Volume and Volume-Based Indicators

Volume is the only series in an OHLCV bar that is not a price, and it is the one most often described in language it cannot support. This lesson separates what volume actually records from what people say it reveals, then builds the two derived measures AmiBroker provides and explains why their absolute values are meaningless.

What volume records, and what it does not

Volume is the number of shares, contracts or lots that changed hands during the bar. That is the entire content of the number.

In particular, it is not a measure of buying. Every executed trade has a buyer and a seller, in equal size, at the same price. There is no such thing as “more buying than selling” in a matched market — there is more aggression on one side, which is a real and different thing, but volume alone cannot tell you which side that was. Order-flow tools that classify trades by whether they hit the bid or lifted the offer exist, and Part 22 covers what AmiBroker offers there; the Volume array is not one of them.

What high volume does say is that a lot of positions changed hands, so a lot of participants re-evaluated at those prices. That is a genuine and useful piece of context. It is a much weaker statement than “the smart money was accumulating”.

The problems with raw volume

⌀Data quality warning

Volume can legitimately be zero. Index symbols, many FX feeds, some continuous futures series and thinly traded instruments carry zero or synthetic volume. Every ratio, every percentage change and every volume-weighted indicator degenerates on those bars — usually silently, occasionally with a division by zero. Look at the raw Volume array before you build anything on it.

It is not comparable between instruments. Ten million shares is enormous for one company and a quiet morning for another. Only volume relative to that instrument’s own history means anything.

It is not comparable across time within one instrument. Turnover in most markets has changed enormously over the decades, and a symbol’s typical volume today may be several times what it was fifteen years ago for reasons that have nothing to do with the current bar.

Split adjustment scales it. AmiBroker’s ASCII importer, when given an ADJCLOSE field, computes a split factor from ADJCLOSE / CLOSE, multiplies the OHLC prices by it and divides the volume by it. That is the right behaviour, and it means the volume figures in an adjusted database are not the share counts that were actually printed on those days. If you are matching against an exchange report, expect a difference.

The last intraday bar is partial. A live 5-minute bar has accumulated only part of its volume, so any “unusual volume” test evaluated intraday can be true at 09:35 and false at 09:40.

Different sources count different things. What a vendor reports as volume — which venues it consolidates, how it treats off-exchange prints — is a property of that vendor. Check its documentation rather than assuming, and never mix two sources in one study.

Average and relative volume

Since only relative volume means anything, the first thing to build is a baseline. There is no dedicated function for it; it is MA applied to the Volume array:

Fragment — not a complete formula

VolAvg = MA( Volume, 50 );                 // mean volume over the last 50 bars
RelVol = Volume / VolAvg;                  // today as a multiple of that mean
Unusual = Volume > 2 * VolAvg;             // the classic filter

Volume distributions are heavily skewed — a handful of enormous days pull the mean up, so “above average volume” happens on well under half of all bars. AmiBroker’s own documentation leans on more robust alternatives in its examples. The Median page’s example is exactly this filter, and the Percentile page’s is the same idea by rank:

Fragment — not a complete formula

// AmiBroker's own documented examples, both more robust than a mean for volume.
AboveTypical = Volume > Median( Volume, 50 );
InTopThird   = Volume > Percentile( Volume, 100, 70 );

Two documented details on those. Median( array, period ) returns the lower middle element when the period is even — it does not average the two middle values; Percentile( array, period, 50 ) does that averaging, at the cost of being slower. And Percentile’s rank argument runs from 0 to 100, not 0 to 1, so Percentile( Volume, 100, 0.7 ) asks for the seven-tenths of one per cent level, not the seventieth percentile. The page warns twice that the function is computation intensive, so keep it out of large optimisation runs.

Guard the division, because MA is Null for the first periods - 1 bars and volume can be zero:

Fragment — not a complete formula

VolAvg    = MA( Volume, 50 );
HasVolume = NOT IsNull( VolAvg ) AND VolAvg > 0;
RelVol    = IIf( HasVolume, Volume / VolAvg, 0 );

OBV and accumulation/distribution

AmiBroker provides two cumulative volume lines. Both take no arguments at all:

Fragment — not a complete formula

Balance = OBV();       // On Balance Volume. OBV( 20 ) is an error.
Flow    = AccDist();   // Accumulation/Distribution line.

