EMA vs Hull Moving Average
The exponential moving average weights recent bars more heavily and lags predictably. The hull moving average combines several weighted averages to cut that lag further, and the price of the reduced lag is that it overshoots and reverses on moves that go nowhere.
One is the standard responsive average; the other was designed specifically to cut the lag further. The second one works, and the interesting question is what it costs.
What each one is
The exponential moving average weights recent bars more heavily, so it follows price more closely than an equally weighted line while still lagging by a predictable amount. The exponential moving average covers it.
The hull moving average combines several weighted averages in a way designed to remove much of that remaining lag. The hull moving average covers the construction.
Both are smoothings of the same closes. Neither can turn before price does; they differ only in how quickly they respond once it has.
Where they differ
How quickly each turns. The hull turns earlier, and at a genuine reversal that is a real advantage worth having.
How often each turns for nothing. The hull also reverses on moves that go nowhere, because a line that reacts faster reacts to everything faster, including noise.
Whether the behaviour is predictable. An exponential line’s lag is a known function of its length. The hull’s overshoot depends on the shape of the recent move, which is harder to anticipate.
What each is used for. The exponential line is usually a trend filter or a dynamic level. The hull is usually used as a direction signal in its own right, read off its colour or its slope.
Where they agree
Both are lagging by construction. Reducing lag is not eliminating it, and neither line contains information that was not already in the closes it averaged.
Both fail in a range. On this site’s shared series direction runs average 2.01 bars with a longest of 11, and short runs produce false turns in either — more in the faster one.
Both cost a round trip per signal acted on — about 2% of the median bar range of 0.493 here — which matters more for the line that produces more signals.
And neither contains a stop. The ninetieth percentile bar range here is 1.101, and the invalidation belongs at structure rather than at a line.
Which one to use
Run the exponential average when you want predictable behaviour. Knowing roughly how late the line will be is worth more than being slightly earlier and not knowing when it will be wrong.
Run the hull when you have counted the extra trades and accepted them. It turns earlier at real reversals, and the only way to know whether that pays is to count both on your own instrument.
Run the hull when your method exits fast anyway. If you are already cutting quickly, an early false turn costs you a round trip rather than a position, which is the cheapest form of that error.
And when the argument for the hull is that it has less lag, run the exponential one. Less lag is the claim, not the benefit — the benefit only exists if the extra turns cost less than the lag saved.
Why less lag is not free
Because responsiveness is not selective. A line cannot react quickly to real turns and slowly to noise; it reacts quickly to changes, and most changes are noise.
And because the extra signals are not spread evenly. They cluster in ranges, which is exactly where a directional line is least useful and most expensive.
How to test the swap honestly
Count the trades both ways over the same window. Not the winners — every signal each line produced. The faster one will produce more, and the size of that gap is the whole cost of the swap.
Use the same length on both. Comparing a short hull against a long exponential line is comparing two lag settings, and whichever one suited the period you picked will look like the better indicator.
Include the ones you would have skipped. A signal you would have ignored because it looked wrong is still a signal the rule produced, and leaving it out is how a test quietly becomes a description of your judgement instead of the tool’s.
And run it across a range as well as a trend. On this site’s shared series direction runs average 2.01 bars, so a window chosen because it trended cleanly will flatter the faster line and tell you almost nothing about the rest of the year.
The original data
Of the 24,971 unique videos in research/search-study-corpus.jsonl, no title compares these two
directly — this pair is constructed from two subjects the corpus covers separately rather than from a
measured comparison. Separately, the hull moving average appears in 83 titles at a median of 1,923
across 67 channels. The counts come from site/corpus_count.py.
83 videos on the hull at a median of 1,923 across 67 channels. Modest coverage and a small audience per video — a specialist tool rather than a widely taught one, which is worth knowing before adopting it as the centre of a method.
The answer to the question on that chart is that you cannot tell yet, and that is the point. The early turn is right sometimes and early-and-wrong the rest of the time — which is the same trade the line was chosen for, arriving in the form nobody wants.
When it fails
The failure is adopting the faster line after seeing it in review, and the review was never fair. Scrolling back, the hull turned earlier at every major reversal on the chart, which looks like a clear improvement. What the eye skips is the far larger number of turns it made in between that led nowhere, because those do not stand out on a chart where you already know which moves mattered. Live, the extra turns arrive at the same rate as the good ones and cost a round trip each.
The second failure is treating either line as a signal on its own. Both lag price.
A third is running both. They are two smoothings of the same closes.
A fourth is placing a stop at the line. It has no relationship to structure.
A fifth is optimising the length after a losing run. That is fitting the past.
And a sixth is comparing them without counting trades. The count is the whole difference.
Related
The exponential moving average covers the standard responsive line. The hull moving average covers the reduced-lag construction. And the moving average crossover covers the signal both are usually traded through.
Every faster average trades lag for false signals, and the hull is the clearest example of the deal. It does turn earlier at real turns. It also turns early at things that are not turns, and you do not get to keep one without the other.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.