Flows Predict Returns. Or Do They?

Question

A researcher finds ETF flows correlate 0.4 with next-month sector returns and concludes: "flows cause returns, trade it." List the distinct non-causal explanations that could produce this correlation, and what you'd check before allocating.

Accepted Answer

Five distinct machines generate correlation without the claimed causation: 1. Common driver (confounding). Momentum or a macro factor lifts both flows and subsequent returns; flows are a passenger, not a pilot. Check: does the flow signal survive after regressing out known factors? 2. Reverse causality plus persistence. Returns cause flows (performance chasing), and returns are autocorrelated at the sector level, so flows merely proxy past returns. The arrow points backward. Check: lead–lag structure; does flow predict return after controlling for past return? 3. Selection / multiple testing. A correlation of 0.4 found after scanning many sectors, lags, and definitions is the maximum of many noisy estimates, expected to be large under pure noise. Check: how many variants were tried; out-of-sample. 4. Regime artifact (nonstationarity). One era, say, a sustained passive-inflow regime, created the comovement; nothing says it persists. Check: subsample stability, rolling correlations. 5. Mechanical/definitional overlap. If flows are imputed from prices (or share creations triggered by price moves), the variables share construction, correlation by bookkeeping, plus possible lookahead in

Flows Predict Returns. Or Do They?

Problem

Your answer

Hints