Back to Docs
T
Traderoid

Z-Score Node

Standardized deviation from mean - fundamental normalization metric

StatisticalNormalizationSignal

Overview

Z-Score is the standardized distance from the mean in units of standard deviation. Z = (Value - Mean) / StdDev. A Z-score of 0 means at the mean; +2 means 2 standard deviations above mean; -3 means 3 below. Z-scores convert any distribution into standard normal for comparison across assets and timeframes.

Z-Score is the foundation of mean reversion trading: extreme Z-scores (|Z| > 2) signal mean reversion opportunities. Use with HalfLifeMeanReversion to quantify how fast reversion occurs. For machine learning, Z-Score normalization standardizes features for better model training.

Formula & Calculation

Z-Score Definition
Z = (X - μ) / σ
X = current value
μ = rolling mean
σ = rolling standard deviation
Interpretation
Z = 0: At mean
Z = ±1: 68% of data within range
Z = ±2: 95% of data within range (extreme zone)
Z = ±3: 99.7% of data (very extreme)

Parameters

ParameterDefaultDescription
lookback20-252Historical window for mean/std
entry_threshold±2.0Entry signal trigger

Common Use Cases

1. Mean Reversion Entry

Buy when Z < -2 (oversold), sell when Z > +2 (overbought). Pair with HalfLifeMeanReversion to confirm reversion speed. Example: SPY drops and Z = -2.5, quick HalfLife (<5 bars) = strong entry.

2. Outlier Detection

|Z| > 3 signals statistical anomaly requiring investigation. Data quality check: extreme Z-scores may indicate data errors or true rare events. Use for alert generation.

3. Feature Normalization

Normalize inputs for ML models: Z-score brings all features to ~(-3, +3) range. Improves neural network convergence and coefficient interpretation. Standard preprocessing step.

4. Cross-Asset Comparison

Compare BTC at Z=2.5 vs SPY at Z=1.8: both extreme but BTC more extreme. Z-scores enable apples-to-apples comparison despite different scales and volatilities.

Advantages & Limitations

Advantages

  • Universal normalization
  • Easy interpretation
  • Works across assets
  • Foundation for statistics
!

Limitations

  • Assumes normality
  • Lagging indicator
  • Backward-looking
  • Fails in regime shifts

Related Nodes