OPEN-SOURCE SCRIPT
AetherEdge - KALMAN | State-Space Trend
🖊️ Overview
AE-KALMAN treats price as a noisy observation and recursively estimates the true underlying trend level and velocity (slope) behind it with a two-state Kalman filter. Rather than averaging the past like a moving average, it runs a predict → observe → correct cycle on every bar. Crucially, the filter re-estimates how noisy the market is from its own forecast residuals, so it adapts automatically as volatility shifts. The lag-versus-smoothness trade-off is resolved probabilistically instead of by a fixed setting — that is the essence of AE-KALMAN.
🔶 Key Features
A slope-colored Kalman trend line estimating level and velocity (slope) simultaneously
Predictive bands whose width adapts to the predictive variance S — not a fixed multiplier
Shock detection via a standardized innovation "surprise" z-score — flags moves and regime breaks the model did not expect, with a marker and background tint
Innovation-based adaptive measurement noise — the filter learns the market's noise level from its own residuals and self-tunes (toggle)
A multi-step forecast ray (level + n·slope) with the projected price labeled
A gold-framed HUD showing trend, velocity, surprise z, Kalman gain, band width, noise, and forecast
No repaint — every signal is gated on bar close
Process/measurement noise is ATR-scaled, aligning automatically with each symbol's volatility
🧠 Technical Architecture
The state-space model is a local linear trend. The state vector is x = [level, slope], with transition F = [[1,1],[0,1]] (level advances by slope; slope is a random walk) and observation H = [1,0] (price = level + measurement noise). Each bar runs the standard recursion:
Predict: level' = level + slope; covariance P' = F·P·Fᵀ + Q
Update: innovation y = price − level'; predictive variance S = P'₁₁ + R; Kalman gain K = P'·Hᵀ / S; state correction x = x' + K·y; covariance P = (I − K·H)·P'
The 2×2 case is implemented in closed form by hand, so it needs no matrix inversion and stays numerically transparent.
Adaptive noise: the observed innovation variance estimates S = P'₁₁ + R, so R ≈ Var(y) − P'₁₁ is tracked online with an EWMA (with a floor). The filter therefore behaves smoothly in quiet conditions and responsively when the market turns turbulent.
Regime adaptation: velocity = slope ÷ ATR is compared against a dead-zone (slopeThr) to classify UP / DOWN / FLAT, while the magnitude of the surprise z = y ÷ √S detects shocks (news, fast moves).
Honest scope: this is a linear-Gaussian state-space filter (Kalman) — not a neural network and not a crystal ball. The forecast is a linear extrapolation of the current state.
⚙️ Recommended Settings & Tuning Guide
The defaults are tuned for crypto. The parameters that matter most are measurement noise (measR), process noise (procLvl / procSlp), band width (bandMult), the flat dead-zone (slopeThr), and the shock threshold (shockThr).
Smoother / slower: raise measR / lower procLvl
Faster / more reactive: raise procLvl / lower measR
procSlp controls how quickly trend direction may change — smaller keeps direction stable
Crypto starting points (tune on your chart):
BTC / ETH (1H–4H): defaults are a solid baseline; raise procLvl toward 0.4 to catch turns faster
SOL / XRP and high-volatility alts: wicky and noisy — use measR 2.0–2.5 and slopeThr 0.03–0.05 to suppress false FLAT/flip reads
Scalping (1–15m): raise procLvl (0.5+) for responsiveness; bandMult 1.5–2.0 for reversion entries
Swing (daily): lower procSlp (0.01–0.02) to stabilize slope; raise measR for a smoother line
For volatile crypto, keep adaptive noise ON; switch it off only when you want a fixed, predictable feel
💡 How to Use in Practice
Trend-following: ride the line color (velocity sign) and the UP/DN signals; treat FLAT as a range/stand-aside state
Mean-reversion: taps of the upper/lower predictive band flag overextension; because band width scales with S, small stretches in low vol and large stretches in high vol are judged on the same footing
S/R flips & breakouts: a decisive close through the Kalman level with a velocity flip marks the start of a regime change
Shock markers: highlight news, liquidations, and fakeouts — useful as an overextension warning or the onset of a volatility expansion
Forecast ray: the extrapolation of current level + velocity, handy for higher-timeframe bias and rough targets
Multi-timeframe: read directional velocity from a higher-timeframe AE-KALMAN and time entries on a lower one; agreement of both line colors is a high-confidence filter
Combinations: pair with AE-STRATA (SMC structure + ML/RL) and act only when Kalman velocity and STRATA's smart-money probability agree
⚠️ Important Notes
Initial convergence: from a large initial uncertainty, the filter takes a few dozen bars to settle — wait for values to stabilize at the start of history or right after switching symbols
Lag vs smoothness is a trade-off: smoother means slower; faster means noisier — there is no universal setting
The forecast is not a guarantee: it is a linear extrapolation (level + n·slope) and will miss sharp turns; treat it as a projection of the current state only
Parameter sensitivity: optimal noise settings vary by timeframe and symbol — always tune on the chart
🚨 Disclaimer
This indicator is provided 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 — please use proper backtesting and disciplined risk management.
AE-KALMAN treats price as a noisy observation and recursively estimates the true underlying trend level and velocity (slope) behind it with a two-state Kalman filter. Rather than averaging the past like a moving average, it runs a predict → observe → correct cycle on every bar. Crucially, the filter re-estimates how noisy the market is from its own forecast residuals, so it adapts automatically as volatility shifts. The lag-versus-smoothness trade-off is resolved probabilistically instead of by a fixed setting — that is the essence of AE-KALMAN.
