Analog forecasting and wobble-aware stops

The idea

Before a move, price usually wobbles against it. A stop placed inside that wobble gets hit by noise, not by a wrong idea. The analog forecast measures the wobble from history: it finds the moments in this pair's past that most resemble right now and looks at what happened after each of them.

The card is titled Similar past situations (analog forecast) on the pair detail page under the consensus card. It needs the analog index, which is built during deep analysis. Until then it says analog index not built; run deep analysis.

How a match is found

For every 15 minutes of history the index stores:

• the normalised shape of the last 48 fifteen-minute closes (12 hours);
• the volume shape over the same window;
• the full indicator state on every timeframe (the same feature vector the crash model uses);
• the forward path on 1h, 4h and 24h measured on 5-minute candles: return, maximum favourable and adverse excursion for a long and for a short, in 5-minute ATR units.

Live, the same representation is built for now and compared with every stored moment. Distance is 0.5 x price shape + 0.3 x indicator state + 0.2 x volume shape. The 60 nearest matches are kept, with two rules: matches must be at least two days apart from each other, and the last two days are excluded so the forecast cannot match itself. Fewer than 10 usable matches gives too few analogs.

Reading the card

The header shows how many analogs were used out of how many indexed moments and when the index was built. Choose the horizon with 1h, 4h or 24h.

Then one panel per side:

• won n% of analogs: how often the analogs ended in that side's favour;
• Wobble (p80): the 80th percentile adverse excursion, in percent, among analogs that ended in the side's favour. Four out of five winners never went further against the trade than this;
• Stop beyond: the wobble times 1.1, placed beyond the current price;
• Target: the median favourable excursion, and after the slash the 75th percentile.

The table Closest matches lists the twelve nearest moments with their similarity and the return after 1h, 4h and 24h in ATR units, coloured green above +0.3 ATR and red below -0.3 ATR.

Where the wobble is used

1. Consensus card. When the card's tight stop sits inside the analog wobble for the called direction on the 4h horizon, the tight stop is widened to the analog stop, the wide stop is set half a 1h ATR beyond it, R:R is recomputed and the note under the card records the change.
2. My Positions advice. The stop suggestion for an open position is never tighter than the analog wobble stop, except when the position is already more than 1 ATR in profit, in which case break-even wins. The advice text says beyond the typical adverse wobble seen in similar historical situations (80th percentile).
3. Exit policies. The wobble_stop policy in the exit-policy backtest places the stop beyond the winners' 80th-percentile adverse excursion measured from pattern outcomes, the same principle applied to pattern signals.

Limits

• The index is rebuilt at each deep analysis; the card refreshes every two minutes.
• Analog odds are not a signal. They describe what followed similar set-ups, with no proof that the resemblance matters. Treat them as a sanity check on the consensus direction and as the source of a realistic stop distance.
• A pair with little history (new listing) has few analogs and wide confidence bands.

Tip: when the consensus says LONG but the analog bias is short with a wide typical-move band, size down or wait. The two methods look at the pair from different angles; disagreement is information.