Articles

How Metrand computes a backtest, rule by rule, and what changed in it and when, with the numbers: fills inside a bar, margin and liquidation, indicators, warm-up and data checks.

• Backtests now trade inside every bar on the exchange's finer candles

  A Metrand backtest on time bars now reads the exchange's own finer candles under them — one-minute candles under a month of hourly bars, coarser ones under longer windows — and trades stops, targets, liquidation and resting orders on them, instead of assuming the worse of a bar's two ways. On three months of Binance futures, 38% of the trades that assumption decided had in fact reached their target first.

• Backtest audit, part 4: checking the backtest itself

  The last part of the Metrand backtest audit adds two checks that run on every change to the engine — that no decision can see anything from after the bar it was taken on, and that a second simulator, written separately from the same rules, settles 4,000 random scenarios exactly as the engine does.

• How the Metrand backtest is tested

  Two checks run on every change to the Metrand backtest engine — a replay with the future replaced, which must leave every earlier decision, bar and trade unchanged, and a second simulator written from the same rules, which must settle 4,000 random scenarios exactly as the engine does — and how to run the first on your own strategy.

• Backtest audit, part 3: how the results are shown

  The third part of the Metrand backtest audit checks every figure a result shows — the ledger's totals, a basket's account, the optimizer's held-out check, the average trade, a year's return, rounding and time zones — and makes each say exactly what was measured.

• How a Metrand basket backtest adds markets into one account

  A Metrand basket replays one strategy on several markets from one account — every market a backtest of its own, added bar by bar into one balance, with the worse case inside a bar, one cross-margined maintenance requirement, and figures that say what the account as a whole did.

• How a Metrand backtest prices liquidation and the largest position

  A Metrand backtest liquidates a leveraged position on its own market's margin schedule, read from the exchange every day — the maintenance margin by position size and the leverage each size allows — refuses the positions and the leverage the exchange would refuse, and names the schedule it used.

• Backtest audit, part 2: the data a backtest is given

  The second part of the Metrand backtest audit checks what goes in — every candle, the funding series, the other markets a strategy reads, the price a maker order rests at and the market's own margin schedule — and makes the engine refuse what an exchange would never have produced, and say what its history lacks.

• What a Metrand backtest accepts as market data

  Every bar a Metrand backtest replays is checked before anything runs — prices, the high-low envelope, volume, open times in milliseconds, one timeframe in time order — and the bars an exchange's history lacks are counted in the report instead of being skipped in silence.

• Backtest audit, part 1: the simulator, the indicators and the warm-up

  We took the Metrand backtest apart and rebuilt what did not hold up — how a bar is walked, when orders fill, how liquidation and funding settle, how indicators start, what an empty value means, and how much history a strategy warms on. What changed and what it does to results.

• How a Metrand backtest fills orders inside a bar

  The rules the Metrand backtest engine uses to decide what happened inside each candle — which extreme came first, when a limit order fills, how stops, targets, liquidation and funding are settled — and why each rule leans against the strategy.

• Indicators and rules: values you can check against TA-Lib and TradingView

  How Metrand computes moving averages, oscillators and the rules built on them — the TA-Lib start of every average, what an empty value means, how all/any/not and crosses treat it, and how much history a strategy needs before its signals count.

• A backtest warms up the way a live run does

  Before its first decision, a Metrand backtest runs the strategy over the same history a live run warms on — 1,000 bars of its timeframe, or a week of candles for range bars — so indicators are ready on the window's first bar and nothing before the window is traded or counted.