Files
QuanTAlib/lib/cycles/phasor/phasor.md
T

215 lines
8.9 KiB
Markdown
Raw Normal View History

# PHASOR: Phasor Analysis (Ehlers)
[Pine Script Implementation of Phasor](https://github.com/mihakralj/pinescript/blob/main/indicators/cycles/phasor.pine)
## Overview and Purpose
The Phasor Analysis indicator, developed by John Ehlers, represents an advanced cycle analysis tool that identifies the phase of the dominant cycle component in a time series through complex signal processing techniques. This sophisticated indicator uses correlation-based methods to determine the real and imaginary components of the signal, converting them to a continuous phase angle that reveals market cycle progression. Unlike traditional oscillators, the Phasor provides unwrapped phase measurements that accumulate continuously, offering unique insights into market timing and cycle behavior.
## Core Concepts
* **Complex Signal Analysis** — Uses real and imaginary components to determine cycle phase
* **Correlation-Based Detection** — Employs Ehlers' correlation method for robust phase estimation
* **Unwrapped Phase Tracking** — Provides continuous phase accumulation without discontinuities
* **Anti-Regression Logic** — Prevents phase angle from moving backward under specific conditions
Market Applications:
* **Cycle Timing** — Precise identification of cycle peaks and troughs
* **Market Regime Analysis** — Distinguishes between trending and cycling market conditions
* **Turning Point Detection** — Advanced warning system for potential market reversals
## Common Settings and Parameters
| Parameter | Default | Function | When to Adjust |
| ------ | ------ | ------ | ------ |
| Period | 28 | Fixed cycle period for correlation analysis | Match to expected dominant cycle length |
| Source | Close | Price series for phase calculation | Use typical price or other smoothed series |
| Show Derived Period | false | Display calculated period from phase rate | Enable for adaptive period analysis |
| Show Trend State | false | Display trend/cycle state variable | Enable for regime identification |
## Calculation and Mathematical Foundation
**Technical Formula:**
**Stage 1: Correlation Analysis**
For period $n$ and source $x_t$:
Real component correlation with cosine wave:
$$R = \frac{n \sum x_t \cos\left(\frac{2\pi t}{n}\right) - \sum x_t \sum \cos\left(\frac{2\pi t}{n}\right)}{\sqrt{D_{cos}}}$$
Imaginary component correlation with negative sine wave:
$$I = \frac{n \sum x_t \left(-\sin\left(\frac{2\pi t}{n}\right)\right) - \sum x_t \sum \left(-\sin\left(\frac{2\pi t}{n}\right)\right)}{\sqrt{D_{sin}}}$$
where $D_{cos}$ and $D_{sin}$ are normalization denominators.
**Stage 2: Phase Angle Conversion**
$$\theta_{raw} = \begin{cases}
90° - \arctan\left(\frac{I}{R}\right) \cdot \frac{180°}{\pi} & \text{if } R \neq 0 \\
0° & \text{if } R = 0, I > 0 \\
180° & \text{if } R = 0, I \leq 0
\end{cases}$$
**Stage 3: Phase Unwrapping**
$$\theta_{unwrapped}(t) = \theta_{unwrapped}(t-1) + \Delta\theta$$
where $\Delta\theta$ is the normalized phase difference.
**Stage 4: Ehlers' Anti-Regression Condition**
$$\theta_{final}(t) = \begin{cases}
\theta_{final}(t-1) & \text{if regression conditions met} \\
\theta_{unwrapped}(t) & \text{otherwise}
\end{cases}$$
**Derived Calculations:**
Derived Period: $P_{derived} = \frac{360°}{\Delta\theta_{final}}$ (clamped to [1, 60])
Trend State:
$$S_{trend} = \begin{cases}
1 & \text{if } \Delta\theta \leq 6° \text{ and } |\theta| \geq 90° \\
-1 & \text{if } \Delta\theta \leq 6° \text{ and } |\theta| < 90° \\
0 & \text{if } \Delta\theta > 6°
\end{cases}$$
> 🔍 **Technical Note:** The correlation-based approach provides robust phase estimation even in noisy market conditions, while the unwrapping mechanism ensures continuous phase tracking across cycle boundaries.
