Autocorrelation Residual Error
Quick Reference
| Property | Value |
|---|---|
| Dimension | quality |
| Category | statistics |
| Version | v1.0 |
| Output Column | autocorr_residual |
Autocorrelation residual of a signal at specified lag 鈥?measures prediction efficiency
Formula
autocorr(signal, lag)
CDM Inputs
| Column | CDM Table | Description |
|---|---|---|
signal | cdm_* | CDM source table |
Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
lag | integer | 100 | Lag for autocorrelation in milliseconds |
Output
Column: autocorr_residual
Autocorrelation at specified lag (residual)
Market Intuition & Trading Rationale
Autocorrelation residual error measures the serial dependence of a signal: autocorr(signal, lag). High positive autocorrelation means the signal is persistent 鈥?today's value predicts tomorrow's. This is good for trend-following strategies (momentum persists) but bad for signal efficiency (the signal contains redundant information 鈥?each observation isn't independent).
High negative autocorrelation means the signal oscillates 鈥?positive today, negative tomorrow. This is the signature of mean-reverting signals and is desirable for mean-reversion strategies. Near-zero autocorrelation means the signal is independent from observation to observation 鈥?each value carries new information. This is the ideal for feature sets feeding ML models (no redundancy).
The "residual" framing comes from time series analysis: after fitting a model, the residuals should be uncorrelated. Non-zero autocorrelation in the signal itself indicates that a simple autoregressive model could predict it 鈥?the signal isn't fully efficient at incorporating past information.
Usage Cases
- Signal efficiency assessment: autocorr near zero 鈫?signal is efficient (good for ML). autocorr > 0.3 鈫?signal is persistent (good for trend following, redundant for ML). autocorr < -0.3 鈫?signal is mean-reverting (good for reversal strategies).
- Regime classification: Rising autocorrelation in a momentum signal means the trend is becoming more persistent 鈥?increase position size. Falling autocorrelation means the trend is losing persistence 鈥?prepare for reversal or choppy conditions.
- quality context: Used in
microstructure_mean_reversionpack 鈥?autocorrelation of micro_price_deviation tells you whether the deviation is mean-reverting (negative autocorr) or trending (positive autocorr), which determines the appropriate trading strategy.
YAML Definition
name: autocorrelation_residual_error
description: Autocorrelation residual of a signal at specified lag 鈥?measures prediction
efficiency
category: statistics
dimension: quality
version: v0.9.0 (Beta)
required_inputs:
- signal
output_column: autocorr_residual
output_description: Autocorrelation at specified lag (residual)
tags:
- quality
- autocorrelation
- statistics
parameters:
lag:
type: integer
description: Lag for autocorrelation in milliseconds
required: false
default: 100
formula: autocorr(signal, lag)