Features Pricing Knowledge base
Start free trial Log in
live feedconnecting
Knowledge baseCrash Radar

Crash short call and the crash-short backtest

When the consensus card switches to CRASH SHORT, how the backtest picks the risk level and target, and how to read the result.

What a crash short is

A crash warning blocks longs, but it can also be a trade. The question is whether, on this pair, opening a short when the crash risk reached a given level would have paid. The crash-short backtest answers that from history, and when the answer is yes the consensus card turns into a CRASH SHORT call as soon as the live risk crosses that level.

The backtest

It runs during deep analysis (stage backtesting crash-short strategy) and can be re-run through the API. It replays the crash score every 15 minutes over up to 40,000 samples of history (the rule score, combined with the trained models when available) and tests five risk levels: 35%, 45%, 55%, 65% and 75%.

For each level:

  1. A short is opened at the 5-minute close the first time the risk crosses the level. The trade is re-armed only after the risk drops back under the level, so one crossing gives one trade.
  2. The stop is one 1-hour ATR above entry (the same distance as the Void above level of a live warning).
  3. Four targets are tested at 1.0, 1.5, 2.0 and 3.0 ATR below entry (the 2.0 ATR target matches Full confirmation).
  4. If neither stop nor target is hit within four hours the trade is closed at the close of that bar.
  5. Two taker fees are deducted.

Per level the result holds the number of trades, the best target multiple, its win rate, average and total return, and the median and 75th-percentile favourable excursion in ATR units.

How the level is chosen

A level qualifies when it produced at least 15 trades with an average return above 0.05% per trade after fees. Among qualifying levels the one with the highest average return weighted by the trade count (capped at 60) is recommended. If no level qualifies, the pair never gets a crash-short call and the consensus card keeps its normal logic under crash conditions (NO LONGS, LEAN or WAIT).

The live call

When the pair's live crash risk is at or above the recommended level, the consensus card is replaced:

Element Value
Badge GO SHORT with the basis crash short (backtested on this pair)
Headline CRASH SHORT · risk 67% ≥ 55% · on this pair shorting at this level won 61% of 38 times, avg +0.84% per trade (target 1.5× 1h ATR)
Confidence The backtested win rate at that level
Entry range Current price plus or minus 0.15 of a 1h ATR; chase limit half an ATR below
Stop-loss The Void above level of the crash warning; wide stop half an ATR beyond it
TP1 The best target multiple times the 1h ATR below price
TP2 The larger of 1.5x that multiple and the 75th-percentile historical drop
Reasons The live crash risk and level, then the radar's top reasons

The note under the card reads: Stop = the level that voids the crash warning; TP1 = the target that paid best historically at this risk level; TP2 = the 75th-percentile drop.

The crash-short override is applied after the analog stop widening and before the call is logged, so it appears in the live track record and in calibration like any other call.

How to use it

  • Treat the headline numbers as a frequency on a small sample. 38 trades at 61% is useful evidence; 15 trades at 70% is not much better than a guess.
  • Enter inside the entry range. A crash short chased half an ATR lower has already given up most of the target.
  • The stop is above the void level because a move above it also cancels the warning that justified the trade. Do not widen it.
  • Market-wide warnings (BTCUSDT, ETHUSDT) can push alt pairs over their level at the same time. Spread risk rather than shorting every pair.

Note: the override only fires when the pair's own risk, not the global risk shown in the top bar, reaches the recommended level, and only when the 1h ATR is available.

Tip: the Crash alert history on the Crash Radar page tells you how often warnings on this pair were followed by a real 3% drop. A pair with many false alarms rarely gets a recommended level, which is the system working as intended.