Configuring Lag Range

What Is Lag Range?

In a Correlation Sweep, the system tests multiple time offsets (lags) to find where sentiment best correlates with price. The lag range controls how many offsets are tested and how far they extend.

Think of it as setting the search window: are you looking for signals that play out within the same day, or signals that take a week to materialize?

Available Presets

Standard: ±24h (49 lags tested)

The default setting. Tests every hourly offset from -24 hours to +24 hours. This covers:

• Same-day effects (±0–12h).
• Next-day effects (±12–24h).

Best for: most analyses, especially for crypto assets where news cycles are fast and price reactions happen within hours.

Extended: ±72h (145 lags tested)

Tests offsets from -72 hours to +72 hours (3 days each direction). This captures:

• Multi-day effects that ±24h would miss.
• Slower-moving assets where price reactions are delayed (e.g., some commodities, macro events).

Best for: assets with lower liquidity, macro-driven assets, or when ±24h shows no significant signal.

Weekly: ±168h (337 lags tested, 7 days each direction)

Tests offsets spanning a full week in each direction. This is the widest preset and captures:

• Weekly cycles (e.g., end-of-week position adjustments, Monday effects).
• Slow institutional flows where sentiment takes days to translate into orders.

Best for: traditional assets, forex pairs, or when you suspect long-delay effects.

Custom

Set the minimum and maximum lag manually, anywhere from -168 to +168. This lets you focus on a specific range — for example, testing only +12h to +72h if you're interested exclusively in forward-looking signals with 12+ hour lead time.

The Multiple Testing Problem

Here is the critical trade-off: the more lags you test, the more likely you are to find a spurious correlation by chance. Testing 337 lags (weekly range) means 337 opportunities for a random pattern to appear significant.

SentiLab addresses this with FDR correction (Benjamini-Hochberg False Discovery Rate), which adjusts p-values based on the number of tests performed. The more lags you test, the stricter the correction becomes:

• 49 lags (standard) — moderate correction. Moderately strong signals survive.
• 145 lags (extended) — stricter correction. Only stronger signals survive.
• 337 lags (weekly) — strictest correction. Only the genuinely powerful signals survive.

This means wider ranges don't give you "free" extra chances to find a signal — the statistical bar gets higher with each additional lag tested.

Practical Guidance

1. Start with ±24h (Standard) — this covers the most common same-day and next-day effects and has the least FDR penalty. If you find a significant signal here, it's likely real and actionable within a short time window.
2. If no signal at ±24h, try ±72h — some assets respond more slowly to news sentiment, especially those with lower trading volumes or stronger institutional ownership. The extended range may reveal multi-day patterns.
3. Use Weekly (±168h) for specific hypotheses — for example, if you believe that Monday's sentiment predicts Friday's price, the weekly range lets you test this. But use it purposefully, not as a default.
4. Use Custom to narrow, not widen — if you already know sentiment leads by ~6–12 hours for an asset (from a previous Standard run), create a Custom range of +4h to +16h to focus the analysis and minimize FDR correction.

Combining with Other Settings

Lag range works alongside other advanced settings:

• Quality filters — higher quality data with fewer lags = cleaner results.
• Predictive Only — future-oriented articles with a focused lag range can produce very precise signals.
• Permutation Test — provides additional validation regardless of lag range chosen.

Why This Matters

Choosing the right lag range is the difference between finding actionable signals and drowning in statistical noise. A narrow, focused range with strong FDR-surviving results is far more valuable than a wide range that produces dozens of "maybe significant" correlations. Be intentional about your search window.