HFT vs mid-frequency vs low-frequency: how trading horizon shapes the job

Two quant researchers can share a job title, a degree, and a salary band and still do almost unrelated work. One spends the week shaving microseconds off the path from a market data packet to an order. The other spends it arguing about whether a signal that holds positions for three days survives transaction costs. The difference is the trading horizon, and it shapes the technology, the research process, the risk profile, and the skills a firm screens for. This guide maps the three horizons using regulator documents, academic studies, and the firms' own job postings, with every claim labeled by source type.

Last reviewed 2026-10-04. Job postings change constantly; the ones quoted here were live on the firms' boards on that date. Horizon labels are loose industry convention with no legal definition, and firms apply them inconsistently.

The three horizons, roughly defined

There is no official boundary between "high" and "mid" frequency. The most useful anchors come from how regulators and firms describe the work:

For HFT, the closest thing to a definition is the SEC's 2010 Concept Release on Equity Market Structure, as summarized in a 2014 SEC staff literature review. It lists five characteristics "often" attributed to HFT: extraordinarily high speed order generation and routing, co-location and direct data feeds, "very short time-frames for establishing and liquidating positions", many orders cancelled shortly after submission, and ending the day "in as close to a flat position as possible". The SEC staff stressed that not every HFT firm shows all five. (Official, SEC, 2014.)

High frequency: a race measured in millionths of a second

One of the most detailed public measurements of HFT competition is a study by Matteo Aquilina, Eric Budish and Peter O'Neill using exchange message data for FTSE 100 stocks (published as NBER Working Paper 29011, July 2021). Its abstract reports that latency arbitrage races in FTSE 100 stocks happen "about one per minute per symbol", that the modal race lasts "5-10 millionths of a second", that races account for roughly 20% of trading volume, and that the top 6 firms account for over 80% of race participation. (Research, 2021.)

That is the environment the hardware investment is built for, and firms say so plainly in their own materials (all official, 2026):

• Jump Trading describes "over 100 co-located compute environments near global exchanges" and "custom ASIC, FPGA, and platform development for low-latency and compute-intensive workloads".
• Optiver's FPGA internship ads describe "hardware systems where nanoseconds matter".
• Jane Street hires hardware engineers to build FPGA and ASIC designs in Hardcaml, an OCaml library for describing hardware.
• Tower Research's job ads describe the firm as "a high-frequency proprietary trading firm founded in 1998" and its engineers working on "low-latency programming, FPGA technology, hardware acceleration and machine learning".

Hudson River Trading's engineering blog (October 2025) gives a rare look at how this splits inside one firm. Its "Trading Tech" group, focused on live trading, works in "something like 70% C++ and 30% python", and the post names its biggest technical challenge as "latency". The research and development group is "roughly 70% python and 30% C++", and its biggest challenge is "scale". (Official, HRT, 2025.)

The Sharpe ratio and the ceiling

HFT's appeal shows up in risk-adjusted returns. A CFTC-authorized study by Matthew Baron, Jonathan Brogaard and Andrei Kirilenko (April 2014 draft), covering HFT firms in E-mini S&P 500 futures from August 2010 to August 2012, found a median annualized Sharpe ratio of 4.3, with 25% of firms above 9.10. It also found revenue concentrated in a few top firms, consistent with "winner-takes-all", and that new entrants were less profitable and more likely to exit. (Research, 2014.)

A widely cited data point comes from Virtu Financial's 2014 IPO filing, which stated that the firm had "only one losing trading day" across 1,238 trading days from January 2009 to December 2013. (Official, SEC Form S-1, March 2014.)

Low-frequency strategies look very different. AQR's "A Century of Evidence on Trend-Following Investing" (Hurst, Ooi and Pedersen, 2017) built a trend strategy across 67 markets back to 1880. Per market, the average gross Sharpe ratio was "approximately 0.4". The authors stress consistency across decades and markets; the portfolio relies on many modest signals in combination, with no single strong one carrying the load. (Research, AQR, 2017.)

So why does anyone run low frequency? Capacity. HFT profit is limited by how much volume passes through the market at a fleeting edge; adding capital does little to raise it. The Aquilina, Budish and O'Neill paper puts the total annual stakes of latency arbitrage at "on the order of $5 billion per year in global equity markets alone". Competitors split that prize among themselves, and extra capital cannot enlarge it. A trend or factor strategy that trades monthly pays transaction costs rarely and can absorb far more capital. For scale, Man Group, whose systematic arms include Man AHL (founded in 1987 as a trend-following CTA), reported US$253.6bn in group funds under management at 30 June 2026. (Official, Man Group, 2026; the group figure includes non-quant strategies.)

The general shape is the trade-off every quant learns early: shorter horizons give many more independent bets per year and therefore higher Sharpe ratios, but each bet is small and the total capital the strategy can use is capped. Longer horizons scale but give fewer independent observations, lower Sharpe ratios, and deeper drawdowns. This is a general principle that applies across firms.

