Every term, a student asks me some version of the same question: “If I build a fast enough algorithm, can I do high frequency trading?” My answer is always the same. Speed is useless if you are fast at reading the wrong information. A Formula One car with a blindfolded driver is not quick. It is just an expensive accident.
So let us ask the better question: what data feeds are necessary to run a high frequency trading system effectively? In forex, this matters more than in almost any other market. There is no single central exchange, prices come from several venues at once, and the “price” on your retail chart is often a broker’s smoothed version of something messier underneath.
In this lecture-in-print, I will go through the feeds a serious HFT operation depends on, from raw order book depth to the humble clock signal. I will explain why each one matters, what it costs you to ignore it, and where beginners waste money. By the end you will know what to ask a data vendor, and also when to walk away.
Table of Contents
- Why Data Is the Raw Fuel of High Frequency Trading
- What Data Feeds Are Necessary to Run a High Frequency Trading System Effectively?
- Cross-Market, News and Reference Data Feeds
- Time, Latency and the Plumbing Nobody Brags About
- The Feeds That Watch You: Drop Copy, Risk and History
- Honest Warnings Before You Spend a Penny
- Frequently Asked Questions
Why Data Is the Raw Fuel of High Frequency Trading
A high frequency trading system does three things repeatedly, often thousands of times a second. It observes the market, decides whether something is mispriced or about to move, and acts before the opportunity disappears. Of those three steps, observation comes first, and a decision cannot be better than the data behind it.
Forex adds a complication. According to the Bank for International Settlements Triennial Survey, global FX turnover averaged around $9.6 trillion a day in April 2025. That liquidity does not sit in one place. It is spread across interdealer platforms such as EBS and LSEG FX Matching, ECNs such as Cboe FX, Currenex and LMAX Exchange, bank single-dealer streams, and the CME’s currency futures. Each venue shows a slightly different picture of the market at a slightly different moment.
This is why understanding your data feeds matters. If you misread them, your system will:
- Trade on stale prices, buying from someone who already knows the price has moved.
- Misjudge liquidity, sending orders larger than the market can actually absorb.
- Produce fictional backtests, because the historical data does not match what you will see live.
Keep that last point in mind. It comes back later, usually at the point where it is most expensive.
What Data Feeds Are Necessary to Run a High Frequency Trading System Effectively?
Let us start with the core: the market data feeds that describe prices and liquidity directly. Consider these the non-negotiables. Without them you are not running HFT. You are running an elaborate guessing game.
1. Top-of-Book (Level 1) Quotes
Level 1 data gives you the best bid, best ask and their sizes at each moment. It is the minimum requirement and the fastest thing to process. Many simple latency arbitrage and market-making strategies start here. On its own, though, it hides almost everything interesting about supply and demand.
2. Market Depth (Level 2) and the Full Order Book
Depth data shows several price levels on each side of the book, not just the best one. There are two flavours worth knowing:
- Market-by-price (MBP): aggregated size at each price level. Simpler and lighter on bandwidth.
- Market-by-order (MBO): every individual order, with its own ID, addition, modification and cancellation. Heavier, but it lets you estimate your queue position and spot behaviour such as rapid quote cancellations.
Here is an example. Suppose EUR/USD shows 1.08502 bid with €5 million, but the next two levels down are thin. A depth-aware system knows a single large sell order could push the price down several pips, and it can adjust its quotes before that happens. A Level 1 system finds out afterwards.
3. Trade Prints (Executed Transactions)
Quotes show what people are willing to do. Trades show what they actually did. Trade feeds tell you the aggressor side, price and size of executions, which is essential for measuring order flow imbalance. Be warned: in FX, trade transparency varies a lot between venues, and some only publish delayed or aggregated trades.
4. Multi-Venue Aggregation
Because forex is fragmented, a serious system subscribes to several venues directly and builds its own consolidated book. Relying on one broker’s aggregated stream means relying on their timing, their filtering and their choice of liquidity providers.
