Accelerator Oscillator Tested: Price Turned Before It on 63% to 67% of Turns, and 0 of 20 Runs Passed

The Accelerator Oscillator is the second oscillator in Bill Williams’ trading system, next to the Awesome Oscillator. Williams’ idea was that price is the last thing to change: first the driving force slows down, then momentum turns, then price. The AO is meant to show momentum, and the AC, being the AO minus its own average, is meant to show the change in momentum, so it should warn earliest.

MetaTrader 4 and 5 ship it under Indicators, Bill Williams, Accelerator Oscillator. It has no inputs at all, only colours. Most guides repeat Williams’ rules: buy on two green bars above zero, three green bars below zero, or on the zero cross. All of that can be counted, so I counted it.

How is the Accelerator Oscillator calculated?

Subtracting an average from a line takes away its slow part and leaves the wiggle. So the AC is a much twitchier line than the AO it is made from: on our charts the two correlated at only 0.19, on all 4 of them. The full head-to-head is in Accelerator vs Awesome Oscillator.

How was it tested?

• Data: EURUSD M5 from June 3 to August 26, 2026, 16,926 bars; gold M5 from June 16 to August 26, 13,736 bars. M15 built from the same bars.
• Five rules: the zero cross; the zero cross only on the EMA 200 side; any colour change (buy on the first green bar after red, sell on the first red after green); Williams’ rule (above zero, buy on two green bars in a row after a red; below zero, wait for three; sells the mirror image); and Williams’ rule taken only on the side of the AO.
• Exit and pass: stop 1.5 ATR, target 3 ATR, 48 bars at most, one trade at a time, the same exit as in the forex indicators list, ranked. Cost 1 pip on EURUSD and $0.30 an ounce on gold, from spread and swap costs in MT4. A run passes if it makes money after cost, in both halves (split July 15), and beats the 95th percentile of 200 random-entry runs with the same exit, as in algorithmic rule vs random entry.
• Price turn: a bar whose close is the lowest or highest of the 6 bars before and 6 bars after it. That is only known afterwards, which is fine here: it is used to check timing, not to trade.

How often does the AC change colour?

The AC crossed zero 2.6 to 2.9 times as often as the AO on the same chart (35.6 against 13.5 a day on EURUSD M5) and changed colour about 30% more often. A colour change came roughly every 20 minutes on M5, the same rhythm as the RVI’s signal-line crosses. Whatever the timeframe, a colour lasted a median 4 bars.

The last row is the one that matters. On none of the 4 charts did a new colour tell you much about the next hour on M5 or the next three hours on M15: price followed it 44% to 52% of the time.

Does the AC really turn before price?

This is the claim the whole indicator rests on, and the data did not back it. About two times in three, the close had already made its low or high before the AC bar changed colour. That is not surprising once you look at the formula: the AC is built from averages of past bars, so it can only react to a turn that has started. It does react sooner than the AO, which was late on 69% to 77% of turns, but sooner than a lagging line is still not before price. The same trade-off between delay and false signals is measured in indicator lag vs whipsaw.

And roughly half of all colour changes on EURUSD M5, and two thirds on gold, came nowhere near a real turn. Those are the wiggles.

Did any AC rule make money?

Trades in brackets; the random rows are for the same number of trades as the zero cross. 19 of 20 runs lost after cost, and 0 of 20 passed. On EURUSD M5 the zero cross made +0.16 pips a trade before the spread, so the signal was a hair better than random and the 1-pip cost turned it into a loss.

Williams’ own rule, in one of its two forms, was the worst of the five on 3 of 4 charts. Waiting for two or three green bars means the entry comes 2 to 3 bars after the AC low, which itself came about 2 bars after the price low. By then much of the short swing it was meant to catch is gone, and on M15 it lost more than 2 pips a trade on EURUSD.

The one profitable run, the zero cross on the EMA 200 side on gold M15, made $1.39 an ounce a trade in the first half and lost $0.13 in the second. Gold trended in this sample, and the luckiest 5% of random runs made $3.16 or more with the same number of trades, so +$0.38 is well inside luck.

What would I do with the Accelerator Oscillator?

• Do not trade its colours on M5. At 61 to 68 changes a day they are mostly noise, and every rule lost on EURUSD M5.
• Do not read it as an early warning. Price usually turned 1 to 2 bars before it did. If you want to see a turn, the chart shows it first.
• If you keep it on the chart, compare it with the AO. The better of the AO’s zero cross and saucer beat the best AC rule on all 4 charts; see Accelerator vs Awesome Oscillator.
• Slower periods did better than the default. The built-in AC cannot be changed, but a copy can; 24 settings per chart are in Accelerator Oscillator settings for MT4.

Frequently asked questions

What is the Accelerator Oscillator?

A Bill Williams indicator that shows the Awesome Oscillator minus its 5-bar average. It is meant to show whether momentum is speeding up (green bars) or slowing down (red bars).

How do you trade the Accelerator Oscillator?

Williams’ rule buys on two green bars in a row above zero or three below zero, and sells on the mirror image. In this test that lost after cost on all 4 charts, from $0.61 an ounce on gold M5 to 2.25 pips on EURUSD M15.

Does the Accelerator Oscillator lead price?

Not on our data. Within 6 bars of a price turn, the close made its low or high first 63% to 67% of the time, and the AC changed colour a median 1 to 2 bars later.

What is the difference between the AC and the AO?

The AC is the AO minus its own 5-bar average. That makes it turn earlier than the AO, about 53% to 56% of the time, and makes it far noisier: it crossed zero 2.6 to 2.9 times as often.

Does the Accelerator Oscillator repaint?

No. It uses closed high and low prices. The current bar’s value and colour change until the bar closes, so all signals here were taken on closed bars.