qkt

From hunch to honest backtest, fast. Then the same file goes live. No rewrite, no second language. Free, open source, pre-1.0.

[qkt @tryqkt]

Ported a strategy? Check the trade list before the P&L. Same bars and sides first, then fills, then costs. A port that disagrees on entries won't agree on money.

#algotrading #tradingstrategy

[qkt @tryqkt]

One failed check and a strategy isn't listed on qkt Labs. The failure report goes privately to the seller.

https://t.co/ntOI9r1TC6

#backtesting #quant

[qkt @tryqkt]

Every strategy is market neutral until the market does something.

#quant #tradingmemes

[qkt @tryqkt]

In January 2021, about 140% of GameStop's public float had been sold short.

More shares short than shares available to trade. When the price rose, covering pushed it higher still.

#trading #quant

[qkt @tryqkt]

MT5 builds 6274 and newer stop under Wine 10.

These builds trigger a false debugger dialog on virtual displays. You can test a newer build by setting MT5_TERMINAL_SOURCE to empty during your docker build.

Build 6244 remains the stable default.

https://t.co/I6ig2ZiYK0

[qkt @tryqkt]

Got a 502 unknown_outcome from mt5-gateway?

A 502 unknown_outcome from mt5-gateway means the broker may have accepted the order: check positions before retrying.

Double check before you fire again.

Here's how: https://t.co/GeZOwIdHYx

#MT5 #backtesting

[qkt @tryqkt]

Buy a strategy on qkt Labs and you get the files, not a promise.

The exact strategy file, the exact cost profile and every raw report, each with its SHA-256. Re-run it on the same data with qkt, get the same trades.

https://t.co/n06lrURi88

#backtesting #tradingstrategy

[qkt @tryqkt]

Every report behind a qkt Labs listing must match the fingerprint the engine recorded when it wrote it. A report edited afterwards fails the gate.

#tradingstrategy #backtesting

[qkt @tryqkt]

qkt to pandas, the lines that carry most strategies:

ema(x, 20) → ema(x, 20)
x CROSSES ABOVE y → x > y and x.shift() <= y.shift()
highest(x, 20) → x.shift().rolling(20).max()
CLOSE gold → flat

#XAUUSD #MT5

[qkt @tryqkt]

Ever misplaced a decimal point when sizing a trade?

SIZING 0.5 PCT RISK sizes every trade so that hitting the stop costs half a percent of equity, avoiding decimal errors.

The engine requires a bracket with a stop for it to compile.

#algotrading #backtesting

[qkt @tryqkt]

qkt walkforward reruns the sweep on a rolling year, then scores that year's pick on the next 90 days.

20 folds. Mean in-sample Sharpe 1.31. Mean out-of-sample: 0.06.

[qkt @tryqkt]

A parameter sweep always finds a winner. That's the problem.

Here's how to check whether the winner is real in qkt: sweep, read the warning, then walk it forward onto windows the sweep never saw. Losses stay in.

#backtesting #algotrading

[qkt @tryqkt]

The docs put the catch plainly: 100 combos with 50 trades each will find about 5 winners by chance alone.

The best row of any sweep is also the row most likely to be luck. The only fair test is data it never saw.

[qkt @tryqkt]

The world's highest share price belongs to Berkshire Hathaway's class A stock: $700,000 in August 2024.

The reason is simple. The board has historically been opposed to stock splits.

#quant #markets

[qkt @tryqkt]

VPS bill: $12 a month.

Hours spent making sure the VPS is fine: uncountable, also unpaid.

#algotrading #quant

[qkt @tryqkt]

Past performance, simulated or real, is not a reliable indicator of future results. Test on a demo account before risking real money.
https://t.co/ntOI9r1TC6

[qkt @tryqkt]

Every number on qkt Labs comes from a simulation.

The methodology page ends with its own limits section. Here it is, one limit at a time.

#backtesting #tradingstrategy

[qkt @tryqkt]

A strategy picked from many candidates can look better than it is.

The out-of-sample, walk-forward and deflated Sharpe figures exist to show that risk, not to remove it.