Multiple timeframe analysis

What the top down sequence is 

Top down analysis is a working order, not a technique. The same charts are read in every case; what the convention fixes is the direction of travel through them, from the widest window to the narrowest, and the requirement that each pass be completed before the next one is opened. The direction is the whole of the proposal, because reading the narrowest chart first and then looking for a wider chart that agrees with it is a different procedure that happens to use the same three windows, and it is the procedure the fourth section of this lesson is about.

As it is usually reported in technical literature, the sequence runs in three passes, each answering a narrower question than the one before it.

1. The widest pass. The highest timeframe in the chosen set is described in the terms the earlier lessons in this module established: trending or ranging, where the swing points sit, where the levels that have been revisited most often lie. Nothing narrower is consulted. The output of this pass is a description of a context, and it is a description of past prices at one resolution, nothing more.
2. The intermediate pass. The middle timeframe is described in the same vocabulary, and the two descriptions are then compared. The comparison has three possible outcomes rather than two: the descriptions can agree, they can conflict, or the middle chart can be sideways, which is neither agreement nor conflict and is the most common of the three to mishandle.
3. The narrowest pass. The lowest timeframe is described last. Conventionally this pass is where the precise levels used to place orders are read, because a level read off a weekly chart is several days wide by the time it is drawn on a five minute one. The narrowest chart carries the finest levels and the least context, which is exactly the trade the sequence is built around.

Why the timeframes disagree 

Disagreement between timeframes is not a fault in the data, a defect in the platform or a sign that something has been read incorrectly. It follows directly from the aggregation rule established earlier in this module. A higher timeframe candle keeps the first open, the last close, the highest high and the lowest low of everything inside its window, and discards the order in which those prices were visited. A description built on the surviving four values and a description built on the discarded sequence are answering different questions, so they can differ without either being false.

Anyone describing the weekly chart in that block reports a market that rose. Anyone describing the daily chart reports a market that fell for most of the period under review and then recovered sharply. Both descriptions are accurate accounts of the same prices, and the disagreement is not resolvable by looking harder, because the information the weekly candle would need in order to agree is precisely the information it deleted when it was drawn.

What alignment is, and what it is not 

Alignment is the name given to the case where the descriptions produced at two or more timeframes point the same way. Technical traditions treat it as a condition worth recording and disagree sharply about what, if anything, it indicates. What can be stated without dispute is narrower than the word usually suggests: alignment is a statement about past prices at two resolutions, and the arithmetic that produced both descriptions is arithmetic over the same underlying quotes.

That last point is the one most often lost. The candles on a lower timeframe are not additional history. They are the same history rendered finer, so two charts agreeing is not two independent observations agreeing. It is one body of data described twice, and counting it as two pieces of evidence double counts it in exactly the way lesson two of this module set out. A weekly chart and a daily chart of the same instrument are about as independent of one another as a photograph and a crop of the same photograph.

The failure this workflow is most prone to 

Because timeframes disagree by construction, a chart menu is a supply of contradictory descriptions of one instrument, all of them arithmetically correct. The failure mode follows immediately. A view formed on one chart, followed by a search across the others until one is found that agrees with it, produces agreement almost every time, and the agreement carries no information at all, because the search was going to terminate the moment it appeared.

The two directions produce the identical count, which is the point of computing both. Whatever direction is preferred before the charts are opened, the great majority of possible arrangements contain a timeframe that supports it. A procedure that adds timeframes until one agrees is therefore not testing a view. It is enumerating until it finds the answer it began with, and it will almost always find it.

The remedy conventionally proposed is procedural rather than analytical: the set of timeframes and the order they are read in are fixed in advance, in writing, and the set does not change because of what any one of them shows. Under that convention a conflict between two charts is recorded as a conflict rather than resolved by consulting a third. The convention has an obvious limit, and it is worth stating plainly. Nothing in the price history tells anyone which set to fix, so the choice of set is itself a judgement made before the evidence, and fixing it in advance removes the option of changing it later without removing the arbitrariness of having chosen it.

How far apart the timeframes are conventionally set 

A convention widely repeated in technical writing places adjacent timeframes a factor of roughly four to six apart. The argument offered for it is symmetrical: charts spaced closer than that are near duplicates of one another and the second adds little, while charts spaced much further apart leave a span of price history that neither renders in any detail. No authority establishes the factor, no evidence for it is published here, and the timeframe menus that platforms ship are not spaced by it, which is a reasonable indication of how settled it is.

Timeframe, holding period and cost 

The timeframe a description is drawn on tends to set the holding period, because a level read off a five minute chart is reached, and passed, in minutes, while a level read off a weekly chart may take months to reach. That connection is what makes timeframe choice a cost question as well as a chart question, and the earlier costs module supplies the mechanism: cost is charged per round trip, so the number of round trips taken to cover a given distance decides how much of that distance the costs consume.

Read across the rows, the pattern is arithmetic rather than opinion. Cost is proportional to the number of round trips and independent of how far price moved, so the shorter the working timeframe the larger the share of any given distance the costs account for, in the favourable direction and in the adverse one alike. The adverse rows matter as much as the favourable ones: costs add to a loss on exactly the same terms as they subtract from a gain, and they are charged whether the distance was covered in the expected direction or not.

Where practitioners disagree 

The first disagreement is whether the higher timeframe should govern the lower at all. One tradition holds that the wider chart sets the context and the narrower chart only the timing, on the grounds that a description resting on more history rests on more evidence. Another answers that this is a hierarchy imported from outside the data: every timeframe is the same quote stream aggregated differently, more history is not more independent history, and a weekly candle cannot constrain an hourly one because it is made of them. Neither position can be settled from the chart, since both are consistent with every chart that has ever been drawn.

The second concerns how many timeframes belong in the set. Three is the number most often repeated, and the argument for it is that two leave no way to break a tie while four or more make a contradiction inevitable. The objection is that a tie between two descriptions of the same data cannot be broken by a third description of that same data, because a majority vote among dependent observations is not a vote. Practitioners who accept the objection tend to work from two charts and treat conflicts as a reason to record uncertainty rather than to seek a casting vote.

The third is about what a conflict is for. One school treats an unresolved conflict between timeframes as information in its own right, a signal that the descriptions available are not stable, and leaves it unresolved. Another treats resolution as the purpose of the exercise and reads a conflict as an instruction to look again at a finer resolution. The second position quietly reintroduces the search described earlier in this lesson, since looking again at a finer resolution until the conflict disappears is the same procedure with a more respectable name, and this is the point in the workflow at which the two are hardest to tell apart.