Do ETF Flows Predict Where Bitcoin Goes Next?

Executive summary

US spot Bitcoin ETFs haven’t turned TradFi investors into dip buyers. In 7 of the 8 Bitcoin sell-offs since the ETFs launched in January 2024, the ETFs saw net outflows on the way down. Sell-offs have become slower and less violent, but ETF investors aren’t the dip buyers people think they are.

Two questions, answered.

Do ETF flows explain Bitcoin’s moves on a given day? Partly. Flows line up with about a sixth of a typical day’s price move (correlation 0.41). Price usually moves first, and we can’t tell whether flows caused the rest.

Do big flows signal a move to come? Not reliably. Big flow days in general carry no edge. The one setup worth watching is a big outflow on a day when the price barely moved: in 8 of 12 cases, BTC fell over the next 20 days, but all of those came from three sell-offs.

1. ETF investors sell into sell-offs and buy into rallies. Net outflows came in 7 of 8 falls of 15% or more, and net inflows came in all 9 rallies. ETF investors bought on the day of the peak at 7 of 8 tops and sold on the day of the low at 7 of 8 bottoms. The two most recent sell-offs saw the heaviest selling of the ETF era: −$6.9B (October 2025 to February 2026) and −$8.2B (May to June 2026).

2. Flows follow price; they don't lead it. Flows and price move together strongly on the same day (correlation 0.41), but price usually moves first. Flows correlate 0.54 with BTC's prior five-day return and 0.03 with the next five days.

3. ETF investors sell once Bitcoin falls about 7%. A 5% dip barely moves flows. A 7% fall from the 20-day high flips them from about +$74m a day to −$72m within a week. Rallies of any size, from 5% up, draw inflows.

4. Sell-offs are smoother, not shallower. The median sell-off falls about as far as before (−21.3%) but more slowly: 0.69% a day, against 1.00% in 2022–23. Sell-offs of 15% or more now come about 3 times a year, against 4 to 8 before, and days with a 7%+ drop fell from about 6 a year to 2.

5. IBIT carries no extra signal. The largest fund's flows predict prices no better than the rest.

6. Big flows on quiet days are an occasional warning, not a signal. Big outflows on days BTC barely moved preceded sharp falls in three sell-offs but rallies in four other cases. Big inflows on quiet days show no edge since 2025.

About a fifth of ETF flow appears to be hedged through the CME basis trade. Accounting for it doesn't change these conclusions.

The question

A common view since the spot ETFs launched is that institutional money has changed how Bitcoin sells off. Long-term institutional buyers use the ETFs to buy dips, which softens declines and speeds up recoveries.

We tested that view against the flow data, asking three things:

• Do ETF investors buy or sell during sell-offs?

• Do their flows lead price moves, or follow them?

• Have sell-offs actually changed since the ETFs launched, and if so, can ETF buying explain it?

Data and method

We compared daily net flows for all US spot Bitcoin ETFs with Coinbase BTC prices from January 11, 2024, to October 16, 2026 (694 trading days), using 2021 and 2022–23 as pre-ETF comparison periods.

A sell-off is a fall of 15% or more from a peak close to a low close, and a rally is a rise of 15% or more from a low. The ETF era has eight sell-offs and nine rallies. Full definitions and robustness checks are in the appendix.

Why do total flows include GBTC? When money moved from GBTC into cheaper ETFs, the outflow and the matching inflow canceled out in the total. Leaving GBTC out would falsely count that rotated money as new buying.

1. ETF investors sell into sell-offs and buy into rallies

The ETFs saw net outflows in 7 of the 8 ETF-era sell-offs. The selling has grown in each of the last three: −$3.4B in January–April 2025, −$6.9B from October 2025 to February 2026, then −$8.2B in May–June 2026, when money left on 88% of trading days.

• The one exception was May–July 2024. $2.2B flowed in while Mt. Gox repayments and German government sales pushed the price down. A known supply overhang drove that sell-off, not a broad loss of confidence, and ETF buyers absorbed part of it.

• Outflows often get heavier into the low. In 5 of 8 sell-offs, the last five trading days before the bottom saw the biggest outflows. In May–June 2026, outflows averaged −$412m a day in that final week.

