OPEN-SOURCE SCRIPT
AetherEdge - VECTOR | Conformal Forecast
🖊️ Overview
AE-VECTOR forecasts the move over the next H bars and attaches a calibrated prediction interval to the target. A point estimate alone cannot be trusted or sized. AE-VECTOR learns the expected move with online regression and uses conformal prediction to draw a band around the target within which the realized value lands roughly 1 − α of the time. Its essence is quantifying the uncertainty of the forecast, not just its center.
🔶 Key Features
Online linear regression that learns the H-bar move (in ATR units) — m̂
A calibrated target band via conformal regression (target ± q·ATR, coverage ≈ 1 − α)
The band is volatility-normalized, so it widens in high vol and tightens in low vol automatically
A target cone at the right edge (projection line + band box + target/coverage label)
A highly selective signal that fires only when decisive (even the pessimistic band end is beyond price)
A gold HUD showing direction, predicted move, target, band width, coverage, and decisiveness
No repaint — signals gate on bar close; labels use the H-bar-lagged realized move
🧠 Technical Architecture
Regression: trend slope, RSI momentum, short momentum, range position, and volatility regime are standardized (EWMA), and a linear model m̂ = w·x + b is trained by SGD (squared loss, L2). The label is the realized H-bar move (ATR units); the current forecast uses no future data.
Conformal target band: the realized absolute residuals of past forecasts (in ATR units, i.e. volatility-normalized) are stored in a rolling calibration set. For a new forecast, the (1 − α) empirical quantile q gives the band target ± q·ATR. By the split (inductive) conformal guarantee, the realized value falls inside this band with coverage calibrated to about 1 − α.
Honest scope: a linear regressor plus inductive conformal prediction — not a neural network, and not a guarantee of the point estimate. What is guaranteed is the band's coverage.
⚙️ Recommended Settings & Tuning Guide
Key parameters: H (horizon), α (band coverage = 1 − α), moveThr (minimum predicted move), lr (learning rate), warmup.
Lower α → wider, higher-coverage band (conservative); higher → tighter band
H: small (5–8) for scalping, large (20–40) for swing
Crypto starting points (tune on your chart):
BTC / ETH (1H–4H): defaults are the baseline (H = 10, α = 0.10)
SOL / XRP and high-vol alts: moveThr ≈ 0.8 to filter small noisy forecasts; α = 0.08 for a slightly wider band
Scalping (5–15m): H = 5–8, lr = 0.02 for fast adaptation
Swing (daily): H = 20–40, raise calibN to stabilize the quantile estimate
Requiring "decisive" narrows to confident targets where even the pessimistic band end favors the trade
💡 How to Use in Practice
As a target: the cone's center line is the target, the band its uncertainty — an objective basis for take-profit levels and R:R
Reading band width: narrow = stable forecast (model confident), wide = uncertain (stand aside or downsize)
Decisive signals: a calibrated, strong condition that the target sits beyond price even pessimistically — useful at the start of a move
What coverage means: a 90% band has historically contained ~90% of realized values — a calibrated expectation
Multi-timeframe: target direction from a higher-timeframe VECTOR, timing on a lower one
Combinations: pair with AE-QUORUM's directional probability or AE-HELM's trade management, and design exits with VECTOR's band
⚠️ Important Notes
Learning period: until both the regression and the calibration spin up (warmup + minCalib bars), the target shows "warming" and no signals fire
Learning reset: changing inputs, symbol, or timeframe re-learns the internal state
The band is guaranteed, not the center: the point estimate can be wrong, and during shocks the band can briefly under-cover
Linear extrapolation has limits: it is weak to sharp regime changes and unforeseen news — treat it as a projection of the current state
🚨 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 trading decisions are your own — use proper backtesting and disciplined risk management.
AE-VECTOR forecasts the move over the next H bars and attaches a calibrated prediction interval to the target. A point estimate alone cannot be trusted or sized. AE-VECTOR learns the expected move with online regression and uses conformal prediction to draw a band around the target within which the realized value lands roughly 1 − α of the time. Its essence is quantifying the uncertainty of the forecast, not just its center.
🔶 Key Features
Online linear regression that learns the H-bar move (in ATR units) — m̂
A calibrated target band via conformal regression (target ± q·ATR, coverage ≈ 1 − α)
The band is volatility-normalized, so it widens in high vol and tightens in low vol automatically
A target cone at the right edge (projection line + band box + target/coverage label)
A highly selective signal that fires only when decisive (even the pessimistic band end is beyond price)
A gold HUD showing direction, predicted move, target, band width, coverage, and decisiveness
No repaint — signals gate on bar close; labels use the H-bar-lagged realized move
🧠 Technical Architecture
Regression: trend slope, RSI momentum, short momentum, range position, and volatility regime are standardized (EWMA), and a linear model m̂ = w·x + b is trained by SGD (squared loss, L2). The label is the realized H-bar move (ATR units); the current forecast uses no future data.
Conformal target band: the realized absolute residuals of past forecasts (in ATR units, i.e. volatility-normalized) are stored in a rolling calibration set. For a new forecast, the (1 − α) empirical quantile q gives the band target ± q·ATR. By the split (inductive) conformal guarantee, the realized value falls inside this band with coverage calibrated to about 1 − α.
Honest scope: a linear regressor plus inductive conformal prediction — not a neural network, and not a guarantee of the point estimate. What is guaranteed is the band's coverage.
⚙️ Recommended Settings & Tuning Guide
Key parameters: H (horizon), α (band coverage = 1 − α), moveThr (minimum predicted move), lr (learning rate), warmup.
Lower α → wider, higher-coverage band (conservative); higher → tighter band
H: small (5–8) for scalping, large (20–40) for swing
Crypto starting points (tune on your chart):
BTC / ETH (1H–4H): defaults are the baseline (H = 10, α = 0.10)
SOL / XRP and high-vol alts: moveThr ≈ 0.8 to filter small noisy forecasts; α = 0.08 for a slightly wider band
Scalping (5–15m): H = 5–8, lr = 0.02 for fast adaptation
Swing (daily): H = 20–40, raise calibN to stabilize the quantile estimate
Requiring "decisive" narrows to confident targets where even the pessimistic band end favors the trade
💡 How to Use in Practice
As a target: the cone's center line is the target, the band its uncertainty — an objective basis for take-profit levels and R:R
Reading band width: narrow = stable forecast (model confident), wide = uncertain (stand aside or downsize)
Decisive signals: a calibrated, strong condition that the target sits beyond price even pessimistically — useful at the start of a move
What coverage means: a 90% band has historically contained ~90% of realized values — a calibrated expectation
Multi-timeframe: target direction from a higher-timeframe VECTOR, timing on a lower one
Combinations: pair with AE-QUORUM's directional probability or AE-HELM's trade management, and design exits with VECTOR's band
⚠️ Important Notes
Learning period: until both the regression and the calibration spin up (warmup + minCalib bars), the target shows "warming" and no signals fire
Learning reset: changing inputs, symbol, or timeframe re-learns the internal state
The band is guaranteed, not the center: the point estimate can be wrong, and during shocks the band can briefly under-cover
Linear extrapolation has limits: it is weak to sharp regime changes and unforeseen news — treat it as a projection of the current state
🚨 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 trading decisions are your own — use proper backtesting 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.