Cross-Market, News and Reference Data Feeds
Next come the feeds that explain why prices are about to move. Currencies do not trade in a vacuum. Students who forget this tend to rediscover it very abruptly, usually at 13:30 London time on the first Friday of the month.
Currency Futures and Related Markets
CME currency futures, such as the 6E Euro FX contract, are published over the exchange’s MDP 3.0 multicast feed. At certain times of day, futures can lead spot prices by tiny fractions of a second. Many HFT strategies watch them as an early signal. Interest rate futures, government bond yields and equity index futures are also useful, because a sudden move in US Treasury yields often reaches USD pairs moments later.
Correlated Currency Crosses
Triangular relationships are classic HFT territory. If EUR/USD and USD/JPY move, EUR/JPY should move consistently with them. To check that relationship in real time you need fast feeds on all three legs, not just the pair you trade.
Machine-Readable News and Economic Releases
Specialist vendors deliver economic figures, such as Non-Farm Payrolls or CPI, as structured data within milliseconds of release. A human reading a headline is already far too late. Even if you choose not to trade the news, your system needs the calendar so it can widen spreads or step aside when volatility arrives.
Reference and Static Data
This is the dull but critical category:
- Minimum tick sizes and lot sizes per venue
- Trading hours, holidays and settlement (value) dates
- Instrument identifiers and symbol mappings between venues
- Fee schedules and credit limits
A mismatched symbol mapping between two venues once cost a trading desk I advised a full afternoon of very confused reconciliations. Nobody puts reference data in the brochure, but it causes a remarkable amount of trouble.
Time, Latency and the Plumbing Nobody Brags About
Having good data is one thing. Receiving it quickly, and knowing exactly when you received it, is another. This is where most of the infrastructure budget goes.
Time Synchronisation Feeds
Your servers need a precise, trustworthy clock, usually distributed via Precision Time Protocol (PTP, IEEE 1588) from a GPS-disciplined source. In Europe, MiFID II’s RTS 25 requires firms using high frequency algorithmic techniques to keep business clocks within 100 microseconds of UTC. The practical reason is just as important as the regulation. If timestamps from two venues are not comparable, you cannot tell which price came first, and your entire edge may be an artefact of clock drift.
Feed Protocols
- FIX: the industry-standard messaging protocol, maintained by the FIX Trading Community. It is widely supported and readable, but text-based and comparatively slow.
- Native binary feeds: venue-specific formats that are much faster to parse. Several FX ECNs offer them alongside FIX.
- Multicast: one message is broadcast to all subscribers at once, which is common for exchange futures data.
Colocation
Major FX matching engines sit in data centres such as Equinix LD4 near Slough, NY4 in Secaucus and TY3 in Tokyo. Light in fibre travels at roughly 200 kilometres per millisecond, so a London-to-New York round trip costs tens of milliseconds whatever your code does. HFT firms place servers in the same building as the matching engine and receive the feed over short cross-connects. Physics, unlike your broker, does not negotiate.
The Feeds That Watch You: Drop Copy, Risk and History
So far we have covered data about the market. An effective system also needs data about itself. Students often skip this, and accountants generally do not forgive it.
Execution Reports and Drop Copy
Every order you send produces acknowledgements, fills, partial fills and rejections. A drop copy is an independent, real-time copy of your executions sent to a separate system, usually your risk monitor. If your trading engine crashes or miscounts its position, the drop copy is the independent check that something has gone wrong.
Real-Time Risk and Position Data
Your system must know its exposure, P&L and credit usage at every moment. Pre-trade risk checks, such as maximum order size or maximum position, depend on accurate, low-latency position data. Knight Capital’s 2012 loss of around $440 million in roughly 45 minutes, in US equities, is the textbook example of what happens when automated systems run without effective controls.