Rallies are the mirror image. ETFs saw net inflows in all 9 rallies of 15% or more, and in 7 of 9 the inflows were heaviest in the final five days before the peak. 

Flows and price move together, so this alone isn’t surprising. The more useful question is whether ETF investors buy at the bottom and sell at the top.

Mostly not. On the day of the low, ETF flows were negative in 7 of 8 sell-offs. The exception is July 2024, where the low fell on a weekend and the first trading day was already part of the rebound. Inflows returned within 10 trading days in 6 of 8 cases, but only once the rebound was under way: BTC rose 5% to 14% in the five trading days after every low. April 2025 is the clearest case: outflows averaged −$40m a day while BTC rose 9.7%, then inflows reached +$436m a day. In the two sell-offs since, flows stayed flat or negative while the price recovered.

Tops show the same pattern in reverse. Flows were positive on the peak day in 7 of 8 cases. One of the largest inflow days of the ETF era, +$1.2B, came on the day of the October 2025 top, and BTC fell 4% to 19% over the next five trading days.

These results use a 15% move to define a sell-off or rally, and they also hold at 10% and 20% (see Appendix: robustness). Where the peak fell on a Sunday, the "peak day" flow is the next trading day. 

2. ETF flows follow price, they don’t lead it

Flows and price move together on the same day (correlation 0.41, a moderate link), but price usually moves first. Flows line up with what BTC has already done: 0.54 with the prior five days and 0.40 with the prior day. 

In other words, flows respond more to what the price has already done than to what it’s doing that day. They barely line up with what comes next: 0.09 the next day and 0.03 the next five days, close to no link at all. That’s the pattern of momentum money, which buys after a rise and sells after a fall.

The pattern shows up even within the day. On the biggest inflow days, BTC had usually already risen before US markets opened: up 1.2% overnight on average, against 0.2% on a normal day. Before the biggest outflow days, it had already fallen 0.9%.

• Big flow days come after big moves. Before the top 10% of inflow days, BTC had risen 7.3% over the previous five days on average, and it was up in 97% of cases. Before the top 10% of outflow days, it had fallen 4.4%.

• Big flow days don’t carry forward. After the top 10% of inflow days, BTC rose over the next 20 days 51% of the time, against 57% for any day. After the top 10% of outflow days, the next five days were flat.

The evidence points mostly to investors reacting to price, not driving it.

3. Selling accelerates once Bitcoin falls about 7%

ETF investors aren’t shaken by small dips, but once a fall reaches about 7%, they start selling. We marked every time BTC first closed a given distance below its 20-day high, from 5% to 10%, and compared average daily ETF flows in the five trading days before and after.

• At 5%, inflows slow but keep coming. Flows fell from +$62m to +$27m a day and got worse in only 17 of 34 cases, a 50-50.

• From about 6%, flows turn marginally negative. They went from +$99m to −$10m a day.

• At 7%, the flip is the sharpest and most consistent. Flows went from +$74m to −$72m day and got worse in 19 of 24 cases.

The selling burst is short. After a 7% break, flows were back near zero (+$6m a day) the following week and positive (+$56m) the week after.

Rallies work differently. Investors buy into any move up. We ran the same test on the first close a given distance above the 20-day low, and inflows rose after a rise of every size, from 5% up. The buying also lasts longer: after a break higher, inflows stayed above normal the following week. ETF investors chase rallies of any size but only sell real breaks.

4. Sell-offs are smoother, not shallower

Since the ETFs launched, sell-offs come less often and take longer to play out, and crash days are rare. But once one is under way, it falls about as far as before. The median ETF-era sell-off fell 21.3%, against 24.0% in 2021 and 19.4% in 2022–23.

• Fewer sell-offs. Falls of 15% or more came about 2.9 times a year in the ETF era, against 4.4 in 2022–23 and 8 in 2021. Falls of 10% or more came 5.1 times a year, against 6.9 and 15.

• Slower declines. The median ETF-era sell-off fell 0.69% a day, against 1.00% in 2022–23 and 1.37% in 2021.

• Lower volatility. Annualized volatility was 47%, against 55% and 81%.

• Fewer crash days. Days with a 7%+ drop fell to about 2 a year, from 6 in 2022–23 and 15 in 2021.

