Constructing unsigned true range for directional models

Two measurements come first

A directional-movement construction begins with two measurements: the true range of price movement and the portion of that movement that was directional. True range is assembled first. The directional share is a later measurement and is left aside here.

The largest of three distances

True range is defined as the largest of three distances: today's high minus today's low, today's high minus yesterday's close, and today's low minus yesterday's close. That is the current high-low span together with the two unsigned distances from the prior close to today's high and today's low, keeping only the largest.

The first true-range cell belongs on the second data row because the comparison needs yesterday's close as well as today's high and low.

Keep the quantity unsigned

True range must be stored as a positive quantity. A gap can make a close-to-extreme difference large in magnitude but signed, and a signed value would distort the maximum if it were left unsigned.

Absolute-value conversion is applied to each close-to-extreme gap, not to the outer maximum. It is omitted from the high-low span because that span cannot be negative.

Check the first cell on a control series

A 14-session wheat sample is used as a control series. When named ranges and the three-way maximum are assembled correctly, the first computed true-range value is 3. The sample is used only to check that the programmed comparison matches the definition.

Editorial. Do not attach an average directional index forecast or an ATR position sizing filter until that first control value is in place.

Store the result before the named range moves

After a true-range formula is evaluated, converting that cell to a stored value prevents later movement of the named inputs from rewriting every earlier result to the newest calculation.

A range walk then moves the named input cells down one row after each true-range write so the same comparison can be reused on the next session.

End the loop without stacking calls

A processing loop can stop when the next close cell is blank. That blank-cell stop lets an actively updated price series run without a pre-counted number of rows.

A branch placed inside a called drop routine is treated as a new nested call on each loop. An environment that tracks 32 nested levels can fail near 30 rows even when a 14-row test succeeds.

The nesting failure is avoided by branching only after the drop routine returns, or by clearing the call stack with a restart before branching back to the true-range evaluation.