FAQ
Questions, answered
From getting a backtest right to buying a licence. Each question has its own page, with the short answer first and the detail after it.
Getting systematic research right
Why don't my backtest results match live trading?Usually because the backtest assumed cheaper costs and easier fills than the market gives, used information that was not available at the time, or tested a slightly different strategy from the one that went live. Each cause can be found and closed before real money is at stake.How do I know if my backtest is overfit?You cannot prove a backtest is not overfit, but you can test for the signs. An overfit result holds at one parameter set and collapses at its neighbours, fails on data it was not tuned on, rests on a handful of trades, or disappears once realistic costs are applied. Check each one, on a tool that gives the same answer every run.Why do I get different backtest results each time I run it?Because something in the run is not fixed. Usually it is unseeded randomness, data that changed between runs, an order of operations that is not guaranteed, or a library or setting that drifted. Run the same backtest twice, compare the trade logs, and find the first trade that differs. A seeded engine gives byte-identical output, which leaves your own code and data as the only things to check.How realistic do slippage and spread need to be in a backtest?Realistic enough that the result survives them. Set spread, slippage and commission per instrument from what your broker actually charges, let spread and slippage widen when the market moves fast, and then rerun at one and a half and two times those costs. A strategy whose edge disappears under slightly higher costs was never a reliable edge.How do I manage many research scripts without results drifting apart?Separate what changes between studies from what must not. Each study should contain only its strategy logic, while fills, costs, data handling and metrics live in one shared, versioned engine that every study calls. Record the engine version, configuration, seed and data with every result, so any two results can be compared like for like.Can one laptop run serious quant research and live trading?For mid-frequency strategies on bars, yes. Research on a 2017 four-core laptop can process hundreds of thousands of bars per second, and a live order pipeline on the same laptop can keep latency in the tens of microseconds with one strategy per core. The limits are cores, uptime and the network, not raw speed. Keep live strategies on their own cores and plan for the machine to restart.What does professional-grade trading infrastructure actually include?Less hardware than people expect and more discipline. Research that gives the same result every run, fill and cost rules written down, risk checks kept separate from strategy code, one ordered path for every order, a record of every decision, live monitoring, safe restarts, and performance claims backed by published data. None of it requires a large team or a data centre.
Buy a licence30-day trial or annual licence. Key and kit by email on payment.