Historical Tick Data
Finally, you need clean, timestamped historical data, ideally captured from the same feeds you trade live. Backtesting an HFT strategy on one-minute bars is like assessing a sprinter from a photograph taken once an hour. You need tick-level, and ideally order-book-level, history to simulate queue position, latency and partial fills realistically.
Quick summary of the feed stack:
- Level 1 quotes from every venue you trade
- Order book depth (MBP or MBO)
- Trade prints
- Futures and correlated market data
- Machine-readable news and economic calendar
- Reference and static data
- Precision time synchronisation
- Execution reports, drop copy and risk data
- Historical tick and depth data for research
Honest Warnings Before You Spend a Penny
Now for the part the sales brochures leave out. I want you to go into this with your eyes open, not with your wallet open.
- Retail feeds are not HFT feeds. The price stream from a typical retail MT4 or MT5 broker is filtered, aggregated and often throttled. It is perfectly adequate for swing trading and entirely unsuitable for true HFT.
- Last look distorts what you see. Many FX liquidity providers can reject your order after it reaches them. A quote you can see is not always a quote you can trade. Measure your fill ratios, not just the prices on screen.
- Costs grow quickly. Direct venue feeds, cross-connects, colocation racks and data licences can easily run to thousands of dollars per month before you place a single trade.
- Garbage in, garbage out. Feeds produce gaps, duplicates and out-of-sequence messages. Your feed handler must detect them and recover, not quietly carry on.
- Backtest-to-live mismatch. If your historical data came from vendor A and you trade on vendor B’s feed, your backtest is describing a different market from the one you trade.
Questions to Test Your Understanding
Question: Why might a strategy that is profitable on Level 1 historical data lose money live?
Answer: Level 1 history ignores queue position and depth, so the backtest assumes fills you would never receive in practice.
Question: Why does a strategy trading only EUR/USD need USD/JPY data?
Answer: Cross-pair relationships and shared-dollar moves often signal EUR/USD changes before they appear in the pair itself.
Your Next Steps
- Read the FIX protocol basics and build a simple message parser.
- Download free sample tick data and reconstruct a simple order book from it.
- Measure the latency between your own machine and your broker’s server with timestamped logs.
- Only then compare professional data vendors, armed with specific questions about timestamps, depth and gap recovery.
Frequently Asked Questions
Can I run high frequency trading on a retail broker’s data feed?
Not in any meaningful sense. Retail feeds are usually aggregated and throttled, and too far from the matching engine. They suit automated strategies on slower timeframes, not true HFT competing on microseconds.
What is the difference between Level 1 and Level 2 data?
Level 1 shows only the best bid and ask with their sizes. Level 2 shows multiple price levels of the order book, which reveals liquidity depth and potential price impact.
Why is time synchronisation so important in HFT?
Without precise, synchronised clocks you cannot reliably order events across venues. That makes it impossible to measure latency, trust your backtests or meet regulatory timestamping rules such as MiFID II RTS 25.
Is the FIX protocol fast enough for high frequency trading?
FIX is fast enough for many algorithmic strategies, but the most latency-sensitive firms prefer native binary or multicast feeds because they are quicker to transmit and parse.
How much historical data do I need for HFT research?
Quality matters more than quantity. Several months of clean tick and depth data from your actual trading venues will teach you more than years of minute bars from an unrelated source.
Conclusion
So, what data feeds are necessary to run a high frequency trading system effectively? At the core sit Level 1 quotes, order book depth and trade prints from every venue you trade. Around them you add futures and correlated markets, machine-readable news, reference data, precise time synchronisation, execution and drop copy feeds, and clean historical tick data. Each one covers a blind spot, and in HFT a blind spot is where losses build up.
Remember the blindfolded Formula One driver. Speed is the visible part, but data is what makes it safe to use. Start small: parse a FIX message, rebuild an order book, and measure your own latency. Learn the plumbing before you buy the car, and you will understand this market far better than most people who try to trade it.