Relative strength index events depend on the chosen input combination

What a conventional relative strength index compares

A conventional relative strength index compares a market's own upward and downward closes over a defined lookback, most often 14 periods. It is read as a momentum oscillator bounded between 0 and 100.

How the lookback remembers price

After the first lookback window, the original averaging method carries the prior up-average and down-average forward, so both series keep a numeric value even on a session with no up close or no down close. Recalculating the oscillator from only the most recent 14 sessions, rather than from those carried-forward averages, is presented as a way to keep the reading on current data.

On a historical S&P 500 chart, the original 14-period oscillator produced fewer 30-up crossings and fewer 70-down crossings than a recent-window simple-average version of the same lookback. Replacing that simple recent-window average with an exponentially smoothed moving average also increased the number of 70 and 30 crossings relative to the original oscillator during a sideways range on the same series.

Volume, average price, and a two-day span

Adding volume to the modified oscillator is described as a way to reflect whether activity is entering or leaving a market. On a historical Treasury bond series, the volume-inclusive line marked bottoms more distinctly than the original oscillator.

Substituting an average price for the close, whether the high-low mean, the three-price mean, or the four-price mean, produced little difference among those averages. On General Motors, the high-low mean still produced false bullish divergences during a decline.

Pairing the current close with the prior session's open spans two days of price action. On historical Treasury bonds, that two-day price span produced a smoother oscillator line than the close-only original.

Overlay checks and a weekly direction filter

Overlaying the two-day variation on the original oscillator showed a recurring configuration in which the two-day line standing above the original line coincided with topping or downward-trend conditions. A later drop of the two-day line through 30 coincided with the end of a correction on the illustrated equity chart.

Because both the original oscillator and the variants are scaled from 0 to 100, they can print false divergences in trends. Applying the same 14-period exponentially smoothed variant on weekly bars as a direction filter reduced daily whipsaw and false divergences on the NYSE in a historical sample, including a stretch when daily readings crossed 70 while the weekly line stayed up.

Editorial reading: a daily 30 or 70 event, or a mismatch between a price extreme and an oscillator extreme, is a property of the chosen lookback memory, smoother, input series, and sampling interval. Specify that set before treating the chart condition as a complete reading, and use the weekly sampling of the same oscillator to accept or reject the daily case.