What the Ichimoku cloud is

Five lines from one price series 

The full name of the system is ichimoku kinko hyo, and it comes from a Japanese newspaper publication of the late nineteen sixties, which is why every component carries a Japanese name alongside an English one. The five are the conversion line, also called the tenkan sen; the base line, also called the kijun sen; leading span A and leading span B, together called the senkou spans; and the lagging span, also called the chikou span. Charting software uses the two vocabularies interchangeably and sometimes mixes them in a single settings panel, so both are worth holding.

The inputs are the ones every earlier lesson in this module has used, and there are no others: the high, the low and the close of each bar on whatever timeframe the chart is set to. Nothing external is consulted. No volume, no second instrument, no volatility measure, no rate of change. Four of the five lines are the same single operation applied over different lookback windows, and the fifth is a closing price copied to a different position on the horizontal axis. The apparent complexity of the display comes from where the lines are drawn, not from what is calculated.

The midpoint, not the mean 

The operation underneath four of the five lines is one sentence long. The highest high reached over a stated number of periods is added to the lowest low reached over the same periods, and the result is halved. The conversion line applies it over nine periods and the base line applies it over twenty six, and that is the whole difference between them.

This is not the construction the moving average lesson set out, and the distinction is the most consequential fact in this lesson. A mean uses every value in its window, so every bar contributes and a change in any one of them changes the result. A midpoint of a range uses exactly two values, the extreme at the top and the extreme at the bottom, and discards everything that happened between them. Every bar that is neither the highest nor the lowest contributes nothing at all.

Two properties follow directly, and both are visible on any chart carrying the system. The lines are flat for long stretches, because a range whose extremes have not changed produces the same midpoint bar after bar however much price moves inside it. And when the lines do move, they tend to move in steps rather than smoothly, because a midpoint changes only when a new extreme is set or an old extreme drops out of the back of the window. A line built from a mean drifts. A line built from a range midpoint sits still and then jumps.

The two edges of the cloud 

Leading span A is the midpoint of the two lines already described: the conversion line plus the base line, halved. It is a midpoint of two midpoints, and it introduces no new data. Leading span B is the same range midpoint operation as before, applied over fifty two periods, which is the longest window in the system. Both are then plotted twenty six periods to the right of the bar whose data produced them, and the shaded region between the two is the cloud.

The cloud, or kumo, is that shaded region and nothing more. Its thickness at any point is the vertical distance between two range midpoints calculated over different windows, and its shading conventionally records which of the two spans is on top. A thick cloud describes two midpoints that are far apart, which happens when a short window and a long window cover ranges centred at very different levels. A thin cloud describes two midpoints that are close together. The words thick and thin are descriptions of a gap between two calculated numbers, in the price units of the instrument, and they carry no other content.

What the displacement does 

Displacement is the plainest operation in the whole system and the most frequently misread. A value is calculated on one bar in the ordinary way, and then it is drawn at a different horizontal position. The number is not altered. No extrapolation is performed, no trend is continued, no model of the coming bars is fitted. The line drawn to the right of the last price bar is a copy of numbers that were finished the moment the bars behind it printed.

The lagging span is the same operation in the opposite direction, applied to the simplest possible input. The closing price of the current bar is drawn twenty six periods to the left, above the bars that printed twenty six periods ago. No average is taken and no range is measured. It is one closing price, moved.

Why the region drawn ahead is not a forecast 

The chart shows a shaded region occupying the space to the right of the last price bar, where no bars exist yet. That placement is what makes the system look unlike every other overlay in this module, and it is the source of the one misreading worth spending a section on. The region is not a projection of price. Every value in it was calculated from bars that have already printed, using the arithmetic set out above, and each was simply drawn twenty six positions further right than the bar it came from. The worked block above showed both edges of one point of that region being finished at bar one hundred and drawn at bar one hundred and twenty six.

The consequence is exact. Whatever prices print over the coming bars, the shape already drawn in that region cannot change by a single unit, because none of the coming bars is an input to any value in it. The region will extend further right as new bars print and new values are calculated, and its right hand end will keep moving away, but the part already drawn is fixed. A shape that cannot respond to the thing it is drawn over is not a statement about that thing.

This is the leading and lagging distinction of the earlier lesson, and the system does not escape it. A construction is lagging when every term in it is a price that has already printed, and all five of these lines meet that description without exception. Displacement changes where a number is drawn on the horizontal axis. It does not change what went into the number, and nothing about a drawing position can add information that the arithmetic did not contain. The visual grammar of a chart suggests that anything drawn in the space to the right concerns that space. In this construction it concerns the past, drawn in an unusual place.

The lines move on bars where price does not 

A range midpoint changes for either of two reasons. A new extreme can be set at the front of the window, which is a change caused by the newest bar. Or an old extreme can drop out of the back of the window as it rolls forward, which is a change caused by a bar leaving, and that second case requires nothing whatsoever to happen in the market. The block below rolls a window forward by one bar on which the closing price is identical to the bar before it.

The arithmetic in that block is correct at every step and the data are real. A bar printed, it was identical to the bar before it, and the line moved a full unit. It moved because a high set six bars earlier stopped being inside the window, which is a fact about the calendar rather than about the market. This is a property of every fixed length lookback and not a defect of any particular implementation.

Where practitioners disagree 

The first argument is about the three windows. Nine, twenty six and fifty two are inherited from a publication written when the Japanese trading week ran to six days, so twenty six periods approximated a month of trading days and fifty two approximated two. On a five day week they approximate nothing in particular. One camp adjusts the windows so that they correspond to the same spans of calendar time, and a second replies that any setting fitted to past data was chosen because it described history well, which is not a property that transfers, and that the one merit of the inherited numbers, that a great many chartists are looking at them, is the only merit available. Neither position has displaced the other, and charting software ships the inherited values while exposing the inputs.

The second is narrower and entirely mechanical. Implementations differ over whether a displacement of twenty six periods means twenty six bars beyond the current one or twenty six bars counting the current one as the first, and the two conventions place every displaced line one bar apart. The difference is small, it is visible at the right hand edge of the chart where the cloud is drawn, and it means the same instrument on the same timeframe can carry cloud edges at slightly different horizontal positions in two different packages. Any description of an event occurring at a precise bar therefore depends on which counting convention the package uses.

The third is the substantive one. Technical traditions describe the edges of the cloud as levels of interest in the sense the support and resistance lesson set out, and treat a thick cloud as a wider such zone. The structural objection is that the support and resistance lesson derived its levels from places where transactions clustered, whereas a cloud edge is the midpoint of a range, drawn at a position on the time axis where no transaction occurred at that price on that bar, because the number was produced twenty six bars earlier. Adherents answer that a level watched by enough participants becomes a level regardless of how it was derived, which is an argument about attention rather than about the arithmetic. Both positions are held by practitioners in good faith, and this page reports no figure for how often either description has fitted, because no verified figure exists for it.

What survives all three arguments is small and worth stating plainly. The system is four range midpoints and one closing price, calculated from bars that have already printed, with three of the five drawn at a horizontal position other than the one they were calculated at. Everything in the display is a rearrangement of the high, the low and the close already visible on the chart, and the rearrangement adds no observation that the chart did not already contain.