Quant Memo
Core

Matching Horizon to the Cost Budget

A signal that decays fast enough to need daily trading is only worth trading if daily turnover fits inside what transaction costs allow. The horizon a strategy trades at has to reconcile the signal's decay with the cost budget, not just the signal's peak IC.

Prerequisites: The IC-Horizon Profile

A signal with a sharp, fast-decaying IC-horizon profile might genuinely be the strongest signal a desk has — and still be the wrong one to trade at full frequency, because capturing that fast decay means rebalancing often, and rebalancing often means paying transaction costs often. The horizon a strategy actually trades at is a negotiation between how fast the signal decays and how much the cost budget can absorb, not simply "trade at the peak of the IC curve."

Why the peak of the curve isn't the answer alone

Suppose an IC-horizon profile peaks at 5 days. Trading to capture that peak means rebalancing the book roughly every 5 days for every position, which for a portfolio with turnover-sensitive costs may be perfectly affordable — or may not, depending on how liquid the names are and how wide the spreads and market-impact costs run for the position sizes involved. A slower-decaying signal with a lower peak IC, but one that only needs rebalancing every 20 days, can end up delivering more net return, because a smaller gross edge captured cheaply beats a larger gross edge captured expensively.

The right comparison is never gross IC against gross IC. It is net-of-cost return per unit of risk, computed at whatever horizon each signal would actually need to be traded at to realise the IC it reports.

A worked comparison

SignalPeak ICHorizon of peakRound-trip cost per rebalanceRebalances per month at that horizonRough monthly cost drag
A (fast)0.063 days15 bps~7~105 bps
B (slow)0.0320 days15 bps~1~15 bps

Signal A's raw edge looks twice as strong. But capturing it means paying roughly seven times as many round-trips a month at the same per-trip cost, and the monthly cost drag alone — before even converting IC into an expected net return — is seven times higher. Whether A still wins net of costs depends on how much of that 0.06 IC translates into actual basis points of return per trade; in a fair number of real cases, B's lower gross number survives the trip through costs better than A's higher one does.

A signal's tradeable horizon is set by where its cost-adjusted return is maximised, not where its raw information coefficient peaks. A sharp, fast-decaying edge can be genuinely inferior, after costs, to a duller, slower one that is cheaper to capture.

Before comparing two signals' Sharpe ratios, check whether both were backtested with the same cost assumptions at each one's own natural trading horizon. Comparing a fast signal's gross Sharpe to a slow signal's net Sharpe is an easy way to draw the wrong conclusion without anyone intending to.

Related concepts

Practice in interviews

Further reading

  • Chincarini & Kim, Quantitative Equity Portfolio Management
ShareTwitterLinkedIn