The Millisecond Manifesto: High-Frequency Trading in 2026

1. High-Frequency Trading Starts with Execution Quality

A quoted EUR/USD price can change between a trading decision and the arrival of an order. The result might be a different fill, a partial fill, or a rejection. Understanding that sequence is more useful than assuming every disappointing execution was caused by a faster competitor.

High-frequency trading (HFT) is a subset of algorithmic trading that depends on rapid market-data processing and order handling. Automated strategies on hourly or daily charts are algorithmic too, but they do not automatically belong to HFT. The relevant infrastructure, data resolution, costs, and testing methods depend on the strategy’s holding period and the market it trades.

News-driven volatility can widen spreads and change available liquidity. A useful execution review therefore separates the trading signal from the path an order follows: market-data delivery, strategy evaluation, order submission, broker or venue processing, and the eventual fill. An attractive chart entry does not establish that the same price and quantity were executable.

LuxAlgo Quant, the coding agent built into LuxAlgo charts, helps turn an idea into explicit rules and a chart-based backtest. Native charts and Orderflow tools provide visual market context. These are useful research tools, while measuring subsecond broker latency, queue position, and actual execution requires appropriate order and market-data records.

2. Broker Execution: Measure the Order Lifecycle

A broker’s execution model is relevant, but labels such as A-book, B-book, ECN, or STP are not a substitute for evidence. Internalization means some flow is handled within the dealer’s business rather than every order being passed through in the same way. Hedging, commissions, spreads, and aggregate exposure make the economics more complex than a simple claim that every client loss becomes broker profit.

Review the execution policy for the specific account and instrument. Identify the contractual counterparty, how prices are formed, which order types are available, and the rules for partial fills, rejections, and cancellations. A provider can use different arrangements across products or account types. A marketing description alone does not establish the route taken by your order.

Slippage is the difference between a reference price and the execution price. Choose the reference before comparing results: the price when the strategy made its decision is different from the price when the order reached the venue. Keep buy and sell conventions consistent, and record favorable as well as unfavorable outcomes.

Delays can arise in your connection, application, broker systems, liquidity provision, or venue processing. A slow fill does not identify which component caused it. Likewise, a fast acknowledgement confirms receipt or a status change; it does not necessarily mean the trade has filled. Preserve order identifiers and timestamps so that each stage can be reconciled.

Strategy permissions are a separate question from execution quality. Check the account’s current rules for scalping, high message rates, latency-sensitive trading, and automation before assuming a technically possible workflow is permitted. Section 5 gives a documented provider example.

For a chart-based baseline, use Quant to define entries, exits, and risk assumptions, then inspect the generated logic and trade list. Those backtests do not measure a broker’s actual order-processing delay.

3. Data Centers, Network Latency, and a Better Comparison Table

Physical distance and network routing affect communication time, but they are only part of execution. Equinix operates NY4 in Secaucus, New Jersey, and LD4 in Slough, England. These locations do not establish where a particular broker account’s server, bridge, or liquidity connection is hosted.

A VPS near a verified endpoint may reduce network round-trip time and variability, often called jitter. It cannot remove application queues, liquidity checks, or matching constraints. Test the endpoint your account actually uses and record the route, hosting location, instrument, session, and observation period. One unusually low ping is a weak basis for selecting infrastructure.

Use a measurement sheet instead of comparing unsourced broker latency figures. Keep the account, order size, instrument, and test conditions comparable; record sample counts and tail outcomes alongside averages.

Separate quiet periods from volatile sessions, and compare order types independently. A market order and a resting limit order face different tradeoffs. Preserve unusually slow responses rather than deleting them as inconvenient outliers; a strategy may be most vulnerable precisely when normal execution conditions break down.

Chart overlays and structure tools help describe a setup. They do not explain every fill. Keep the research record alongside the broker’s execution records, and use manual trade journaling to review realized outcomes without treating it as a replacement for detailed subsecond logs.

4. Last Look: Price, Rejections, and Hold Time

In electronic FX trading, “last look” gives a liquidity provider a final opportunity to accept or reject a request at its quoted price. The Global Foreign Exchange Committee’s guidance says the process should be fair and predictable and used for price and validity checks. Its disclosure framework helps participants understand and evaluate how requests are handled.

The often-cited $25-per-million figure needs context. LMAX’s FX transaction-cost study examined third-party aggregator data covering more than seven million trades sent to seven last-look and firm liquidity providers during 2016. LMAX reported a hold-time cost of $25 per million at 100 milliseconds unless the fill rate was 100%. This is a result from that study, not a universal fee or a current benchmark for every account.

