1992issue C051-15
Constructing a double-smoothed true strength index
Build the true-strength-index as a construction sequence: form one-period-momentum, apply double-exponential-smoothing to the signed series and to its absolute value, then read a slow pair and a fast pair as separate length choices rather than as a finished system.
- One-period-momentum is the signed change from the prior close to the current close, so an up-move is positive and a down-move is negative.
- The true-strength-index divides double-smoothed one-period-momentum by double-smoothed absolute momentum and scales the ratio by one hundred so the reading stays comparable when the price level changes.
- The divergence-indicator is the unnormalized numerator; stretching or shortening the first smoothing changes whether that curve tracks price or shows divergences.
- A fast-slow-pairing states regime with a longer double-smoothing pair and timing with a shorter pair, so each window can be inspected before a crossover is treated as a rule.
Start from one-period momentum
One-period-momentum is the current close minus the prior close. An up-move is a positive value and a down-move is a negative value.
That signed change is the series later averaged in both the numerator and the denominator.
Write the two-pass smoother
The true-strength-index is built by applying two successive exponential averages to one-period-momentum in the numerator. The same double-exponential-smoothing is applied to the absolute value of that momentum in the denominator. The ratio is then multiplied by 100.
Double-exponential-smoothing means a second exponential average applied to the output of a first exponential average of the same series. The smoothing-constant is the exponential weight estimated as two divided by one plus the chosen lookback length.
Isolate the absolute-value normalizer
The absolute-value denominator is included so the output stays in a bounded range, illustrated as plus and minus 100. That shared numeric scale remains comparable when the price level changes.
The unnormalized numerator of the same construction is the divergence-indicator: successive exponential averages of one-period-momentum without the absolute-value scale.
Read the first and second lengths
One illustrated specification uses first and second smoothings of 25 and 13 periods. A 7-period exponential average of the index is then used as a signal-line, marking candidate turning points by crossover with the bounded series.
A 300-period first smoothing of momentum with the second length set to 1 produces a curve that follows the shape of price. The bounded index compresses amplitude and can show divergences the unnormalized series does not.
Adding a 9-period second exponential pass to that 300-period momentum series removes short fluctuations while keeping turning points close to those of a comparable exponential average of the close.
Reducing the first window from 300 to 100 periods with a 9-period second pass introduces divergences that were absent at the longer window. The unnormalized and bounded series then share the same shape except for scale.
True strength index from the 14-then-3 calculation worksheet

The first pass is a 14-day EMA (α = 0.1333) and the second pass is a 3-day EMA (α = 0.50). The sidebar reports TSI as column F divided by column H and does not apply the ×100 factor written in the article formula. The 4 February 1991 row lists only formulas and is omitted.
Split a slow regime layer from a fast timing layer
One illustrated two-layer procedure uses a fast-slow-pairing. Trend is defined with double smoothing of 100 and 20 periods. Entries and exits are selected with a faster pair of 20 and 6 periods.
An alternative construction keeps the 20-and-6 index for timing and defines trend as a 20-period exponential average of that same fast series.
Suggested construction pairs for experimentation are 20 and 6, 40 and 20, and 80 and 40. They are presented as lag-versus-smoothness trade-offs rather than universal settings.
All readings on this track · 51 readings
- 1984Half-cycle differencing for momentum signals
- 1987Relative strength evaluation under competing optimization criteria
- 1988Weekly MACD as a two-clock momentum confirmation stack
- 1989Equal-weight zero-cross from smoothed spreads
- 1989Testing relative-strength-index reversal rules against trend continuation
- 1989Smoothed three-day futures filter for index option bounces
- 1989Cycle-length windows for momentum, Relative Strength Index, and stochastic construction
- 1991Filtered rally magnitude as a bull-regime breakout
- 1992Constructing true strength from double-smoothed momentum
- 1992Constructing a double-smoothed true strength index
- 1993When momentum structure and breadth break together
- 1993Constructing double-smoothed range and momentum oscillators
- 1993Constructing a two-parameter relative momentum index
- 1993Building a bounded momentum oscillator with RSI smoothing
- 1993Two-speed oscillators with divergence and trendline gates
- 1993Constructing a relative momentum index from the relative strength index
- 1994Constructing daily advance-decline breadth tools
- 1994Building a composite regime score from monetary climate and weekly trend
- 1994Averaging Relative Strength Index and the stochastic oscillator into one reversal oscillator
- 1995Dividend-yield regression as a hold versus momentum gate
- 1996Jump and hold filters for long-term Treasury yield direction
- 1997Constructing extendedness from a 10 percent swing filter
- 1997A range-expansion oscillator that can refuse its own stretch
- 1997RSI trend permission and Fibonacci pullback rules
- 1998Nested midpoint construction for a range-normalized oscillator
- 1999Evaluating Relative Strength Index momentum with zero-line and threshold rules
- 1999Treat RSI and momentum as three mechanical procedures
- 2000Thrust strength figure from moving-average swings
- 2001Constructing a non-range-bound balance of market power score
- 2004Cleaned breadth oscillator and new-high divergence: a swing-market case file
- 2004Evaluating advance-issues-momentum on a fixed-symbol-basket
- 2004Constructing a trend filter from two adjacent high-low windows
- 2006A dollar-versus-commodity extreme as a regime case
- 2008Zero-centered stochastic bands and bracket stops
- 2012Evaluating engulfing momentum across hold windows
- 2012Staged stops as one mechanical entry and exit procedure
- 2012Stacking a relative-strength-index forecast, a trend filter, and long-only momentum
- 2013Constructing fair-value filters from averages and momentum
- 2014Constructing a multi-window slope divergence entry
- 2015Bandedge trend filter construction with inverse crossover rules
- 2017Opposite rules for index price and volatility momentum
- 2018A three-state overlay that colors a trend only after the line clears the bar
- 2018Half-cycle relative-strength index with a Fisher map for cyclic reversals
- 2018Emotion as a rule input when momentum breaks
- 2018Two-bar body expansion as a momentum breakout construction
- 2019Pair a two-day high breakout with a volume-weighted exit on the same chart
- 2020Building reflex and trendflex cross and extreme entry rules
- 2020Construct a dual-series price momentum oscillator overlay
- 2020A multi-timeframe stochastic as a panel of weekly voters
- 2020Centerline crossovers that compare index momentums
- 2020Multi-timeframe stochastic voting as one mechanical rule