Ensemble Voting of Signals

Ensemble Voting of Signals seats nine classic studies on a panel and lets them vote. Each enabled component — an EMA cross, price against a long SMA, RSI, MACD, Stochastic %K, Bollinger %B, Donchian position, OBV trend and MFI — casts a bullish, bearish or neutral verdict every bar; votes are weighted, netted, scaled to a -100 to +100 score, smoothed and banded into ratings. It is ensemble voting of signals in its transparent form: every voter, weight and verdict is inspectable.

How to Trade the Ensemble Voting of Signals?

• Rating bands: a consensus beyond the Signal Threshold rates Buy or Sell; beyond the Strong Threshold, Strong Buy or Strong Sell — each transition is alertable.
• Agreement %: the share of active weight on the dominant side — a Buy at 55% agreement is a split panel; at 90%, a chorus.
• Neutral votes: bounded voters near their midline abstain rather than force a direction — dashes are genuine indecision.

As elsewhere in the library's machine-learning family, value comes from diversity: unlike components can disagree for informative reasons, near-duplicates cannot.

Ensemble Voting of Signals Settings

• Signal Threshold (default 20): net consensus required before the panel rates Buy or Sell.
• Strong Threshold (default 60): the supermajority boundary for Strong ratings.
• Consensus Smoothing (default 3): EMA applied to the raw tally; set 1 to read it unsmoothed.
• Oscillator Neutral Zone (default 5): half-width around the 50 midline inside which bounded voters vote neutral.
• Components: every voter ships enabled with weight 1; lengths default to EMA Cross 20/50, Price vs SMA 200, RSI 14, MACD 12/26/9, Stochastic %K 14/3, Bollinger %B 20 (multiplier 2), Donchian 20, OBV EMA 20, MFI 14.
• Show Dashboard (default enabled), plus a Gradient Fill style toggle (default on).

Frequently Asked Questions

How is this different from a Random Forest?

A Random Forest learns its voters, training many randomized trees on features. This panel polls fixed, hand-picked indicator verdicts with user-set weights — nothing is fitted, costing adaptivity but keeping every vote explainable.

Why does the rating lag the raw tally?

The consensus is a 3-bar EMA of the tally by default, so a score hovering at a band edge does not flick the rating on and off. Set Consensus Smoothing to 1 for the raw vote, flicker included.

Should every component keep its default weight?

Only if the panel stays diverse. Two components reading nearly the same thing act like one voter with double weight; down-weight or disable overlaps and let independent families carry the tally.