What "r=0.77" between CS2 player count and skin prices really means
Most articles about CS2 skin economics cite a single correlation number — usually r=0.77 — as evidence the market is rational and player-driven. But there are three different ways to measure that correlation, and they tell wildly different stories. This post unpacks all three with 6 years of monthly snapshots, and explains why the rolling correlation has been near zero for the last 18 months.
Why correlation gets cited (and misused) here
The CS2 skin market lacks a clean fundamental anchor. Equities have earnings; bonds have interest payments; commodities have industrial demand. Skins have... vibes. So when someone needs a credibility prop, they reach for player count — the most obvious measurable variable in the gaming side of the equation — and report a correlation with skin prices. The headline number that keeps circulating is r ≈ 0.77, with the implied story: more players → more demand → higher prices, therefore the market is rational.
The problem is that "correlation between player count and skin prices" is ambiguous. There are at least three legitimate ways to compute it, and they don't agree:
- Levels correlation: Pearson(player_count_t, skin_index_t) — both as raw monthly numbers
- Returns correlation: Pearson(player_change_t, skin_return_t) — month-to-month percent changes
- Rolling correlation: 12-month windowed Pearson, recomputed each month — captures time-varying relationship
Skin Trackers publishes all three. Their values today, computed against the STI 500 (broad market index) and SteamCharts monthly average player count:
| Measure | Correlation (r) | What it captures |
|---|---|---|
| Levels | ~0.82 | Long-term trend co-movement |
| Returns (month-over-month) | ~0.26 | Actual co-movement in noise |
| Rolling 12m on returns | ~0 (recent 18mo) | Time-varying relationship |
Three numbers, three stories. Conflating them — usually by quoting only the levels number — produces noise dressed up as analysis.
What r=0.82 in levels actually means
When you correlate two upward-trending time series, you almost always get a high correlation. It's a statistical artifact of the trend, not evidence of a dynamic relationship. A famous example: US per-capita cheese consumption correlates with the number of people who died by becoming tangled in their bedsheets, at r ≈ 0.95 (Tyler Vigen's spurious correlations). Both went up in the 2000s. Cheese isn't killing people.
For CS2 specifically, the levels correlation captures three overlapping trends:
- 2020-2021: Pandemic + QE → both gaming participation and alt-asset speculation surged
- 2022-2023: Player count stagnated; skin prices stagnated. Co-movement in flatness.
- 2024-2026: CS2 launch doubled player count; broad skin market grew but at much lower velocity. Correlation in trend, divergence in magnitude.
The 0.82 number says "both grew over 6 years." It does not say "players drive skin prices." That inferential leap requires the returns-level analysis.
What r=0.26 in returns tells us
When you compute Pearson on month-over-month changes, you get a much weaker number — about 0.26. This is the correlation that an investor actually cares about: when player count moves up 5% in a month, do skin prices also move up that month?
A correlation of 0.26 means about 7% of month-to-month variance in skin returns is explained by player count changes (r²). The other 93% is something else — Operation drops, tournaments, currency moves, sticker capsule revaluations, contraband-tier rallies, F2P weeks, structural decay of discontinued cases. The skin market has its own dynamics, and player count is one weak input among many.
To put this in context: r = 0.26 is roughly what you'd see between any two assets that share macro exposure but aren't directly related (e.g., gold and oil: Fed FEDS paper 2017 shows their returns correlation around 0.20-0.30 in normal regimes). It's real but it's noise-level for single-asset trading decisions.
The rolling correlation: when the relationship broke
Computing a 12-month rolling Pearson on returns reveals something the static numbers hide:
- 2021-2022: Rolling correlation hovered around 0.5-0.7 — the strongest period. Both player base and skin market responded to the same QE- era liquidity drivers.
- 2023: Started declining as the bull cycle ended for skins (Selic up, global QT) but players stayed flat.
- Late 2023 onwards: CS2 launch. Player count nearly doubled overnight (~1M to ~1.8M peak). The discontinuity broke the historical relationship — rolling correlation collapsed to near 0 and has stayed there.
This is a structural break, not noise. The post-CS2 skin market developed its own internal dynamics (new collectibles, contraband revaluation, secondary market maturation) that decoupled from raw player count. Pre-CS2 player count was a useful weak predictor; post-CS2 it isn't.
The lurking variable problem: was it ever "causal"?
Even the 2021-2022 rolling correlation of 0.5-0.7 may not reflect a direct player-to-price causal channel. The more parsimonious explanation is global liquidity as a common driver:
- QE era (2020-2021) → cheap money → gaming spending up + alt-asset speculation up
- QT era (2022-2024) → tight money → discretionary spending down + risk assets compressed
Skin prices and player count both correlate with global liquidity. They look like they correlate with each other only because they share that common driver. Disentangling requires controlling for liquidity (Fed balance sheet, BRL/USD, M2, credit spreads) — which most CS2 economy writeups don't do.
The Skin Trackers methodology page documents this caveat explicitly. We publish the correlations because they're useful descriptive statistics, but we don't claim them as causal evidence. See /en/methodology for the full disclosure.
What this means for someone analyzing the skin market
- Never quote a single correlation number without specifying levels, returns, or rolling. The three are not interchangeable.
- Treat the 0.82 in levels as descriptive, not predictive. Both series trended up. So did your house value, college tuition, and the S&P 500. Co-movement in trend is not a relationship.
- Use the 0.26 in returns for allocation decisions. If you're thinking about skins as a diversifier in your portfolio, this is the number that matters. Low correlation with traditional assets is the genuine portfolio benefit.
- Watch the rolling correlation. If it starts trending up again, the market may be reverting to player-count sensitivity. If it stays flat, the market has matured into its own dynamics. Either way, the rolling number is the leading indicator.
