How the Alpha Score works

The Alpha Score (0–100) ranks how strongly a skin stands out across five measured market signals, compared with every other skin SkinsAlpha tracks at the same moment. It is not a probability of profit and not financial advice.

Alpha = 0.25 × Momentum + 0.25 × Liquidity + 0.20 × Value + 0.20 × Supply/Demand + 0.10 × Risk Quality

Each factor is a percentile from 0 to 100: a Momentum of 90 means stronger momentum than 90% of tracked skins.

The five factors

Momentum (25%). Price change over 7, 30 and 90 days from SkinsAlpha's own hourly history. Until we have 7 days of our own data for a skin, we use the marketplace's reported median sale prices for the last 7, 30 and 90 days, and say so on the skin page.

Liquidity (25%). How easy it is to buy and sell at a predictable price: 40% sales in the last 24 hours, 30% dollar volume over 30 days, 20% number of listings, 10% turnover (sales ÷ listings).

Value (20%). How far the cheapest listing sits below the skin's fair value. In v1 fair value is the 30-day median sale price. Listings far from recent sales (often a different pattern or float) are flagged and lower Confidence.

Supply / Demand (20%). Rewards a rising sales pace (last 7 days vs the 30-day average) and falling listings; penalises the reverse.

Risk Quality (10%). 100 = calmer. Penalises a wide 30-day price range, thin trading, sudden one-day spikes, few listings and a short history.

Confidence is separate

Confidence tells you how much to trust the score: history depth, number of marketplaces, number of sales, liquidity and data freshness. Bands: 0–39 Low, 40–69 Medium, 70–84 High, 85–100 Very High. A factor we can't measure honestly is shown as “n/a”, counts as neutral (50), and lowers Confidence. We never invent numbers.

Data

Prices and sales come from the official Skinport public API, refreshed every hour. Every snapshot is kept, never overwritten. Item images are Valve's official images served from Steam's CDN. Referral links never influence the score.

Does it work? The public backtest

Every day we store each skin's score and later record what the price did. Results are published here, good or bad.

Model version 1.0. Weights and methods may change; every change is versioned and dated here.