On-balance volume (OBV)

Definition

OBV (Joseph Granville’s construction) is a running total that adds each bar’s volume on up-closes and subtracts it on down-closes — a cumulative gauge of whether volume is arriving on strength or weakness. The premise: volume precedes price — accumulation shows in OBV before it shows in price. The premise is folklore in its strong form; the volume-return literature gives it partial, qualified support.

How it works / structure

• Formula: OBV_t = OBV_{t−1} + V_t if C_t > C_{t−1}; − V_t if C_t < C_{t−1}; unchanged if equal.
• Parameters (engine-executable): bar interval, and the signal convention — OBV trend vs price trend (confirmation/divergence, mechanically pinned), OBV breakout ahead of price breakout, or OBV slope as a filter on price signals.
• Properties: the level is meaningless (it depends on the series start); only slope and divergence carry candidate information. All-or-nothing assignment (a +0.01 close books the entire day’s volume as accumulation) is the construction’s crudeness — dollar-volume and signed-by-return variants refine it.

When it applies

Volume-confirmation filtering of price signals (breakouts with rising OBV vs without — strategy-breakout confirmation); divergence scanning as hypothesis generation. On this platform OBV is a secondary filter, never a primary signal — its evidence base is too thin to carry a thesis alone.

Risk profile & failure modes

• Binary assignment noise: near-flat closes flip entire volume days between accumulation and distribution; OBV paths are sensitive to rounding-scale price moves.
• Divergence subjectivity: the standard failure — pinned mechanical definitions or unfalsifiable.
• Volume data quirks: consolidated vs primary-exchange volume, dark-pool prints, and split adjustments all move OBV; the series is only as clean as its volume feed (ms-liquidity).
• Strong-form folklore: “smart money shows in OBV first” is an uncited claim; label and replay.

Evidence & limits

Granville (1963) is the construction source. The academic volume-return literature offers adjacent, not direct, support: Gervais-Kaniel-Mingelgrin (2001) documented a high-volume return premium (unusual volume predicting short-horizon outperformance) — consistent with volume carrying information, but not a test of OBV’s cumulative construction. Direct OBV profitability studies are sparse; the platform treats every OBV convention as unproven-until-replayed.

Falsifiable-thesis examples

Illustrations only, not signals:

• “Breakouts on universe U accompanied by 20-day OBV uptrend outperform those without, over 20 sessions, this year in replay” — falsified by the two buckets.
• “X’s mechanically-defined OBV divergence this week precedes a negative 10-session return” — falsified at the mark.

Cross-references

• Better-grounded volume tools: indicator-vwap, indicator-volume-profile, rel_volume concept
• Habitat: strategy-breakout (confirmation role)
• Data caveats: ms-liquidity, ms-corporate-actions
• Method: lens-technical (divergence discipline)