Session windows, volume weights, and divergence in RSI construction

What classroom RSI maps

Classic relative strength index maps the ratio of average up-close changes to average down-close changes onto a 0-100 scale, with a 14-period lookback used as the common classroom example. That map is a 0-100 momentum transform of average upward price changes against average downward price changes over a stated sampling interval and lookback.

Lock the lookback to actual sessions

If the averages are taken over a fixed count of up days and down days rather than the same number of calendar sessions, a one-sided move can pull in data far older than the advertised lookback. On the NASDAQ Index from 8 January 1985 to 12 March 1985 there were only 14 down days across 43 trading days.

Recomputing the same ratio on only the most recent 14 trading sessions keeps the broad contour of the original oscillator but changes how many turns appear and how sharply they are drawn. The lookback window is the set of observations admitted into the average: an event-count of up or down days is not the same as a calendar run of the same length.

Choose how new data is weighted

Substituting an exponential average for an arithmetic average inside the RSI ratio assigns more influence to the newest observations. That exponential average is a smoothing choice, not a change in the advertised span.

Pick the price sample

The price sample is the increment fed into the ratio. Replacing close-to-close increments with midrange, typical price, or mixed open, high, low, and close differences changes smoothness. A same-session close-versus-open increment was presented as one of the few variants whose contour emphasized peaks as well as troughs.

Volume weight changes the size of each move

A volume-weighted RSI multiplies each up day and each down day by that session's volume before the ratio is formed, so a price change on heavy volume moves the oscillator more than the same price change on light volume. Participation then changes the size of the oscillator move.

In a year-long equity comparison, a marked low on the volume-weighted oscillator coincided with reduced volume in early October 1984.

Reread the bands and divergence

The examples drew horizontal guides at 30 and 70 because those were the original overbought and oversold references, and stated that those levels should be retuned for the stock or commodity under study. Those overbought-oversold bands are instrument-specific marks on the 0-100 scale, not universal constants.

A standard reading of RSI is price-indicator divergence: price making a higher high or lower low while the oscillator does not. Divergence is that chart condition, a new high or low in price that the oscillator fails to confirm. A contemporaneous commodity study reported that pattern on 47.4% of examined cases, slightly less than half the time.