Correlation measures whether two series moved together. It establishes nothing about why. The standard alternatives to causation — reverse causation, a common third cause, and coincidence — are all consistent with any correlation you can compute.
- A causes B, the interpretation usually assumed.
- B causes A, which the correlation cannot distinguish.
- C causes both, the most common real explanation in markets.
- Coincidence, which is guaranteed to appear if you test enough pairs.
Lead/lag adds a further trap. Measuring a lag by scanning many candidate lags and keeping the best one guarantees finding something, because the best of many random results is not random. VILIQ measures lead/lag over a bounded window with the relationship stated in advance, and reports the measured lag alongside the strength so both can be judged.
Common belief
"Global liquidity leads Bitcoin by 90 days, as the chart shows."
What is actually true
A specific lag that fits historical data well is often the best of many lags tested. Published estimates vary widely and change across regimes. VILIQ measures the lag in the current window rather than adopting a fixed figure, precisely because a fixed figure does not survive out of sample.
The practical standard: prefer relationships with a mechanism you can state before looking at the data. A correlation supported by a plausible mechanism and stable across regimes is evidence. A correlation found by search, with the mechanism invented afterwards, is a hypothesis at best.