Report date: 2026-10-22 · Sources checked: October 24, 2026 (Korea time). Prices, balances and product terms refer to the cited reporting cutoff.
A large inflow into Bitcoin ETFs can look like a preview of the next rally. DWF Labs’ October 22 study instead asks which moved first. It examined 694 trading sessions from January 11, 2024 through October 16, 2026.
Daily net flows had a correlation of 0.54 with price changes over the previous five sessions, compared with 0.03 for the following five. Correlation measures how two variables move together; it does not by itself establish cause and effect.
On this page
A correlation of 0.54 is not a 54% hit rate
Correlation summarizes a linear relationship between numerical series. A value of 0.54 does not mean a trading signal was correct 54% of the time. Similarly, 0.03 does not mean a 3% probability of a Bitcoin rally. Forecast accuracy requires a separately defined prediction rule and a test of its results.
The revealing comparison is the direction of time. The same flow series has a substantially different relationship with preceding and subsequent returns. Describing money that arrived after a rally as money that predicted it reverses the sequence.
For example, a Monday rally followed by a large Tuesday inflow does not establish a Wednesday gain. Investors may be responding to Monday, or another variable may affect both series. This is an illustration of the inference problem, not a specific session identified by the study.
The price may already have moved
Within this sample, flows tended to react to existing price trends. That provides weak support for treating a large daily inflow as a forecast of next week’s return.
It does not make flow data meaningless. The figures still describe demand through a particular investment channel. Their descriptive role should be separated from a predictive one.
Define a test before inspecting the outcome
A forecasting test needs a definition of a large inflow, an executable entry time and a fixed evaluation horizon. Selecting only the best-looking interval after seeing the results can produce a misleading impression of usefulness. Publication delays and trading costs also matter.
| Test question | Why it matters |
|---|---|
| Was the information available then? | Later revisions cannot be used as contemporaneous signals |
| When does the hypothetical trade start? | Entry cannot be assumed before the flow release |
| Are the rules consistent across periods? | This limits selecting only favorable examples |
| Does a difference remain after costs? | A statistical relationship is different from realized profit |
This is a framework for reading research, not a proposed trading strategy. A sample of 694 sessions does not represent every future market environment. Whether a conclusion survives other periods and regimes remains a separate question.
Match dates and definitions
Check the observation window, price provider and definition of net flows. A US fund closing time and a continuously traded Bitcoin price need aligned cutoffs before they can be treated as observations from the same day.
When recording a new inflow, put the preceding price move next to it. That simple comparison helps distinguish an explanation of a rally that has already happened from evidence about one that has yet to occur.
Overlapping observations also matter. Adjacent “next five-session” returns share several dates. A daily dataset is not automatically the same number of independent experiments. This is a methodological question, not a claim that an error was found in the report.
Check whether a few extreme sessions dominate a relationship and whether it survives splitting the sample into different periods. A weak average can coexist with different behavior in a particular regime, while unusual observations can create an apparently strong association.
Record the scope of the conclusion: weak general short-horizon predictive evidence in this sample is different from proving that flows can never influence prices under any circumstances.
Change the question you ask of the data
Weak predictive evidence does not make the dataset useless. It can still help describe how sensitively flows respond to price or whether demand through a particular investment channel persists. Asking what followed a rally is different from asking what will cause tomorrow’s rally.
For a personal research log, record publication time alongside preceding returns. This allows a later review of whether optimism about a large inflow arrived after much of the move had already happened. The practical lesson is to establish chronology before calling a figure a leading signal.
Sources and further material
Related reading
US spot Bitcoin ETFs lose $635 million in one session
Glassnode’s $87,700 Bitcoin cost basis: reading a potential resistance zone
