Market participants

Who is in the flow: market makers, prop HFT firms, agency execution, institutions, and retail. Knowing whose order you’re likely trading against is half of adverse selection.

The idea

Why does it matter who is on the other side?

Because a liquidity provider's entire P&L is the spread it earns minus what it loses to better-informed counterparties: adverse selection. Fill an uninformed retail order and you keep the spread; fill an informed order and you have just bought something that is about to fall. Same quote, opposite outcome. The only difference is who took it.

Intuition first: a market maker is an insurer who must quote before knowing who will trade with them. Against uninformed flow the spread is pure profit; against informed flow it is a premium that does not cover the claim. You cannot see the counterparty's identity, so you infer it from the flow's behaviour and price accordingly. This is the practical heart of Glosten–Milgrom (1985) and Kyle (1985). The single organising axis of this whole page is toxicity: how adversely-selecting a stream of flow is. Low-toxicity flow (retail) is the prize; high-toxicity flow (informed) is the tax. Every participant type below is really a position on that axis, and your job as a market maker is to fill more of the prize and less of the tax, by venue choice, quoting, and toxicity detection (PIN/VPIN).

The honest framing: you do not get to choose your counterparties directly, but the venue, instrument and time of day strongly shape the mix you face. Reading that mix (knowing whether you are in a retail-heavy or an informed-heavy pool) is as important as any model. This is the macro view; adverse selection is the micro mechanics.

Retail: the prize (and the PFOF wholesalers who fill it)

Retail traders place small, mostly uninformed orders; they rarely trade on information a market maker lacks, so their flow has low toxicity and is the most profitable to fill. That is exactly why wholesalers pay retail brokers for it (payment for order flow) and internalise it off-exchange. Retail flow is the prize the whole PFOF economy is built to capture.

Intuition first: a retail buy of 100 shares because someone read a headline is almost never the trade that picks you off. Aggregated across millions of users it is close to noise: the textbook "uninformed" or "noise" trader of Kyle (1985). Quoting against it, you keep the spread far more often than you get run over. This is why it is bought: the retail-wholesaler / PFOF model exists precisely because this flow is the least-toxic, most-makeable order flow in the market. Wholesalers pay brokers for exclusive access, internalise it with price improvement on the NBBO, and profit on the spread. "Zero commission" retail trading is funded here.

The 2026 nuance, and retail is not uniformly dumb: a slice of retail is now fast, options-savvy and herding (social-media-coordinated momentum, zero-day options). In bursts this flow becomes informational and moves the market, flipping briefly from prize to hazard. The aggregate is still low-toxicity, but the tails are fatter than the textbook says. Read the mix, not the label. For the crypto and Polymarket reader: the "retail" cohort on an open venue is your largest, most makeable counterparty too, but it sits next to whales and bots with no wholesaler layer in between, so you face it more directly. See crypto market making.

Institutional / asset managers: the elephants execution algos serve

Institutional asset managers (pension funds, mutual funds, insurers) trade in large size to express slow, fundamental views. Their orders are too big to dump at once without moving the price, so they are sliced over time by execution algorithms. To a market maker they are the elephants: big, directional, and therefore moderately toxic when you are on the wrong side of one.

Intuition first: when a fund decides to buy $\pounds 200\text{m}$ of a stock over a week, every slice it sends is a small piece of a large, persistent buying pressure. If you keep selling to it you are repeatedly on the losing side of a move: moderately informed flow, not because the fund is fast, but because it is big and one-directional. This is the market impact story seen from the maker's side. This is the participant the entire execution-algorithms section exists to serve and protect: VWAP/TWAP/POV and Almgren–Chriss slice the elephant's order to minimise its own impact and hide its footprint from exactly the market makers and HFTs trying to detect it. The institution and the HFT are in a quiet game: one hiding size, the other inferring it (see schedule gaming).

Toxicity verdict: moderate and detectable. A single slice is small, but the persistence of a large parent order is a signal: order-flow imbalance and VPIN are partly about spotting an elephant in the room. Fill its flow profitably when you can detect and skew against the direction; lose to it when you cannot.

Other HFTs and market makers: your fast, informed competition

Other HFTs and market makers are the fastest, best-informed participants in the book, and increasingly, the ones on the other side of your trades. When you trade against another market maker, neither of you is the dumb money; whoever is a fraction of a second slower or a touch worse-informed loses. This is the most toxic, lowest-margin flow there is.

