Skip to main content
Track Trend filter
33 / 33
Search the archiveLibrary
Library

2020issue C1248-53

Constructing a relative-strength oscillator with a rank-agreement trend filter

Form a relative-strength reading from signed close-to-close changes, then apply a rank-agreement trend filter in place of smoothing. Published constructions keep a matching lookback on both stages and chart the unfiltered series beside the filtered series.

  • The first stage forms a relative-strength oscillator from accumulated close-to-close advances and declines over a stated lookback.
  • The second stage applies a rank-agreement trend filter so oscillator noise can be reduced without a smoothing filter that would delay the reading.
  • Published constructions use a matching 14-observation lookback on both stages and display the unfiltered series, the filtered series, and a zero line.
  • Threshold selection sits outside the core construction. The same rank-agreement step can be applied to other series, including the close.
Entries in this reading2 entries
Current trackTrend filter137 readingsRelative Strength Index186 readings

Form the relative-strength oscillator

The first stage forms a relative-strength reading from signed close-to-close changes. That oscillator is the normalized difference of accumulated up and down close-to-close changes over a fixed lookback.

The oscillator accumulates close-to-close advances and declines over a stated lookback and, when their sum is nonzero, equals the difference of those sums divided by their total.

Published constructions use a 14-observation lookback for this oscillator.

Replace smoothing with rank agreement

The trend filter is a rank-agreement overlay that scores how closely an input series follows a positively sloped reference over a second lookback.

The filter tallies pairwise sign comparisons of the oscillator against a rising-time reference and divides that count by half the lookback times one less than the lookback.

Published constructions use a matching 14-observation lookback for the trend filter.

The rank-agreement step is built to suppress oscillator noise without a smoothing filter that would delay the reading. The trend-filter output is constrained to the interval from -1 to +1, so it denoises oscillators rather than producing a smoothed price path.

What the charts show

Charted constructions display the unfiltered relative-strength series, the filtered series, and a zero line together.

What stays outside the core construction

Threshold selection for the filtered oscillator was left outside the core construction. Demonstration add-ons used a 5-bar exponential signal line or long-period bands, including a 200-bar illustration.

The same rank-agreement construction can be applied to series other than the relative-strength oscillator, including the close, where the result can resemble the unfiltered oscillator.

Daily SPY: 14-bar MyRSI beside the matching rank-agreement NET filter

The noisier trace is the 14-bar relative-strength oscillator built from signed close-to-close changes; the smoother trace is that same series after the 14-bar Kendall rank-agreement filter. NET holds the swing and skips most of the zero-line whip that MyRSI prints after the March 2020 break and the mid-year pullbacks. Readings were taken from the thinkorswim daily SPY subgraph, which plots both series on a −1 to +1 scale; the last on-screen values are 0.66 for MyRSI and 0.09 for NET.
The noisier trace is the 14-bar relative-strength oscillator built from signed close-to-close changes; the smoother trace is that same series after the 14-bar Kendall rank-agreement filter. NET holds the swing and skips most of the zero-line whip that MyRSI prints after the March 2020 break and the mid-year pullbacks. Readings were taken from the thinkorswim daily SPY subgraph, which plots both series on a −1 to +1 scale; the last on-screen values are 0.66 for MyRSI and 0.09 for NET.SPY · Daily · 2020-02-03T00:00:00.000Z to 2020-10-16T00:00:00.000Z

Published defaults RSILength = 14 and NETLength = 14. Turning points are placed to about 0.1 on the −1 to +1 scale. Dates between the labeled months are anchored to visible price landmarks (February peak, 23 March crash, September break). Endpoints 0.66 and 0.09 are the last values printed on the study axis.

Educational research material, not investment advice. Historical source context does not establish present-day performance.
33 of 33 in the Trend filter track
1984Track finished · Next track: TrendlineConstructing the slow stochastic from a five-session range106 readings
All readings on this track · 33 readings
  1. 1988Opening-range brackets, a two-bar trend filter, and bounded stops
  2. 1990Bezier-curve price trend filter
  3. 1992Constructing a damping-index trend filter
  4. 1992Building a random walk index trend filter
  5. 1992Phase diagrams for moving-average trend filters
  6. 1993Volume-weighted change smoothing and trend ranking
  7. 1993Concurrent highest-low filter with a largest-low-fall trigger
  8. 1994Unit-invariant trend filters and the c-test
  9. 1995Constructing cup and cap entries with a three-bar net line
  10. 1997Why a daily timing evaluation depends on interval, lookbacks, and the fitting objective
  11. 2001A volume budget clock for trend-segment construction
  12. 2001Keep three jobs separate when you test a composite score
  13. 2002Evaluating the weekly four-percent close filter as a market-state procedure
  14. 2003Constructing a confirmed zigzag trend filter
  15. 2004Decompose high, low, and close into separate forecast streams
  16. 2005Three-state moving-average breakout bar coloring
  17. 2005Constructing a volume and move-adjusted trend filter
  18. 2005A fifty-day average breakout as a trend permission filter
  19. 2005Current-bar inclusion can mute a stochastic channel break
  20. 2006A stochastic oscillator gated by a long-term exponential average
  21. 2010A construction test for a modified volume-price trend filter
  22. 2011Constructing a Spearman rank trend filter
  23. 2013Constructing a repeated-median slope as a resistant trend filter
  24. 2014Combining a relative-strength index and trend filters for oversold setups
  25. 2014Price-rooted lookbacks for a relative strength index, a moving average, and a trend filter
  26. 2015Evaluating next-session intermarket range forecasts
  27. 2018Read the intermarket weight matrix first, then the predicted moving-average filter
  28. 2018Constructing the stiffness trend filter from moving-average holds
  29. 2018The averaging kernel and the lagged trend gate are separate specifications
  30. 2019A trend filter is not ready to compare until portfolio constraints are written down
  31. 2019Lookback, threshold, and position-capacity for a stiffness trend-filter
  32. 2020Combining a trend filter with a moving average and a stochastic oscillator
  33. 2020Constructing a relative-strength oscillator with a rank-agreement trend filter
By fieldIndicators and filters682 readingsBy year202085 readings
All 137 readings tagged Trend filter
Also on Trend filter5 readings
1997Why a daily timing evaluation depends on interval, lookbacks, and the fitting objective1997Rational group structure with a trend screen, RSI, and bands2004Decompose high, low, and close into separate forecast streams2005Current-bar inclusion can mute a stochastic channel break2006A dominant-cycle split into a trend filter and residual Relative Strength Index