## Interpretation Details
* **Phasor Angle (Primary Output):**
* **+90°**: Potential cycle peak region
* **0°**: Mid-cycle ascending phase
* **-90°**: Potential cycle trough region
* **±180°**: Mid-cycle descending phase
* **Phase Progression:**
* Continuous upward movement → Normal cycle progression
* Phase stalling → Potential cycle extension or trend development
* Rapid phase changes → Cycle compression or volatility spike
* **Derived Period Analysis:**
* Period < 10 → High-frequency cycle dominance
* Period 15-40 → Typical swing trading cycles
* Period > 50 → Trending market conditions
* **Trend State Variable:**
* **+1**: Long trend conditions (slow phase change in extreme zones)
* **-1**: Short trend or consolidation (slow phase change in neutral zones)
* **0**: Active cycling (normal phase change rate)
## Applications
* **Cycle-Based Trading:**
* Enter long positions near -90° crossings (cycle troughs)
* Enter short positions near +90° crossings (cycle peaks)
* Exit positions during mid-cycle phases (0°, ±180°)
* **Market Timing:**
* Use phase acceleration for early trend detection
* Monitor derived period for cycle length changes
* Combine with trend state for regime-appropriate strategies
* **Risk Management:**
* Adjust position sizes based on cycle clarity (derived period stability)
* Implement different risk parameters for trending vs. cycling regimes
* Use phase velocity for stop-loss placement timing
## Performance Profile
### Operation Count (Streaming Mode, per Bar)
For default period $n = 28$:
| Operation | Count | Cost (cycles) | Subtotal |
| :--- | :---: | :---: | :---: |
| ADD/SUB | 6n + 15 ≈ 183 | 1 | 183 |
| MUL | 4n + 8 ≈ 120 | 3 | 360 |
| DIV | 3 | 15 | 45 |
| SQRT | 2 | 15 | 30 |
| SIN | n = 28 | 40 | 1,120 |
| COS | n = 28 | 40 | 1,120 |
| ATAN | 1 | 80 | 80 |
| CMP | 8 | 1 | 8 |
| **Total** | **~390** | — | **~2,946 cycles** |
**Breakdown:**
- **Correlation sums** (over n bars): O(n) operations
- sin/cos reference wave generation: n SIN + n COS = ~2,240 cycles
- Σ(x·cos), Σ(x·sin), Σx, Σcos, Σsin: 4n MUL + 5n ADD = ~364 cycles
- **Correlation coefficients** (R, I):
- Numerators: 4 MUL + 4 ADD/SUB = ~16 cycles
- Denominators (D_cos, D_sin): 6 MUL + 4 SUB + 2 SQRT = ~68 cycles
- Final division: 2 DIV = ~30 cycles
- **Phase angle conversion**: 1 DIV + 1 ATAN + quadrant logic = ~98 cycles
- **Phase unwrapping**: 4 ADD/SUB + comparisons = ~10 cycles
- **Anti-regression check**: comparisons + conditional = ~5 cycles
- **Derived outputs** (optional):
- Derived period: 1 DIV + clamp = ~20 cycles
- Trend state: comparisons = ~5 cycles
### Complexity Analysis
| Mode | Complexity | Notes |
| :--- | :---: | :--- |
| Streaming | O(n) | Correlation requires full period window scan |
| Batch | O(m·n) | m bars × n period = quadratic in total work |
**Memory**: ~(8n + 80) bytes ≈ 304 bytes for n=28 (correlation buffers + phase state)
### SIMD Analysis
| Optimization | Applicable | Notes |
| :--- | :---: | :--- |
| AVX2 vectorization | Partial | Correlation sums vectorizable; phase logic scalar |
| FMA | ✅ | Correlation products: `x[i] * cos[i] + sum` |
| SVML trig | ✅ | Precompute sin/cos tables for fixed period |
| Batch parallelism | ❌ | Phase unwrapping requires sequential processing |
**Optimization strategies:**
1. **Precompute trig tables**: For fixed period, sin/cos values are constant
- Reduces 2,240 cycles to ~56 cycles (table lookup)
- **Optimized total**: ~762 cycles (3.9× speedup)
2. **Running sums**: Maintain Σx, Σ(x·cos), Σ(x·sin) incrementally
- Update: add new term, subtract oldest = O(1) per bar
- **Streaming optimized**: ~200 cycles with precomputed trig + running sums
3. **SIMD correlation**: AVX2 can process 4 doubles simultaneously
- Correlation sums: ~90 cycles (vs 364 scalar)
### Quality Metrics
| Metric | Score | Notes |
| :--- | :---: | :--- |
| **Accuracy** | 8/10 | Robust correlation-based phase estimation |
| **Timeliness** | 6/10 | Inherent lag from period-length lookback |
| **Noise Immunity** | 8/10 | Correlation averaging filters noise well |
| **Adaptability** | 5/10 | Fixed period assumption limits adaptation |
## Limitations and Considerations
* **Parameter Sensitivity:**
* Fixed period assumption may not match actual market cycles
* Requires cycle period optimization for different markets and timeframes
* Performance degrades when multiple cycles interfere
* **Computational Complexity:**
* Correlation calculations over full period windows
* Multiple mathematical transformations increase processing requirements
* Real-time implementation requires efficient algorithms
* **Market Conditions:**
* Most effective in markets with clear cyclical behavior
* May provide false signals during strong trending periods
* Requires sufficient historical data for correlation analysis
Complementary Indicators:
* MESA Adaptive Moving Average (cycle-based smoothing)
* Dominant Cycle Period indicators
* Detrended Price Oscillator (cycle identification)
## References
1. Ehlers, J.F. "Cycle Analytics for Traders." Wiley, 2013.
2. Ehlers, J.F. "Cybernetic Analysis for Stocks and Futures." Wiley, 2004.