High-Frequency Trading for Retail — Is It Realistic?

A reader wrote in last month to say he had paid three thousand dollars for an "HFT bot for retail traders" course, and was wondering why the bot had bled money for six straight months instead of producing the promised returns. The honest answer is short: high-frequency trading is not a strategy you can buy in a PDF or switch on inside MetaTrader. HFT is a contest about nanoseconds, a fibre cable, and a server rack physically next to an exchange — barriers no online course can move. I am writing this to save the next person three thousand dollars.

What high-frequency trading actually is

HFT is algorithmic trading where the edge is measured in microseconds. Firms such as Citadel Securities, Jane Street, XTX Markets and Virtu Financial mainly do two things. The first is market making — quoting continuous bids and offers across hundreds of instruments at once and harvesting the spread multiplied by enormous volume. The second is statistical arbitrage, capturing short-lived mispricings between related instruments before the market closes them.

The Bank for International Settlements, in its 2011 paper "High-frequency trading in the foreign exchange market", defined HFT through three properties: a very short holding period, a very high ratio of orders submitted to trades executed, and a tendency to flatten positions by the close of each session. That is not a description of "scalping" the way the retail world uses the word. It is the description of an entirely different business.

Why retail traders cannot do HFT — the infrastructural asymmetry

The real barriers to genuine HFT are physical and regulatory, not intellectual. The first is colocation: HFT firms rent rack space inside the exchange's own data centre, for example CME in Aurora, Illinois, or NYSE in Mahwah, New Jersey. An order then travels a few metres of cable instead of a thousand kilometres. Annual rent for a single rack runs into hundreds of thousands of dollars.

The second barrier is hardware. The critical decision path does not run through a general-purpose CPU but through field-programmable gate arrays that decode the exchange feed and emit orders in single-digit microseconds. The third is raw exchange data feeds and the maker-taker rebate schedules available to designated market makers, neither of which exists for a retail account. The fourth is a team of physicists, mathematicians and computer scientists whose annual salaries are a multiple of your trading deposit. The total cost of standing up a serious HFT platform climbs into tens, and at the largest firms hundreds, of millions of dollars.

Your home MetaTrader 5 over WiFi to an offshore broker has an order execution latency measured in tens of milliseconds — an eternity in the HFT world. The "HFT-ready VPS for five dollars a month" advertisement is, quite literally, marketing nonsense, because the network hop between that VPS and your broker, and then between the broker and the actual exchange, already rules you out of any microsecond-scale game.

How HFT affects you anyway, even if you never run an algorithm

This is where the irony ends and the useful content starts. In normal market conditions the presence of tier-one market makers and specialised HFT firms compresses spreads. The reason you see fractional-pip spreads on EUR/USD in the European session is that somebody is quoting both sides of the book continuously and competing for your order flow.

The problem arrives in stress events. On 15 January 2015 the Swiss National Bank removed its EUR/CHF floor at 1.20 and the market collapsed by double-digit percent in seconds. Market-maker algorithms calibrated for a normal-shaped distribution simply withdrew their quotes. Liquidity vanished, and trades that should have closed by stop-loss at 1.18 actually filled at 0.90. A milder version of the same dynamic showed up in the GBP/USD flash crash on 7 October 2016. The lesson for a retail trader is plain: in tail events the assumption that "an HFT will always quote me a price" breaks down, and your stop-loss is no longer a stop-loss but a stop-market filled at a dramatically worse level.

What you can realistically do instead of chasing HFT

A simple illustrative example shows the scale of the problem. Suppose your algorithm recognises some microstructure pattern and can predict a half-pip move with 55 percent probability. The expected profit per signal is roughly five hundredths of a pip. Citadel or Virtu make real money from such an algorithm because they execute tens of millions of orders a day, in single microseconds, with effectively zero exchange commission. With your real costs — a half-pip spread, a commission, an execution latency in milliseconds and routine slippage — the same edge is negative, regardless of how clever the idea is.

Sensible retail strategies live on an entirely different time axis. Manual scalping in a one-to-fifteen-minute window depends on technical structure, not on network latency. Medium-frequency algorithmic strategies on M5 to H1 timeframes, written in MQL5 or Python, compete on statistical edge rather than on hardware. It is a different game, one in which a single retail trader actually has a chance of finishing in the black — and I tried to describe it honestly in the piece on first steps in algorithmic trading.

"The world's financial markets had been transformed into a system of haves and have-nots, where the haves paid for the right to exploit the have-nots — and the have-nots had no idea." — Michael Lewis, Flash Boys: A Wall Street Revolt, W.W. Norton, 2014

Red flags — the "retail HFT" course as a business model

Anyone selling you a "retail HFT bot" for a few hundred or a few thousand dollars is doing one of three things. The first is selling terminology — relabelling an ordinary M1 or M5 scalper as an "HFT algorithm" and counting on you not having read the BIS definition. The second is showing a backtest curve constructed after the fact, with fitted parameters and a built-in lookahead bias. The third, in the worst version, is running a signal pyramid in which the most recent subscribers receive entries deliberately engineered so that earlier members can close their positions into them.

National regulators such as the FCA, BaFin, KNF and CySEC publish public warning lists, and many of the firms on them market themselves with the word "algorithm" or "HFT" as a badge of technological credibility. None of that means algorithmic trading is itself a scam — it means the label "HFT" is now used as a sales sticker by people who have never seen a real FPGA up close.

Your next step — how to stop wasting time on a false promise

1. Take an honest inventory of whatever HFT-branded or "algorithmic edge" product you may already own. Open a quiet text file and write down how much you paid, how much you actually earned or lost in live trading, and how many months the strategy ran. Putting the raw number on paper is uncomfortable, but it prices your lesson and helps prevent you from buying the next version of the same promise.
2. Verify any vendor against the regulator of the jurisdiction in which it claims to operate. Check the FCA register, the ESMA database, the CySEC list and, for Polish-speaking readers, the KNF public warnings page. Absence is not by itself proof of legitimacy, but a presence on a warning list ends the conversation immediately, and the chances of recovering money at that point are close to zero.
3. Pick one sensible algorithmic path and give yourself six months of real learning before committing any meaningful capital. Install MetaTrader 5 or a Python environment with pandas, code a simple momentum or mean-reversion strategy on historical data, push it through walk-forward validation, and only then think about live deployment. Skipping this stage means buying a product you do not yet have the vocabulary to evaluate.
4. Accept once and for all that your edge as a retail trader is cognitive and organisational, not hardware-based. A better risk plan, the discipline of a real trading journal, choosing a timeframe where hardware no longer matters (M30 and higher), and a deliberate choice of broker execution model all create measurable advantages after a year. Microseconds will not, and never will.