The weekly net flow figures of Bitcoin ETFs are commonly interpreted as a barometer of institutional conviction in Bitcoin. But week by week, they measure something else: a hidden interest-rate trade that is repeatedly turned on and off. IOSG Ventures published a detailed analysis explaining how to identify this trade, its true size, and why it is quietly leaving the market.
What is cash-and-carry arbitrage?
Cash-and-carry is a classic trade: when futures prices are higher than spot (e.g., spot $100, 3-month futures $103), a trader can simultaneously buy the spot (often via an ETF) and short the equivalent futures. Regardless of price movement, the trader locks in the $3 spread (annualized as the basis). The trader is delta-neutral — indifferent to Bitcoin's direction — as long as the basis exceeds the risk-free rate (e.g., T-bills).
Why does this pollute the headline number?
The first leg of the trade — buying spot — is commonly executed by buying an ETF. Thus, a delta-neutral arbitrageur appears as an ETF inflow, indistinguishable from a true believer. When large amounts of cash-and-carry are established, inflows look strong, telling a story of “institutions piling in,” even though the capital is hedged and will reverse as soon as the trade becomes unprofitable.
Data evidence: weekly flows track futures shorts, not price
By comparing weekly ETF flows with changes in CME leveraged funds’ short positions, a tight co-movement emerges. Roughly half of the week-to-week variance in ETF flows can be explained solely by the increase in fund shorts (correlation 0.70). Bitcoin’s weekly return has no statistically significant predictive power for ETF flows. This means weekly flows are primarily driven by the switching on and off of the arbitrage trade, not by price performance or shifts in conviction.
However, it is crucial to distinguish between variance and levels. Of the approximately $55 billion cumulative net inflow into Bitcoin ETFs, the current net arbitrage position is only about $1 billion. The rest is stable directional buying, averaging roughly $400 million per week, which, compounded over two years, forms the bulk of the “mountain.” Thus, ETF flows overstate the volatility of conviction, not its level. The weekly ups and downs are mostly “rented” arbitrage capital, while the assets that remain are largely “owned.”
This trade is leaving
Leveraged funds’ short positions have steadily declined from a peak of ~$14 billion at end-2024 to ~$4.5 billion, a two-year trend. In June, shorts halved further from ~$6.4 billion to ~$4.3 billion, coinciding with daily ETF outflows of $300–500 million. Looking only at the headline, this looks like a panicked capitulation. But with futures data, it is merely a routine unwind of a trade that is no longer profitable. When the basis compresses below T-bill yields, demand evaporates — true believers do not care about the basis, but this weekly “demand” clearly does.
Comparison with Ethereum ETFs
The same test applied to Ethereum ETFs shows weaker characteristics: a looser correlation with futures shorts and almost no underlying directional buying. This is because holding spot Ethereum means giving up ~3–4% staking yield, often making the basis negative. Arbitrage trades cannot cross the hurdle rate. Ethereum ETFs are smaller, noisier, and lack both strong conviction buying and robust arbitrage support.
How to read ETF flows going forward
When the basis is rich, expect “institutional demand” to appear strong and heavily hedged — do not mistake it for conviction. When the basis compresses, expect inflows and shorts to fade together — do not interpret the resulting outflows as a market verdict against Bitcoin. Two numbers to watch: the basis annualized vs. T-bill rate, and CME leveraged funds’ net shorts. They will tell you how much of the next “demand” headline is real.
IOSG’s analysis does not change the core message: week by week, Bitcoin ETF “demand” is primarily a hidden interest-rate trade, not conviction. The genuine buying is real, patient, and now the vast majority of what remains. The “rented” part has been going home for two years.