We can’t credit this directly to ETF flows. ETF flows were negative in most of these sell-offs, so ETF buying didn’t cushion the falls, and volatility was already falling before the ETFs launched. The ETFs may still have helped indirectly. They brought in a wider, more long-term group of holders, deeper liquidity, and regulated ways to hedge, such as CME futures and, since late 2024, options on IBIT. Together, these made Bitcoin a larger, more mature asset class.

5. IBIT doesn't carry more signal than the rest

BlackRock’s IBIT is the fund most people quote. It took in $65.6B of net inflows since launch, more than the whole ETF complex’s $57.0B, because the other funds, mainly GBTC, had net outflows. A natural guess is that its flows say more about where Bitcoin is heading. The tests below find no sign of that.

• No better at predicting price. IBIT's daily flow correlates 0.08 with BTC's next-day return and 0.04 with the next five days, the same as the ETF complex as a whole.

• Big IBIT inflows come after rallies, not before them. IBIT's top 10% of inflow days followed a five-day BTC gain of 7.4% on average, up in 88% of cases. BTC then rose over the next 20 days only 47% of the time, against 57% for any day. On days when IBIT took in big money while the other funds sold (15 days), BTC fell 0.86% on average over the next five days.

• Big IBIT outflows aren't a reliable warning either. They mostly come inside sell-offs already under way, and BTC still rose over the next 20 days in 54% of cases.

Many point to IBIT’s +$503M inflow on August 20, 2026, after which BTC rallied to about $87k, as evidence that IBIT flows lead price.  But the inflow came after BTC had already risen from about $62.9k to $72.7k in four trading days, and the next three big IBIT inflow days (September 3, 21 and 22) were each followed by a 3% to 6% fall over five days. Single cases like this are easy to find in either direction. Across all 68 big IBIT inflow days, the pattern doesn’t hold.

6. Big flows on quiet days: an occasional warning, not a reliable signal

Traders often look for a big ETF flow on a day when price barely moves, on the idea that price will catch up with the flow later. We took the top and bottom 10% of daily flows (inflows of $507M or more, outflows of $296M or more) on days when BTC moved less than 1% (4pm to 4pm New York). That gives 13 big-inflow days and 12 big-outflow days.

• Big outflows on quiet days are the one setup worth watching. Over the next 20 days, BTC did 10.7 percentage points worse than on other days with a similar prior trend. The result holds if “quiet” is widened to a move under 2% (−8.4 points).

• But it rests on a few sell-offs. The 12 cases fall into seven separate periods. Three preceded falls of 4% to 29% over 20 days: October 2025, the slide into the February 2026 low (five of the cases), and May 2026.

• Big inflows on quiet days carry no real edge. The strong average comes from late 2024, when the whole market was rallying around the US election. Since 2025, the seven cases averaged +1.3% over 20 days, below the +3.0% for any day.

• Nothing shows up in the first few days. Any effect appears over 10 to 20 days, not the next day.

IBIT alone again adds no edge. After big IBIT outflows on quiet days, BTC did 6.7 points worse than similar-trend days over 20 days (against 10.7 for all ETFs) and fell in only 7 of 14 cases, a coin flip. Big IBIT inflows on quiet days showed a small edge (+3.4 points) that fades when we widen “quiet” to a move under 2%. The warning works best on total ETF flows.

Big inflows on quiet days may become even harder to read. Since late July 2025, investors can move BTC they already hold into an ETF in kind. That counts as an inflow, but it adds no buying on the market.

The CME basis trade

Some ETF flows aren’t directional bets. In the basis trade, a hedge fund buys ETF shares and shorts CME Bitcoin futures to earn the price gap, so its flows are not directional bets.

In weekly CFTC data, the “leveraged funds” category (mostly hedge funds) added about 0.21 BTC of CME short for every 1 BTC of ETF inflow (correlation −0.50 across 144 weeks). That suggests hedge funds hedged about a fifth of ETF flow this way, though the data can’t confirm that the same funds hold both sides. Their net short peaked near 116k BTC in December 2024 and is now about 29k BTC.

Does it affect our earlier findings? Not much.

• Jan–Apr 2025: ETF outflows (41,500 BTC) roughly matched hedge funds closing CME shorts (44,600 BTC), so that sell-off’s outflows may be mostly basis unwind.

