1. What Is Signal Half-Life?

Every edge has a shelf life. A signal that looks brilliant the day you discover it is usually decaying from the moment you trade it — the only real question is how fast. Signal half-life puts a number on that decay, and it's one of the most consequential numbers in systematic trading: it tells you how long a model's edge is worth holding before it's mostly gone.

More formally: take a prediction signal s_t (say, an ML-predicted excess return) and bucket observations by signal strength. For a given bucket, track the cumulative forward return over a holding horizon h.

As you extend h, the incremental edge erodes. The half-life is the horizon at which cumulative return reaches roughly 50% of its maximum — equivalently, the point where the marginal information from holding one more day drops below the threshold that makes it worth the risk and cost.

2. Measuring Decay Curves

Practical method:

1. Sort daily observations by signal value into quantiles (e.g. top 10%)
2. For each entry date t, track realised returns for h = 1,...,H days
3. Compute average cumulative return for each h
4. Plot decay curves: E[R_{0→h}^(top)] vs h

You typically see strong positive edge for small h, flattening or reversal beyond a certain horizon.

3. Half-Life for ML vs Simple Models

Research use-case: compare decay curves for:

• ML models (e.g. XGBoost, LSTM)
• Simple models (e.g. 12-month momentum, value)

Questions: Are ML signals shorter-lived or longer-lived? Do ML signals mostly capture microstructure/short-horizon effects? Does model complexity correlate with shorter half-life?

4. Stability Across Regimes and Universes

Half-life is not constant:

• in high-vol regimes, edges may compress (shorter half-life)
• in calm regimes, edges may extend
• in small caps, microstructure edges may be strong but short-lived
• in large caps, slower fundamental edges may dominate

Thus we measure half-life by regime, by asset bucket, and by horizon type.

5. Decay and Portfolio Construction

Knowing half-life helps:

• choose optimal holding periods
• avoid holding positions beyond their edge horizon
• design staggered rebalancing policies
• combine multiple signals with different half-lives

Example: ML microstructure signals (half-life 1–2 days) → trade with high turnover, tight risk. Medium-term factor signals (half-life 20–60 days) → trade with lower turnover, larger sizing.

6. How This Maps to volarixs

Decay analysis only works if you keep the raw material: a model's predictions over time, the horizon each one was made for, and the market regime it was made in. That's exactly what volarixs records. Every experiment stores multi-horizon predictions — 1-day, 5-day, 21-day, 63-day and beyond — for each model, and carries the regime context those predictions were generated under.

That history is the foundation a decay curve is built on. From it, you can measure how persistence actually varies rather than assuming a single number:

• per experiment, per model, and per prediction horizon
• and, because regime labels travel with every run, per regime

The goal is to move the question from “Is this model predictive?” to “How long does this edge persist, for which assets, and under which regimes?” — the version that actually governs how long you hold a position and how often you rebalance. That's the lens volarixs is built around, and the explorer above is a simplified, hands-on version of the same idea.