Cumulative Volume Delta (CVD) is a running total of signed volume intended to represent buying pressure minus selling pressure. Depending on the implementation, that sign comes from bid/ask trade data or an estimate; neither version can tell you why participants traded or guarantee where price goes next.
That distinction matters. A CVD line may come from actual bid/ask trade data, or from an estimate built from lower-timeframe price direction. The two are not interchangeable, and a neat divergence is not evidence of an edge by itself.
What Cumulative Volume Delta actually measures
In a trade-classified implementation, volume delta for one interval is the volume executed at the ask minus the volume executed at the bid. CVD adds that delta to a running total:
bar delta = ask volume - bid volume
CVD now = CVD before + bar deltaSierra Chart documents Cumulative Delta as the cumulative difference between Ask Volume and Bid Volume. TradingView's CVD documentation likewise accumulates a signed volume delta over a chosen period, although its standard calculation estimates the sign from lower-timeframe bars rather than historical bid/ask prints.
“At the ask” means a buyer accepted the displayed offer; “at the bid” means a seller accepted the displayed bid (CME Group's market-order definition, Nasdaq's market-structure guide). CVD therefore describes aggressive execution, not every intention in the order book. Resting limit orders, cancellations and hidden liquidity are not captured by the simple subtraction. That narrower scope is consistent with Sierra Chart's bid/ask-volume definition and is important because the academic measure studied by Cont, Kukanov and Stoikov also includes changes in best-quote depth from limit orders and cancellations.
True delta and estimated delta are different data
The label “CVD” does not guarantee one calculation. A platform with historical bid/ask trade volume can accumulate the actual classified prints supplied by its data service. Sierra Chart states that its study needs historical BidVolume and AskVolume and may be incomplete when the service does not provide them (Sierra Chart documentation).
Other implementations infer the sign. TradingView says its indicator examines lower-timeframe bars, assigns volume as positive or negative from price direction, then accumulates the result. It also says a lower intrabar timeframe gives more precision but less historical coverage (TradingView CVD calculation). These facts lead to a practical rule: before comparing two CVD charts, identify the feed, trade-classification method, intrabar resolution and reset period.
| Question | Why it changes the line |
|---|---|
| Actual bid/ask prints or inferred direction? | The same trade stream can be classified differently |
| One venue or consolidated data? | The visible volume may cover a different market |
| Session reset or continuous total? | Changing the anchor changes every later CVD value |
| Which session and timezone? | The reset can cut the tape at a different point |
| Tick data or lower-timeframe bars? | Aggregation can hide the sequence inside each bar |
The reset is not cosmetic. Both Sierra Chart and TradingView allow accumulation over a chart, trading day or specified period. A divergence can appear or disappear when the anchor changes because the baseline has changed.
How traders read CVD with price
CVD is usually read beside price, not alone. The common observations are confirmation and divergence:
| Price | CVD | Descriptive reading |
|---|---|---|
| Rising | Rising | Price and classified aggressive buying agree |
| Falling | Falling | Price and classified aggressive selling agree |
| Rising | Falling | Price advances despite net aggressive selling |
| Falling | Rising | Price declines despite net aggressive buying |
The final two rows are divergences. Traders may describe them as absorption, exhaustion or a possible reversal. Those labels are hypotheses, not facts. Price can rise while CVD falls if passive buyers absorb aggressive selling, but the same picture can also arise from an incomplete venue, an estimation method or a reset choice.
There is a second interpretive trap. An aggressive buyer crosses the spread to trade now; that does not mean the buyer is informed or will profit. Hasbrouck's study of trades and quote revisions found that trade impact depends on trade size and market conditions, while Cont, Kukanov and Stoikov found that price response to order-flow imbalance varied with market depth (Hasbrouck paper, order-book event study). Signed volume without liquidity context is an incomplete explanation.
What the evidence supports, and what it does not
Market-microstructure research supports a relationship between order flow, liquidity and short-horizon price changes. It does not establish a universal CVD-divergence trading rule.
Cont, Kukanov and Stoikov studied order-book events for a sample of US stocks and reported a short-interval relationship between price changes and an order-flow imbalance that includes market orders, limit orders and cancellations. Hasbrouck modelled trades together with quote revisions and found that the price impact of a trade was not a fixed quantity. Both support the modest proposition that signed flow contains information within a market-microstructure setting.
