OathraForge Developers

Statistics

  • Calmar, last 36 monthsGrowth a year over the last 36 months of the traded window, divided by the worst drawdown inside those months.
  • Cash over the same periodWhat the same money would have earned in cash at the run's risk-free rate.
  • Chain verdictHow far the combination got and what ended it. A combination whose passers were never graded is never filed as if it had been.
  • Compound annual growthThe yearly growth rate that takes the first equity to the last over the traded window.
  • Cost multiple at break-evenHow many times the costs this run paid it could take before the average trade makes nothing.
  • Costs by kindWhat the run paid in each of the four costs, and how much of the gross profit that was.
  • Deflated SharpeThe chance the Sharpe is real after allowing for how many settings were tried. A raw Sharpe does not know that.
  • Deflated SharpeWhether the Sharpe survives having been picked as the best of many.
  • Deposit load, by share of barsHow the deposit load was spread: the share of time spent in each band, the last band being 90 and above.
  • Down captureHow much of the instrument's bad days the strategy took.
  • Gate 4, costKills a setting whose costs eat too much of what its trades make, or that dies when the spread widens.
  • Gate 5, sampleKills a setting that history could not produce enough evidence about.
  • Gate 6, bounded tailFlags a hidden tail for review. It never kills a setting: the person owns this verdict.
  • Gate 7, parameter budgetScores how many knobs the search could turn: every one multiplies the ways the optimiser can fool itself.
  • Geometric holding period returnThe one factor that, repeated for every trade, carries the starting balance to the final one.
  • Give-back, meanHow much of the best price the trade had it handed back before closing, in money.
  • Give-back, totalHow much of the best price the trade had it handed back before closing, in money.
  • Give-back, worstHow much of the best price the trade had it handed back before closing, in money.
  • Gross lossSum of the trades that lost money, after all costs.
  • Gross profitSum of the trades that made money, after all costs.
  • In-sample profit factorThe same ratio on the data the search chose this setting from. On its own it says little: every setting in a sweep looks good in sample.
  • Information ratioThe yearly return above the instrument for each unit of how far the strategy strayed from it.
  • Interval on a statisticThe range a statistic could plausibly have had on other draws of the same trades, and whether it includes the value that means no edge.
  • Is the average trade above zeroWhether the average trade is more than luck. The t figure assumes independent trades, and says so.
  • K-fold pass rateHow many of the consecutive folds of the search period the setting passed.
  • K-fold pass rateWhether the setting holds up on most of the consecutive pieces of the search period.
  • MAR, full windowGrowth a year over the whole traded window, divided by its worst drawdown.
  • Margin in use, mostThe most margin the run held at once, with the bar it happened on.
  • Margin level, lowestHow close the account came to a margin call: the lowest level reached, with the bar it happened on.
  • Maximum deposit loadHow much of the account the open positions tied up at the most.
  • Maximum drawdownThe worst fall from a peak to a later low on the out-of-sample equity. Negative; -1 means the account blew up.
  • Mean RWhat the average trade made, in units of the risk it was sized to take.
  • Minimum margin levelHow close the account came to a margin call: the lowest equity as a percent of the margin held.
  • Monthly gridHow the account moved month by month.
  • Most positions open at onceThe most legs open together, with the first bar it happened on and the legs then open.
  • Most risk open at onceThe most the open legs were sized to lose at once, in money and as a share of the equity then.
  • Most risk open at once, with correlationThe most risk open at once once instruments that move together are counted as one.
  • Payoff ratioAverage win divided by average loss.
  • Payoff ratio, out of sampleAverage win divided by average loss on the out-of-sample trades.
  • PlateauWhether the settings next to this one work too: a setting that works alone is fitting noise.
  • Plateau scoreHow many of the settings next to this one also work. A setting that works alone is fitting noise.
  • PositionsHow many positions the fills add up to. An add to a position is not a new one.
  • Positions open, averageHow many legs were open on an average bar of the test.
  • Probability of backtest overfittingHow often the best setting on one half of the history lands below the middle on the other half. Above 0.5, the search is choosing noise.
  • Probability of backtest overfittingA low figure is good: it says the winner of the search is not just the luckiest of many.
  • Profit factorTotal won divided by total lost, after commission, spread and slippage.