• The two largest sell-offs: the unwind can explain at most about a quarter of the outflow. The rest was directional selling.

• Today: with the short down to about 29,000 BTC, less of current flow is the basis trade, so flows are a cleaner read of real demand than in 2024.

What it means

Institutional money through the ETFs hasn’t made Bitcoin safer in a sell-off. It has added a large group of investors who behave like momentum traders.

• ETF flows confirm moves; they don’t predict them. Flows tell you what investors are doing in response to price, which is useful for gauging conviction. They carry almost no information about the next week’s direction.

• ETF holders add to selling once a fall is under way. After a 7% break from the 20-day high, outflows typically follow within days. A break of that size is a reasonable point to expect ETF selling to start.

• The investor base is less sticky than the dip-buying view assumes. The two largest outflows of the ETF era came after October 2025, and most of that selling was directional, not basis-trade unwinding. Buyers also didn’t return quickly after those lows.

• Smoother sell-offs need another explanation. Declines are slower and crash days rarer, but ETF flows were negative through most of them. If the ETFs stabilized sell-offs, they did it through market structure (a broader holder base, deeper liquidity, and regulated hedging), not dip-buying.

• Flows are becoming a cleaner signal. With the CME basis trade largely unwound, a bigger share of today’s ETF flow reflects directional demand than it did in 2024.

Limitations

• Small sample. Eight sell-offs and 24 price breaks are enough for a clear pattern, not a statistical proof. The report uses counts ("7 of 8") rather than model language.

• Correlation, not cause. The strong same-day link (0.41) can't be fully split into cause and effect. The timing tests show that flows don't lead price, not that they never add pressure on the day.

• The basis trade is estimated, not removed. CFTC data is weekly, and "leveraged funds" CME shorts include hedges other than the basis trade. The basis-trade figures are best read as an upper bound.

• Daily flow data. ETF flows are reported once a day, so we can't see when during the day investors bought or sold.

• One price source. Prices are Coinbase BTC-USD. Other major venues trade within a few basis points, so this is unlikely to change the results.

Appendix: data and method in detail

Why CME only. This report is about institutional money, and institutions hedge ETF positions on CME, the regulated US futures market. The data supports this: weekly changes in CME open interest track ETF flows (correlation 0.50), while open interest on offshore exchanges such as Bybit and OKX does not (0.02). CME data also identifies trader types, which offshore exchanges don't publish. 

Appendix: robustness

The core results hold when we change how a sell-off or a rally is defined. First, sell-offs:

Then rallies, and the flows on the day of each top and low:

• Outflows in sell-offs hold at every threshold.

• Inflows in rallies hold at every threshold too, and so does the pattern at the turns: ETFs had inflows on the peak day at most tops and outflows on the low day at most lows, whichever definition we use.

• Outflows heaviest into the low is common but not the rule. It holds in only half of the 10% sell-offs, so the report says "often".

• Timing. Re-running the lead–lag test on UTC daily closes instead of 4pm New York prices gives the same picture: prior 5 days 0.39, prior day 0.38, same day 0.41, next day 0.05, next 5 days 0.03.

• GBTC. Excluding GBTC, outflows came in 5 of 8 sell-offs. The two differences are March and April 2024, when rotation from GBTC into cheaper ETFs makes the other funds look like they had new inflows. Total flows are the right measure for the reasons given in Data and method.

Appendix: the eight sell-offs

Every ETF-era fall of 15% or more, most recent first. "CME shorts closed" is the drop in hedge funds' net CME short during the sell-off, a rough measure of basis-trade unwinding.

Appendix: every big flow on a quiet day

Days in the top or bottom 10% of total ETF flows when BTC moved less than 1% (4pm to 4pm New York), oldest first.

Sources

• US spot Bitcoin ETF daily net flows, US$m, 11 Jan 2024 – 16 Oct 2026: Farside Investors.

• BTC-USD daily closes (1 Jan 2021 – 19 Oct 2026) and hourly closes (1 Jan 2022 – 19 Oct 2026): Coinbase Exchange public API.

• CME Bitcoin futures positioning: CFTC Traders in Financial Futures, weekly, Jan 2024 – Oct 2026.