Neither paper tests the retail rule “trade when price and CVD diverge.” Their variables are also not identical to a basic ask-volume-minus-bid-volume line. The two studies instead model price response through trades, quotes, depth and broader order-book events, which is a warning against converting one contemporaneous line into a simple forecast (Hasbrouck, Cont, Kukanov and Stoikov).
The honest conclusion is narrower than the sales pitch. CVD is a useful description of classified trading pressure. Whether a particular confirmation, divergence or threshold predicts anything after costs is an empirical question for that instrument, feed and session.
Where CVD analysis breaks
CVD becomes fragile when its inputs or labels stay vague.
- Feed dependence: one venue's volume need not represent the whole market.
- Classification error: inferred up/down volume is not the same as bid/ask trade data.
- Anchor dependence: changing the reset period changes the cumulative path.
- Liquidity blindness: the same signed volume can have different price impact at different depth.
- Hindsight: “obvious divergence” has no stable meaning until the pivots and comparison window are fixed.
- Execution: a short-lived order-flow signal may disappear under spread, slippage and latency.
The first four limitations follow directly from the implementation differences documented by Sierra Chart and TradingView, and from the depth-dependent price impact in Cont, Kukanov and Stoikov and Hasbrouck. The last two are strategy-design problems. A chart annotation made after the move is not a rule.
How you'd actually test it
A CVD backtest must reconstruct the information that existed at the decision time. Testing a line calculated later from a different feed answers a different question.
Freeze these choices before looking at results:
- Name the instrument, venue, data provider, session and timezone.
- State whether delta uses true bid/ask prints or an estimator.
- Fix the intrabar resolution and the CVD reset rule.
- Define the event: level, slope, zero cross, confirmation or divergence.
- For divergence, define both price pivots, both CVD pivots, the maximum separation and the lookback without future bars.
- Specify entry timing, exit, stop loss and position sizing.
- Charge spread, slippage, commission and financing where applicable.
- Separate design data from untouched out-of-sample data.
Then test components rather than one polished rule:
| Test | Question |
|---|---|
| Price-only baseline vs price plus CVD | Does CVD add information? |
| True trade classification vs estimated delta | Is the result a feed artefact? |
| Session reset vs another pre-declared anchor | Is the result stable to the baseline? |
| Confirmation vs divergence | Which claim is actually being tested? |
| Gross vs net | Does the short-horizon effect survive costs? |
| In-sample vs out-of-sample | Did the pattern generalise? |
Report trade count, expectancy, drawdown, exposure and turnover, not just return. Check nearby parameters and different sessions. The realbacktesting guides to parameter sensitivity, time-of-day filters and out-of-sample testing cover the three easiest ways to fool yourself here.
realbacktesting is a trading-software studio for cTrader built around reproducible backtests. For CVD, reproducibility starts with the feed and classification rule: another researcher should be able to rebuild the same line before judging the strategy (how realbacktesting verifies a cTrader backtest).
Frequently asked
What does CVD mean in trading?
CVD means Cumulative Volume Delta. It is a running total of signed volume intended to show whether classified aggressive buying or selling has accumulated over the chosen period.
Is CVD the same on every platform?
No. Platforms can use actual bid/ask trade volume or estimate direction from lower-timeframe bars, and they can reset the total at different anchors. Check the calculation and data feed before comparing values.
Does CVD divergence predict a reversal?
Not by definition. Divergence describes disagreement between price and CVD; reversal is a separate forecast that needs fixed rules, realistic costs and out-of-sample evidence.
Can CVD be used in forex?
Only with a clear understanding of what the feed represents. Nasdaq describes how US equity quotes are consolidated across exchanges, while CME distinguishes retail FX/CFD feeds from a centralized exchange order book (Nasdaq market-structure guide, CME comparison). A broker or venue feed can therefore represent a narrower slice of activity, so the source must be part of the hypothesis.
Takeaway
CVD can make aggressive execution visible. Until the feed, reset and signal are fixed, the line is a description, not a tested edge.