Mid frequency: the crowded middle

Between the two sits mid frequency, where several firms known for speed are now openly hiring. Candidates on Blind summarize the classic split this way: "HFT is mostly making their revenue from low latency market making strategies whereas quant hedge funds are using mid-frequency strats like long/short or stat arb." A reply from a user labeled Citadel adds that quant funds run a lot more than stat arb or mid-frequency strategies. (Self-reported, Blind, March 2023.)

That line is blurring, because firms known for speed now advertise mid-frequency work in their own postings (all official, live on 2026-10-04):

• HRT has a "Quantitative Researcher (Mid-Freq)" role in New York for "developing mid-frequency systematic trading strategies", covering "alpha generation, portfolio construction/optimization and trade execution algorithms". It asks for 3+ years of stat-arb experience and lists a base salary range of $175,000 to $300,000 (New York, 2026). HRT's 2027 internship ad says interns can "rotate between our high- and mid-frequency trading teams, as well as our machine learning teams".
• DRW says it is "building a new systematic, mid-frequency trading business" combining "diverse data, modern machine learning, and high-performance software systems". Its Cumberland Systematic team runs "mid and high-frequency fully systematic strategies in the delta one space".
• Jump says its research "spans every asset class and a full range of time horizons, from high frequency to strategies that hold for days and weeks".
• IMC is hiring machine learning researchers for "high to mid frequency delta one trading strategies" in Asia Pacific.
• Five Rings asks lateral researchers for experience finding "intraday signals for mid- to high-frequency trading".
• Squarepoint advertises a senior quant researcher role in equity mid/low frequency with a minimum base of $150,000 in New York, asking for "proven ability to run successful long term strategies on Equities" (2026).

The hiring follows the money. In September 2025 FT Alphaville reported that HRT houses its mid-frequency signals in a separate unit called Prism, which reportedly made more than $2bn of profit in 2024, and that mid-frequency trading had grown to about 25 to 30% of Tower Research's revenues, up from under 10% two to three years earlier, according to a person familiar with the matter. The same piece puts mid-frequency at holding periods of roughly one to five days. (Reported, September 2025.)

In MFT the binding constraint moves from latency to data and compute. XTX Markets, which describes using machine learning "to produce price forecasts for over 53,000 financial instruments", says it runs "over 25,000 GPUs in our research cluster" and "over 1 Exabyte of usable storage" (official, XTX website, 2026; XTX does not publish a horizon breakdown). eFinancialCareers reported that HRT's UK service entity spent £391m on property, plant and equipment in its 2025 financial year, about £2.4m per UK employee, and that now-expired HRT ads cited "very high GPU-to-research ratios". (Reported, eFinancialCareers, September 2026.)

How firms organize around horizon

Horizon often decides team structure too. HRT's blog says its algo teams "are very roughly organized by the time horizon at which they trade, and then by asset class and region", and that the group is not a "pod" system "where trading teams are siloed with IP barriers" (official, 2025). Tower's ads describe the opposite emphasis: the firm empowers "portfolio managers to build their teams and strategies independently" on shared infrastructure (official, 2026). Neither model is tied to one horizon, but it is worth asking which one you are joining, because it decides who owns your research.

What each horizon rewards

High frequency. Systems programming (C++ dominates live trading at HRT by its own estimate), networking, Linux, and for some roles hardware design. On the research side: market microstructure, order book data, and fast probabilistic reasoning. An eFinancialCareers piece on quant developers quoted a practitioner saying a developer "is more interested in the actual low latency implementation, getting as efficient as possible in the code from market data signal to trade signal." (Reported, 2023.) If this is your target, the quant developer prep page covers the C++ and systems side, and the market making game drills the quoting intuition behind liquidity provision.

Mid frequency. Statistics on large, noisy datasets, machine learning, Python plus enough C++ to productionize, and the unglamorous parts: transaction cost models, portfolio construction, execution. DRW's Singapore equities posting lists "models for intraday alpha, risk, transaction costs, liquidity, and portfolio construction" and asks for 4+ years of stat-arb or systematic research experience (official, 2026). Several of the mid-frequency postings above ask for years of experience; graduates more often enter through general researcher programs and internships like HRT's and Jump's.

Low frequency. Econometrics, economic intuition for why a premium should persist, and strict discipline against overfitting, since a monthly strategy produces far fewer independent observations than an intraday one. AQR's trend paper needed data back to 1880 to make its point. Jump's research internship ad puts the shared ethos well: "One excellent, fully understood result is worth more here than a dozen shallow ideas" (official, 2026).

Using this in an interview

When a firm says it trades "across time horizons", ask which horizon your team trades and what the binding constraint is: latency, data, or capacity. The answer tells you whether to expect a systems interview, a statistics interview, or both. Firm-by-firm processes are in our guides, for example Hudson River Trading and Jump Trading.

QuantReady is an independent prep platform and is not affiliated with, endorsed by, or sponsored by any firm named here. All trademarks belong to their respective owners. Nothing here describes any firm's actual strategies beyond what the firm publishes. Job ads are quoted as published and may change or close. If something is out of date, tell us via /contact.

QuantReady is an independent prep platform and is not affiliated with, endorsed by, or sponsored by any firm we write about. All company names and trademarks belong to their respective owners. Read how we source our content.