Case EV vs Case Price Dynamics

Investigation into how CS2 case prices and the expected value (EV) of their contents evolve over time. Tests whether case prices lead or lag the EV of their contents — an analogue to spot-forward basis or NAV deviations in traditional markets.

Quick Start

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

The dashboard works immediately — all analysis is precomputed into data/precomputed/.

The Signal

Each CS2 case contains skins at known rarity tiers with known unboxing probabilities. At any time t, the expected value of opening a case is:

EV(t) = Sum over items [ P(rarity) / N_tier * P(wear) * P(stattrak) * median_price(item, wear, t) ]

The basis is EV(t) - CasePrice(t) - KeyCost. When positive, opening is +EV. When negative, the case trades at a premium to its contents.

Key Findings

Finding	Detail
No case has positive EV	Every case costs more to open (case + $2.49 key) than the expected value of its contents
Best EV/Cost ratio is 0.60x	Huntsman Weapon Case is the least negative, returning ~$0.60 per $1.00 spent
Price leads EV in 27/42 cases	Speculative demand drives case prices first, contents adjust later
Fees kill short-horizon strategies	5% round-trip on Buff163 destroys any edge at < 14-day rebalance
7-day hold period is binding	Steam's trade lock eliminates short-term mean-reversion trades

No case is worth opening on EV

Cost = CasePrice + $2.49 key. All 42 cases are negative EV.

Case	EV	Cost	EV/Cost
Huntsman Weapon Case	$7.77	$13.03	0.60x
Falchion	$2.49	$4.32	0.58x
Snakebite	$1.75	$3.08	0.57x
Shadow	$2.43	$4.30	0.57x
...
eSports 2013	$11.00	$69.15	0.16x
CSGO Weapon Case	$22.40	$175.94	0.13x

Analysis Modules

The dashboard runs 10 quantitative analysis modules per case, per timescale:

• Core Statistics — correlation, return distributions, spread z-score
• Efficiency — lead-lag / Granger-like cross-correlation
• Volatility — rolling vol, AR(1), mean-reversion half-life
• Cross-Section — EV/Price ratio time series
• Hurst Exponent — trending vs mean-reverting regime classification
• Autocorrelation — return predictability structure
• Cointegration — ADF test, error-correction speed
• Regime Detection — structural break identification
• Trading Signals — z-score bands with buy/sell thresholds
• Liquidity — absolute return proxy for bid-ask spread

Regenerating from Raw Data

# Download raw price data (requires Google Drive access)
pip install gdown
python setup_data.py

# Regenerate all 42 case analysis JSONs
python src/precompute.py

Data

• Source: PriceEmpire API aggregating 70+ marketplaces
• Period: 2021-03-24 to 2026-03-24 (5 years daily)
• Cases: 42 (all major CS2 cases)
• Items: 2,565 skins, knives, and gloves across all cases
• Providers: 13-14 exchanges per item (Buff163, Steam, CSFloat, DMarket, etc.)

Part of CS2 Quant Research

Christian Garry — CS2 Quant Research Series