CVD & divergence

It is 14:25 ET (02:25 ICT, winter). NQ has ground lower all afternoon and just printed a new low at 20,421.25, under the 13:35 low of 20,428.00. Your CVD panel tells a different story: the session’s cumulative delta was −3,100 at the first low and is only −2,950 at the second. Price went lower; the running tally of selling did not.

Many traders call that a buy signal. This lesson shows what the CVD line records, how to read a divergence between two swings, and what the data says when the reversal claim is tested.

Key terms

Cumulative Volume Delta (CVD) (delta tích lũy): the running total of bar delta from an anchor point Definition. If three bars print +200, −150 and +400, CVD reads +200, +50, +450. Bar delta is one bar’s net aggression; CVD is the sum since the anchor.

Divergence (phân kỳ): two swings of the same kind (two lows or two highs) where price and CVD disagree about which swing went further Definition.

Term you need: delta. Delta is volume traded at the ask minus volume traded at the bid. Positive delta means buyers lifted offers more than sellers hit bids. M3.04 covers it in full.

Chart first: two lows, two readings

How it works

Slope, not level

The level of CVD is arbitrary: it depends on where counting started. A reading of −2,950 means only that sellers have been net more aggressive since the anchor. The slope carries the information. Rising price with rising CVD says new buying aggression arrived over that stretch; price still rising while CVD flattens says the rise came on less net buying (the M3.04 thinning, summed over many bars).

Read this as a description of the bars you are looking at, not a forecast. Delta is closely associated with the same bar’s price change (in one public NQ study, 1-minute delta explained about 44% of the same bar’s move and close to nothing of the next bar’s) PlausibleOne trader's study. Using CVD to call what comes next has not held up: in the same study, net delta for a whole session did not predict the next session’s direction Lore.

The anchor decides what you compare

CVD depends on its starting point: the 18:00 ET Globex open, the 09:30 ET regular-session open, the left edge of the chart, or a custom time. The same two swings can show a divergence on one anchor and none on another if a reset falls between them. Overnight and the regular session are different auctions with different participants, so a sensible house default is to reset at the session you trade and only compare swings inside one anchor House rule. Write the anchor on your trade sheet.

Two divergence shapes

Compare two swings of the same kind, inside one anchor:

• New extreme on weaker CVD (often called exhaustion divergence). Bullish: price makes a lower low, CVD makes a higher low. Bearish: price makes a higher high, CVD makes a lower high. The second push went further with less net aggression.
• Held extreme on stronger CVD (often called absorption divergence). Bullish: CVD makes a lower low while price holds a higher low. Bearish: CVD makes a higher high while price holds a lower high. More aggression went in and price moved less, the M3.02 pattern spread over a swing.

These are descriptions Definition. The popular explanations (sellers running out of fuel, passive buyers soaking up the dips) are possible stories, not mechanisms anyone has shown. Divergences also repeat: in a strong trend the same shape can print several times while price keeps going.

Myth / hypothesis: “A CVD divergence warns that the move is ending, so trade the reversal.” Tested on NQ, divergence did no better than the same swing points without it. Treat it as a hypothesis you can log and test, not as a warning that carries odds.

Why it is still worth knowing

You still need the vocabulary. CVD and its divergences describe the auction compactly: who pressed, and whether the pressure grew or shrank between two swings. They mark where to pay attention, give you a cheap invalidation point (a fresh extreme that erases the divergence), and they are exactly the kind of claim the course teaches you to test: write the rule, compare it with a baseline, then decide.

What the data says

Two independent checks, one large and one small, point the same way: on NQ a divergence did not pick turns better than the swings around it. That is “no edge shown”, not “proven useless”, and one narrower variant is still being tested (see “What we don’t know yet”).

Worked example: the afternoon lows, long side

Synthetic session, built to the chart above, not a real date. NQ, 5-minute bars, tick 0.25 points, $5 per tick per contract, CVD anchored at 09:30 ET.

Yesterday’s value-area low, 20,420.00, is on your map. Price falls to 20,428.00 at 13:35 with CVD at −3,100, bounces to 20,441.50, then pushes to 20,421.25 at 14:25 with CVD at −2,950. Lower low in price, higher low in CVD, a little above a mapped level. You write “bullish divergence, 14:25, RTH anchor” on the sheet. That is a note, not an entry.

