High-frequency trading vs retail algo trading: what differs
High-frequency trading earns tiny amounts per trade from speed and queue position, a race retail traders cannot win. Retail algo traders do better testing strategies on minute to daily bars, where each trade's edge can be large enough to pay costs.
By Soren Huang · Published · updated · Editorial policy
Contents
- What is high-frequency algo trading?
- How does a high-frequency trading algorithm make money?
- Can a retail trader compete with HFT on speed?
- What can a bar-based backtest tell you about fills?
- How much do slippage and costs limit short-horizon strategies?
- Which bar lengths make sense for a retail algo trader?
- Frequently asked questions
- References
What is high-frequency algo trading?
High-frequency algo trading is automated trading where firms earn tiny amounts per trade by reacting in microseconds, usually as market makers with servers next to the exchange. A retail futures trader cannot win that speed race. The realistic path is slower algorithmic strategies on minute to daily bars, where each trade's edge can be large enough to pay costs.
High-frequency trading (HFT) is a form of algorithmic trading in which computers send, change and cancel orders in fractions of a millisecond and usually hold positions for seconds or less. Algorithmic trading is the wider term. It covers any trading where fixed rules decide entries and exits and a computer executes them. All HFT is algorithmic trading, but most algorithmic trading is not high frequency.
| Aspect | HFT firm | Retail algo trader |
|---|---|---|
| Typical holding time | Seconds or less | Minutes to weeks |
| Source of the edge | Being first, earning the spread, small price differences between markets | Price behaviour over many bars: trends, reversals, session patterns |
| What decides the result | Speed and place in the order queue | Whether the average trade is larger than its costs |
| Infrastructure | Servers in the exchange data centre, direct market data feeds | A computer, a broker connection, a trading platform |
| Data needed to test | Every order book message | Bar data, from one minute to one day |
| How it is tested | Simulation of the order book and the queue | Bar-based backtest with costs, then out-of-sample checks |
How does a high-frequency trading algorithm make money?
Most HFT profit comes from many very small gains, and each one depends on being faster and earlier in line than other traders.
Latency is the time between an event, such as a price change, and the moment your reaction reaches the exchange. HFT firms measure latency in microseconds, which are millionths of a second.
Co-location is renting space for your servers inside or next to the exchange's data centre, so the cable to the matching engine is as short as possible. The matching engine is the exchange computer that pairs buy and sell orders.
Market making is quoting a buy price and a sell price at the same time and earning the difference between them, the spread, when both sides trade. In liquid index futures the spread is often a single tick, the smallest price step the contract allows. The futures contract calculator shows what one tick is worth in your contract.
Queue position is your place in the line of orders that wait at the same price. Many futures markets fill orders at one price in the order they arrived. This rule is called price-time priority. An order at the front of the line gets filled first.
A market maker's main risk is getting filled just before the price moves against the quote. Speed solves this in two ways. A fast firm reaches the front of the queue, so its orders fill on normal trading flow. A fast firm also cancels old quotes before slower traders can trade against them. For more on how such firms are set up, see how quantitative trading firms work and what retail traders can copy.
Can a retail trader compete with HFT on speed?
No. A retail order passes through several delays that a co-located firm has already removed. Picture people buying the last tickets for a concert. Some stand at the ticket window. You send your request by post. Your request may be well written, but the tickets are gone before it arrives.
- The market data you see on screen has already travelled from the exchange to your broker or data vendor, and then to you.
- Your platform needs time to calculate the signal and create the order.
- The order travels over the internet to the broker.
- The broker runs risk checks and sends the order on to the exchange.
A rented server near the exchange makes some of these steps shorter. It does not move you to the front of the queue, and it does not let you cancel quotes as fast as a firm that works next to the matching engine. Any strategy that profits from reacting to a price change before others react will be taken by traders who are already faster.
Retail traders have other advantages. You can hold a trade for hours or days, so a delay of milliseconds does not matter. With small size, a market order in a liquid contract usually fills close to the quoted price. You also do not need to trade all day to cover the cost of servers and staff.
What can a bar-based backtest tell you about fills?
A bar-based backtest models fills well when the strategy decides on completed bars and trades on the next bar. It cannot model queue position or the order of price moves inside a bar. A bar-based backtest is a simulation that only knows the open, high, low and close of each bar, and often the volume. It does not see the individual orders that traded inside that bar.
| Situation | What bar data can model | What bar data cannot model |
|---|---|---|
| Market order after a signal at the bar close | A fill at the next bar's open plus a fixed slippage | The actual spread at that moment |
| Limit order at a price | Whether price reached that level | Whether your order was far enough forward in the queue to be filled |
| Stop and target inside the same bar | That both levels were touched | Which level was touched first |
| Large order | A fixed cost per contract | How far your own order moves the price |
| Holds of a few seconds | Little of use | The effect of latency on every fill |
The limit order row matters most. Think of a queue at a bakery counter. The shop sells bread at your price, but it runs out before your turn. A bar backtest only sees that bread was sold at your price, so it records that you got some. In real trading, price often touches your limit, fills the orders ahead of yours and then turns away.
