← All writing

Markets9 min read

Trading the gap in CS:GO skins

The first market I ever traded was a video game's. How instant-cashout sites created a knowledge gap, how I later played spreads between North American and Chinese marketplaces, and why the setup itself was the moat.

Long before order books and perpetual futures, the first market I traded was made of gun skins.

In Counter-Strike: Global Offensive (now Counter-Strike 2), weapons can have cosmetic skins. They do nothing for gameplay, but they're rare, tradeable between players, and some are worth more than a car. That combination produced a real market, with real liquidity, real spreads, and very uneven information. This post is about how I found the gap in it, twice.

How the market is shaped

That one detail, Steam money being stuck in Steam, creates most of the structure:

  • The Steam Market has the most visible prices, but those prices are in a currency you can't cash out, after a cut taken on every sale. The cut is 15% on top of what the seller receives (a 5% Steam fee plus a 10% Counter-Strike fee), so a seller keeps about 87% of the listed price. Steam also caps a wallet at $2,000 and a single listing at about $1,800, so the most expensive skins can't trade there at all: their prices are set off Steam, by definition.
  • Instant-trade and cashout sites, CS.Money being the best known at the time, ran bots that would take your skin immediately and give you another skin or cash. Instant liquidity, at a spread.
  • Peer-to-peer marketplaces let players sell to each other for real money, slower but closer to fair value.
Where the same skin can be pricedEach venue offers a different trade-off between speed, cash-out, and price. The gaps between them are the opportunity.

Round one: the knowledge gap

In high school, the first edge I found wasn't technical. It was informational.

Most players thought of a skin's value as "what it says on the Steam Market" or "what an instant site offers me right now". Far fewer compared those against each other. The instant sites were valuable to people who wanted liquidity now, and they priced that convenience in. Some players wanted a specific skin and would overpay to get it immediately. Others wanted out and would sell below what a patient buyer would pay.

So the trade was straightforward to describe:

  • buy from people who valued speed over price
  • sell to people who valued a specific item over price
  • know the prices across venues better than either side did

An instant site is, mechanically, a market maker. Its bots quote a price to buy your skin and a higher price to sell you one, and the difference between those two quotes pays for the inventory the site holds and the risk that prices move while it does. Anyone who sells into the bot crosses that spread. Buying from that player at a price between the bot's two quotes, and selling to someone who would otherwise have bought from the bot, leaves both of them better off than the bot would have, and the space between the quotes is where my margin came from.

I started by doing it by hand, then wrote scripts to watch prices across sites and flag when the same item was meaningfully cheaper in one place than another. The edge wasn't speed. It was simply knowing more than the person on the other side of the trade.

Round two: the cross-border spread

Years later, in university, I came back to the market with better tools and a bigger idea.

By then the most important prices weren't in North America at all. Buff163 and YouPin are Chinese marketplaces with enormous volume, and a lot of price discovery for high-value skins happens there. CSFloat had become the common peer-to-peer marketplace for North American traders. The same item could sit at meaningfully different prices on each, and because skins move through Steam, an item bought on one could be sold on another.

The catch is that the Chinese marketplaces aren't built for people outside China. Getting set up meant:

  1. Step 1A Chinese phone numberThese platforms are built for users in mainland China, so registration and verification are tied to a mainland Chinese phone number and Chinese services that a North American setup doesn't cover.
  2. Step 2An Alipay accountDeposits and withdrawals run through Alipay, which needs its own verification before it can move money in and out.
  3. Step 3Marketplace accounts and verificationEach marketplace has its own account, its own verification, and its own rules for listing and withdrawing.
  4. Step 4Currency and settlementPrices are in yuan, so every spread has to be computed after currency conversion and each platform's fees.
  5. Step 5Steam's own constraintsItems move through Steam trades, with trade holds and protections that lock an item for a period after it moves.

That list is the real reason the spread existed. None of the steps is hard on its own. Together, they were enough friction that most North American traders simply didn't bother. The setup was the moat.

Automating it

Once the accounts existed, the rest was an engineering problem, and a satisfying one:

  • Price feeds from each marketplace, normalized into one currency and one item naming scheme, since the same skin is described slightly differently on each site
  • Net spreads, not headline spreads. A gap only counts after both platforms' fees, the currency conversion, and the cost of having money tied up while an item is locked
  • Holding time as a cost. Because items can be locked after a trade, a spread that looks great but takes a week to realize competes with everything else that capital could do in that week
  • Execution automated across listing, buying, and moving items, so the process ran without me babysitting each trade
One cross-market trade, end to endA spread only counts once it survives fees, currency conversion, and the time the item is locked.

