Cost to Hold · No. 0005 · Monday 17 August 2026 · six minutes

Prediction markets are the closest thing crypto has to a genuinely new instrument, and they have already made the same structural choice perps made. Today: what the volume data actually shows, why both venues are now quoting against their own users, what happened the last time an industry did that, and a bitcoin tape where one venue is paying longs 13.81% while another charges them 9.45%.

---

Prediction markets found the vital few. Now they are becoming the house.

Joseph Juran's useful idea was that in most systems a small number of causes produce most of the effect — the vital few against the trivial many — and that the discipline is knowing which is which before you allocate effort. Prediction markets have found their vital few. What matters is what happened to everything else.

Since July 2024, sports, politics and crypto have accounted for 91% of trading volume on Kalshi and 90% on Polymarket. Sports alone is 80% of Kalshi. Pew Research One Polymarket contract, the World Cup winner, has taken more than $3 billion by itself.

The promise was never sports. It was that anything with a resolvable outcome becomes a price: infrastructure timelines, drug approvals, policy dates, whether a specific bill passes by a specific Thursday. Those markets exist. Go and open one. You will find a spread you would not cross, a book three orders deep, and a resolution months out. It is priced and it is untradeable, which are different things.

That is not a failure of interest, it is what happens when liquidity has to come from someone. In election contracts the top three market makers supply roughly 70% of the liquidity, and 80% of volume trades within half a percent of mid. Trade The Outcome Where those three choose to quote, the market is excellent. Where they do not, there is no market — only a listing.

So the venue becomes the counterparty

Kalshi runs an internal desk, Kalshi Trading, that places bids on its own exchange. The stated purpose is supporting liquidity. A proposed class action alleges it sets lines that disadvantage customers. Polymarket has been recruiting for an internal market-making team that may trade against its own users. TechCrunch · coin360

There is no conspiracy in that, which is almost worse. If three firms supply most of the depth and none of them wants the long tail, the venue either leaves the tail unquoted or quotes it itself. Both are bad and one of them looks like growth. Every party behaves rationally and the counterparty on your obscure market becomes the house.

There is already a name for this, and we know what it does

In contracts for difference and retail foreign exchange, a broker does one of two things with your order. It can A-book it, passing the risk to the market and earning the spread. Or it can B-book it and keep the other side itself. If you lose, the broker keeps your money. If you win, it pays you from its own pocket.

Stated plainly: a B-book operator's revenue is its customers' losses.

That is not an accusation, it is the mechanism, and it is entirely legal. ESMA and the FCA both permit it, on condition the conflict is managed, exposure is monitored in real time, and the operator publishes what happens to its customers. Liquidity Finder

Which is where a number from a few issues ago comes back. European CFD brokers are required to print the share of their retail accounts that lose money, recalculated quarterly, and ESMA's own analysis put it at 74% to 89%, with average losses per client between €1,600 and €29,000.

That is the outcome distribution of a market where the house can take the other side. Not fraud, not a scandal — the arrangement working as designed, measured by the regulator that permits it, and printed on the advertising by law.

Prediction markets are moving toward that structure and print nothing at all.

The obvious conclusion is half wrong

The instinct is that users will see the house winning, feel it, and leave — that the whole category implodes.

The first half of that is right. The second half is not, and the CFD industry is the proof. B-booking did not kill contracts for difference. They are still legal, still large, and still sold to retail every day, under a banner saying most customers lose money. What actually followed was disclosure, leverage caps, and a slow migration of serious flow toward venues that do not take the other side.

So the realistic path is not collapse. It is a warning label and a split: people indifferent to their counterparty stay, people who care leave, and they take the flow that market makers actually want with them.

That is the thing to watch. Not whether prediction markets survive — they will. Whether the serious end of the order flow stays.

The alternative is not a better house. It is no house.

If the problem is that three firms will quote and none of them wants the tail, there are two answers. Have the venue quote it, which is where Kalshi and Polymarket are heading. Or make quoting the tail automatic.

XO Market is trying the second: anyone can create a market, against Kalshi and Polymarket deciding internally what gets listed. As reported, it is running 1,400-plus user-created markets and roughly $420M of cumulative volume off a $6M seed round. CoinDesk

The mechanism is the part worth looking at. Rather than waiting for a market maker, it uses a liquidity-sensitive LMSR — an automated maker that will quote any market at any time and deepen as participation grows. Nobody has to decide your obscure question deserves attention; the algorithm quotes it anyway.