The published construction of On Balance Volume is a running total: the bar’s volume is added when the close is higher than the previous close and subtracted when it is lower. The accumulation/distribution line weights each bar’s volume by where the close sat within the bar’s range before accumulating it, so a close at the high contributes its full volume and a close in the middle contributes little.

∴Research note

AmiBroker’s OBV and AccDist pages are both stubs: a one-line description, an empty EXAMPLE box and an empty SEE ALSO. Neither publishes its accumulation rule. So the descriptions above are the published definitions of the indicators, not statements about AmiBroker’s code, and you should not quote a formula for either as “official”. If an exact rule matters to your work, reconstruct a candidate from Cum and Volume and compare it against the built-in on your own data, the way the previous lesson compared BBandTop against MA plus StDev.

Why their level is meaningless

This is the single most important practical fact about both functions.

They are cumulative from the first bar loaded. Not from the start of the year, not from some anchor date — from whatever bar happens to be the earliest one AmiBroker has in memory for that symbol. Change the database’s history depth, change the Analysis range, run the same formula on a symbol with a longer history, and the absolute value changes completely.

✕Common mistake

Writing Buy = OBV() > 5000000; or comparing the OBV values of two different symbols. Both are meaningless. Only the shape of the line — its slope, its turns, its behaviour relative to price over the same window — carries any content, and even that is relative to the window you are looking at. If you need a threshold, threshold something normalised instead, such as the slope of OBV divided by average volume.

AccDist has an additional degenerate case: when a bar’s high equals its low, the “where did the close sit in the range” term has a zero denominator. That happens on limit-locked futures, on halted or suspended stocks, and on padded non-trading days.

Money Flow Index

MFI( periods = 14 ) is the volume-weighted relative of RSI — bounded 0 to 100, default period 14, no input array. Its SEE ALSO points at rsi(), which tells you the intended mental model, but the page does not document its internal smoothing. So do not claim MFI uses Wilder’s smoothing; only RSI and ATR have that documented. Everything said about RSI’s saturation in trends applies here too, with the added requirement that the volume it is weighting by has to be real.

Price-volume claims, and how weak the evidence usually is

Three claims dominate the literature:

1. Volume “confirms” price — a move on high volume is more trustworthy than one on low volume.
2. A breakout on high volume is more likely to hold than one on low volume.
3. Extreme volume marks exhaustion and therefore turning points.

Every one of them is testable, and almost none of the writing that asserts them tests them. Look at what each statement leaves out.

“High volume” is defined after the fact. High relative to what baseline, over what window, measured as a multiple or a percentile? Change that definition and you change which bars qualify. If the definition is chosen after looking at the chart, the claim is unfalsifiable.

There is no horizon. “More likely to hold” over the next three days, three weeks or three years are entirely different assertions with entirely different answers.

There is no base rate. If breakouts hold 45 per cent of the time generally, then high-volume breakouts holding 48 per cent of the time is a small effect that needs a sample size before it means anything — and high-volume breakouts holding 45 per cent of the time means volume added nothing whatsoever.

The confound is enormous. High-volume bars are usually also high-range bars, and volatility clusters. A test that finds bigger subsequent moves after high-volume days may be measuring nothing but that clustering. Any honest volume study needs to ask whether the effect survives after controlling for the size of the move on the signal bar itself.

?Reality check

Claim three — the exhaustion claim — deserves particular scepticism, because it is symmetrical in a way that makes it very hard to be wrong. A huge-volume day near a high is called a buying climax if price then falls, and a breakaway day if price then rises. The same bar, labelled after the outcome. Whenever a piece of technical vocabulary has two names distinguished only by what happened next, it is describing the past rather than classifying the present.

Part 12 contains a full reality check on high-volume breakouts, built with the Analysis tools you will have by then. What you can do now is stop repeating the claims as though they were settled, and notice how rarely the writing that asserts them names a universe, a period or a base rate.