Intuition first: filling another HFT is filling someone with the same models, the same data and possibly a faster connection. There is no uninformed-flow premium to earn; it is a zero-sum speed-and-information contest. You only want this trade when you are the faster or better-informed side; otherwise it is pure adverse selection. The 2026 reality: as algos populate more of the book, a rising share of every market maker's counterparties are themselves algos. The pool of pure uninformed flow per market maker shrinks, which is one structural reason per-trade edges thin and the economics push toward scale. More machines, less easy flow.

This is also the latency-arbitrage arena: when two fast participants react to the same event, the faster one picks off the slower one's stale quote. Against this cohort, speed is the edge, which is why colocation and FPGA are table stakes in the classic venues and why a slow newcomer is simply someone else's profit. Toxicity verdict: highest. Trade other HFTs only when you have a genuine speed or information advantage; otherwise this flow is the tax in its purest form.

Informed / toxic flow: the tax

Informed (toxic) flow is any counterparty trading on information you do not have: a fundamental signal, a faster feed, a leaked event. Filling it means you systematically buy just before prices fall and sell just before they rise. It is the single biggest risk to a liquidity provider and the reason the spread exists at all. Detecting it is half the job.

Intuition first: the spread is not a fee, it is compensation for the risk that whoever takes your quote knows more than you. Glosten–Milgrom (1985) makes this exact: the bid-ask spread is the price of adverse selection. Widen it and you lose less to the informed but win less from the uninformed; the whole art of spread vs adverse selection is balancing the two. "Informed" is a spectrum, not a category: a fundamental fund acting on research, a faster firm reacting to a news event microseconds before you, a participant who saw a related market move first. What unites them is that their arrival predicts the next price move against you, which is precisely what order-flow imbalance, PIN/VPIN and trade-sign inference try to measure in real time.

The defensive playbook is recognition, not avoidance (you cannot dodge it entirely): widen quotes when toxicity rises, skew away from the pressure, reduce size, and in the limit pull quotes. A market maker who cannot detect rising toxicity is one who keeps quoting tight into a stampede and is carried out. This is the core risk the whole order-flow-information topic is built to manage. Toxicity verdict: by definition the highest, and the one that defines the spread. You do not eliminate it; you price it.

Hedgers: the price-insensitive flow

Hedgers trade to offset a risk elsewhere (a corporate hedging FX exposure, an option dealer delta-hedging) not to express a view on price. Their flow is price-insensitive and predictable in shape (often around expiries or fixings) rather than informed about direction, so it is low-to-moderate toxicity and sometimes a structurally exploitable pattern.

Intuition first: a hedger does not care whether the price is "right"; they need to be flat a risk by a deadline, so they trade regardless. That makes their flow uninformed about direction but concentrated in time (around the close, an expiry, a fixing): predictable plumbing rather than a signal about value. Why it matters to a market maker: hedging flow is mostly makeable like retail (low directional toxicity), but its concentration creates recurring event windows (the option-expiry "pin", the FX fixing, the index-rebalance flow) where large, predictable, price-insensitive orders arrive. Knowing the calendar of hedging flow is itself an edge (see scheduled vs unscheduled events).

The caution: hedging flow tied to a derivative can be informed indirectly. A wave of delta-hedging can signal where a large option position sits and force a reflexive move (gamma effects). So "hedger" is low-toxicity on average but not always inert. As ever, read the flow, not the label. Toxicity verdict: low-to-moderate, and often a calendar pattern more than a risk.

Worked example

A stylised hour on a market maker's book, illustrative and as of 2026, sorted by what each counterparty did to the P&L. Retail (including via a wholesaler) is about 45% of fills at low toxicity: it keeps roughly the full spread and is the bedrock of the day's P&L. Hedgers (expiry and fixing flow) are about 10% at low-to-moderate toxicity, mostly makeable, concentrated around the close. Institutional slices (the elephants) are about 20% at moderate toxicity, a small loss when on the wrong side of a parent order, profitable when skewed correctly. Other HFTs and market makers are about 20% at high toxicity, near break-even at best, a loss whenever a fraction slower. Informed, news-driven flow is about 5% at the highest toxicity: the bulk of the day's losses, a small share doing large damage.

How the mix changes the business: shift this book onto a venue with more informed flow (say, a fast institutional dark pool) and the red rows grow, the maker must widen spreads, and the economics get harder. Shift it onto a retail-heavy venue and the green rows grow. The same model is profitable or not depending entirely on the participant mix, which is the whole point of the page, and the practical heart of the going-independent question: the open venues you can actually access (crypto, prediction markets) have a thinner institutional layer and a different toxicity profile, so reading their mix is the first job. Numbers are synthetic, rounded and illustrative; real mixes vary by venue, instrument and hour. The durable fact is the Pareto shape (a small slice of toxic flow drives most of the losses) not the exact percentages. Educational only, not investment advice.