The other side of the trade

A stock token is worth its stock only because someone keeps it there. That someone is an arbitrager, who needs two tools: money, and stock to borrow. Robinhood Chain has always had the first. Borro supplies the second. Here is why, with the numbers we could measure.

Two directions, two tools. When a token trades below its stock, the fix is to buy it. That takes only money, and anyone with USDG can do it in seconds. When a token trades above its stock, the fix is to sell it. If you already hold some, you can. If you do not, you have to borrow it first. That is why every stock market runs on a loan desk. At the end of 2025, €41.5 trillion of securities were available to lend worldwide and €3.5 trillion were on loan, and lenders were paid nearly $15 billion that year. Robinhood Chain had the stocks and no desk.

What the pools show. We compared the pool Borro trades through for each of its nine stocks with that stock's Chainlink price, every hour from August 15 to October 15, in market hours only. The feed itself moves only on a 0.5% change, so we counted a gap only when it was wider than that plus the pool's own fee: a gap someone could actually have traded. Over 8,554 stock-hours, the tokens sat above the feed 56% of the time. Gaps past that line came up for 86 hours on the expensive side and 12 on the cheap side. No cheap gap lasted more than two hours. Expensive ones ran for up to seven. On September 4 the GME token sat between 0.6% and 3.3% above GME for seven straight hours, while Hyperliquid's GME perp agreed with the feed.

Today the gap is mostly small. Most of those hours came in the pools' first month. From September 15 to October 15 the pools tracked their feeds closely: the token sat above the feed in 2,473 stock-hours and below it in 2,414, and gaps past the line added up to 9 hours rich and 3 cheap. That is good news, but not the whole story. On October 14, MSTR fell from about $151 to $144 between 13:45 and 16:15 UTC, on Chainlink's feed and on Hyperliquid's MSTR perp alike. The token's pool did not follow. From 14:15 to 17:00 UTC you could sell MSTR tokens into it for 1.0% to 2.1% more than the stock was worth, after the pool's fee. In that window the pool took 23 buys worth $14,145 and only 5 sells. Closing the gap would have taken selling about $64,000 of MSTR. Borro's MSTR vault had $520 to lend.

What a premium costs. Whoever buys at a premium loses it when it closes. Anything that values the token at its pool price marks it too high, so a loan against it rests on a price that is not there. And a premium held up by a thin pool falls harder when it breaks.

The same gap, on perps. Stock perps for Robinhood Chain users trade on Lighter and Arcus, and their funding shows which way the market leans. Over the 30 days to October 15, on Lighter's seven perps that match Borro's stocks, longs paid shorts in 5,018 of 5,040 hours. Outside the quiet band where its funding rests, at 0.0004% an hour, the perp sat above its index for 1,467 hours and below it for 26. On Arcus's nine, funding ran above its base rate, SOFR plus 0.5% or about 4.4% a year, in 3,044 hours and below it in 1,780.

A perp market maker who sells to a crowd of buyers ends up short the perp. The hedge is to buy the token, which takes only money. When the crowd sells, the maker ends up long the perp, and the hedge is to sell a token it does not own. With nothing to borrow, it widens its quotes or carries the risk. The hours when shorts paid were rare and sharp: Lighter's worst COIN hour ran at the equivalent of minus 100% a year, and Arcus's worst PLTR hour at minus 115%. That is the pattern you would expect when one hedge is missing.

What the research says. Markets have run this experiment. When the US banned short sales of financial stocks in September 2008, shorting in the large banned stocks fell about 77%, and their relative effective spreads widened by 32 basis points on average. Options market makers, expecting to lose short selling as a hedge, nearly stopped quoting, and spreads on options on banned stocks ran more than $0.96 wider than their controls. Across 30 countries, the bans of 2008 and 2009 hurt liquidity, slowed price discovery and were at best neutral for prices.

What Borro adds. On Borro, a trader borrows real stock tokens from the lenders' vault, sells them into the pool and holds the dollars, in one transaction. That sale is the missing direction. The sale money and the trader's deposit sit in the Steakhouse USDG vault while the short is open, earning about 4.5% a year at today's rate. For a perp desk the hedge now works both ways. When longs pay, short the perp, hold the token and lend it on Borro for rent. When shorts pay, go long the perp and short the token on Borro. The rent a shorter pays goes to the lenders, less Borro's 10%: the price of the other side of the trade.

The fine print. The peg holds only as deep as the lending supply. Today Borro's nine stock vaults hold about $6,860 of stock, and almost none of it is on loan; against a gap like MSTR's, that is a rounding error. Rent floats with demand. A short needs enough collateral or it is closed. Weekend hours, when the feeds pause and a pool's move is news rather than a mispricing, are left out of every count above. Nothing in this note is advice.

Sources. Pool prices and feeds: Borro's own read-only calls to Robinhood Chain (chain id 4663), hourly from August 15 to October 15, 2026, every 15 minutes from October 5, and the MSTR pool's depth and trades on October 14. Perp prices: Hyperliquid's public API. Funding: Lighter's API and funding docs for its Robinhood Chain instance and Arcus's API and funding docs, September 15 to October 15, 2026. Steakhouse USDG rate: DefiLlama, October 15, 2026. Securities lending: ISLA Securities Lending Market Report, H2 2025. The 2008 ban: Boehmer, Jones and Zhang, Shackling Short Sellers, Review of Financial Studies, 2013; Battalio, Mehran and Schultz, Market Declines: What Is Accomplished by Banning Short-Selling?, New York Fed, 2012; Beber and Pagano, Short-Selling Bans Around the World, Journal of Finance, 2013.