Order Flow Autocorrelation

Quick Reference

Property	Value
Dimension	stability
Category	order_flow
Version	v0.9.0 (Beta)
Output Column	ofi_autocorrelation

Order flow autocorrelation: corr(ofi, prev(ofi, 1), window=20)

Formula

rolling_corr(ofi, lag(ofi, lag), window)

CDM Inputs

Column	CDM Table	Description
best_bid_size	cdm_*	CDM source table
best_ask_size	cdm_*	CDM source table
ofi	cdm_*	CDM source table

Parameters

Parameter	Type	Default	Description
window	integer [2, 200]	20	Window size for correlation calculation
lag	integer [1, 50]	1	Lag for autocorrelation (e.g., 1 for correlation with previous value)

Output

Column: ofi_autocorrelation

Autocorrelation of order flow imbalance with 1-period lag

Market Intuition & Trading Rationale

Order flow autocorrelation measures the persistence of OFI direction: corr(ofi[t], ofi[t-lag]). It answers: is the current order flow direction similar to recent flow? High positive autocorrelation means buying pressure tends to follow buying pressure — the flow is persistent and likely informed (an institution working a large order). Low or negative autocorrelation means flow flips direction frequently — noise trading or market-making activity.

This is the stability counterpart to flow_persistence_score (signal dimension). The signal feature captures the magnitude of persistence; this stability feature captures its reliability. A flow persistence score of 0.8 is impressive, but if the order_flow_autocorrelation is only 0.1, that persistence score is unreliable — it fluctuates too much to trade on.

Usage Cases

• Informed flow confirmation: High OFI magnitude + high order_flow_autocorrelation = genuine informed flow. High OFI magnitude + low autocorrelation = noisy flow, possibly a single large order that won't persist.
• Execution program detection: Sustained positive autocorrelation (> 0.5) over multiple windows indicates an institutional algorithm is actively working an order. Join the direction; exit when autocorrelation drops.
• stability context: Foundational stability feature used across flow-based feature sets — every OFI-based signal has its autocorrelation measured to distinguish persistent from transient flow.

YAML Definition

name: order_flow_autocorrelation
description: 'Order flow autocorrelation: corr(ofi, prev(ofi, 1), window=20)'
category: order_flow
version: v0.9.0 (Beta)
dimension: stability
status: Pre-release
required_inputs:
  - best_bid_size
  - best_ask_size
  - ofi
output_column: ofi_autocorrelation
output_description: Autocorrelation of order flow imbalance with 1-period lag
tags:
  - autocorrelation
  - ofi
  - order_flow
  - correlation
  - time_series
  - microstructure
parameters:
  window:
    type: integer
    description: Window size for correlation calculation
    required: false
    default: 20
    constraints:
      min: 2
      max: 200
  lag:
    type: integer
    description: Lag for autocorrelation (e.g., 1 for correlation with previous value)
    required: false
    default: 1
    constraints:
      min: 1
      max: 50
formula: rolling_corr(ofi, lag(ofi, lag), window)