The plan does not need the divergence to be right:

• If a 5-minute bar closes back above 20,426.00 (the high of the 14:25 bar) with positive bar delta, long 1 contract at 20,426.25.
• Stop 20,419.75, six ticks under the new low and under the level: 6.50 points, 26 ticks, $130 per contract before costs.
• Target 2R, 13 points, at 20,439.25, just under the 13:55 bounce high.
• Invalid if a 5-minute bar closes below 20,421.25 first. The new low erases the divergence and puts price through the level; you stay flat.

The trigger and stop come from the level and the bars, a house rule House rule, not from the divergence. In this synthetic session the 14:40 bar closes at 20,427.50 on delta +190; the order fills at 20,426.25 and the target trades at 15:15. That is one ending, not the expected one.

The mirror and the failure

Short side. Flip the figure. Price makes a higher high on a lower CVD high under a mapped level. The plan mirrors: short only on a close back below the high bar’s low with negative delta, stop a few ticks above the new high, invalid on a close above it.

The failure. Another synthetic afternoon, same shape: lows at 20,428.00 (CVD −3,100) and 20,421.25 (CVD −2,950). The trigger bar never comes. At 14:50 sellers press again and price breaks to 20,409.50 with CVD at −3,600, a lower low on both lines. The divergence is gone, the level is broken, and the trader who bought the 14:25 low at 20,421.50 with the same stop loses 7 ticks, $35 plus costs. Trend days print this pattern again and again. On the data in the boxes, divergences fail about as often as ordinary swings do, which is what no edge looks like.

Do it on the platform

Prop-firm note. The stop under the new low is the number your evaluation’s loss limits see. Size from the stop distance, not from how clean the divergence looks; the Prop Firm course covers how daily and maximum loss limits interact with stop distance.

Common mistakes

Common mistake: Treating a divergence as an entry. “Bullish divergence, buy.” On NQ, divergence swings reversed no more often than ordinary swings. The entry, the stop and the invalidation have to come from somewhere else: the level and the bars that follow.

Common mistake: Treating a base rate as an edge. “Most of my divergences reversed” means little until you know how often ordinary swings at the same places reversed, scored with the same rule. Without that baseline, a 60% hit rate may be the level working, or the day.

Common mistake: Comparing across an anchor. A low at 08:50 ET and a low at 10:15 ET sit on different CVD counts when the anchor is 09:30. The “divergence” between them is the reset, not the market.

Drill: log divergences against ordinary swings

Over five sessions of NQ replay, with CVD anchored at 09:30 ET, mark every confirmed swing low and swing high on 5-minute bars that forms within 8 points of one of your pre-market levels. Before you look at what happens next, write your outcome rule: for example, “reversal” means 10 points away from the swing within 12 bars before 10 points through it.

1. For each swing, log time (ET), level, price, CVD, and whether it forms a divergence with the previous swing of the same kind inside the same anchor.
2. Score every swing with your written rule.
3. Compare the reversal rate of divergence swings with the reversal rate of the other swings at the same levels.

What we don’t know yet

• Whether divergence into a level’s first touch behaves differently from divergence in general. The public study’s variant showed +4.6 pp [−0.0, +9.2], n = 567, with a larger holdout number, still graded Lore as an open lead; M5.04 covers it.
• Our lab’s own graded test of divergence on at least 60 regular sessions of NQ: not yet run; the six-day study above is preliminary.
• Whether divergence interacts with the options regime (dealer gamma context): not yet measured.

Recap

• CVD is the running total of bar delta from an anchor; read its slope, not its level, and write the anchor down.
• Divergence describes two swings: a new price extreme on weaker CVD, or a held extreme on stronger CVD.
• As a reversal signal, divergence is graded Lore on NQ: one large public study and one small lab check found no edge over ordinary swings.
• Keep it as a note on the sheet; take trigger, stop and invalidation from the level and the bars that follow.

Further reading

• Order Flow, Measured: All claims. A public, pre-registered practitioner study on NQ, 2018–2026. Read the divergence rows (DA-01, DA-02), the session-delta row (DA-05) and the first-touch divergence variant.
• Cont, Kukanov and Stoikov (2014), “The Price Impact of Order Book Events”, Journal of Financial Econometrics 12(1). Order-flow imbalance explains price changes over the same interval; that is explanation, not forecast.
• Our lab: a CVD divergence event study on six days of NQ footprint data (November 2026), which found no divergence group that beat ordinary swings, and the delta and CVD concept notes, which grade forecasting with delta Lore.

Next: M3.06 moves from delta over time to delta by price (the delta profile), and applies the same test: describe first, then check the baseline.

Educational content only. Not financial advice. Futures trading involves substantial risk of loss.