KILLED · not causal
Hypothesis
Transfer entropy from coin A to coin B measures directed information flow; trading laggards when leaders move captures a predictable spillover.
Math — transfer entropy
$$ T_{A\to B} = \sum p(b_{t+1},b_t,a_t)\,\log\frac{p(b_{t+1}\mid b_t,a_t)}{p(b_{t+1}\mid b_t)} $$
Method
Estimate pairwise TE across 40 coins, trade detected laggards on leader moves, costs + latency applied.
Results
| Detected lead-lag in-sample | present |
| Lag horizon | 1–2 bars |
| Net after fees at that horizon | − |
| Direction stability out-of-sample | flips |
Detected information flow is real but the exploitable lag is 1–2 bars and unstable — by the time TE is estimated, the spillover is priced, and the direction of the A→B arrow flips out-of-sample. Net negative after cost. Killed (cf. our separately-killed lead-lag work).
Estimating that A leads B takes a window of data; within that window the lead is already arbitraged. Measurable ≠ tradeable, once latency is honest.