QuanTAlib/lib/volume/aobv/Aobv.md

# AOBV: Archer On-Balance Volume

> *OBV told me what was happening. AOBV told me when to act.*

| Property         | Value                            |
| ---------------- | -------------------------------- |
| **Category**     | Volume                        |
| **Inputs**       | OHLCV bar (TBar)                          |
| **Parameters**   | None                      |
| **Outputs**      | Multiple series (LastFast, LastSlow)                       |
| **Output range** | Unbounded                     |
| **Warmup**       | `> SlowPeriod` bars                          |
| **PineScript**   | [aobv.pine](aobv.pine)                       |

- Archer On-Balance Volume (AOBV) applies dual exponential smoothing to the classic On-Balance Volume indicator, creating a responsive yet noise-filt...
- No configurable parameters; computation is stateless per bar.
- Validated against TA-Lib, Skender, and Tulip reference implementations where available.

Archer On-Balance Volume (AOBV) applies dual exponential smoothing to the classic On-Balance Volume indicator, creating a responsive yet noise-filtered momentum signal. The intersection of fast and slow EMAs provides actionable crossover signals while preserving OBV's core insight: volume precedes price.

Developed by EverGet (known as "Archer" in the TradingView community), AOBV addresses OBV's fundamental weakness—its sensitivity to single high-volume bars that can distort the cumulative reading. By smoothing with EMAs of period 4 (fast) and 14 (slow), AOBV filters noise while maintaining responsiveness to genuine accumulation/distribution shifts.

## Historical Context

On-Balance Volume (OBV) was introduced by Joseph Granville in his 1963 book "Granville's New Key to Stock Market Profits." The premise was elegant: volume is the fuel that drives price moves. If price rises on high volume, the smart money is accumulating. If it falls on high volume, they're distributing.

Traditional OBV has one critical flaw: it's cumulative and unbounded, making a single aberrant volume bar (earnings, news events) create permanent distortion. AOBV solves this by applying EMAs—not to smooth the OBV value itself, but to create a dual-line system where crossovers filter false signals.

The choice of periods 4 and 14 follows the Fibonacci-adjacent philosophy common in technical analysis. Period 4 captures roughly a week of market action; period 14 represents roughly three weeks. This creates natural separation between short-term noise and medium-term trends.

## Architecture & Physics

AOBV is a three-stage pipeline:

### 1. OBV Accumulation

The foundation is standard OBV logic:

- If today's close > yesterday's close: add volume
- If today's close < yesterday's close: subtract volume
- If closes are equal: add nothing

This creates a running sum that rises during accumulation and falls during distribution.

### 2. Fast EMA (Period 4)

$$
\alpha_{fast} = \frac{2}{4 + 1} = 0.4
$$

The fast EMA responds quickly to OBV changes, capturing short-term accumulation/distribution shifts.

### 3. Slow EMA (Period 14)

$$
\alpha_{slow} = \frac{2}{14 + 1} \approx 0.1333
$$

The slow EMA provides the trend baseline. When fast crosses above slow, it signals strengthening accumulation; crossing below signals distribution.

## Mathematical Foundation

### OBV Calculation

$$
OBV_t = \begin{cases}
OBV_{t-1} + V_t & \text{if } C_t > C_{t-1} \\
OBV_{t-1} - V_t & \text{if } C_t < C_{t-1} \\
OBV_{t-1} & \text{if } C_t = C_{t-1}
\end{cases}
$$

where:

- $C_t$ = Close price at time t
- $V_t$ = Volume at time t

### EMA with Warmup Compensation

Standard EMA suffers from initialization bias. AOBV uses exponential compensation:

$$
\beta_{fast} = 1 - \alpha_{fast} = 0.6
$$

$$
\beta_{slow} = 1 - \alpha_{slow} \approx 0.8667
$$

For each bar, the compensation factor evolves:

$$
e_{fast,t} = e_{fast,t-1} \times \beta_{fast}
$$

$$
c_{fast,t} = \frac{1}{1 - e_{fast,t}}
$$

The compensated EMA:

$$
EMA_{raw,t} = \alpha \cdot OBV_t + (1 - \alpha) \cdot EMA_{raw,t-1}
$$

$$
EMA_{compensated,t} = EMA_{raw,t} \times c_t
$$

This eliminates warmup bias, providing accurate values from the first bar.

