RSI: what 'overbought' does and does not mean

RSI compares the size of recent gains to the size of recent losses over a lookback window — 14 days here, the standard setting. It averages the up moves, averages the down moves, and expresses the ratio on a 0–100 scale. An RSI of 70 means up moves have dominated down moves over the last fortnight by a wide margin; 30 means the reverse.

The important structural detail is that RSI measures the character of recent moves, not their absolute size or the price level. A stock can grind to a new all-time high with an unremarkable RSI if the climb was gradual, and a stock can post a high RSI while still well below its old highs if the recent bounce was sharp.

The overbought and oversold thresholds are inherited defaults from the indicator's original 1970s description. They are not statistical boundaries and they are not calibrated per stock. A low-volatility utility and a high-beta semiconductor will hit 70 under completely different circumstances.

The most common mistake is reading 'overbought' as 'due to fall'. It means the recent balance of gains to losses has been lopsided — which is what a strong uptrend looks like. In a powerful trend RSI can sit above 70 for weeks while price keeps rising, and every day of that stretch looks like a warning to someone reading the threshold as a rule.

Between 30 and 70 the indicator gets a score of exactly zero here — not a small positive or negative lean. That range covers most of the time for most stocks, and squeezing a directional opinion out of an RSI of 54 would be inventing information that the calculation does not contain.

Note also the direction of the scoring: a stretched RSI counts against the composite, not for it. An RSI above 70 contributes a bearish lean because the indicator is flagging that the move is extended. That is a stated modelling choice, and in a sustained uptrend it will be wrong for a while.