NVIDIA Liquidity Alpha: Alternative Data Factor Model

Project Overview

This project implements an institutional-grade factor model to predict Liquidity Regimes for NVIDIA (NVDA) using alternative data. By analyzing search volume across Retail, Consumer, and Enterprise clusters, the model isolates a latent "Retail Hype" signal that acts as a leading indicator for trading volume.

Key Results (Alpha)

• Information Ratio (IR): 1.64 (Investable Grade > 1.0)
• Global IC: 0.46
• Regime Efficiency: The signal exhibits high conviction (Correlation > 0.69) approximately 46% of the time.

Methodology

1. Data Engineering: Aggregated 15 distinct Google Trends keywords into a unified dataset, aligned to Weekly-Friday financial reporting frequency.
2. Signal Processing: Applied a Hodrick-Prescott Filter ($\lambda=14400$) to de-trend the search data and isolate cyclical demand shocks.
3. Dimensionality Reduction: Utilized PCA (Principal Component Analysis) to extract the First Principal Component (PC1), effectively filtering idiosyncratic noise and isolating the common "Retail Liquidity Factor."
4. Performance Attribution: Calculated rolling Information Coefficients (IC) to identify predictive regimes.

Tech Stack

• Language: Python
• Libraries: statsmodels (HP Filter), scikit-learn (PCA), pandas, yfinance
• Techniques: Unsupervised Learning, Time-Series Filtering, Factor Analysis

How to Run

1. Clone the repository:

   git clone [https://github.com/YOUR_USERNAME/NVIDIA-Liquidity-Alpha.git](https://github.com/YOUR_USERNAME/NVIDIA-Liquidity-Alpha.git)

2. Install dependencies:

   pip install -r requirements.txt

3. Run the model:

   python model.py

Factor Loadings (Why it works)

The model identified that "Retail Speculation" keywords (e.g., Buy NVDA, Jensen Huang) are significantly stronger predictors of liquidity than "Enterprise" keywords (e.g., H100), validating the behavioral finance hypothesis of retail-driven volume.

---

Author: Mohamad El Jouzou