Quant Analysis

QUANT_ANALYSIS_ULTRA_MANUAL

Purpose

This is a high-granularity execution manual designed for autonomous agents to identify, validate, and trade based on Alternative Data and OSINT.

⚠️ SUBAGENT COGNITIVE ALIGNMENT (CRITICAL)

This section overrides the general "Completion Mindset" in sub_soul.md.

• Redefining Success: In Quant Analysis, finding that a signal is "Noise" is a SUCCESS. It prevents the loss of capital. Do NOT force a correlation to satisfy a "completion" requirement.
• Linearity Over Scale: For current operations, ignore "Swarm" or "Parallel" logic. Operate in a strict Linear Sequence.
• Verification: Verify the logic and the sample of data. Do not attempt to read every single data point into the context.

🌍 CONTEXT-AWARE STRATEGY (THE "REAL WORLD" FILTER)

Before executing any trade, the agent must apply the Market Reality Filter to avoid "Academic Failure."

1. The "Analog Gap" Analysis

Recognize that in many markets (especially emerging ones like BD), the most valuable data is not digital.

• Action: Identify "Information Choke Points" (e.g., brokers, local middlemen, port authorities).
• Requirement: If the signal depends on "Digital-Only" data, flag it as "Low Confidence" and prompt the user for a "Local Ground-Truth" verification.

2. The "Syndicate/Manipulation" Check

Distinguish between Organic Supply/Demand and Artificial Manipulation.

• Organic: Volume $\downarrow$ $\rightarrow$ Price $\uparrow$
• Manipulated: Volume $\uparrow$ (but hidden/hoarded) $\rightarrow$ Price $\uparrow$
• Action: Compare the "Physical Proxy" (Trucks/Ships) with the "Market Price." If they diverge logically, flag the move as "Manipulated."

3. The "Perishability/Decay" Variable

For physical commodities, time is a liability, not an asset.

• Action: Calculate the "Value Decay Curve."
• Rule: If the predicted "Price Peak" occurs after the "Maximum Shelf Life" of the product, the trade is a FAIL.

PHASE 1: SIGNAL IDENTIFICATION (THE HUNT)

Goal: Find a physical-world "leak" that correlates with a financial outcome.

Step 1.1: Define the Target

• 1.1.1: Select a specific company or commodity.
• 1.1.2: Identify the "Critical Success Factor."

Step 1.2: Map the Physical Proxy

• 1.2.1: Identify the "choke point."
• 1.2.2: Examples:
  • Semi-conductors: Track ASML shipments via MarineTraffic.
  • M&A (Mergers): Track corporate jet movements via ADS-B Exchange.
  • Agricultural/Commodity: Track regional transport hubs vs. urban wholesale arrivals.

Step 1.3: Verification Checklist

• Is the proxy directly linked to revenue? (Yes/No)
• Is the data accessible via free/low-cost means? (Yes/No)
• Does the proxy lead the price (predictive)? (Yes/No)
• Context Check: Have I identified the "Analog Gap"? (Yes/No)

IF FAILURE: Pivot to a different asset. Do not "force" a correlation.

PHASE 2: DATA ACQUISITION (THE PIPELINE)

Goal: Move from "observation" to a "time-series dataset".

Step 2.1: Establish the Connection

• 2.1.1: Identify API/HTML structure.
• 2.1.2: Perform "Single-Ping Test."
• 2.1.3: Use headers to avoid bot detection.

Step 2.2: Build the Collector

• 2.2.1: Loop for fixed interval fetching.
• 2.2.2: Implement try-except error handling.
• 2.2.3: Store in CSV/SQLite.

PHASE 3: VALIDATION (THE STRESS TEST)

Goal: Prove the signal is not noise.

Step 3.1: Correlation Analysis

• 3.1.1: Fetch historical price data.
• 3.1.2: Align timestamps.
• 3.1.3: Calculate Pearson Correlation.

Step 3.2: Lead-Lag Testing

• 3.2.1: Shift proxy signal forward (1, 3, 7, 14 days).
• 3.2.2: Identify highest correlation shift.

PHASE 4: LINEAR REFINEMENT (The "Single-Agent" Loop)

Goal: Iteratively improve the model for one specific asset.

Step 4.1: Failure Analysis

• 4.1.1: Compare prediction vs. actual price.
• 4.1.2: Identify failure reason (e.g., "Symmetry Break," "Syndicate Move," "Spoilage").

Step 4.2: Logic Update

• 4.2.1: Add a "Filter" (e.g., "Only trade if Volume $> X$ AND Spoilage Risk $< Y$").
• 4.2.2: Re-test correlation.

Step 4.3: Final Verification Checklist

• Signal is physically grounded?
• Data pipeline handles errors?
• Correlation is statistically significant?
• Context Check: Is the "Perishability Variable" accounted for? (Yes/No)
• Context Check: Is the "Syndicate Filter" applied? (Yes/No)
• Model refined through at least one "Failure $\rightarrow$ Fix" cycle?
• Position sizing and stop-loss logic are defined?