← PositionFlow.ai

Methodology

How PositionFlow computes positioning signals for ES futures. Every rule is explicit, every threshold is documented, and the full logic is open for audit.

1. Data source

PositionFlow uses the Traders in Financial Futures (TFF) report published weekly by the U.S. Commodity Futures Trading Commission (CFTC). Specifically:

Report TFF — Futures Only
Market S&P 500 Consolidated (13874+)
Covers E-mini + classic S&P 500 futures combined
Published Every Friday at 3:30 PM ET
As-of date Positions as of Tuesday close
History 2011 to present (~793 weekly reports)

The CFTC data is public domain. PositionFlow ingests it automatically every Friday and runs the signal engine within minutes of publication.

2. Why TFF, not Legacy COT

Most COT analysis uses the Legacy report. For ES futures, Legacy COT has a critical flaw: it groups hedge funds and pension funds together under "Non-Commercial." These two groups trade with completely different horizons and motivations.

Legacy COT

Non-Commercial = hedge funds + pension funds + CTAs, all in one bucket. Long-term structural positions dilute the crowding signal from fast money.

TFF Report

Separates Leveraged Funds (hedge funds, CTAs) from Asset Managers (pension funds, endowments). The separation is the signal.

Trader categories in TFF

Category Who Role in signal
Leveraged Funds Hedge funds, CTAs, managed money Primary signal. Fast money — crowded positioning precedes mean reversion.
Asset Managers Pension funds, endowments, insurance Context only. Structural, slow-moving. Confirms or opposes the LF extreme.
Dealer/Intermediary Banks, swap dealers Not used. Primarily hedging client flow.
Other Reportables Everyone else above reporting threshold Not used. Too heterogeneous for signal extraction.

3. Core thesis

When Leveraged Funds positioning reaches a historical extreme — measured by 5-year percentile rank — it signals crowding. Crowded positioning is unstable and tends to unwind, creating a directional lean for the opposite move.

This is a contrarian signal. When hedge funds and CTAs are positioned at extreme levels relative to the last 5 years, the trade is "crowded." Crowded trades unwind — sometimes gradually, sometimes violently.

The signal does not predict timing. It identifies a positioning environment where the probability of mean reversion is elevated. It's one input among many, not a standalone trading system.

4. Percentile computation

Raw position counts are meaningless without context. 500,000 Leveraged Funds net short contracts might be extreme in 2015 and normal in 2026 as open interest grows. Percentile rank normalizes across time.

How it works

For each weekly report, PositionFlow computes the percentile rank of the current Leveraged Funds net position relative to two lookback windows:

Window Lookback Purpose
52-week ~52 reports Recent context. Is positioning extreme relative to the past year?
5-year ~260 reports Primary signal window. Is positioning extreme relative to a full market cycle?

Percentile formula

percentile = (count of historical values ≤ current value) / (total historical values) × 100

A value at the 85th percentile means 85% of readings over that window were at or below the current level. A minimum of 10 data points is required; otherwise, the percentile defaults to 50 (neutral).

The same percentile computation is applied to Asset Manager net positioning over 52 weeks, used for the institutional context layer.

5. Signal rules

The signal engine evaluates a fixed set of rules in order. Every rule that fires is included in the rules_triggered array so you can audit exactly what produced the signal.

Step 1 — Primary signal (Leveraged Funds 5yr extreme)

Condition Result Rule tag
LF net percentile (5yr) ≥ 80 bearish_lean lev_funds_net_above_80th_percentile_5yr
LF net percentile (5yr) ≤ 20 bullish_lean lev_funds_net_below_20th_percentile_5yr
Between 20 and 80 neutral No rules triggered

The 5-year window is the primary trigger. If positioning is within the 20th–80th percentile range, no signal is generated regardless of other factors.

Step 2 — 52-week alignment

If a 5-year extreme is detected, the engine checks whether the 52-week percentile confirms it:

Condition Rule tag
Bearish lean + LF 52wk percentile ≥ 80 lev_funds_net_above_80th_percentile_52wk
Bullish lean + LF 52wk percentile ≤ 20 lev_funds_net_below_20th_percentile_52wk

52-week alignment means the extreme isn't just unusual over 5 years — it's also unusual in the recent context. This increases confidence.

Step 3 — Asset Manager context

Asset Manager positioning is evaluated on a 52-week percentile basis to determine whether institutional money confirms or opposes the Leveraged Funds extreme:

AM 52wk percentile During bearish_lean During bullish_lean
< 35 (AM positioned short) Opposing (strengthens reversal) Supportive
35 – 65 Neutral — no context signal
> 65 (AM positioned long) Supportive Opposing (strengthens reversal)
Why opposition strengthens the signal: When Leveraged Funds are crowded in one direction and Asset Managers are positioned the opposite way, the fast money (LF) is out on a limb without structural support. This increases the probability and potential magnitude of mean reversion.

6. Signal strength

Signal strength is determined by how many confirming factors align. It escalates in three levels:

Strength Requirements
Low 5-year extreme detected only. No 52-week alignment, no Asset Manager context.
Medium 5-year extreme + Asset Manager opposition (or support).
High 5-year extreme + 52-week alignment + Asset Manager opposition (or support). All three factors firing together.

When positioning is in the neutral zone (20th–80th percentile), signal strength is none and no bias is assigned.

Pipeline summary

1 Check LF net 5yr percentile against thresholds (20 / 80)
↓
2 If extreme: check LF net 52wk percentile alignment
↓
3 Evaluate Asset Manager 52wk percentile for context
↓
4 Compute strength from rule count → output signal

7. API output fields

Every API response includes these signal fields. Nothing is hidden — what fires is what you see.

Field Values Description
bias neutral, bullish_lean, bearish_lean Directional lean from LF positioning extreme
signal_strength none, low, medium, high How many confirming factors align
signal_type no_signal, leveraged_fund_extreme What generated the signal
institutional_context asset_managers_neutral, ..._opposing, ..._supportive Asset Manager positioning relative to LF extreme
rules_triggered Array of rule tags Every rule that fired, for full auditability
extreme_flag true / false Whether a 5yr extreme was detected
extreme_side null, leveraged_funds_long, leveraged_funds_short Which side is crowded
signal_version v1 Bumped when thresholds or logic change

8. Thresholds reference

All thresholds used by the v1 signal engine, in one place:

Parameter Value Used for
LF extreme high 80th percentile Crowded long detection (bearish_lean)
LF extreme low 20th percentile Crowded short detection (bullish_lean)
AM neutral low 35th percentile Below this, AM context is non-neutral
AM neutral high 65th percentile Above this, AM context is non-neutral
52-week window 52 reports Recent percentile context
5-year window 260 reports Primary signal percentile window
Minimum data points 10 Below this, percentile defaults to 50

When any of these thresholds change, signal_version is bumped so consumers can detect the change.

9. Limitations

This is not investment advice. PositionFlow provides positioning context as one input for your analysis. It is not a standalone trading system and should never be treated as one.