TRIX lookback and momentum derivative parameters

Two constructions of the same curve

TRIX is a triple-smoothed exponential average of an ordered price series whose period-to-period change is read as a filtered trend oscillator.

One construction applies three successive exponential averages to a closing-price series and then takes the period-to-period percentage change of the final smooth.

A second construction takes the one-period difference of a triple-exponentially-smoothed logarithm of closes and scales that difference by 10000 for display. The two constructions are described as producing nearly identical curves.

Lookback period as a cycle cutoff

Each extra exponential average further damps short fluctuations. Triple exponential smoothing is intended to remove cycles shorter than the chosen average length, illustrated with a 14-period setting.

The lookback period is that exponential-average length. It sets which shorter cycles the oscillator is meant to ignore.

Positive TRIX readings are treated as a rising filtered trend and negative readings as a falling filtered trend.

Two timing rules on the same smooth

One momentum-strategy rule fires when TRIX crosses a 3-period moving average of itself, illustrated on a 14-period TRIX.

The first difference of TRIX is TRIX momentum, treated as a second derivative of the triple-smoothed average. Peaks of that series mark a slowing of the filtered trend.

One coded rule takes reversals of that difference against a 3-period simple average while the difference is still on the opposite side of zero. In that rule, the opposite side of zero is the signal level the reversal must still meet.

The moving-average cross of TRIX is described as lagging turns by two to five sessions. First-difference turns arrived earlier in the illustrated comparisons, in some cases by as much as a week.

One search and a forward window

System optimization searched TRIX length from 5 to 21 in steps of 4 for both the cross rule and the first-difference rule. The lookback stays inside one entry-and-exit procedure.

After in-sample optimization, four illustrated names were checked in a later window. That forward window produced a similar count and character of signals, offered as support that a series oscillator personality can persist.

The same oscillator rules were described as aligning more cleanly with turns on a cyclical series than on a more persistently trending series. Series personality is the tendency of one market series to stay more cyclical or more trending across adjacent samples.