LMAX also describes its own venue as using firm liquidity without a last-look hold window. That distinction does not imply that every order must fill or that network and processing latency disappear. Compare the complete execution outcome: quoted price, fill probability, partial quantities, timing, and the cost of retrying or missing an intended trade.

A rejected order’s opportunity cost depends on the strategy and the chosen comparison horizon. Buying immediately elsewhere, waiting, or abandoning the trade can produce different outcomes. Define that treatment in advance so the transaction-cost analysis reflects the workflow you actually intend to run.

5. AI in Routing, Surveillance, and Strategy Research

AI can serve different roles across a trading workflow. A coding agent translates and revises strategy logic; an execution model may estimate liquidity or inform routing decisions. One capability does not imply the other. Quod Financial’s machine-learning material, for example, describes predictive agents and adaptive routing as part of its own execution architecture.

Potential applications include:

• Routing: using liquidity and execution observations to inform where or how orders are sent.
• Surveillance: flagging unusual activity or behavior that a provider’s rules restrict.
• Reliability: detecting abnormal system behavior and supporting operational monitoring.

These are uses to evaluate, not evidence that an AI system guarantees better fills. The objective, training data, live feedback, and failure controls all matter. Historical performance alone cannot demonstrate how a model will behave during a new market disruption.

Account restrictions also vary. FundedNext’s published policy lists HFT, tick scalping, latency trading, and hyperactivity among prohibited strategies. This is a provider-specific rule; it should not be generalized into a claim that all algorithmic trading is prohibited or that all HFT is manipulative. Check the terms that apply to your own service and account.

In LuxAlgo, Quant can help create a clearly specified bar-based strategy and revise its assumptions. Review what the code actually does, run it, and examine the trades before interpreting the summary. If your intended edge depends on milliseconds, a candle-based result leaves important execution questions unanswered.

6. Arbitrage Infrastructure and Cost-Aware Testing

Arbitrage research requires comparable instruments, executable prices, consistent units, and realistic costs. A difference between two displayed quotes is not enough: the prices may refer to different times, sizes, contracts, or execution conditions. Any proposed advantage must survive fees, uncertain fills, and the risk that only one side of a multi-leg trade completes.

FIX is an open standard for financial messaging, including orders, executions, and market data. It provides a common language for communicating with supported counterparties. Access still depends on the provider’s service, permissions, and implementation; using FIX alone does not guarantee a faster route or a better price.

Build the infrastructure review around three practical questions:

• Connectivity: Does the provider support the required messages, order types, rate limits, and recovery behavior for this account?
• Hosting: Does a measured change in network location improve the complete order lifecycle under representative conditions?
• Cost and incomplete execution: Does the idea remain viable after commissions, benchmark-relative execution costs, rejected requests, and partial fills?

For execution benchmarks, VWAP summarizes volume-weighted prices over a defined period. It can be useful for evaluating certain execution objectives, but it is not a promise that a strategy could fill at that value. The SEC’s 2020 report on algorithmic trading discusses VWAP and other execution algorithms. Match the benchmark and time horizon to the trade’s purpose.

A Practical LuxAlgo Research Workflow

1. Make the idea explicit. Ask Quant to specify the chart interval, entry condition, order timing, exits, and position sizing. Resolve ambiguous instructions before testing.
2. Inspect the implementation. Review the generated code and verify selected trades against the chart. Check when information becomes available and whether the simulated order timing matches the intended rule.
3. Test cost sensitivity. Use the native strategy Inputs and Properties to review parameters, commission, slippage, and sizing. These assumptions support sensitivity testing; they do not recreate a broker’s full order book or last-look process.
4. Preserve the research baseline. Star a backtest run to retain its script, symbol, interval, inputs, and properties. Separate later revisions from the original result.
5. Compare with actual execution evidence. Record the broker’s order lifecycle and realized trades. Investigate differences between simulated and observed outcomes before attributing them to the signal or infrastructure.

Plans and Next Steps

Start with the workflow your strategy needs, then compare the current LuxAlgo plans and features. Quant allowances, chart limits, historical data, and tool access vary by plan. Keep software subscription costs separate from broker commissions, data charges, and hosting expenses when evaluating the total cost of a trading process.

Speed matters when the proposed strategy depends on speed. For other time horizons, precise rules, realistic assumptions, and repeatable review may be more relevant than a lower ping. Use LuxAlgo to develop and examine the chart-based logic, and use measured execution records to evaluate what happens between an intended trade and its actual fill.