1994issue C071-15
A comparable group-trend ledger from published ranks
Published group ranks become a shared ledger when one least-squares line is fit to each series, slope is turned into a signed angle, and every group is sorted on the same lookbacks. Rank rotation then assigns unique order ranks instead of leaving the comparison as a chart-reading impression.
- Group relative strength is a published 1-through-99 rank from a capitalization-weighted comparison of group price performance versus other groups over the prior six months.
- A linear-regression trendline through dated ranks supplies a constant slope. Group speed is that slope as a signed angle, and group acceleration is the short-lookback angle minus the long-lookback angle.
- Because every group uses the same lookback intervals, rank rotation sorts all 197 groups by descending trend-angle size and assigns unique ranks from 1 to 197 that ignore absolute score levels.
- The ledger keeps four parallel sorts. After a group is selected, the construction inspects the largest-capitalization names in that group, or a corresponding sector fund.
Group relative strength is constructed as a 1-through-99 rank from a capitalization-weighted comparison of group price performance versus other groups over the prior six months. That published series is the input the ledger scores before any trendline is fit.
Fit one least-squares line to dated ranks
The construction treats a group-rank series as N dated points and fits a least-squares line y = ax + b, with x as days elapsed and y as the integer rank between 1 and 99. That straight line is the linear-regression trendline. It minimizes squared vertical residuals, and its slope is the constant rate of change of approximated rank.
Turn slope into group speed and acceleration
On the fitted line the rate of change of approximated rank is the constant slope a, and group speed is the signed angle arctan(a) relative to the time axis. The sign and size of that angle describe the direction and steepness of rank change.
Group acceleration is the short-lookback speed angle minus the long-lookback speed angle, so its unit is degrees and its sign marks steepening versus flattening of the rank trend. A lookback interval is a fixed count of daily rank observations used to estimate a trendline, commonly a longer window and a shorter window together.
Sort every group on the same lookbacks
Because every group is scored on the same lookbacks, rotation is implemented by sorting the 197 groups by descending trend-angle size and assigning unique ranks from 1 to 197 that ignore absolute score levels. Rank rotation is that shared sort: the order rank, not the absolute score, assigns place.
The ranking ledger uses four parallel sorts: group speed, group acceleration, the mean of ranks over the lookback, and the product of speed and that mean. Illustrated construction windows pair a 20-week trendline with a six-week trendline when ranking groups on speed-times-mean and on acceleration.
Express the group signal in names or a sector fund
After a group is selected from the ranked ledger, the construction next inspects the largest-capitalization names in that group, or a corresponding sector fund, as the practical expression of the group signal.
All readings on this track · 19 readings
- 1987A mechanical rank-rotation sleeve for monthly fund leaders
- 1989Rank rotation in a five-name no-load sleeve
- 1990Cycle-tested five-year fund rank rotation
- 1991Blue-chip rank rotation by relative-strength-index slope
- 1992Currency rank rotation and intermarket timing
- 1992Rank rotation and relative strength for portfolio construction
- 1994MACD crossovers then short-horizon rank rotation
- 1994A comparable group-trend ledger from published ranks
- 1997Normalized yield rank rotation as a full portfolio procedure
- 1997Constructing an investor preference index from two capitalization-weighted series
- 1998Constructing anchored momentum from a centered average
- 2000Rank rotation, a stop-loss order, and Relative Strength Index in fund switching
- 2003A one-fund daily rank is a two-sleeve construction problem
- 2004Evaluate rank rotation only where persistence already exists
- 2004Sector fund rank rotation with regression and trailing stops
- 2006Evaluating equal-weight annual yield-rank rotation
- 2007Weekly preferred-symbol reselection for mechanical trend systems
- 2011Portfolio capacity and entry pacing for mechanical systems
- 2011Rank rotation as a testable ETF construction procedure