Reality Check
May 25, 2026

Buying at RSI 30: A Buy on the Index, a Trap on the Stock

We tested the most-taught rule in charting across the S&P 500 and ten big single stocks. On the index, oversold is a real edge. On single stocks it is a coin flip, and on the best stock of the decade it lost money.

It is the first rule most traders ever learn: when the 14-day RSI drops below 30, the asset is "oversold," and oversold means buy. The logic is mean reversion. Price has fallen too far, too fast, and is due to snap back. We put it to the test the honest way. Not "did it go up", because in a long bull market almost everything goes up, but "did buying at RSI 30 beat simply buying on a random day", across the S&P 500 and ten of the largest single stocks. The answer splits cleanly along one line: it works on the index, and it is a trap on the single stock.

On the index, oversold is a real (if small) edge

Median return and percent of the time higher over the next 20 trading days, buying on a random day versus buying the day RSI crossed below 30. Total-return basis.
Buy any dayBuy at RSI 30Edge
S&P 500+1.4%, 66% up+2.9%, 73% up+1.4pp
SPY+1.6%, 68% up+2.8%, 73% up+1.2pp
Nasdaq 100+1.9%, 66% up+1.5%, 56% up-0.4pp

On the S&P 500, buying the RSI-30 dip returned a median 2.9% over the next month and finished higher 73% of the time, a clear step up from the 1.4% you would have made buying on a random day. A diversified index cannot go to zero, so when the whole basket gets pushed to oversold it is usually broad fear overshooting, and it bounces. (The more concentrated, tech-heavy Nasdaq was already a wash, a hint of what is coming.)

On single stocks, it stops working, and on NVDA it backfired

The same test on ten large single stocks, sorted by edge. "Edge" is the RSI-30 median minus the buy-any-day median, over 20 trading days.
StockBuy any dayBuy at RSI 30Edge
TSLA+2.2%+4.5%+2.3pp
GOOGL+1.9%+3.6%+1.8pp
META+2.5%+4.3%+1.7pp
MSFT+1.7%+2.0%+0.3pp
AAPL+2.6%+0.9%-1.7pp
NFLX+2.7%+1.3%-1.4pp
AMZN+2.6%+0.3%-2.3pp
NVDA+3.1%-6.2%-9.2pp

It is a coin flip. Three names rewarded the dip-buyer (TSLA, GOOGL, META), four punished them (AAPL, NFLX, AMZN, and worst of all NVDA). Pool the eight together and buying at RSI 30 returned a median 1.5% over the next month, below the roughly 2.4% you would have made buying these same names on a random day. On a single stock, "oversold" is, on average, a slightly worse-than-nothing entry.

And then there is NVDA. The single best-performing large stock of the era, the one a buy-the-dipper would most want to own, is the one where buying at RSI 30 lost the most: a median of minus 6.2% over the next month, higher only 36% of the time. NVDA's oversold moments were not bargains. They were the middle of its most violent drawdowns, late 2018 and the whole of 2022, when the stock had already fallen hard and the knife kept falling. Buying it "oversold" meant catching that knife, over and over.

On an index, RSI 30 is broad fear that fades. On a single stock, it is just as often the market correctly pricing in something real, and a stock that is oversold can get a great deal more oversold.

The trend filter does not save it

The usual fix is to buy oversold only when the stock is still in an uptrend, above its 200-day average, so you are buying a dip rather than a collapse. It helps for some names and hurts for others, and across the eight single stocks it simply washes out: oversold-in-an-uptrend returned a median 1.4% over the next month, oversold-in-a-downtrend 1.5%, all but identical. For the index, oversold bounced in both uptrends and downtrends, because the index cannot go to zero. For a single stock, even the trend filter cannot reliably tell a dip from a knife.

Why the same rule means opposite things

RSI does not know what it is pointed at. On the S&P 500, an oversold reading aggregates 500 companies, and it takes broad, indiscriminate selling to drag the whole basket down there, the kind of fear that overshoots and reverts. On a single stock, an oversold reading usually means that one company has a specific problem, an earnings miss, a guidance cut, a broken story, and the market is busy repricing it. That is precisely the situation where the decline is justified and tends to continue. Diversification is the whole difference. It is what turns "oversold" from a reason to worry into a reason to buy.

What the Deep Dive Showed

The RSI-30 buy, tested on the index and ten big single stocks, against the only benchmark that matters: buying on a random day.

The index

Oversold works

Buying the S&P 500 at RSI 30 returned a median 2.9% over the next month and was higher 73% of the time, a real step up from buying on a random day.

Single stocks

A coin flip, below baseline

Pooled across eight big names, the RSI-30 buy returned a median 1.5%, under the roughly 2.4% you would have made buying them on a random day. No edge, and often a penalty.

The cautionary tale

NVDA lost money

The best large stock of the decade was the worst RSI-30 buy: a median minus 6.2% over the next month, higher only 36% of the time. Oversold meant the middle of a crash.

The trend filter

Does not rescue it

Only buying oversold in an uptrend washed out across the single stocks: 1.4% in uptrends versus 1.5% in downtrends. It cannot tell a dip from a knife.

How we tested it

Three Takeaways

The same indicator, two opposite meanings, depending on what it is measuring.

1

Oversold is not the same as cheap

On a single stock, RSI 30 often means the market is correctly repricing bad news, not that the price overshot. The rule confuses "fell a lot" with "fell too far," and only one of those bounces.

2

Diversification is what makes mean reversion work

An index reverts because no single problem can break all of it at once. A stock has no such protection. Apply the index trader's oversold rule to a single name and you are catching knives.

3

Always test against the right baseline

In a bull market, "it went up after I bought" proves nothing, because it went up after almost any day you bought. The only honest question is whether the signal beat buying at random. For single stocks at RSI 30, it did not.

More from Reality Check

We test popular signals the honest way: every instance counted, every result measured against a plain baseline. See what else held up, and what did not.

Browse Reality Check

Comments

Loading comments…