I’ve been experimenting heavily with quantitative concepts lately, and I wanted to share a highly effective way to filter out market noise when scalping the lower timeframes (M1 and M5). It involves a concept borrowed from data science called Principal Component Analysis (PCA).
What is a PCA Trend Vector?
In data science, PCA is used to reduce the dimensions of a massive dataset while preserving its core variance. In simple terms, it finds the "path of least resistance" or the strongest underlying pattern in a chaotic scatterplot of data.
When we apply a simplified 2D PCA (Price vs. Time) to Forex, the First Principal Component gives us a Trend Vector. Instead of looking at erratic, noisy candlesticks that constantly fake out, this vector calculates the exact mathematical line of maximum variance over a given period.
If you are scalping, you don't care about the noise; you only care about the core vector of the market at that exact moment.
How to Use it for Scalping
The strategy here is not to use the PCA vector as an entry trigger, but as a strict directional filter.
Timeframe: M1 or M5.
The Rule: When the PCA Vector histogram is green (positive slope), you only look for long setups. When it is red (negative slope), you only look for short setups.
The Entry: Wait for price to pull back against the vector, and use an oscillator (like a stochastic dipping below 20 in an upward vector) to snipe the entry.
Because the PCA vector mathematically cuts out the random walk of the market, you'll find that your pullbacks result in far fewer stop-outs.
The MT4 Code
True multi-dimensional PCA requires matrix algebra libraries (usually via Python integration), but in a 2-variable environment (Price and Time), the 1st Principal Component mathematically converges with the slope of a Linear Regression line.
Here is a lightweight custom indicator I wrote for MT4. It calculates the core vector slope and plots it as a histogram.
How to install: Open MetaEditor, create a new Custom Indicator named PCA_TrendVector, paste this code, and compile.
Code: Select all
//+------------------------------------------------------------------+
//| PCA_TrendVector.mq4 |
//| Approximation of 1D PCA (Regression) |
//+------------------------------------------------------------------+
#property copyright "Forum Community"
#property strict
#property indicator_separate_window
#property indicator_buffers 2
#property indicator_color1 clrLime
#property indicator_color2 clrRed
#property indicator_width1 2
#property indicator_width2 2
input int PCAPeriod = 20; // Period for the PCA Vector
double UpVector[];
double DnVector[];
int OnInit() {
SetIndexBuffer(0, UpVector);
SetIndexStyle(0, DRAW_HISTOGRAM);
SetIndexBuffer(1, DnVector);
SetIndexStyle(1, DRAW_HISTOGRAM);
IndicatorShortName("PCA Trend Vector (" + IntegerToString(PCAPeriod) + ")");
return(INIT_SUCCEEDED);
}
int OnCalculate(const int rates_total, const int prev_calculated,
const datetime &time[], const double &open[],
const double &high[], const double &low[],
const double &close[], const long &tick_volume[],
const long &volume[], const int &spread[]) {
int limit = rates_total - prev_calculated;
if(limit == 0) limit = 1;
if(prev_calculated == 0) limit = rates_total - PCAPeriod - 1;
for(int i = limit; i >= 0; i--) {
double sumX = 0, sumY = 0, sumXY = 0, sumX2 = 0;
// Loop backwards to align time logically (oldest to newest in the period)
for(int j = 0; j < PCAPeriod; j++) {
double y = close[i + PCAPeriod - 1 - j];
double x = j;
sumX += x;
sumY += y;
sumXY += x * y;
sumX2 += x * x;
}
// Calculate the Primary Component Vector (Slope)
double denominator = (PCAPeriod * sumX2) - (sumX * sumX);
double vector_slope = 0;
if(denominator != 0) {
vector_slope = ((PCAPeriod * sumXY) - (sumX * sumY)) / denominator;
}
UpVector[i] = 0;
DnVector[i] = 0;
if(vector_slope > 0) {
UpVector[i] = vector_slope;
} else if(vector_slope < 0) {
DnVector[i] = vector_slope;
}
}
return(rates_total);
}