FAQ · Finding a tool for a specific need

What's a good commercial research engine for mid-frequency strategies on OHLCV bars?

A good one gives byte-identical results for the same inputs and seed, publishes the rules it fills orders by, models bid, ask, spread and slippage, runs fast enough to sweep parameters, keeps your strategy on your own machine and has a path to live trading. Reamer Research is built for this case. It costs $20 per seat per month, cancel any time.

A good one gives byte-identical results for the same inputs and seed, publishes the rules it fills orders by, and models bid, ask, spread and slippage the way your broker charges them. It also runs fast enough to sweep parameters, keeps your strategy on your own machine and has a path to live trading. Reamer Research is a commercial Python and C++ engine built for exactly this case: systematic strategies on time, tick, volume, dollar or custom bars, held from minutes to days.

The list matters more than any one name. Below is what to check in any engine, the questions that separate a serious one from a script with a price on it, and how Reamer Research answers each.

What mid-frequency strategies on bars need

Mid-frequency means positions held from minutes to days, decided on bars rather than on every quote. That puts the weight on a few things:

  • Repeatable results. The same data, settings and seed should give the same output every run. Otherwise a change in results could be the strategy or the tool, and you cannot tell which. See What makes a backtest deterministic, and why does it matter?
  • Fills you can read. The rules for when an order fills and at what price should be written down: market, limit and stop orders, stops and targets on the same position, gaps, and order lifetimes. A result is only as trustworthy as those rules.
  • Costs that match the broker. Bid and ask rather than one price, spread and slippage that can vary, commission, overnight financing, margin. At mid-frequency, costs decide whether a marginal edge survives. See How realistic do slippage and spread need to be in a backtest?
  • What happens inside a bar. Stops, targets and limits often trigger within a bar, so the engine needs a stated way to decide what was reached first.
  • Portfolios. Several instruments in one run, with shared equity and margin, and costs set for each one.
  • Speed for sweeps. One backtest is never the test. Checking that a result holds across nearby parameters means hundreds or thousands of runs. See How do I know if my backtest is overfit?
  • The full record. Every order, fill and closed trade, not just a final number, so any result can be traced.
  • A route to live. The research has to lead somewhere. See What software do independent quants use to go from research to live trading?

Questions to ask any commercial engine

  1. Can I read the fill rules before trusting a result? If not, every number it produces is a claim you cannot check.
  2. If I run the same test twice, do I get the same bytes? Run it and compare the output files.
  3. Can I reproduce the published speed on my own hardware? A benchmark you cannot rerun is marketing.
  4. What leaves my machine? Strategy code and data are your edge. Know exactly what the tool sends, and where.
  5. Which languages can drive it? It should fit the code you already write, not ask you to rewrite it.
  6. What is it built for? The kinds of strategy, data and platforms it serves, stated up front, so you can check the fit in minutes.
  7. What does it cost each year, per person, and can I try it on my own strategy first?

How Reamer Research answers them

Reamer Research is a deterministic backtesting engine you call from Python or C++, built on a stable C ABI that any other language can wrap. Your program hands it bars, settings and a strategy, and gets back the full result. Everything below is documented in the kit:

