Gaussian Signals and Candle
- 指标
- 版本: 1.0
- 激活: 5
The band on nearly every channel indicator you can download is sized by an ATR, while the line running down the middle of it comes from something else entirely. Two filters, two different lags, stapled together. The join is where the false signals live.
This one runs the true range through the same Gaussian filter as the price, at the same pole count and the same period. The band and the centre line turn on the same delay, because they are the same filter fed different data.
Where the usual versions drift
| The theory | What most versions implement | What this does |
|---|---|---|
| A Gaussian filter has a pole count, and each pole trades lag for smoothness | Poles fixed at 4, or no setting exposed at all | Factor 1 to 9, with the binomial weights computed for whichever one you pick |
| Band width should react on the same delay as the centre line | ATR(14) times a multiplier: a different filter with a different lag | True range goes through the same filter, same poles, same period |
| A trend tool should be allowed to say "no opinion" | Two colours, always committed to one side | Three candle states, and the middle one is the honest one |
| Entering from below is not the same trade as entering from above | One fixed threshold in both cases | The threshold moves depending on where the bar opened |
| A reversal needs more evidence than a continuation | Symmetric conditions both ways | Red needs the open and the close under the lower band; green needs only the close over the upper |
Row two is the one that costs money. When the band comes from ATR and the line comes from a smoother, the two disagree hardest exactly when volatility turns, which is exactly when you're deciding whether to act. You get a band that has already widened while the line hasn't moved, or the reverse.
What it draws
The channel. Three lines: the Gaussian filter itself and a band either side. Width comes from the filtered true range scaled by the Sensitivity input, so it widens in fast markets and tightens in quiet ones with nothing for you to adjust.
Coloured candles. Green when the close is above the upper band. Red when the open and the close are both under the lower band. Neutral for everything else, which on most charts is most of the time. That neutral state is deliberate: price inside the band means the filter has no read, and colouring it anyway would be making something up.
Buy and Sell labels. Each sits on its own indicator buffer as well as a text label on the chart, so an EA can read the signal through iCustom without scraping objects. One signal per direction: after a Buy there is no second Buy until a Sell has happened.
It draws no entry, stop or target levels. No zones, no order blocks, no panel, no dashboard, and it reads only the chart it is placed on.
The signal rule worth knowing
The threshold for a Buy is not fixed. If the bar opened below the centre line, a Buy fires when the close crosses the centre line. If the bar opened at or above it, the Buy waits for a close through the upper band.
Coming up from underneath, it takes the midline cross. Already above it, and it wants the extra proof of a band break. Sell is not quite the mirror, and that is the part worth knowing. If the bar opened above the upper band, a Sell fires on a close back through that upper band; otherwise it waits for a close through the lower band. The pivot for a Buy is the centre line, the pivot for a Sell is the upper band, so a Sell triggers a little more readily from stretched territory than a Buy does from the bottom of a range. You can see it on any chart with a decent swing: labels near the bottom of a move sit on the middle line, labels inside a run sit on the outer band.
Settings
| Input | Default | What it does |
|---|---|---|
| Factor | 6 | Pole count, 1 to 9. Lower turns faster and rougher, higher is smoother and later |
| Agility % | 55 | Filter period. Sets how wide the whole thing breathes |
| Sensitivity | 28 | Band width multiplier applied to the filtered true range |
| Source | Low | Price the filter runs on. Close and Median are the usual alternatives |
| 2nd calculation method | off | Lag-reduction variant. Faster, noisier, off by default for a reason |
| Fast response mode | off | Blends the N-pole and 1-pole outputs. Same trade-off |
| Notification type | none | Popup, push, email or Telegram |
If it feels late, drop the Factor before you touch the period.
Reading it
| On the chart | Read it as |
|---|---|
| Green candles | Close is clear of the upper band, trend extended |
| Red candles | Whole body under the lower band, the down move has commitment |
| Neutral candles | Inside the band, no read, and this is normal |
| Label on the centre line | Cross from the other side of the filter |
| Label on the outer band | Continuation break while already on that side |
| Bands flat and narrow | Range, and the labels here are the weak ones |
Long: wait for a Buy label, check the band is sloping rather than flat, then drop to your execution chart and take your own trigger. The centre line is a reasonable place to trail behind. Short is the same inverted.
Skip it when the bands are horizontal and price is crossing the middle every few bars. It will keep printing labels in that state, alternating Buy, Sell, Buy, Sell as price saws through. That's not a fault to fix, it's a range.
What this version can't do
It's noisy in a range, and you should see the number rather than find out later. Over the last 500 bars of USDJPY H4 it printed 63 labels. XAUUSD H1 over the same span printed 37. Most of the difference comes from two sideways stretches. Trade every label and you'll churn.
There is no correct pole count for every symbol. Factor 6 with period 55 is a fair starting point on H1 and H4 and it's what ships. You'll want to change it per symbol and timeframe, and anyone who tells you their trend setting works everywhere is selling something.
The source defaults to Low, not Close, which sits the channel slightly below where you might expect. That's intentional, and it's an input.
No multi-timeframe. It reads the chart it's on. Nothing else.
You won't find a win rate table in this listing, and I'd treat any indicator listing that has one with some care.
Honestly, the part that ate the most time wasn't the filter recursion. It was a hard-coded table of nine weights, 36, 84, 126 and so on, that I typed out in full before noticing they were just binomial coefficients. Two lines instead of forty, and I only proved it by running both versions side by side over a few thousand bars and diffing the output.
Questions you'll have
Does Gaussian Signals & Candle repaint? Closed bars, no. Every value comes from finished bars, and the state deciding whether the next label is allowed is stored per bar rather than in a running variable, so a recalculation reproduces the same history. The forming bar moves until it closes, like every trailing calculation ever written.
Gaussian Signals & Candle on M1? You can, and I wouldn't. The band is sized from true range, and on M1 that's mostly spread and noise, so you get labels with no move behind them.
Gaussian Signals & Candle best settings? Factor 6, period 55, sensitivity 28 is the shipped set and a fair start on H1 and H4. The period changes how wide the whole thing breathes; the factor changes how sharply it turns.
Gaussian Signals & Candle vs SuperTrend or a Keltner channel? Both of those size their band from ATR and put a different calculation in the middle. This uses one filter for both, which is the whole design. Whether you prefer that is a real question, and one chart will answer it.
Can an EA read the signals? Yes. Buy and Sell each have their own buffer, non-zero on the signal bar and zero otherwise.
MT4? There's an MT4 build listed separately. Same filter and same signal rules, with two differences: MQL4 has no coloured-candle plot, so the MT4 version gives you the channel and the labels without candle colouring, and its alerts cover popup, push and email but not Telegram.
Support
Message me on MQL5 with a screenshot of the chart, plus symbol, timeframe and your Factor and period. That's normally enough for me to tell you what you're looking at. Updates are free.
Analysis tool. It places no trades and makes no profitability claim. Past price behaviour is not a guide to future price behaviour.
