What counts as an engulfing candle?
An engulfing candle is one whose body completely covers the body of the preceding candle while closing in the opposite direction: a bullish engulfing closes above the previous open after a down candle, a bearish engulfing closes below the previous open after an up candle.
The specification sounds airtight and contains at least three choices. Does the engulfing apply to bodies only, or must the wicks be covered too? Must the prior candle be of opposite colour, or does any candle count? Must there be a prior move for the pattern to reverse, and if so, how is that move defined?
Each choice changes the population substantially. Requiring full range engulfment rather than body engulfment can cut the sample by a large factor and select for very different bars — typically much larger ones, which means the rule quietly becomes a volatility filter.
None of these choices is wrong. What matters is stating which one you used, because two people reporting engulfing statistics with different definitions are describing different patterns.
Does an engulfing candle actually mean a reversal?
The evidence for candle patterns in isolation is generally weak, and engulfing bars are no exception: on most instruments the difference between what follows an engulfing bar and what follows an average bar is small enough to be easily swamped by trading costs.
The mechanism is not implausible, which is why the pattern persists. A large opposite-direction body means the period ended with decisive movement against the prior candle, and decisive movement is at least some evidence about the balance of pressure.
But the effect has to clear a real hurdle to be tradeable. A pattern that improves the odds of the next ten bars going your way by a percentage point or two is genuine and worthless once spread and slippage are applied, and most published candle statistics are quoted without costs.
The more productive question is not whether engulfing bars matter in general, but whether they matter in a specific context you have independently identified. That question has a much better track record.
How do you measure a candle pattern against the base rate?
Measure it by computing the same forward outcome for every bar in the sample and then comparing the pattern subset against the whole, which is the only way to see whether the pattern selected anything.
Concretely: define an outcome — did price move X ATR in your favour before moving X ATR against, within N bars — and compute it for every bar in the history. Then split the bars into pattern and non-pattern, and compare the rates. The pattern's value is the gap between the two, and the gap is what has to survive costs.
Do this per instrument before pooling. An effect that appears on one pair and reverses on three is what splitting on an uninformative variable looks like, and pooling would have hidden that by averaging.
Report the count of pattern occurrences alongside the rate. Strict engulfing definitions on higher timeframes can produce surprisingly few instances over a decade, and a five-point difference on ninety occurrences is not a finding.
Can engulfing candles be combined with other filters?
Combining them with a location filter is where candle patterns tend to earn their place, because the location supplies the selectivity that the candle shape does not have.
The natural pairings are structural: an engulfing bar at a tested level, at the edge of a range, or immediately after a sweep of a prior extreme. In each case the level answers where and the candle answers when, and the combination has far fewer occurrences and a far more specific claim behind it.
The risk is that each filter you add is another parameter, and combinations multiply fast. Two levels of context, three candle definitions and four horizons is twenty-four trials, and the best of twenty-four looks convincing whether or not anything is there.
The defence is the same as always: fix the whole combination in one place, grade it once on data you have not seen, and report how many combinations you tried before settling. Nothing else reliably distinguishes a finding from a search.