Skip to main content
Track Mean reversion
15 / 36
Search the archiveLibrary
Library

2007issue C111-3

Two-period relative strength index versus a one-week universe baseline

Editorial view: lock a liquidity-filtered universe and a one-week hold before treating oscillator length as settled. A January 1, 1995 through December 31, 2006 sample then asked whether two-period relative strength index extremes sat on a different next-week distribution than the 0.25 percent universe baseline.

  • Editorial stance: treat the oscillator lookback as a hypothesis after the liquidity screen and one-week hold are locked, not as a default setting.
  • The historical sample kept names priced above 5 dollars with a 100-day average volume above 250,000 shares and used the 0.25 percent one-week average change of all qualifying names as the comparison baseline.
  • Names with a two-period reading below 2 had a one-week average change of 0.88 percent, while names above 98 had a one-week average change of -0.17 percent. The fourteen-period setting did not show a distinguishable next-week advantage.
  • A 200-day moving-average trend filter, consecutive days below 2, and an intraday decline of 1 to 3 percent were proposed only after that short-horizon comparison.
Entries in this reading3 entries
Current trackMean reversion43 readingsTrend filter137 readingsRelative Strength Index186 readings

Treat the lookback as a hypothesis

This archive article takes an editorial position. A popular oscillator lookback is a hypothesis, not a default. The useful order is to lock a liquidity-filtered universe and a one-week holding window first, then ask whether a two-period extreme sits on a different next-week distribution than the universe average. A long-horizon trend filter is layered on only after that comparison is specified.

The relative strength index is a bounded oscillator that compares recent up-close magnitude with recent down-close magnitude over a chosen lookback so that short-horizon extremes can be compared with a longer default setting. Mean reversion, in this procedure, treats two-period oscillator readings below 2 as oversold candidates and readings above 98 as overbought names to avoid or fade, then exits after five trading days.

The filtered universe and the one-week baseline

The historical sample ran from January 1, 1995 through December 31, 2006. It kept only names priced above 5 dollars with a 100-day average volume above 250,000 shares. More than eight million trades were reviewed.

The one-week baseline is the average five-session percentage change of every name that clears that price and volume screen. In the filtered universe the one-week average change was 0.25 percent. Editorial note: that figure is the explicit comparison for oscillator-selected groups in this sample, not a claim about later markets.

Two-period extremes versus the fourteen-period default

A two-period extreme is a two-period relative strength index close below 2 or above 98. Oversold was defined as a two-period relative strength index below 2, and overbought as a reading above 98.

Names with a two-period reading below 2 had a one-week average change of 0.88 percent in that sample, while names above 98 had a one-week average change of -0.17 percent. The same historical sample compared the two-period relative strength index with the common fourteen-period setting and reported that the longer default did not show a distinguishable next-week advantage.

Two-period RSI extremes versus the one-week universe baseline

A two-period RSI close below 2 was followed by a 0.88 percent average gain the next week, more than three times the 0.25 percent gain of the liquidity-filtered universe, while a close above 98 was followed by a 0.17 percent average loss. Those three percentages are the authors’ stated January 1995–December 2006 results, taken from the article prose and the printed one-week benchmark table, not from the decorative or unlabeled candle drawings.
A two-period RSI close below 2 was followed by a 0.88 percent average gain the next week, more than three times the 0.25 percent gain of the liquidity-filtered universe, while a close above 98 was followed by a 0.17 percent average loss. Those three percentages are the authors’ stated January 1995–December 2006 results, taken from the article prose and the printed one-week benchmark table, not from the decorative or unlabeled candle drawings.US stocks priced above $5 with 100-day average volume over 250,000 shares · One-week hold (five trading days) · 1995-01-01T00:00:00.000Z to 2006-12-31T00:00:00.000Z

Universe filter: price above $5 and 100-day average volume above 250,000 shares; more than eight million trades. Hold is five trading days. The 14-period RSI is described as having little or no edge, but no 14-period return is reported.

A trend filter comes after the oscillator comparison

A trend filter is a long-horizon moving-average screen, here a 200-day average, used only after the short-horizon oscillator comparison is specified, to keep mean-reversion tests inside a broader directional regime. Editorial reading: that order keeps the first question intact, namely whether the two-period extreme differs from the one-week baseline.

A 200-day moving-average trend filter, consecutive days below 2, and an intraday decline of 1 to 3 percent were proposed as further tests around the same short-horizon oscillator.

Educational research material, not investment advice. Historical source context does not establish present-day performance.
15 of 36 in the Mean reversion track
20081-4 pp.Next on Mean reversionBuilding ETF mean-reversion entries with a two-bar washoutKeep the book to diversified exchange-traded funds, hold three names at a time, and treat an oversized decline as a snapback candidate only while the fund remains above a long average of closes.
All readings on this track · 36 readings
  1. 1986A futures fade as one range, order, and secrecy procedure
  2. 1992Constructing the mass-index range-reversal procedure
  3. 1993Switch trend following and mean reversion with an equity-curve filter
  4. 1994Evaluating weekly trend-following and mean-reversion timing rules
  5. 1996Dual-horizon bands for a precious-metals cash switch
  6. 1997Constructing a moving regression oscillator
  7. 1997Regime-dependent long and short rules in mechanical systems
  8. 2002A same-session pair book with a morning-fixed volatility envelope
  9. 2004Combining noncorrelated trend and reversion systems
  10. 2004Failed-breakout overlays on trending markets
  11. 2004Rank rotation after a path split, then Robustness testing
  12. 2004Range-bound tape as a filter for trend and oscillator rules
  13. 2005A moving-average short pullback that is only in scope in a decline
  14. 2006Constructing an adaptive price zone from a double-smoothed range
  15. 2007Two-period relative strength index versus a one-week universe baseline
  16. 2008Building ETF mean-reversion entries with a two-bar washout
  17. 2008Rebuild a short-period stochastic as a premier stochastic oscillator
  18. 2008A three-market regime map for equity bounces and dollar cycles
  19. 2009Option trade adjustment as one testable procedure
  20. 2010Implied volatility as a May 2010 market-regime lab for the S&P 500
  21. 2011Treat a large one-day move as a classified event
  22. 2011Long-call exits, volatility regimes, and spread assignment
  23. 2011Pairing same-horizon oscillators with a walk filter
  24. 2012Two-bar band extreme entries with trailing stops
  25. 2012An eight-month average as a monthly gate for high-yield bonds
  26. 2014Complete the checklist before the trade
  27. 2014Coded rules should face one test, not a kinder sample
  28. 2015Build a mean-reversion basket from one correlation path
  29. 2015Index dip reversion is horizon and regime dependent
  30. 2016Treat the end of a trend as a handoff, not a broken system
  31. 2017A testable half-swing pullback for trend continuation
  32. 2017Evaluating four swing detection rules for mean reversion
  33. 2018Intraday breakout and mean reversion as one rule set
  34. 2018Evaluating rare consecutive-close mean-reversion entries
  35. 2020Moving-average baselines, price vetoes, and mean reversion
  36. 2020Two-dimensional FX scaling for trend and reversal systems
By fieldIndicators and filters682 readingsBy year200773 readings
All 43 readings tagged Mean reversion
Also on Mean reversion5 readings
2018Constructing an adaptive filter for adoption-cycle reversals1997Constructing a moving regression oscillator2008Building ETF mean-reversion entries with a two-bar washout2011Pairing same-horizon oscillators with a walk filter2014Complete the checklist before the trade