🔶 Key Features
A slope-colored Kalman trend line estimating level and velocity (slope) simultaneously
Predictive bands whose width adapts to the predictive variance S — not a fixed multiplier
Shock detection via a standardized innovation "surprise" z-score — flags moves and regime breaks the model did not expect, with a marker and background tint
Innovation-based adaptive measurement noise — the filter learns the market's noise level from its own residuals and self-tunes (toggle)
A multi-step forecast ray (level + n·slope) with the projected price labeled
A gold-framed HUD showing trend, velocity, surprise z, Kalman gain, band width, noise, and forecast
No repaint — every signal is gated on bar close
Process/measurement noise is ATR-scaled, aligning automatically with each symbol's volatility
🧠 Technical Architecture
The state-space model is a local linear trend. The state vector is x = [level, slope], with transition F = [[1,1],[0,1]] (level advances by slope; slope is a random walk) and observation H = [1,0] (price = level + measurement noise). Each bar runs the standard recursion:
Predict: level' = level + slope; covariance P' = F·P·Fᵀ + Q
Update: innovation y = price − level'; predictive variance S = P'₁₁ + R; Kalman gain K = P'·Hᵀ / S; state correction x = x' + K·y; covariance P = (I − K·H)·P'
The 2×2 case is implemented in closed form by hand, so it needs no matrix inversion and stays numerically transparent.
Adaptive noise: the observed innovation variance estimates S = P'₁₁ + R, so R ≈ Var(y) − P'₁₁ is tracked online with an EWMA (with a floor). The filter therefore behaves smoothly in quiet conditions and responsively when the market turns turbulent.
Regime adaptation: velocity = slope ÷ ATR is compared against a dead-zone (slopeThr) to classify UP / DOWN / FLAT, while the magnitude of the surprise z = y ÷ √S detects shocks (news, fast moves).
Honest scope: this is a linear-Gaussian state-space filter (Kalman) — not a neural network and not a crystal ball. The forecast is a linear extrapolation of the current state.
⚙️ Recommended Settings & Tuning Guide
The defaults are tuned for crypto. The parameters that matter most are measurement noise (measR), process noise (procLvl / procSlp), band width (bandMult), the flat dead-zone (slopeThr), and the shock threshold (shockThr).
Smoother / slower: raise measR / lower procLvl
Faster / more reactive: raise procLvl / lower measR
procSlp controls how quickly trend direction may change — smaller keeps direction stable
Crypto starting points (tune on your chart):
BTC / ETH (1H–4H): defaults are a solid baseline; raise procLvl toward 0.4 to catch turns faster
SOL / XRP and high-volatility alts: wicky and noisy — use measR 2.0–2.5 and slopeThr 0.03–0.05 to suppress false FLAT/flip reads
Scalping (1–15m): raise procLvl (0.5+) for responsiveness; bandMult 1.5–2.0 for reversion entries
Swing (daily): lower procSlp (0.01–0.02) to stabilize slope; raise measR for a smoother line
For volatile crypto, keep adaptive noise ON; switch it off only when you want a fixed, predictable feel
💡 How to Use in Practice
Trend-following: ride the line color (velocity sign) and the UP/DN signals; treat FLAT as a range/stand-aside state
Mean-reversion: taps of the upper/lower predictive band flag overextension; because band width scales with S, small stretches in low vol and large stretches in high vol are judged on the same footing
S/R flips & breakouts: a decisive close through the Kalman level with a velocity flip marks the start of a regime change
Shock markers: highlight news, liquidations, and fakeouts — useful as an overextension warning or the onset of a volatility expansion
Forecast ray: the extrapolation of current level + velocity, handy for higher-timeframe bias and rough targets
Multi-timeframe: read directional velocity from a higher-timeframe AE-KALMAN and time entries on a lower one; agreement of both line colors is a high-confidence filter
Combinations: pair with AE-STRATA (SMC structure + ML/RL) and act only when Kalman velocity and STRATA's smart-money probability agree
⚠️ Important Notes
Initial convergence: from a large initial uncertainty, the filter takes a few dozen bars to settle — wait for values to stabilize at the start of history or right after switching symbols
Lag vs smoothness is a trade-off: smoother means slower; faster means noisier — there is no universal setting
The forecast is not a guarantee: it is a linear extrapolation (level + n·slope) and will miss sharp turns; treat it as a projection of the current state only
Parameter sensitivity: optimal noise settings vary by timeframe and symbol — always tune on the chart
🚨 Disclaimer
This indicator is provided 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 — please use proper backtesting and disciplined risk management.
開源腳本
秉持TradingView一貫精神，這個腳本的創作者將其設為開源，以便交易者檢視並驗證其功能。向作者致敬！您可以免費使用此腳本，但請注意，重新發佈代碼需遵守我們的社群規範。
免責聲明
這些資訊和出版物並非旨在提供，也不構成TradingView提供或認可的任何形式的財務、投資、交易或其他類型的建議或推薦。請閱讀使用條款以了解更多資訊。
開源腳本
秉持TradingView一貫精神，這個腳本的創作者將其設為開源，以便交易者檢視並驗證其功能。向作者致敬！您可以免費使用此腳本，但請注意，重新發佈代碼需遵守我們的社群規範。
免責聲明
這些資訊和出版物並非旨在提供，也不構成TradingView提供或認可的任何形式的財務、投資、交易或其他類型的建議或推薦。請閱讀使用條款以了解更多資訊。