KDJ Indicator: Why the J Line Is Not on the K and D Scale
Three lines load with this indicator, and the third is not measured on the same scale as the other two. The arithmetic producing it appears on the pages a reader is most likely to find; the range it implies is stated on none of them, and one of those pages gives the indicator a range its own formula cannot produce.
Key takeaways
- K and D come from the stochastic oscillator, which MetaQuotes defines as the close expressed as a percentage of the recent high to low range.
- That definition bounds K and D between 0 and 100, while the published J formula of three K minus two D bounds J between -200 and 300.
- Rearranged, J is the K line displaced by twice the distance between K and D, so a J reading reports a gap rather than a level.
- Two readable explanations publish different calculations for K and D under the same 9, 3, 3 label, and neither mentions the other.
- One smooths with plain averages, the other with a recursive form seeded at 50, so the earliest values depend on how much history was loaded.
Table of contents
- What the Three Lines Are Built From
- The J Line Does Not Sit on the Same Scale as K and D
- Two Published Recipes for K and D, Under One Label
- Where the Recursive Version Starts
- What a J Reading Above 100 Actually Says
- Which Numbers Are Missing Here, and Why
- Who This Is Not For
- Checking Which Version Your Platform Runs
What the Three Lines Are Built From
The first two lines are not new. MetaQuotes sets out the calculation in the MetaTrader 5 help for the stochastic oscillator: the main line takes the closing price, subtracts the lowest low seen in the lookback window, divides that by the gap between the same window’s two extremes, and multiplies by 100.
The second line is described in the same document as a moving average of the first. Both answer one question, which is where the close finished inside the recent range, the second more slowly than the first.
KDJ keeps both and adds a third. The two pages on this topic that carry a full calculation publish the same expression for it: J equals three times K minus two times D. That single line is where the indicator stops behaving like the stochastic oscillator it is built from, and most of what follows comes out of it.
The J Line Does Not Sit on the Same Scale as K and D
A closing price always falls inside the extremes of a window that contains its own bar. The numerator of the stochastic can therefore never exceed its denominator and never fall below zero, which fixes K between 0 and 100. D is an average of K values, and an average of numbers between 0 and 100 stays between 0 and 100.
J inherits neither bound. Push the published formula to its corners: with K at 100 and D at 0, three times 100 minus two times 0 gives 300, and with K at 0 and D at 100 the same expression gives -200. The line can run across a span of 500 points while K and D are confined to 100.
The same expression rearranges into something more useful than its corners. Three K minus two D is K plus twice the quantity K minus D, so J is the K line displaced by double the gap between K and D. When the two underlying lines sit on top of each other, J equals them exactly; when they separate, J moves away at twice the speed of the separation.
A chart drawing all three on one axis therefore has to make a choice: either J is clipped where it leaves the panel, or the panel rescales and the 80 and 20 gridlines stop meaning what they mean on a stochastic. One of the two readable explanations states a 0 to 100 range for the indicator as a whole, on the same page that publishes three K minus two D.
Two Published Recipes for K and D, Under One Label
Both readable pages label the settings 9, 3, 3. Underneath that label they describe two different calculations.
The first averages the last three raw stochastic values to get K, then averages the last three K values to get D. A value that leaves the window leaves the calculation completely.
The second builds K as two thirds of the previous K plus one third of the current raw value, and D as two thirds of the previous D plus one third of the current K. Every past bar keeps a share of the result, shrinking as it recedes but never reaching zero.
The two produce different numbers from identical prices: one form forgets on a schedule and the other never quite forgets. MetaQuotes documents the stochastic second line as a simple moving average of the first, matching the shape of the first recipe rather than the second. Neither page acknowledges that the other calculation exists.
| Question a reader has to settle | Averaging form | Recursive form |
|---|---|---|
| How K is built | Average of the last three raw stochastic values | Two thirds of the previous K plus one third of the current raw value |
| Weight on a bar ten periods back | None | Small but never zero |
| Value needed before the first reading | None beyond the window itself | A starting figure, stated as 50 |
| Does loaded history change early values | No | Yes, until the seed decays |
Where the Recursive Version Starts
A calculation that refers to its own previous value has to begin somewhere. The page that publishes the recursive form states that 50 is used for K and D when no previous reading exists, which is the midpoint of their range.
