forex-scalping.com

Efficiency ratio over 10 bars: a smooth path, a jagged one and a choppy one, by hand

Kaufman's efficiency ratio is simple: the net change over N bars divided by the sum of each bar's absolute change. It runs from 0 (all noise) to 1 (a straight line).

Three 10 bar paths, closes in pips above the start, all ending 9 pips higher.

A: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9. Net 9. Path length 9. ER = 1.00.

B: 0, 3, 1, 5, 2, 6, 4, 8, 5, 9. Steps: +3, -2, +4, -3, +4, -2, +4, -3, +4, absolute sum 29. ER = 9/29 = 0.31.

C: 0, 6, -2, 7, -3, 8, -1, 9, 0, 9. Steps: +6, -8, +9, -10, +11, -9, +10, -9, +9, absolute sum 81. ER = 9/81 = 0.11.

On a chart, A and C both finish up 9 pips. One has a route that a trend follower could use and the other has swings of 10 pips each way that would have hit most stops. The ratio tells them apart where a net change cannot.

The by hand examples make the ratio very intuitive. On M5 I find an efficiency ratio above about 0.4 over 10 bars marks a usable trend for scalping, while values below 0.2 suggest chop. Using it as a filter rather than a signal works well, for example only taking breakout trades when the ratio is rising.

Kaufman's adaptive moving average turns the ratio into a smoothing constant: sc = (ER x (fast - slow) + slow) squared, with fast = 2/(2+1) = 0.667 and slow = 2/(30+1) = 0.0645. At ER = 1, sc = 0.444, like an EMA of length 3.5. At ER = 0.31, sc = 0.063, like a length of 31. At ER = 0 sc = 0.0042, a length near 480, which hardly moves. So in the choppy path C the average stays almost flat and ignores the swings, while in the smooth path A it hugs price. That is the whole design: follow when it trends, stay put when it does not.

Setting a threshold needs the base rate. For a pure random walk, the average 10 bar ER is about 0.32, and in a simulation of 200,000 random windows I ran, 47% had an ER above 0.3 and 22% above 0.5. So a filter of ER over 0.3 passes nearly half of random noise windows, and is not selective. A level of 0.5 passes about one in five. Real markets have some trending in them, but the benchmark helps: a threshold only means something if it is well above what randomness gives. I would test 0.5 and 0.6 rather than 0.3 before drawing any conclusions.

Illustrative test, with made up results to show the bookkeeping. Take 200 breakout entries. Without a filter they show +0.02R average, +4R in total. With ER above 0.5 at the entry, 43 trades remain, averaging +0.20R, +8.6R in total. Fewer trades and more R, which looks attractive. The question is whether it holds on data you did not use to choose 0.5. Split the sample: pick the threshold on the first half and report the result on the second. If the edge from the filter vanishes in the second half, the first half result was fitting.

In a spreadsheet, with closes in C1 to C11, the ER over 10 bars is =ABS(C11-C1)/SUMPRODUCT(ABS(C2:C11-C1:C10)). Drag it down and it updates for each new bar. Note the window: 10 bars need 11 closes, since the sum covers 10 steps. On M5 that is 50 minutes, on H1 ten hours. A single news bar can dominate the denominator and the numerator alike, so a 25 pip bar early in the window makes the ratio look strong for ten bars, and then falls out of the window and the ratio drops abruptly. That jump on the exit of a big bar is a property of the window, not a change in the market.

On a calendar they look like twins: red icon, same time slot, same pairs reacting. On the screen they behave differently.

NFP is a bundle. The headline jobs number, the unemployment rate, average hourly earnings and revisions all land together, and they can point in different directions. That is why the first spike so often gets reversed. The market reads one line, then another, then argues with itself for several minutes, and price tends to be choppy and two-sided.

CPI is narrower. Headline and core inflation arrive together, but the question is mostly one-dimensional: does it change the expected path of rates? A clear miss in core tends to produce a cleaner, more one-directional repricing because there is less to argue about. Gold and yields also respond more directly to it, which can make dollar pairs follow through.

Another difference is what the market is positioned for. When growth fears dominate, a jobs number gets read through that lens. When inflation is the main worry, CPI becomes the day's main event and NFP the sidebar. The same release can matter very differently depending on the story the market is currently telling.

None of this is a rule, just a tendency that shows up often enough to be worth checking against past releases on whichever pair gets traded.

Does CPI really give cleaner follow-through than NFP on your pairs, or does it look the opposite on your charts?