RSI: what it is, and whether it works

5,594 of 26,513 out-of-sample tests beat simply buying and holding the same asset — 21.1%. On the other 20,919 it did not. That is above the 20.1% rate across all 382 indicators we test (one pooled rate over all 660,005 tests we have run, not a mean of the per-indicator rates).

One test is one variant on one asset on one timeframe. We tested 15 variants of RSI, so an asset that carries several of them contributes several tests — this is the family's record across everything we ran, not a count of assets. Those variants are not alike: individually they beat buy-and-hold between 17.3% and 27.7% of the time, and 21.1% is the pooled rate over all of them, not a property every version of this indicator has. The table below gives each one separately. "Beat" means a higher return than holding that same asset over that same window. Measured out-of-sample, on data the setup was not chosen on.

Over a lookback window, conventionally 14 periods, every bar's change is split into gains and losses. The average gain is the sum of the up moves divided by the LOOKBACK LENGTH — not by the number of up bars — and the average loss likewise, so a window with few but large up moves is not flattered. Because both are divided by the same length, the relative strength ratio is simply the sum of gains over the sum of losses; the index rescales that ratio onto a 0-100 axis. After the first window Wilder smooths it recursively rather than recomputing a plain average — each new bar carries the previous average forward as (previous x 13 + new) / 14 — which is why two platforms seeded on different start dates can print slightly different values for a while. Because it is bounded at both ends it can be read as a level rather than only as a direction, which distinguishes it from an unbounded indicator such as MACD.

How it's read. The conventional levels are 70 and 30: above 70 is read as overbought, below 30 as oversold. A second reading is divergence — price making a new high while RSI does not — taken as momentum failing to confirm the move. Both readings are conventions, not properties of the formula.

Where it struggles by design. In a sustained trend RSI can sit above 70 or below 30 for a long period without reversing, so the overbought reading fights the trend for as long as it lasts. Its sensitivity is entirely a function of the lookback: a short window makes it cross the thresholds constantly, a long one makes it almost never cross them.

Origin: Introduced by J. Welles Wilder Jr. in 1978, in New Concepts in Technical Trading Systems.