Displaying volume in AmiBroker

Volume needs its own pane, and it is conventionally a histogram. Plot’s colour argument accepts an array as well as a single value, so you can colour each bar by whether the close rose:

Fragment — not a complete formula

VolAvg    = MA( Volume, 50 );
BarTone   = IIf( Close >= Ref( Close, -1 ), colorGreen, colorRed );

Plot( Volume, "Volume",       BarTone,   styleHistogram | styleThick );
Plot( VolAvg, "50-bar mean",  colorBlue, styleLine | styleThick );

The average line matters more than the colouring. Colour alone is a poor carrier of meaning — some readers cannot distinguish those two hues — and the horizontal reference line is what actually lets you judge whether a bar was unusual, rather than whether it merely looks tall next to its neighbours.

For OBV and AccDist, give each its own pane and never label the axis as though the numbers mean something. What you are reading is the slope.

Turning this into a testable question

On my universe, over my chosen period, is the average absolute 10-day forward price change larger after bars whose volume exceeded three times its own 50-bar median than after all other bars — and does any difference survive when I compare only against other bars with a similar true range?

The first clause is the ordinary version of the test. The second clause is what makes it worth running: it is the attempt to separate a volume effect from the volatility clustering that would produce a positive answer even if volume carried no information at all. Designing that second comparison properly is harder than the first, and it is the kind of work that distinguishes a result you can rely on from one you merely like.

Recap

Volume counts what traded; it does not measure buying, because every trade has both sides. Raw volume is not comparable between symbols, across eras, or across data sources, and split adjustment rescales it. Build a baseline with MA( Volume, n ), and prefer Median or Percentile because the distribution is skewed — both appear in AmiBroker’s own documented examples. OBV() and AccDist() take no arguments and accumulate from the first bar loaded, so their absolute level is an artefact of your database; read slope, never level. MFI is volume-weighted momentum with an undocumented smoothing that needs real volume to mean anything. And the popular price-volume claims are testable, rarely tested, and heavily confounded by the fact that busy days are also volatile days.

Check your understanding

Question 1. Why is `Buy = OBV() > 10000000;` a mistake?

• OBV requires a period argument
• OBV accumulates from the first bar loaded, so its absolute level depends on how much history is in the database
• OBV is only valid on intraday data
• OBV returns a percentage, so the threshold should be below 100

Show the answer and why

Answer: OBV accumulates from the first bar loaded, so its absolute level depends on how much history is in the database

Both OBV and AccDist are running totals with no anchor date. Change the history depth or the analysis range and every value changes. Only the shape of the line carries content, and comparisons between symbols are meaningless.

Question 2. Which of these is a documented AmiBroker example of a volume filter?

• Filter = Volume > VolumeMA( 50 );
• Filter = Volume > Median( Volume, 50 );
• Filter = OBV( 50 ) > 0;
• Filter = Percentile( Volume, 0.7 );

Show the answer and why

Answer: Filter = Volume > Median( Volume, 50 );

That line is the example on AmiBroker’s Median page. There is no VolumeMA function; OBV takes no arguments; and Percentile requires an array, a period and a rank between 0 and 100.

Question 3. A study finds that days with volume above three times the 50-day median are followed by larger absolute price moves. What is the most important objection to raise before believing it says anything about volume?

• The 50-day median is too short a window
• High-volume bars tend to be high-range bars and volatility clusters, so the result may be measuring volatility rather than volume
• Volume data is unreliable on most feeds
• Absolute returns cannot be averaged

Show the answer and why

Answer: High-volume bars tend to be high-range bars and volatility clusters, so the result may be measuring volatility rather than volume

The confound is the point. To attribute the effect to volume you have to show it survives when comparing bars of similar range, otherwise you have rediscovered that volatile days follow volatile days.

Question 4. Which statements about the Volume array in AFL are correct? Select all that apply.

• Volume and V refer to the same array, and identifiers are not case-sensitive
• Volume can legitimately be zero on some symbols
• AmiBroker provides a dedicated volume moving-average function
• Split adjustment via ADJCLOSE divides the imported volume by the split factor

Show the answer and why

Answer: Volume and V refer to the same array, and identifiers are not case-sensitive, Volume can legitimately be zero on some symbols, Split adjustment via ADJCLOSE divides the imported volume by the split factor

There is no VolumeMA or MAV built-in — an average of volume is just MA applied to the Volume array. The other three are documented in the language reference and the ASCII importer chapter.

Check my answersClear