CS2 Portfolio Efficient Frontier

Mean-variance optimization across every tradeable CS2 item. Applies Markowitz portfolio theory to 22,100 digital assets spanning skins, knives, gloves, and cases with 5 years of daily multi-exchange price data.

Quick Start

python -m http.server 8000
# Open http://localhost:8000

The dashboard works immediately using precomputed data in data/precomputed/.

What It Computes

1. Per-item statistics (22,100 items)

For every skin+wear combination with sufficient liquidity:

• Annualized return and volatility
• Sharpe ratio
• Maximum drawdown
• Skewness and kurtosis
• Last observed price

2. Covariance estimation

With 22,100 items and ~1,800 trading days, the sample covariance matrix is severely rank-deficient. We use Ledoit-Wolf shrinkage to produce a well-conditioned estimate, then cap the frontier computation at the top 500 items by Sharpe ratio.

3. Efficient frontier (80 points)

Long-only mean-variance optimization solved at 80 target return levels using sequential quadratic programming.

4. Special portfolios

Portfolio	Return	Risk	Description
Equal Weight	-0.1%	0.7%	Naive 1/N allocation across all items
Min Variance	-0.1%	0.5%	Minimum risk portfolio
Max Sharpe	0.2%	0.9%	Tangency portfolio (Sharpe = 0.21)
Risk Parity	-0.1%	0.7%	Equal risk contribution

5. Correlation analysis

• Sector-average correlations (rifle-rifle, knife-pistol, etc.)
• Most and least correlated pairs
• Identifies diversification opportunities

Item Distribution

Type	Count	Description
Rifle	2,102	AK-47, M4A4, AWP, etc.
Pistol	1,865	Glock, USP-S, Desert Eagle, etc.
Knife	1,651	Karambit, Butterfly, Bayonet, etc.
SMG	1,298	MAC-10, MP9, P90, etc.
Shotgun	720	Nova, XM1014, MAG-7
Glove	358	All glove types
MG	228	M249, Negev
Other	13,878	Cases, stickers, patches, agents, etc.

Regenerating from Raw Data

# Option A: Point directly at local CSGO data warehouse
python src/precompute_frontier.py --prices-dir /path/to/CSGO/Data/processed

# Option B: Download from Google Drive (requires access)
pip install gdown
python setup_data.py
python src/precompute_frontier.py --prices-dir data/prices/processed

# Options
python src/precompute_frontier.py --min-days 300 --max-items 500 --skip-st

Data

• Source: PriceEmpire API — 70+ marketplace aggregation
• Period: 2021-03-24 to 2026-03-24 (1,804 trading days)
• Items scanned: 2,656 unique skins (25,193 item+wear combos before ST filter)
• Items in frontier: 500 (top by absolute Sharpe after Ledoit-Wolf shrinkage)

Part of CS2 Quant Research

Christian Garry — CS2 Quant Research Series