  • Profit factor of a periodThe profit factor of the trades that closed in one week, month, year or weekday.
  • Share of trades without a stopThe share of the trades that had no stop, so no R.
  • Sharpe of a held curveThe Sharpe of the strategy and of the same money held in each instrument, on the same days.
  • Sharpe, dailyThe return per unit of day-to-day swing, scaled to a year.
  • Sharpe, per tradeMean return of a closing fill on the balance it met, divided by the standard deviation of those returns. Not scaled to a year. On one run it can differ from the daily Sharpe by a factor of two.
  • Short trades wonThe share of short trades that made money.
  • SkewWhether the big results lean to the upside (positive) or the downside (negative).
  • Skew of the trade resultsWhether the big results lean to the upside (positive) or the downside (negative).
  • Sortino, dailyThe return per unit of downside swing, scaled to a year.
  • Sortino, per tradeThe return per unit of downside swing, per trade. Not scaled to a year.
  • Span in daysHow long the traded window is, in calendar days.
  • SQNTharp's System Quality Number: the square root of the number of trades with an R, capped at 100, times the mean R, divided by the standard deviation of R. 2.0 to 2.5 is average, 2.5 to 3.0 good, 3.0 to 5.0 excellent, above 5.0 superb. A figure without the cap is not called SQN.
  • Tail ratioHow big the best days run against the worst, over the same series: above 1, the good days are bigger.
  • Tracking errorHow far the strategy's daily returns strayed from the instrument's, scaled to a year.
  • TradesHow many trades the out-of-sample test made.
  • TradesHow many closing fills the run made. A position closed in pieces counts once per piece.
  • Trades by directionHow many trades were long and how many were short.
  • Trades in the best winning runHow many trades made up the winning run that earned the most.
  • Trades in the worst losing runHow many trades made up the losing run that cost the most.
  • Trades past minus one RTrades whose loss went past the risk they were sized to take: a stop that let more through.
  • Trades with an R of exactly zeroTrades that made or lost exactly nothing, whose planned risk cannot be worked back from their R.
  • Trades without a stopA trade with no stop has no stop distance, so it has no R. These trades are counted here and left out of every R figure.
  • Trading daysThe days the data traded on, which the yearly constant is measured from.
  • Trading days a yearThe trading days in a year this run's data contains, which every annualised figure uses.
  • Treynor ratioThe strategy's yearly excess return for each unit of beta.
  • Up captureHow much of the instrument's good days the strategy kept.
  • Walk-forward pass rateHow many of the consecutive later windows inside the search period the setting passed.
  • Walk-forward pass rateWhether the setting keeps working in the period straight after the one it was fitted on, again and again.
  • Weekly gridHow the account moved week by week.
  • What a trade isEvery count and average here counts closing fills, so a position closed in pieces counts once per piece.
  • Win rateShare of trades that made money.
  • Win rate, out of sampleShare of the out-of-sample trades that made money.
  • worst drawdown in money, barThe worst fall from a peak to a later low in money, on the equity curve this figure names. Negative; zero when the curve never fell.
  • worst drawdown in money, dailyThe worst fall from a peak to a later low in money, on the equity curve this figure names. Negative; zero when the curve never fell.
  • worst drawdown in money, intrabarThe worst fall from a peak to a later low in money, on the equity curve this figure names. Negative; zero when the curve never fell.
  • worst drawdown, barThe worst fall from a peak to a later low, as a share of the peak, on the equity curve this figure names. Negative; zero when the curve never fell.
  • worst drawdown, dailyThe worst fall from a peak to a later low, as a share of the peak, on the equity curve this figure names. Negative; zero when the curve never fell.
  • worst drawdown, intrabarThe worst fall from a peak to a later low, as a share of the peak, on the equity curve this figure names. Negative; zero when the curve never fell.
  • Worst floating lossThe deepest unrealised loss the open positions reached, with the first bar it happened on.
  • Worst losing runWhat the worst losing run cost in total, as a negative number.
  • Worst loss, share of equityThe largest loss on one trade as a share of the account equity when that trade opened.
  • Worst RThe worst trade in units of its planned risk. Below -1, a stop let more through than planned.
  • Z-score of the runsWhether wins and losses come in streaks: negative when they clump, positive when they alternate.