous, tradeable price to everyone at once; others make you ask a dealer for a price before you can trade at all. Which model a market uses shapes almost everything about how liquidity and information move through it.

Every order lives its whole life as a sequence of states, new, partially filled, filled, cancelled, rejected, and an execution report is the message that tells you which state it just moved to.

The same trade can be scored as a 22 bps loss or a 40 bps win depending on what you compare it to. Picking the benchmark is not a reporting decision, it is an instruction to the algo about what to optimise.

Every child order you send is evidence. If your slices are the same size, on the same venue, at the same cadence, other people work out what you are doing and the price moves before you get there. Leakage is the part of your cost that never comes back.

Every order is a trade-off between paying the spread now for a guaranteed fill (aggressive) and waiting in the queue for a better price that might never come (passive). The right choice depends on urgency and how likely you are to actually get filled.

You cannot backtest an execution algorithm the way you backtest a signal, your own order changes the very book you're trading against. Simulating execution means modelling queue position, fill probability, and impact, not just replaying historical prices.

Trading depresses the price, and then the book heals, new liquidity slowly refills the levels you just ate. Obizhaeva-Wang models that healing explicitly, which is what lets it say something Almgren-Chriss can't: trading fast and then pausing is different from trading slowly and steadily.