lev-os Analyzing Alpha Signal Decay
When To Use
- Evaluating whether an alpha signal retains predictive power across different holding periods
- Determining optimal rebalance frequency by measuring how quickly signal strength deteriorates
- Estimating strategy capacity limits before market impact erodes expected returns
- Comparing decay profiles across candidate factors during signal selection or portfolio construction
- Assessing whether a live signal has degraded relative to its backtested decay curve (alpha erosion monitoring)
Inputs To Gather
- Signal return series: Period-by-period returns for portfolios sorted by the signal (e.g., decile or quintile long/short spreads)
- Holding-period returns matrix: Returns measured at 1-day, 5-day, 10-day, 21-day, 63-day, and 126-day forward windows
- Turnover data: Portfolio turnover rate at each rebalance frequency; round-trip transaction cost estimate (bps)
- Universe and dates: Investable universe definition, backtest start/end dates, any regime breaks or structural changes
- AUM or notional size: Current or projected strategy size for capacity analysis
- Market impact model parameters: Participation rate assumptions, ADV percentiles, spread estimates [VERIFY — impact model choice varies by asset class and execution venue]
Workflow
Construct the decay curve
- Compute the information coefficient (IC) or long/short spread return at each forward horizon (1d through 126d)
- Plot IC vs. holding period; fit an exponential decay model IC(t) = IC₀ · e^(−λt) to estimate the decay constant λ
- Derive the signal half-life: t½ = ln(2) / λ
- Report confidence intervals around half-life using bootstrapped IC series
Assess turnover cost breakeven
- Calculate implied turnover at each rebalance frequency from the signal's rank-change rate
- Multiply turnover by estimated round-trip cost to get the cost drag per period
- Identify the breakeven rebalance frequency where gross alpha minus cost drag is maximized
- Flag if optimal rebalance is faster than operationally feasible settlement or execution cycles [VERIFY — settlement cycles differ by market and instrument type]
Estimate capacity constraints
- For each rebalance, estimate total dollars traded per period
- Apply market impact model (e.g., square-root impact: cost ∝ √(participation rate)) to compute expected slippage at increasing AUM levels
- Find the AUM level where net-of-impact alpha falls below a minimum threshold (e.g., 0 bps or a target Sharpe hurdle)
- Produce a capacity curve: net alpha vs. AUM
Diagnose decay regime and crowding risk
- Compare decay profile across sub-periods (pre/post publication, different volatility regimes)
- Check if half-life has shortened over time — an indicator of signal crowding or information diffusion
- Cross-reference with short-interest, ETF flow, or factor-crowding proxies if available
- Note any structural breaks where decay accelerated (new data vendors, regulatory changes, competing strategy launches)
Synthesize findings into actionable recommendations
- Recommend hold period and rebalance cadence consistent with the decay profile
- State capacity ceiling with assumptions made explicit
- Identify risk: if half-life is near or below the cost-breakeven horizon, flag the signal as capacity-constrained or potentially non-viable at scale
- Suggest mitigants (partial rebalancing, execution algorithm tuning, blending with slower-decay signals)
Output
The deliverable is a structured Signal Decay Analysis Report containing:
- Signal summary table: signal name, universe, backtest period, IC₀, half-life, decay constant
- Decay curve chart: IC or spread return vs. forward horizon with fitted exponential overlay
- Turnover-cost analysis: table of rebalance frequency vs. gross alpha, cost drag, and net alpha
- Capacity curve: net alpha vs. AUM chart with annotated capacity ceiling
- Regime comparison: sub-period half-lives and any crowding indicators
- Recommendation: optimal rebalance cadence, maximum deployable AUM, and risk flags
Quality Checks
- Verify that IC calculations use point-in-time data with no lookahead bias in signal construction
- Confirm transaction cost assumptions reflect realistic execution for the asset class and AUM level [VERIFY — cost estimates should be calibrated to actual fills or TCA data where available]
- Ensure half-life estimate is stable across bootstrap samples; flag if confidence interval spans more than 2× the point estimate
- Check that capacity estimate accounts for correlated liquidation risk (other funds trading the same signal)
- Validate that decay curve uses non-overlapping or bias-adjusted overlapping returns to avoid autocorrelation inflation
- Cross-check turnover calculations against actual portfolio rebalance logs if the signal is already in production