An algorithmic trading firm uses programs to make trading decisions, quote prices or execute orders. These are different jobs: a bank, market maker or fund can use automation without sharing the same business model. Before judging a provider, separate what its algorithm does, whose capital bears the risk and whether the firm acts as your counterparty. The examples below show why these distinctions matter.

Which roles does algorithmic trading cover?

A market maker quotes buying and selling prices and manages the resulting exposure. An execution algorithm implements a specified purchase or sale, often by splitting it into smaller orders; it need not choose the investment direction. A decision-making algorithm instead uses defined inputs to decide when and what to trade. The BIS reports distinguish these functions, which can coexist within one institution.

HFT is a subset of algorithmic trading, associated with short holding periods and low latency. A program executing an order over a longer period is not automatically HFT. A hedge fund may use quantitative strategies without being a market maker, and an online prop challenge does not by itself identify an institutional trading firm. These labels answer different questions, so do not treat them as interchangeable descriptions of a provider.

Provider capabilities to check

Banks and non-bank providers can both use technology and risk capital. Check pricing, credit access, inventory controls and client relationships. XTX Markets describes FX spot liquidity and algorithmic execution among its client services, with banks, funds and retail brokers among its clients. That is the provider’s description of its offering; it neither ranks the whole market nor proves which firm executed a particular order on your retail account.

Citadel Securities LLC’s October 2021 FX disclosures supply a different detail: in the described transactions it acts as principal for its own account, rather than as the counterparty’s agent. Pricing can vary with transaction size, prevailing conditions and inventory or risk appetite. The document expressly excludes other affiliates and does not describe every individual execution. Use it to identify that disclosed arrangement, not to infer terms for an entire group.

A 20% share and 25% ratio can describe the same executions

Record the denominator before comparing algorithm reports. The name “volume share” alone does not resolve it.

Scroll sideways to see the full diagram

Same volume, two denominatorsIn one fictional window and data source, an algorithm executes EUR 10,000 and other participants EUR 40,000. Its share of the EUR 50,000 total is 20%; its ratio to others’ volume is 25%. These are two custom measures of the same executions.THE SAME EXECUTION SETOwn: EUR 10,000 · others: EUR 40,000Q = 10,000M = 40,000Share of the entire setQ / (Q + M) = 10,000 / 50,000 = 20%Ratio to others’ executionsQ / M = 10,000 / 40,000 = 25%Difference: 5 percentage points.Same source and window.Only the denominator changes.

A complete fictional EUR/USD execution set from one source in one defined window: own executions Q = EUR 10,000 and others M = EUR 40,000. Each execution is counted once in EUR; unfilled orders are excluded. Both measures are defined only for this exercise, not as any provider’s POV parameter or the global FX market share. The BIS — FX execution algorithms and market functioning report describes percentage-of-volume algorithms using estimated volume and data limitations in fragmented FX markets.

Both numbers match their definitions. They do not describe different execution outcomes.

The first report divides own volume by the total: 10,000 / 50,000. The second uses only others’ executions: 10,000 / 40,000. The five-percentage-point difference adds no trades. Comparing reports requires matching denominators, windows, units and sources. Neither calculation gives the completed fraction of a parent order.

How they make markets with algorithms and speed

A market maker’s spread is not guaranteed profit. Take two alternative fictional scenarios, excluding commissions and other costs: it buys EUR 10,000 at USD 1.1000 per euro. Selling the same amount at 1.1002 produces USD 2; closing after a decline to 1.0995 instead loses USD 5. The exposure between the transactions matters. These are our arithmetic examples, not observations of any firm’s results.

An execution algorithm solves another problem. A fictional EUR 100,000 purchase is split into EUR 40,000 at 1.1000 and EUR 60,000 at 1.1002 USD per euro. The total cost is USD 110,012 and the volume-weighted price is 1.10012. That is USD 12 above a purchase entirely at the initial reference price, but the example does not establish that all the required volume was available at that price. A comparison needs an attainable benchmark, completion rate and costs.

Co-location can shorten a network path, and internalisation can offset flow within a provider. Neither tells you that inventory risk is zero. Likewise, “prop” requires reading the specific agreement: FTMO’s cited CFD FAQ describes trading on its platform as simulation with fictitious capital, while data may separately be used for the firm’s own live trading. This provider-specific statement is not a definition of every prop program or every institution trading its own capital.

What it changes for spreads and liquidity

Compare costs for a defined venue, client, size and period. More providers do not guarantee a tighter retail spread on every account.

The BIS’s 2020 report examines trade-offs among market impact, market exposure and the chance of completing an order. A quote can change or disappear, but a volatility spike does not prove that every algorithm or bank reacts identically. Compare a defined venue, client access, order size and time window. A record of actual bid, ask and fills is more informative than attributing the change to a named firm without evidence.

Measure liquidity under stated conditions

Record bid, ask, size, venue, client access and timestamp. Quiet conditions do not guarantee deep liquidity or a fill without slippage; stressed conditions do not imply identical behavior by all providers.

An unfinished order also has a measurable consequence. In our fictional model, the intended purchase is EUR 100,000 and the initial reference is 1.1000 USD per euro. Only EUR 60,000 fills at 1.1002. At the chosen end of the window the reference is 1.1010; this is a comparison price, not an assumed executable quote.

Price cost and unfilled quantity are separate
Model componentResultMeaning
EUR 60,000 filledUSD 12Difference from the initial reference
EUR 40,000 unfilledUSD 40Assumed opportunity cost
Combined comparisonUSD 52Not a commission invoice
Completion rate60%40% remains unpurchased

The filled portion costs USD 66,012, or USD 12 above its starting reference. The remaining portion contributes 40,000 × (1.1010 − 1.1000) = USD 40 to the chosen opportunity-cost measure. Reporting only the USD 12 difference hides incomplete execution. The combined USD 52 is not a fee paid or money owed: the actual purchase covered only EUR 60,000. It depends on the benchmark and end time, and proves no alternative execution was available.

What this means for an individual trader

Check the broker’s disclosures and execution records. A tight spread alone does not identify an upstream provider.

Institutional infrastructure differs from a retail setup. A higher timeframe or fundamental view does not automatically establish a retail edge.

If you want to see how the various participants fit together in one place, our market participants guide maps out the full liquidity chain — from banks to funds and algorithmic trading firms.

What to do after closing this article

  1. Read the actual contract and execution policy. B-book is not synonymous with matching equal customer orders, and a provider list does not prove the routing of your trade.
  2. Record spread and depth at synchronized times. Two quotes cannot attribute a change to the retreat of a particular market maker.
  3. Choose a strategy and timeframe using evidence, costs and a loss budget. Hourly charts and fundamentals do not automatically remove algorithmic competition.