  • Byte-identical runs. A fixed rng_seed gives byte-identical output, including the randomised spread and slippage. In the published test, 20 of 20 runs matched by SHA-256.
  • A written execution specification. Fill prices, bid and ask, spread and slippage, commission, margin, order lifetimes (GTC, IOC, GTD), gaps, stop-and-target collisions, overnight swap and futures rolls are each specified, and the engine is tested against that document. It ships with the kit.
  • Inside the bar. A deterministic synthetic tick path runs through every bar. Orders fill at the first tick that meets their condition, and the earlier of a stop and a target wins.
  • Portfolios with per-instrument costs. Many instruments in one run with shared equity and margin. Commission, spread, slippage, swap and bid or ask pricing can be set for each one. Non-price data, such as earnings dates, can be attached to an instrument and is only visible from its own timestamp onward.
  • Fast enough to sweep. On an AMD EPYC 9575F and release 4.6.0, one backtest ran at about 1.83 million bars a second, and its speed varied by only 0.66% (coefficient of variation) across 2,000 runs. A 50-million-bar test, 10,000 tickers over 20 years with 990,000 stop-and-target orders, ran in 15.8 seconds. Each backtest is single-threaded, and separate backtests are safe to run at the same time from your own threads, so a sweep uses every core. A benchmark tool in the kit reruns the measurement on your own machine.
  • Python, with a cost. The Python package installs with pip, takes CSV, pandas or numpy bars, and ships with 17 example strategies and scripts. It is far slower per bar than C++: on a 2017 laptop and release 4.1.2, an 884,130-bar backtest took about 65 seconds from Python.
  • Local. After activation it runs without a connection. Only the licence key, fingerprint and seat id go to the licence server, at activation, deactivation and each monthly renewal; strategies, data and results stay on your machine.
  • The full record. 31 summary metrics, including Sharpe, Sortino, Calmar and drawdown, plus every order, fill and closed trade, written as a versioned JSON report.
  • Full state on every bar. The strategy sees its position in each instrument and every resting order, and can re-price an order, move a position's stop to breakeven or into profit, close everything or cancel everything. Each action reports whether it was applied and why not.
  • A route to live. reamer_relay_* sends the orders your strategy returns to your broker through the open-source receiver over the published protocol, and reads the fills back as position state. The decision logic carries over; indicators move from a bar window to running state.

The method and raw data behind the speed and repeatability figures are public: see Reamer Research findings on EPYC.

Scope

  • Bars, minutes to days. Time, tick, volume, dollar or custom bars, intraday to multi-day.
  • Equities, futures, FX and crypto on bars.
  • Code first. You write the strategy and the program that runs it.
  • Linux x86-64, macOS on Apple Silicon, and Windows through WSL2.
  • Data from any source, or from your Alpaca or Databento account through the bundled fetchers, with equities already adjusted for splits and dividends.

Price

$20 per seat per month, billed monthly, cancel any time, and the licence key arrives by email within minutes. Before paying, the free download runs in Evaluation Mode: one ticker per backtest over at most 30 days of bar time or 50,000 bars. A month is the way to run your own strategy on all of your data before deciding. See pricing.

Next questions

What is Reamer Research?Reamer Research is a deterministic backtesting and live trading engine for Python and C++, with a visual trade diagnostics tool; its relay sends live orders to your broker. You install it with pip or include one C++ header, on your own machine. It is built on a stable C ABI, so any other language that can call C can wrap it, live trading included. It tests a strategy, diagnoses it trade by trade, sweeps it, reports it, and sends its orders live, to your broker through the open-source receiver over the published protocol. Every order fills tick by tick on a seeded tick path the engine generates inside each bar, from bar data alone. Its visual trade diagnostics tool charts every closed trade on its own bars beside every value your strategy printed. The same data, settings and seed give byte-identical results every run. It costs $20 per seat per month, cancel any time.Who are Reamer Research and Reamer Server for?Systematic quants who write their strategies in code and trade on bars, held from minutes to days, whether independent or at a firm. Both products are built for that job: bars of any type, from Python, C++ or any language that can call C, on your own machine, with the data vendor and broker you choose.Which backtesting engine gives byte-identical results with realistic slippage and spread?One whose randomness is seeded and tied to each position in the data, not drawn from a running stream. Then spread and slippage can vary with the market and still repeat exactly for a fixed seed. Reamer Research works this way. In its published test, 20 of 20 runs with spread and slippage noise switched on were byte-identical by SHA-256.How should a backtest fill a stop and a target that both fall inside one bar?Reamer Research settles it on a seeded tick path it generates inside the bar: whichever level the path touches first is the exit, the same way on every run. The bar alone cannot say which was hit first, so a backtest should settle it by one stated rule, apply that rule the same way on every run, and let you check how much the result depends on it. A strategy whose result changes when the rule changes depends on something the data does not contain.Can I backtest futures with roll handling, or FX with swap costs?Yes, but only if the backtest handles two things a plain price series hides. In a continuous futures series, the price jump at each contract roll must not be counted as profit or loss. An FX position held overnight must be charged or credited swap every night it is held. A backtest that skips either can report a result the account would never have produced.

Every answer here restates the documents that ship in the kit, which are authoritative. All questions.

Buy a licenceKey and kit by email on payment.