Information horizon (decay & cadence)

python main.py horizon measures how fast a strategy's alpha decays and turns that into a recommended rebalance cadence and a current/lagged blend. It is read-only — a diagnostic over an existing alpha's decay structure, never a control input.

python main.py horizon \
  --strategy volume_spike \
  --symbols NVDA,AAPL,META,AMD,TSLA,GOOG,MSFT,AMZN \
  --start 2024-01-01 --end 2024-12-31

Information horizon: 'volume_spike' 2024-01-01..2025-02-05
  IC by lag: 1:-0.084  2:+0.090  3:+0.012  4:-0.024  5:-0.029  6:+0.007  7:+0.027  8:+0.072  9:+0.113  10:+0.126
  decay δ 1.177  half-life ∞ (no decay detected)  fit R² 0.18
    CI (±1.96 SE on the fit): [4.8, ∞] periods — the multi-period trading policy
    discounts using the upper bound (conservative against killing a good signal)
  recommended cadence: every 2 periods  (best return horizon ≈ inf)
  lagged blend [diversify]: w_now +0.39  w_lagged +0.61  (ρ 0.28, cost 2.70%/yr → not worth the turnover cost)

Read it as: δ is the per-period IC decay and half-life the periods for IC to halve — with a confidence band (CI), since short histories give noisy decay fits and a point estimate alone invites over-trusting it. recommended cadence is the rebalance interval that maximizes IC(Δt)·√(1/Δt) — trading faster than that pays cost for noise, slower throws away breadth. The lagged blend mixes the current signal with a lagged copy to raise the IR when the two diversify (ρ below δ) or hedge (ρ above δ); it's only recommended when it diversifies and its turnover cost is modest — a high-turnover blend that would lose net of cost is flagged, not recommended.

Options

Flag	Default	Meaning
--strategy	volume_spike	The strategy whose alpha decay is measured.
--source	strategy	Alpha score origin (strategy / signal / scanner).
--symbols	demo universe	The cross-section.
--start / --end	last year	Measurement window.
--benchmark	SPY	Used to strip beta (residual returns).
--max-lag	10	Longest lag (periods) the IC-vs-lag profile is fit over.
--timeframe	1Day	Bar timeframe.
--neutralize-factors	off	Measure the decay of the same factor-neutral alpha you'd deploy — see Ranking by alpha.

Superseded (as a recommendation, not as a report)

The lagged blend here is the original, ad-hoc answer to "should I slow down trading into a decaying signal?" Multi-period trading (allocate --objective utility --policy aim) generalizes it into a continuous decay discount composed with the optimizer's own no-trade band. This report's numbers stay accurate either way; prefer --policy aim for new construction work. See Multi-period trading (engineering) for why the discount formula needs the confidence band this report now surfaces.

The same computation is available to agents as the read-only MCP tool compute_horizon.