OPEN-SOURCE SCRIPT
AetherEdge - Seasonality
🖊️ Overview
Seasonality tools show raw averages and let you fool yourself with three-sample "edges." AE-SEAS measures the time-of-day / day-of-week / month return profile properly, with significance. It learns each bucket's mean return, win-rate, sample count and t-stat, and shrinks small-sample buckets toward the global mean (empirical Bayes) to prevent overfitting — answering "which time periods have a real edge" with data.
🔶 Key Features
Time-of-period profiles — learned mean return and win-rate by hour / weekday / month
Significance (t-stat) — separates a real edge from noise (don't trust raw averages)
Empirical-Bayes shrinkage — small-sample buckets shrink toward the global mean (anti-overfitting)
Current-bucket bias — the shrunk expected return, win-rate and significance for the bucket you're in
Heatmap table — all buckets at a glance, current one highlighted
Signals + alerts on entering a significant-edge bucket; bar-close accumulation — no repaint
🧠 Technical Architecture
Bucket statistics: each time bucket (hour 0–23 / weekday / month) accumulates the bar's return, yielding mean, variance, sample count, win-rate and a t-stat (mean ÷ standard error). Larger |t| = a less-likely-to-be-chance edge.
ML (empirical-Bayes shrinkage = regularized estimation): shrunk mean = global mean + (bucket mean − global mean) × n/(n+k). Low-sample buckets shrink hard toward the global mean; well-sampled buckets approach the raw value. This stops a five-sample bucket from masquerading as a strong signal — regularized estimation, not raw counting, is the "ML."
Timezone: buckets are computed in the selected timezone (Exchange / UTC / NY / London / Tokyo).
Honest scope: regularized per-bucket statistics (empirical-Bayes shrinkage + t-test) (no RL). Not deep learning, not a guarantee.
⚙️ Recommended Settings & Tuning Guide (crypto 15m–4H)
Key parameters: active profile (hour/weekday/month), timezone, shrinkage strength k, significance threshold |t|, min samples.
Profile: "Hour" for crypto intraday, "Weekday" for weekly tendencies, "Month" for the long run (higher timeframes)
Timezone: UTC is standard for crypto; exchange TZ for equities
Higher shrinkage k pulls small samples harder toward the global mean (conservative)
Significance |t| defaults to 2.0 (≈95%); use 2.5–3.0 to be stricter
Longer history / timeframe means more samples per bucket and higher reliability
💡 How to Use in Practice
Use the heatmap table to spot hours/days with a large mean and a ✦ (significant)
Reference the current-bucket bias as the backdrop for the period you're in
Treat entering a significant bull/bear bucket as a trend/caution cue
Ignore buckets with low sample count (n) or small t
Combine with AE-VOL (volatility) or AE-CVD (flow) to layer time × context
⚠️ Important Notes
Seasonality is a historical tendency with no guarantee of persistence (structural shifts break it)
Low-sample buckets remain uncertain even after shrinkage (always check n and t)
Needs a learning period (warmup) and ample history; accumulation resets on timeframe/timezone change
Probability, not a guarantee — always do your own due diligence and use risk management
🚨 Disclaimer
This indicator is for educational and informational purposes only and is not financial advice or a recommendation to buy or sell. No method guarantees future profits; past performance does not indicate future results, and trading carries the risk of loss. All decisions are your own — use proper validation and disciplined risk management.
Seasonality tools show raw averages and let you fool yourself with three-sample "edges." AE-SEAS measures the time-of-day / day-of-week / month return profile properly, with significance. It learns each bucket's mean return, win-rate, sample count and t-stat, and shrinks small-sample buckets toward the global mean (empirical Bayes) to prevent overfitting — answering "which time periods have a real edge" with data.
🔶 Key Features
Time-of-period profiles — learned mean return and win-rate by hour / weekday / month
Significance (t-stat) — separates a real edge from noise (don't trust raw averages)
Empirical-Bayes shrinkage — small-sample buckets shrink toward the global mean (anti-overfitting)
Current-bucket bias — the shrunk expected return, win-rate and significance for the bucket you're in
Heatmap table — all buckets at a glance, current one highlighted
Signals + alerts on entering a significant-edge bucket; bar-close accumulation — no repaint
🧠 Technical Architecture
Bucket statistics: each time bucket (hour 0–23 / weekday / month) accumulates the bar's return, yielding mean, variance, sample count, win-rate and a t-stat (mean ÷ standard error). Larger |t| = a less-likely-to-be-chance edge.
ML (empirical-Bayes shrinkage = regularized estimation): shrunk mean = global mean + (bucket mean − global mean) × n/(n+k). Low-sample buckets shrink hard toward the global mean; well-sampled buckets approach the raw value. This stops a five-sample bucket from masquerading as a strong signal — regularized estimation, not raw counting, is the "ML."
Timezone: buckets are computed in the selected timezone (Exchange / UTC / NY / London / Tokyo).
Honest scope: regularized per-bucket statistics (empirical-Bayes shrinkage + t-test) (no RL). Not deep learning, not a guarantee.
⚙️ Recommended Settings & Tuning Guide (crypto 15m–4H)
Key parameters: active profile (hour/weekday/month), timezone, shrinkage strength k, significance threshold |t|, min samples.
Profile: "Hour" for crypto intraday, "Weekday" for weekly tendencies, "Month" for the long run (higher timeframes)
Timezone: UTC is standard for crypto; exchange TZ for equities
Higher shrinkage k pulls small samples harder toward the global mean (conservative)
Significance |t| defaults to 2.0 (≈95%); use 2.5–3.0 to be stricter
Longer history / timeframe means more samples per bucket and higher reliability
💡 How to Use in Practice
Use the heatmap table to spot hours/days with a large mean and a ✦ (significant)
Reference the current-bucket bias as the backdrop for the period you're in
Treat entering a significant bull/bear bucket as a trend/caution cue
Ignore buckets with low sample count (n) or small t
Combine with AE-VOL (volatility) or AE-CVD (flow) to layer time × context
⚠️ Important Notes
Seasonality is a historical tendency with no guarantee of persistence (structural shifts break it)
Low-sample buckets remain uncertain even after shrinkage (always check n and t)
Needs a learning period (warmup) and ample history; accumulation resets on timeframe/timezone change
Probability, not a guarantee — always do your own due diligence and use risk management
🚨 Disclaimer
This indicator is for educational and informational purposes only and is not financial advice or a recommendation to buy or sell. No method guarantees future profits; past performance does not indicate future results, and trading carries the risk of loss. All decisions are your own — use proper validation and disciplined risk management.
Скрипт с открытым кодом
В истинном духе TradingView, создатель этого скрипта сделал его открытым исходным кодом, чтобы трейдеры могли проверить и убедиться в его функциональности. Браво автору! Вы можете использовать его бесплатно, но помните, что перепубликация кода подчиняется нашим Правилам поведения.
Отказ от ответственности
Информация и публикации не предназначены для предоставления и не являются финансовыми, инвестиционными, торговыми или другими видами советов или рекомендаций, предоставленных или одобренных TradingView. Подробнее читайте в Условиях использования.
Скрипт с открытым кодом
В истинном духе TradingView, создатель этого скрипта сделал его открытым исходным кодом, чтобы трейдеры могли проверить и убедиться в его функциональности. Браво автору! Вы можете использовать его бесплатно, но помните, что перепубликация кода подчиняется нашим Правилам поведения.
Отказ от ответственности
Информация и публикации не предназначены для предоставления и не являются финансовыми, инвестиционными, торговыми или другими видами советов или рекомендаций, предоставленных или одобренных TradingView. Подробнее читайте в Условиях использования.