- Don't mistake correlation for causation. The most likely common driver is global liquidity, not player count itself.
Limitations and caveats
- Sample size: 73 monthly snapshots is decent but not large. Confidence intervals on the rolling correlation are wide. Take specific values with ±0.15 uncertainty.
- Player count proxy: SteamCharts publishes monthly average concurrent players, which is one of several reasonable measures (peak players, unique daily users, hours played). Each gives a different correlation. Skin Trackers uses SteamCharts because it's public and consistent over the full window.
- Index choice: correlations vary across STI tiers. STI 500 (broad) shows the cited numbers. STI 30 (blue chip) shows weaker correlation; STI Cases (containers, disjoint universe) shows much weaker correlation with players.
- Structural breaks: the CS2 launch in late 2023 is a known regime change. Pooling pre-CS2 and post-CS2 data gives misleading averages. Where possible, segment the analysis.
Frequently asked
What is correlation in plain English?
Correlation is a number between -1 and +1 that tells you how strongly two variables move together. +1 means perfect positive (when X goes up, Y always goes up by a proportional amount). -1 means perfect negative (X up, Y down). 0 means no linear relationship. The most common measure is Pearson's correlation coefficient (r), introduced by Karl Pearson in 1895. Critically, correlation does not imply causation — two variables can correlate strongly because of a third lurking variable, sample bias, or pure coincidence.
Why is the r=0.77 number quoted everywhere for CS2 skins?
Because it's the easiest narrative: 'more players → more demand → higher skin prices'. The number originated from a few academic-style writeups around 2024 and has been recycled in popular skin-economy commentary ever since. The problem is that 'r=0.77' is computed by correlating the absolute level of player count against the absolute level of skin prices — both of which trended upward over 2020-2026. Strong trends will manufacture correlation between any two upward-trending series. The deeper question is whether the month-to-month changes correlate, and there the picture is much messier.
What's the difference between correlation in levels vs in returns?
Levels correlation looks at the raw values: did the player count and skin index both rise over the same window? Returns correlation looks at month-to-month percent changes: when player count went up 5% in a month, did skin prices also go up that month? Levels correlation is dominated by long-term trend co-movement; returns correlation captures actual co-movement in the noise. For investors, returns correlation is far more useful — knowing two assets trended up over 6 years doesn't tell you anything about diversification or hedging benefits.
Why did the rolling correlation between players and skins drop to ~0 recently?
Two structural changes broke the historical pattern. First, the CS2 launch in late 2023 brought a permanent regime shift — the player base nearly doubled overnight, a discontinuity that breaks correlation calculations. Second, post-launch the skin market developed its own internal dynamics (new collectibles, contraband revaluation, secondary market maturation) that decoupled from raw player count. The rolling-12m correlation went from ~0.6 in 2022 to near 0 by late 2025. The market is no longer a simple function of player count.
Does correlation imply causation in the skin market?
No, and the CS2 skin case illustrates exactly why. The plausible causal story is 'more players → more demand → higher prices', but the correlation is also consistent with 'global liquidity drove both player count growth (gaming spending) AND skin price growth (alt asset speculation)'. That third variable — global risk appetite during the 2020-2021 QE era — is the more parsimonious explanation. Player count and skin prices both correlate with global liquidity, not necessarily with each other. Disentangling requires controlling for liquidity, which is exactly what skin economy studies don't do.
Should I use correlation to time my skin trades?
Not directly. Even at peak (r=0.82 in levels, ~2022-2023), correlation tells you about co-movement, not predictability. A correlation of 0.82 in levels still leaves 33% of variance unexplained — and importantly, levels correlation says nothing about whether the next month will move together. For trade timing, returns correlation matters, and it sits around 0.26 — meaning ~93% of month-to-month skin price variance is unrelated to player count changes. The signal is real but small. Better predictors of short-term price moves: tournament events, Operation drops, F2P weeks, currency moves.
How do you compute these correlations on the Skin Trackers platform?
Three values, all computed from the same underlying data. Levels: Pearson(monthly_players, monthly_STI_500_index_value) over the 73 snapshots since April 2020 — yields r ≈ 0.82. Returns: Pearson(month_t/month_(t-1) - 1 for both series) — yields r ≈ 0.26. Rolling-12m: 12-month windowed Pearson on returns, evaluated at each month — recent values near 0. Code is open-source in src/lib/correlation.ts; raw data via /api/indices and SteamCharts public API. Reproducibility is the point.
What's the takeaway for someone analyzing the CS2 skin market?
Three things. First, never quote 'r=0.77' without specifying which correlation — levels, returns, or rolling. They tell different stories and conflating them produces noise dressed up as analysis. Second, treat the 0.82-in-levels correlation as evidence of long-term co-movement (both trended up) but not as evidence of dynamic relationship. Third, the rolling-correlation collapse post-CS2 launch is a regime change worth modeling — pre-2023 player count was a useful weak predictor; post-2023 it isn't. Build your analytical frame on the 3-measure decomposition, not on a single number.
Where to go from here
- See the methodology behind the three correlation measures at /en/methodology — with formulas and edge cases.
- All raw STI snapshots and player count series are public via /api/indices. Reproduce these correlations in your own notebook.
- For risk-adjusted return analysis (Sharpe ratio applied to CS2 skins), see /en/blog/sharpe-ratio-cs2-skin-investors.
- For the practical "is my portfolio underperforming the market" angle, see /en/blog/skin-portfolio-vs-sti-30.
Methodology note: correlations computed via Pearson coefficient. Rolling correlation uses 12-month windows on monthly returns. Sample: 73 monthly observations from April 2020 to April 2026. Source code in src/lib/correlation.ts (open), raw data via /api/indices and SteamCharts public API.