The weight that starting figure keeps is two thirds multiplied by itself once per bar: about 0.017 after ten bars, about 0.0003 after twenty. The first handful of readings on a freshly loaded chart carry a visible trace of the seed, and by roughly two dozen bars the trace is smaller than the rounding on the display.
The consequence is narrow. Two traders on the same pair and timeframe, one with three months of history loaded and one with three years, can read different values at the left edge of the shorter chart and identical values further right.
The averaging form has no such behaviour, because it holds nothing from before its window. Here the published descriptions rather than the data decide the line, as they do on an indicator whose signal line is described two ways.
What a J Reading Above 100 Actually Says
Because J is K plus twice the gap between K and D, a J value folds two pieces of information together, and the level of K is only one of them.
Work two cases. With K at 80 and D at 60, J comes out at 120: past 100 while the underlying stochastic sits at a reading many traders would not call extreme. With K at 95 and D also at 95, J comes out at 95: both underlying lines near the top of their scale and J below 100.
The same J value therefore describes different market states, separated by how far K has pulled away from D. A J line above 100 establishes that K is rising faster than its own average, not that price sits further into the upper part of its range than a K reading of 100, which is arithmetically impossible anyway.
Read that way, J is a speed measurement wearing the clothes of a level. Our page on what an overbought reading does and does not mean covers the level side for the bounded lines, and the same caution applies twice over to a line that is not bounded at all.
Which Numbers Are Missing Here, and Why
The readable pages on this indicator carry thresholds: 80 and 20 on one, 70 and 30 on another, 90 and 10 for the J line on a third, along with the levels of 100 and 0 as J signals. Not one of them attributes any of those numbers to a source.
This indicator is not defined by any exchange, regulator or standards body, and no such institution publishes a level for it. Where several pages repeat a figure that no official source states, this site does not restate it, so no threshold appears here as a level meaning anything by itself.
The same applies to results. None of the sources consulted offers evidence for how often a crossing or an extreme reading is followed by a move in either direction.
Who This Is Not For
A trader who wants one number to read against fixed lines will find J uncooperative, because its position depends on the distance between two other lines as much as on the market.
Anyone on a platform that does not disclose which calculation it runs is in a weaker position again: one settings label covers two sets of values, and no reading compares with a reading taken elsewhere until that is settled.
The indicator suits a reader content to treat K and D as the measurement and J as a derived speed line, and who checks the arithmetic on their own chart rather than adopting a threshold from a page that cites nobody.
Checking Which Version Your Platform Runs
Four steps settle it on any chart that prints the three values.
- Set the periods to 9, 3, 3, hover a single bar and write down the K, D and J values it reports.
- Work out three times K minus two times D by hand. If the result does not match the J on screen, the platform is computing J some other way and its documentation is the only place to find out how.
- Add a plain stochastic oscillator with the same periods to the same chart. If its main line matches K bar for bar, the averaging form is running; a K line that is smoother and lags it points to the recursive form.
- Reload the chart with a much longer history. Values at the left edge that shift while later values stay put confirm a seeded recursive calculation.
None of this makes the indicator better or worse than any other oscillator. It establishes what the numbers on screen actually are, and our wider technical toolkit places the result beside the tools it is usually paired with.
Risk notice. This page is educational and describes how one technical indicator is calculated and what its published descriptions state. Nothing here is a recommendation to buy or sell any instrument, no indicator reading is a forecast, and no figure above is presented as a result anyone should expect. Leveraged trading carries a high risk of loss.