Every forecast decays. How fast it decays determines the rebalance frequency, the turnover, the cost, and ultimately whether a signal can be implemented at all. We estimate it before we estimate the return, because a large edge that decays in a day is worth less than a small one that persists for a quarter.
Take the information coefficient — the cross-sectional correlation between the forecast and the realised outcome — and compute it at a sequence of horizons rather than at one. The profile is almost always approximately exponential, and the parameter that matters is the horizon at which half the predictive content is gone.
Half-life sets the rebalance frequency, and rebalance frequency sets turnover, and turnover multiplies every cost the strategy pays. A signal with a two-day half-life must be traded roughly twenty times as often as one with a two-month half-life to capture the same proportion of its edge, and it pays spread and impact on each of those turns.
This is the transfer coefficient in the fundamental law, and it is where most research strategies die between the backtest and the book. The decay profile tells you the answer before the capital is committed.
The same measurement, applied across the instruments we cover, produces very different regimes. Price-derived signals in liquid equities decay in hours to days, because everyone is reading the same tape. Signals built from disclosure — filings, capital formation, ownership and control — decay over weeks to months, because the documents diffuse slowly and few people read them. On-chain settlement data sits between the two: public immediately, but unreconciled, so the edge lives in the reconciliation rather than the access.
That spread is the reason we build where we build. It is not that disclosure-derived signals are stronger; it is that they survive long enough to be implemented net of cost.
Research commentary on measurement. It is not an offer, a solicitation, or investment advice, and it recommends no security, strategy or transaction.