A research team presents a signal with an information coefficient that holds out of sample, and the room treats the work as finished. It is not: a signal is a forecast, and a strategy is what you do about it. Between the two sit position sizing, the risk model, the interaction with everything else already in the book, and the cost of the turnover implied — and most of the distance between a good forecast and a disappointing P&L lives there.
Sizing is the first place edge is lost. Converting a forecast into a position requires a view on conviction and on the volatility of the instrument, and a naive mapping will put the largest positions in the noisiest names, where the forecast is weakest. Risk-adjusted sizing sounds obvious and is applied inconsistently in practice, usually because the research metric — hit rate, IC, backtested Sharpe — never forced the question.
The second is correlation with the existing book. A new signal is evaluated standalone and added to a portfolio that already holds several others. If it correlates with what is there, the marginal contribution is small and the added capacity is smaller still; if it is genuinely orthogonal, it can be worth an allocation even with a weaker standalone number. Evaluating candidates against the incremental risk they bring, rather than against each other, changes which ones get funded.
Turnover is the third, and it is the least forgiving because it compounds. A forecast with a half-life of two days demands a rebalancing cadence that may cost more in spread and impact than the forecast is worth. The honest test is to specify the rebalancing rule and the cost model before the allocation decision, then ask whether the strategy is still interesting. A surprising number are not, which is a finding worth having before capital moves rather than after.
Finally, decay. Signals do not fail suddenly; they erode as others find them and as the market adapts. Monitoring the live information coefficient against its research distribution, and agreeing in advance what level of decay triggers a review or a decommissioning, turns a slow disappointment into a managed decision. The discipline is not in the mathematics — it is in having written the rule down while everyone was still optimistic.