Netting a spread

"Net spread" is easy to say and easy to get wrong, so it's worth writing out. Every venue takes its cut at a different point, some from the seller's proceeds and some on the way out to a bank, and the currency conversion sits in the middle. These are the published fees at the time of writing (fees change, and they weren't always these):

15%Steam Market fee, on top of what the seller receives, paid in wallet credit
2% + 0.5 to 2.5%CSFloat seller fee, plus a withdrawal fee that falls with sales volume
2.5% + 1%Buff163 seller fee, plus withdrawal through Alipay
1%YouPin seller fee
7 daysSteam Trade Protection on received items, since July 2025

Call the buy price B and the sell price S, both in the same currency. Each fee that comes out of the proceeds multiplies them by (1 - fee), so the trade only breaks even when the gross gap clears the product of all of them:

net proceeds = S x (1 - seller fee) x (1 - withdrawal fee) x (1 - FX cost)
break-even:    S / B >= 1 / ((1 - seller fee) x (1 - withdrawal fee) x (1 - FX cost))

CSFloat to Buff163, fees only: 1 / (0.975 x 0.99)  - 1 ≈ 3.6%
Buff163 to CSFloat, fees only: 1 / (0.98 x 0.975)  - 1 ≈ 4.7%  (top withdrawal tier)
                               1 / (0.98 x 0.995)  - 1 ≈ 2.6%  (bottom tier)
Into the Steam Market:         1.15 - 1             = 15%, and the proceeds aren't cash
Gross gap needed just to cover the feesDerived from the published fees above, before any currency conversion cost or holding cost. YouPin's figure is its seller fee only.
Show data
routeBreak-even gross gap
Into Steam Market15%
Buff163 to CSFloat (top tier)4.7%
CSFloat to Buff1633.6%
Buff163 to CSFloat (bottom tier)2.5%
CSFloat to YouPin (seller fee only)1.0%

The Steam bar is why the cross-border trade happened between the cash marketplaces and not through Valve's: a 15% fee wipes out any gap smaller than itself, and what's left is wallet credit.

To see how much a headline gap shrinks, take an illustrative trade (made-up prices, real fees): buy a skin on CSFloat for $1,000, sell it on Buff163 for the yuan equivalent of $1,050, and assume converting the yuan back costs 1%.

sell on Buff163         $1,050.00
- 2.5% seller fee       $1,023.75
- 1% Alipay withdrawal  $1,013.51
- 1% FX (illustrative)  $1,003.38
net gain                $3.38 on $1,000, about 0.3%

A 5% gap on the screen is 0.3% in hand. That's the arithmetic behind "net spreads, not headline spreads": the scanner ranked items on the bottom line, not the top one.

Holding time is a cost, and it's also a risk. The money isn't free while the item is in transit. Under the Trade Protection rules Valve shipped in July 2025, an item received in a trade can't be traded, listed, or sold for 7 days, and the sender can reverse the trade inside that window. So one round trip ties up capital for at least the lock, plus the time to sell, plus any wait for a marketplace to pay out a sale that could still be undone. Two consequences:

  • A spread has to be compared per day, not per trade. As an illustration, a 2% net that takes 10 days is 0.2% per day of capital tied up; a 1% net that clears in 2 days is 0.5% per day. The bigger headline is the worse trade.
  • The price can move while you hold it. Price risk over a holding period grows roughly with the square root of its length. As an estimate with an illustrative 1% daily volatility, a 7-day hold carries about a 2.6% one-standard-deviation move (1% x sqrt(7)), eight times the 0.3% net in the example above. A spread is only worth taking when the net is large next to that number, or the item is one whose price barely moves.

Depth caps the size. The rich side's price is the price of its best listing or bid, not of every unit. Selling five of the same skin walks down the other side's book exactly as it would on any exchange, so the realised spread on the fifth unit is smaller than on the first. The opportunity is real, but it doesn't scale linearly with capital, which is part of why it stayed open.

Where a headline gap goesEach layer comes off the gross gap. What's left has to beat the price risk of holding the item through the lock.

What it taught me

Looking back, almost everything I do now has a version that started here:

  • The quote is not the fill. The price on the screen and the price you actually get are different numbers, and the difference is where the work is.
  • Friction is a moat. Some opportunities exist not because they're hard to see, but because they're tedious to reach. Being willing to do the boring setup is an edge.
  • Net everything. Fees, currency, time locked up. A spread that doesn't survive its costs isn't a spread.

It's the same instinct that later shows up in ranking bridges by what actually arrives, and in charging every cost per trade in my own book.