Jordon Harrell · Quanticator

Verifying trading scripts and claimed backtests — mine included. Author, Trading as a System. Free: the 10 checks I run before I believe a backtest ↓

[Jordon Harrell · Quanticator @Quanticator]

Most backtests fail on arithmetic, not strategy.

60% wins over 25 trades: the 95% interval is roughly 41–77%. A profit factor of 2.0 at 35% wins needs winners 3.7× the size of losers.

The 10 checks I run before I believe a backtest, free on one page:
https://t.co/dTvRvJGXrI

[Jordon Harrell · Quanticator @Quanticator]

@carlquintanilla Before reading it as May 2000: slide the Nvidia peak date a few weeks either way and refit, and run the same overlay on big-tech peaks that recovered. If those fit about as well, the match says little about what comes next.

[Jordon Harrell · Quanticator @Quanticator]

@ChartGuys Part of that gap is the fixed share count. 10 shares short from 100 to 30 make $700. Re-short to keep the whole balance short as it falls and the same smooth slide pays about $2,330, like the long. Chop is what costs that version.

[Jordon Harrell · Quanticator @Quanticator]

@quantbeckman Worth logging the R-squared of that fit on each date. If the risk drivers explain 80% of the raw score's variance, the residual keeps 20% of it, about 45% of its spread, so measure its IC on its own rather than borrowing the raw score's.

[Jordon Harrell · Quanticator @Quanticator]

@DonWedge Worth separating two tests. On 2-month bars the 2002 and 2008 lows printed roughly 70% under the 20-bar average, while the 2025 low held above it. The higher lows held, but the average itself was not the floor in two of the five.

[Jordon Harrell · Quanticator @Quanticator]

@Andrew_Menaker On the automated side, the regime part can be settled ahead of time: tag each backtest trade by regime (trending or ranging, high or low vol) and see which bucket holds the worst drawdown. If one does, that tag is the written off switch.

[Jordon Harrell · Quanticator @Quanticator]

@TraderMorin Expectancy per trade leaves out how often the setup fires. 0.5R on 2 trades a week is 1R; 0.2R on 10 a week is 2R. Take 0.05R of costs off each trade and it is 0.9R vs 1.5R. Compare net R per week, not per trade.

[Jordon Harrell · Quanticator @Quanticator]

@Venu_7_ Of the 7 losers, 5 closed at -9 to -10%. FORM and VSH reached -34% and -48%: 82 points between them vs 46 for the other five combined. I'd log why each got past the usual exit (a gap or a hold); that line does most of the loss control.

[Jordon Harrell · Quanticator @Quanticator]

The whole check: payoff from the profit factor, breakeven from the payoff, expectancy from both, then the cost that zeroes it. Two minutes.

If your round trip is near that number, you hold a fee schedule, not a strategy.

That's one of ten. The rest are in my profile.

[Jordon Harrell · Quanticator @Quanticator]

One number decides whether a backtest is a business, and almost no report prints it: the cost per trade at which the whole thing goes to zero.

It takes four lines of arithmetic from numbers the report does print. Seven posts.

[Jordon Harrell · Quanticator @Quanticator]

This is why "stress-tested at double the costs" answers the wrong question. Double of what?

Solve for the break-even cost instead, then put your own fills next to it. One number, no choices in it, nothing to argue about.

[Jordon Harrell · Quanticator @Quanticator]

@FreedomTradesx The 10% is the number to test. Pull each holding's worst overnight gaps over the past five years; a name that has gapped 30% on earnings turns a 25% position into a 7.5% hit, and a stop underneath fills near the open, not at its level.

[Jordon Harrell · Quanticator @Quanticator]

@panicsellweekly It helps to put expected variance in numbers before the loss: list the backtest's worst single-trade losses in R. A loss that shows up once per 100 trades has about a 78% chance of appearing at least once in a 150-trade year.

[Jordon Harrell · Quanticator @Quanticator]

Commission-free is not cost-free. The fee moved into the spread.

Cross a five basis point spread both ways and you paid five basis points a round trip, with nothing in the commission field to show it. On an eight basis point gross edge you keep three.

Ask for the spread.

[Jordon Harrell · Quanticator @Quanticator]

@nicaqs In the late-2021 to 2022 decline, NVDA had two rallies of 30%+ on daily closes (March, then July to August) before its October 2022 low, about 66% under the peak. Both ended in lower lows, so a sharp bounce alone didn't split the paths.

[Jordon Harrell · Quanticator @Quanticator]

@MikeZaccardi Worth checking the basis: intraday high to an after-hours print usually reads deeper than close to close, and most backtest max-drawdown figures use daily closes. Same move, different number depending on which prices you measure.

[Jordon Harrell · Quanticator @Quanticator]

@SowingAlphaSeed Before sizing the trend piece, pull the months when your biggest factor bets fell together and score the sleeve there on its own. Trend often lags the first leg of a sell-off, and a 10% sleeve up 20% in a bad month only offsets 2 points.