T
Traderoid

Autocorrelation Node

Measures self-correlation at different time lags

StatisticalCorrelationPattern Detection

Overview

Autocorrelation measures how strongly a time series correlates with itself at different time lags. It identifies whether past values have a statistically significant relationship with current values, revealing patterns that repeat at regular intervals (seasonality, cycles, mean reversion).

Understanding autocorrelation is crucial for time series analysis because standard statistical tests often assume independence between observations. High autocorrelation violates this assumption and affects the reliability of statistical models. It also helps identify whether price movements show momentum (positive autocorrelation) or mean reversion (negative autocorrelation).

Formula & Calculation

Autocorrelation Function (ACF)
ACF(k) = Cov(X_t, X_t-k) / Var(X)
Where k is the lag period, Cov is covariance, and Var is variance
Normalized Form (-1 to 1)
ρ(k) = Σ[(X_t - X̄)(X_t+k - X̄)] / Σ[(X_t - X̄)²]
Returns correlation coefficient from -1 to 1 for each lag
Example Interpretation
ACF(1) = 0.75 means today's return is highly correlated (75%) with yesterday's
ACF(5) = 0.15 means today's return has weak correlation with 5 days ago
ACF(20) = -0.20 means 20-period lag shows slight negative autocorrelation (mean reversion)

💡 Lag Selection: Typical lags range from 1-50 periods. Plot ACF values across lags to visualize the correlation structure. Lags beyond the 95% confidence bands are statistically significant.

Parameters

ParameterDefaultDescription
lag1-20Number of periods to calculate correlation. Higher lags show longer-term relationships.
sourceCloseInput data series (typically closing prices or returns)
lookback100Number of periods to use for calculating autocorrelation

Common Use Cases

1. Detect Seasonality & Cycles

Spikes in ACF at specific lags reveal seasonal patterns. For example, if ACF is high at lag-252 (1 year), the market shows seasonal behavior recurring yearly. Lag-5 spikes indicate weekly cycles. Use this to adjust strategies for predictable seasonal patterns.

2. Identify Momentum vs Mean Reversion

Positive ACF indicates momentum (trending behavior) - recent price action predicts future changes. Negative ACF indicates mean reversion - price tends to reverse after moving. ACF near 0 suggests random walk. This determines which trading strategy to use.

3. Test Stationarity & Mean Reversion

ACF that decays slowly suggests non-stationary data (needs differencing). Rapid decay indicates stationary data suitable for many statistical models. Used to validate whether pairs trading (cointegration) will work correctly.

4. Validate Trading Models

Check if residuals from a trading model have significant autocorrelation. If they do, your model is missing important information hidden in the autocorrelation structure. Use to improve model design and incorporate discovered patterns.

Advantages & Limitations

Advantages

  • Reveals hidden patterns and cycles
  • Identifies momentum vs mean reversion
  • No parameters to optimize
  • Mathematically rigorous foundation
  • Useful for validating models
!

Limitations

  • Only works with stationary data
  • Doesn't show causal relationships
  • Requires sufficient historical data
  • Complex interpretation for traders
  • Changes over time (non-stationary)

Tips & Best Practices

📊 Plot ACF Diagrams

Visualize ACF values with confidence bands. Values outside the bands are statistically significant. Look for patterns: sharp drop-off (stationary), slow decay (trending), spikes at regular intervals (seasonal).

🔄 Use ACF on Returns Not Prices

Price autocorrelation is usually high due to trends. Calculate returns (price changes) first, then ACF on returns. This reveals the true autocorrelation structure that trading strategies can exploit.

⚡ Combine with PACF

Partial ACF (PACF) removes effects of intermediate lags, isolating direct relationships. ACF + PACF together determine the best ARIMA parameters for forecasting models.

⚠️ Check for Non-Stationarity

If ACF decays very slowly (stays high for many lags), the data likely isn't stationary. Difference the data (subtract previous value) and recalculate. Most statistical tests require stationary data.

Example Strategy: Autocorrelation-Based Trading

Setup

  • • Calculate ACF for lags 1-20 using daily returns
  • • Monitor ACF(1) and ACF(5) for changes
  • • Identify periods of positive vs negative autocorrelation

Momentum Strategy (Positive ACF)

If ACF(1) > 0.30: Use momentum (trend-following) strategies. Price has momentum - recent changes predict future changes. Use moving average crossovers or breakout systems.

Mean Reversion Strategy (Negative ACF)

If ACF(1) < -0.20: Use mean reversion strategies. Price tends to reverse - recent moves are likely followed by opposite moves. Use RSI divergences, Bollinger Band reversals.

💡 Dynamic Strategy Selection: Calculate ACF monthly and switch strategies based on autocorrelation structure. This adaptive approach often outperforms fixed strategies.

Related Nodes