Diagnosing a Slump: Decay, Crowding or Costs

A strategy that has run well for years starts underperforming. Everyone on the desk wants an explanation, and there are three genuinely different ones worth telling apart, because each points to a different fix. The underlying effect might simply be weaker than it used to be. Other traders might have found the same trade and be competing away the same limited profit. Or the strategy might be exactly as good as it ever was, with the returns now being eaten by execution costs that have quietly risen. Treating all three as the same problem, usually by cutting the strategy's size, is the wrong answer to at least two of them.

Three causes, three signatures

Each cause tends to leave a slightly different fingerprint in the data, if you know where to look.

Cause	What it looks like	What confirms it
Genuine decay	Signal strength itself is weaker; the underlying relationship the strategy relies on is fading	A rolling measure of predictive power (like information coefficient) declines steadily, even measured on paper before any trading costs
Crowding	The signal still predicts correctly, but profits from acting on it have shrunk because more capital is chasing the same trades	Correlation to other known strategies or factors rises; the trade's cost of entry (spread, impact) rises specifically around the moments the strategy trades
Rising costs	The signal and the competitive landscape are both roughly unchanged, but market impact, spreads, or borrow costs have increased	Paper (pre-cost) returns are flat or healthy while realised, cost-inclusive returns decline; realised slippage exceeds backtested assumptions by a growing margin

The most useful single diagnostic is comparing a pre-cost, paper version of the strategy's returns against its realised, cost-inclusive returns over the same slumping period. If both have weakened together, the problem is likely decay or crowding acting on the signal itself. If the paper version still looks healthy while realised returns have fallen, the problem is much more likely to be costs.

Worked example

A merger-arbitrage strategy's returns have declined over the past year relative to its five-year average. The team checks three things. First, they compare the strategy's win rate on deals that closed successfully against its historical average — unchanged, suggesting the underlying edge in deal selection hasn't weakened. Second, they check the average spread the strategy captures at entry — narrower than historically typical, and narrower specifically on the subset of deals where several other well-known arbitrage funds are known to have also taken positions, pointing toward crowding rather than decay. Third, as a control, they check a paper version of the strategy using historical, uncrowded-era spread assumptions applied to this year's deal set — that paper version would have produced returns close to the historical average, confirming the gap is coming from spread compression at entry rather than from a change in the strategy's ability to pick the right deals. The diagnosis points toward crowding, and the response — reducing position concentration in deals with heavy known competitor interest, rather than abandoning the strategy or blindly cutting size across the board — follows directly from correctly identifying the cause.

In practice

• Always separate pre-cost signal quality from realised, cost-inclusive returns before diagnosing a slump; conflating the two is the single most common way to misdiagnose the cause.
• Watch for crowding specifically around your own entry and exit points, not just general market conditions — see Factor Crowding and Spotting A Crowded Trade for how to recognise a trade that's become common knowledge.
• A genuine decay diagnosis should be corroborated across several independent signal-quality measures, not a single noisy statistic, tying back to the broader monitoring discipline in Detecting Decay in a Live Strategy.
• Distinguish a slow slump from a sudden, discrete break in performance — the latter often has a specific triggering event (a data change, a regime shift, a system issue) rather than one of these three gradual causes; see Structural Break or Just Noise in Live PnL?.
• Match the response to the diagnosis. Decay usually calls for research into a refreshed signal or eventual retirement; crowding calls for reduced concentration or a search for a less-followed variant; rising costs calls for a capacity or execution review, not necessarily abandoning the strategy at all.