That solves the listing problem. It does not make the tail free, and here is where it comes back to what this newsletter is about: an automated maker that always quotes does so by charging for it. Thin participation is a wider effective spread, paid on the way in and again on the way out, on a position you hold for months. The tail becomes tradeable at a price rather than untradeable at no price — better, and still a cost nobody puts on the ticket.

One thing to watch rather than applaud: resolution runs through an automated first-pass system with a jury layer for disputes. An automated resolver deciding whether you won is the same question this newsletter asked last week about automated execution — fine until it is wrong, and what matters then is whether there is a record of why it decided what it did.

What this has to do with perps

A perpetual and a prediction contract are closer relatives than they look.

They share the concentration. Ninety percent of prediction volume sits in three categories. On our own tape, bitcoin carries $15.63 billion of open interest across four venues against solana's $1.74 billion. Both depend on a small number of firms to make the tail tradeable, and both are learning those firms are selective.

They differ on expiry, and that difference is the whole subject. A prediction contract resolves. On a known date the thing either happened or it did not, and the price must meet reality. A perpetual never expires, so nothing forces convergence — which is exactly why funding exists. With no settlement to pull the contract back toward the index, a periodic payment does the pulling instead.

So both charge you to hold, and neither quotes it as a cost.

In a prediction market you pay by locking capital at zero yield until resolution. Buy a contract at 40 cents that settles in March and you have committed that money for seven months to earn nothing. Nobody calls it a fee. It is the largest cost in the trade and appears on no statement.

In a perp you pay funding, every eight hours, whether the position moved or not. Today that runs from −13.81% annualised on Hyperliquid to +9.45% on Binance for bitcoin — a twenty-three-point spread on one asset, with Hyperliquid paying longs while the other three charge them.

One instrument charges you visibly and quotes it badly. The other charges you invisibly and does not quote it at all.

The tape

Price and 24h change

	BTC	ETH	SOL
Mark	$64,038	$1,908	$75.98
24h	+1.35%	+1.21%	+0.64%

Funding, per 8h and annualised

	BTC	ETH	SOL
Hyperliquid	−0.0126% · −13.81%	+0.0088% · 9.63%	+0.0009% · 0.97%
Binance	+0.0086% · 9.45%	+0.0063% · 6.89%	−0.0025% · −2.68%
OKX	+0.0021% · 2.34%	+0.0026% · 2.87%	+0.0002% · 0.24%
Bybit	+0.0052% · 5.74%	+0.0043% · 4.69%	−0.0073% · −8.03%

Hyperliquid settles hourly; its rate is shown multiplied by eight so the column compares. Binance and Bybit are read from CoinGecko rather than directly, and their interval is assumed to be 8h — see the venue note below.

Open interest, basis, and change since Friday

	BTC	ETH	SOL
OI, Hyperliquid	$2.78B +2.9%	$1.74B +4.6%	$0.39B +2.1%
OI, Binance	$6.89B −3.8%	$4.57B +1.5%	$0.66B +2.1%
OI, OKX	$2.05B −9.9%	$1.35B −3.6%	$0.22B +1.3%
OI, Bybit	$3.91B −8.9%	$1.52B +5.6%	$0.47B +0.9%
Basis, Hyperliquid	−6.1 bps	−4.7 bps	−5.8 bps
Basis, OKX	−5.1 bps	−4.9 bps	−6.6 bps

Price rose across the board over the weekend while bitcoin open interest fell on three of four venues — positions closed into strength rather than added to it, the mirror of Friday.

Captured 2026-08-17 16:17 UTC · raw snapshot

Venue watch. Binance returns HTTP 451 and Bybit HTTP 403 to direct requests from this location — geographic blocks, not outages. Their columns come from CoinGecko and are marked as such, so the coverage gate records this issue as degraded: two venues first-party, two second-hand.

One question

If your venue is the counterparty to your trade, do you know it — and how did you find out? Reply and say. I will publish what comes back, without names.

The cost corner

A contract that is quoted is not the same as a contract that is tradeable, and the difference only shows up when you try to get out.

The Founder No. 01 — Your exchange is deleting your trading history

---

Not investment advice. Nothing here predicts anything. Numbers are pulled from venue APIs at the time stamped above and may be revised. Third-party figures are linked to their sources.

Written by Joe DeFeo from inside a live systematic book.