Name the window, then combine leaders

Name the calendar window first

The archive workflow builds a combination only after a calendar window is named. A two-column linear-regression screen then ranks which leading indicators currently lead a target, mid-range series are set aside, and the surviving force series are averaged into one composite leading index.

The association between Treasury bonds and a broad equity index is not one number. Recomputing it for every span from 3 days to 300 days produces a different correlation coefficient for each exact horizon. The same bond-equity pair can imply opposite lead stories when one chart covers a short seasonal slice such as May 1987 and another covers the longer overlapping stretch of the first nine months of 1987.

A calendar window is the exact day count and date range that define one seasonal or regime sample. Changing it can reverse the apparent lead, so the window is stated before any ranking begins.

Rank candidates with a linear-regression screen

Scale direction for a candidate leader can be chosen from a two-column linear regression. A correlation coefficient between +0.5 and +1.0 favors a direct scale. A coefficient between -0.5 and -1.0 favors an inverse scale. A mid-range result is a reason to set that series aside for the current window.

That linear-regression screen decides scale direction and ranks candidate leaders before any composite is built. A leading indicator is a market or economic series whose turns are tested as arriving before those of a target index inside the named window.

Rescale into force series

A force series is an affine rescaling of a raw indicator so several series can share a chart with the target without moving reversal dates. Converting raw indicators into comparable force series with an additive and a multiplicative constant changes chart placement and pattern geometry but leaves the calendar dates of reversals unchanged, unlike a moving average that adds lag.

A 20-day moving average introduces a 10-day lag, which can erase a leader whose initial lead is only about five days. Lagging smoothers are a poor substitute for force rescaling.

Leader rank is period-specific

Leader rank belongs to one period, not to a permanent favorite. In 1973-74 one large-cap share price fell 60% while its earnings rose 45%, so earnings were a weak leader then. An inversely scaled oil force later showed a 0.839 correlation with the equity index between 1 October 1990 and 1 February 1991.

Keeping the same favorite indicator through every calendar regime mixes windows where that series is strong with windows where it is weak. A combination must be re-ranked when the seasonal sample changes rather than frozen as a permanent leader.

Average only the surviving forces

A combination built for a named window ranks candidate forces by current regression strength, converts them to a common scale, assigns weights that sum to 10, and averages those weighted forces into one composite leading index.

A second composite that puts more weight on the largest forces can be paired with the first. A line joining their successive turns points toward a later target turn and begins to appear about 2 to 20 days before that turn. The pair is a preliminary screen rather than a lagging crossover. That second series is a secondary guide index, with turns that fall between the first composite and the target.

Date the buy and sell zones

A combined screen can be audited by dating each buy zone and sell zone, growing a theoretical book with the equity benchmark in buy zones and an 8% annual cash rate in sell zones, then comparing that path with the benchmark and counting the share of profitable zones.

A buy zone is a dated interval when the combined screen treats the target as aligned with a benchmark-linked book. A sell zone is a dated interval when the same screen parks the book in a stated cash-rate placeholder.