### Signal Interpretation

- **Fast > Slow**: Bullish momentum, accumulation strengthening
- **Fast < Slow**: Bearish momentum, distribution strengthening
- **Crossover up**: Buy signal
- **Crossover down**: Sell signal
- **Divergence**: Price making new highs/lows while AOBV fails to confirm

## Performance Profile

### Operation Count (Streaming Mode)

| Operation | Count | Notes |
| :--- | :---: | :--- |
| CMP | 2 | Close comparison for OBV direction |
| ADD/SUB | 3 | OBV update, EMA updates |
| MUL | 8 | Alpha/beta calculations, compensation |
| DIV | 2 | Compensation factors |
| FMA | 2 | EMA calculations via FusedMultiplyAdd |
| **Total** | ~17 | Per bar |

### Memory Footprint

| Component | Bytes | Notes |
| :--- | :---: | :--- |
| State struct | ~88 | 11 doubles (OBV, EMAs, betas, compensators, etc.) |
| Previous state | ~88 | For bar correction rollback |
| **Total** | ~176 | Per instance |

### Quality Metrics

| Metric | Score | Notes |
| :--- | :---: | :--- |
| **Accuracy** | 10/10 | Matches PineScript reference exactly |
| **Timeliness** | 8/10 | Fast EMA (period 4) responds within 2-3 bars |
| **Overshoot** | 6/10 | Unbounded like OBV; EMAs dampen but don't eliminate |
| **Smoothness** | 7/10 | EMAs filter noise; dual-line reduces whipsaws |
| **Allocations** | 0 | Zero heap allocations in Update path |

## Validation

| Library | Status | Notes |
| :--- | :---: | :--- |
| **TA-Lib** | N/A | Has OBV but not AOBV |
| **Skender** | N/A | Has OBV but not AOBV |
| **Tulip** | N/A | Has OBV but not AOBV |
| **Ooples** | N/A | Has OBV but not AOBV |
| **TradingView** | ✅ | Reference implementation by EverGet |

AOBV is a proprietary indicator. Validation is performed against internal consistency checks:

- Streaming matches batch calculation
- Span API matches streaming
- Fast EMA is more responsive than slow EMA
- Warmup compensation produces stable early values

## Common Pitfalls

1. **Warmup Period**: AOBV uses warmup compensation, so values are valid from bar 1. However, `IsHot` only returns true after `SlowPeriod` (14) bars to indicate statistical stability.

2. **Scale Interpretation**: AOBV values are in volume units (potentially millions for high-volume stocks). Compare relative changes and crossovers, not absolute values.

3. **Dual Output**: AOBV produces two values (FastEMA, SlowEMA). The `Last` property returns FastEMA as the primary signal, but trading strategies typically use both for crossover detection.

4. **Volume Quality**: Like all volume indicators, AOBV is only as reliable as the underlying volume data. Crypto wash trading, pre/post-market volume, or adjusted historical data can produce misleading signals.

5. **Fixed Parameters**: Unlike configurable indicators, AOBV uses hardcoded periods (4, 14) matching the original specification. This is intentional—the periods were chosen for their signal characteristics.

6. **isNew Parameter**: Bar correction (isNew=false) properly rolls back state. This is critical for live trading where the current bar updates multiple times before closing.

7. **TValue Not Supported**: AOBV requires OHLCV data (TBar). Attempting to call Update(TValue) throws NotSupportedException.

## References

- Granville, J. (1963). *Granville's New Key to Stock Market Profits*. Prentice-Hall.
- EverGet. "Archer On-Balance Volume (AOBV)." TradingView Script Library.
- StockCharts. "On Balance Volume (OBV)." [Technical Indicators](https://school.stockcharts.com/doku.php?id=technical_indicators:on_balance_volume_obv)