Use these rules to keep a bar-based backtest close to what you could actually trade:
- Decide on the close of a bar and execute on the next bar. This is called next-bar execution.
- Count a limit order as filled only when price trades through the limit by at least one tick.
- When the stop and the target both fall inside one bar, assume the stop filled first.
- Choose a bar length at which a typical trade lasts many bars, so moves inside a single bar decide few results.
How much do slippage and costs limit short-horizon strategies?
The shorter the holding time, the smaller the average move per trade, while the cost per trade stays the same. Costs therefore take a larger share of a short-horizon strategy's result. Slippage is the difference between the price your backtest assumed and the price you actually get.
The check needs one subtraction:
net average trade = gross average trade − round-trip cost
The gross average trade is the average result per trade before costs. The round-trip cost is commission and exchange fees for the entry and the exit, plus slippage on both sides. Express both values in the same unit, for example ticks. The trade list in the performance report of AlphaPrime (Valynox) gives the result of each trade, so you can set the average trade against the round-trip cost in the same ticks.
Hypothetical example
Two strategies trade the same contract. The round-trip cost is 2.5 ticks: commission and fees equal to 0.5 tick, plus 1 tick of slippage on each side. Strategy A uses 1-minute bars and has a gross average trade of 3 ticks. Strategy B uses 60-minute bars and has a gross average trade of 20 ticks. After costs, A keeps 0.5 tick per trade, so costs take more than 80% of its gross result. B keeps 17.5 ticks, so costs take 12.5%. Now double the slippage to 2 ticks per side. The round-trip cost becomes 4.5 ticks. A now loses 1.5 ticks per trade. B still keeps 15.5 ticks.
Tick sizes, tick values and exchange fees are set by the exchange, and they can change. The exchange's contract page is the final word on these values. Check it and your broker's fee schedule before you set costs in a backtest. For a full method of modelling commission and slippage, see trading costs in backtests and the average trade test.
Which bar lengths make sense for a retail algo trader?
Bars from one minute to one day give a practical range for a retail futures trader. The longer end of this range leaves more room between the average move and the cost. For each bar length, check the item in the last column before you trust the result.
| Bar length | Typical holding time | Cost as a share of the average trade | What to check |
|---|---|---|---|
| 1 minute | Minutes | High | Fills inside bars, limit order assumptions, results with double slippage |
| 5 to 15 minutes | Under an hour to one session | Moderate | Session open and close, news times |
| 30 to 60 minutes | Hours to days | Lower | Overnight gaps, enough trades per test window |
| Daily | Days to weeks | Low | Fewer trades, so a longer history is needed |
Short bars produce many trades, and many trades look like strong evidence. They also give a search more noise to fit. If you test many rule variants and keep the best one, its backtest is usually better than its future results (Bailey, Borwein, López de Prado and Zhu, 2014). Judge a strategy on data it did not see during the search. Walk-forward analysis does this by tuning on one window and testing on the next one. AlphaPrime runs these checks in its strategy validation with Basic, Standard and Extensive levels.
Put realistic costs into the search from the first run. AlphaPrime's backtesting evaluates each candidate with fees, slippage and next-bar execution, within the trading sessions you choose. As a result, the ranking already reflects the fills a bar-based test can model honestly.
What this means if you use AlphaPrime: search on bar lengths at which a typical trade lasts many bars, set costs before the first run (next-bar execution is a fixed rule), and judge candidates on their out-of-sample results. The speed race belongs to HFT firms. Your research time goes further on strategies whose edge can survive costs.
Frequently asked questions
Can retail traders do high-frequency trading?
Retail traders can trade often, but they cannot do high-frequency trading in the professional sense. HFT depends on co-located servers, direct data feeds and a place at the front of the order queue. A retail order passes through the internet and a broker first, so faster firms take the opportunities before it arrives.
What is the difference between algorithmic trading and high-frequency trading?
Algorithmic trading is any trading where fixed rules decide entries and exits and a computer executes them. High-frequency trading is one type of algorithmic trading that holds positions for seconds or less and earns its edge from speed. A strategy on 15-minute or daily bars is algorithmic trading, but it is not high frequency.
Is scalping on 1-minute bars the same as HFT?
No. Scalping on 1-minute bars holds trades for minutes and does not depend on microsecond reactions. It shares one problem with HFT, though: the average move per trade is small, so commission and slippage take a large share of it. Test it with realistic costs and with double slippage before you trust the result.
Can you backtest an HFT strategy with minute bars?
No. Minute bars do not show the order queue, the spread at each moment or the order of price moves inside the bar. An HFT strategy needs a simulation built on every order book message. Minute bars suit strategies that decide on completed bars and trade on the next bar.
References
- Bailey, D. H., Borwein, J., López de Prado, M., & Zhu, Q. J. (2014). Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance. Notices of the American Mathematical Society, 61(5), 458–471.