RSI Strategy — Overbought/Oversold in Practice (+ Connors' RSI-2)

RSI is probably the most-used oscillator in the world — and, at the same time, the most-often-misused. The textbook version sounds familiar: below 30, "oversold" — buy; above 70, "overbought" — sell. The problem is that when someone finally ran the rule on a large sample, the result was brutal: zero edge.

This isn't an article against RSI. It's an article about the difference between the textbook and the measurement — because the same indicator, used differently (Larry Connors' RSI-2 system), showed surprisingly solid results in historical tests. The difference doesn't lie in the indicator; it lies in the rules around it.

What RSI and the Classic 30/70 Strategy Are

RSI (Relative Strength Index) measures the speed and scale of recent price changes on a 0-100 scale. The standard setting is 14 periods. The textbook interpretation:

• RSI > 70 — the market is "overbought": the up move is considered stretched, a correction is expected,
• RSI < 30 — the market is "oversold": the sell-off is stretched, a bounce is expected.

The classic mechanical strategy: buy when RSI(14) drops below 30; sell/reduce when it crosses above 70. There are also more subtle applications on top of this — chiefly divergences (price makes a new high, RSI doesn't), which signal a weakening move.

The intuition behind 30/70 is sound: markets overreact, and the overreaction shows up in momentum. The devil is in the parameters — as you're about to see, "how much of an overreaction," "on which timeframe," and "in which trend" is the difference between a profitable system and a random one.

What the Numbers Say — Textbook vs. Measurement

This is the main hook of this article. Backtests published by QuantifiedStrategies let you compare both versions.

Classic RSI(14) on the daily timeframe: QuantifiedStrategies checked every combination of entry thresholds (45, 40, 35, 30, 25) and exit thresholds (55 to 75) on the S&P 500 since 1993. The best variant — buy below 45, sell at 75 — produced 30 trades in over 30 years and 8.51% a year versus 10.1% for buy-and-hold, with the same maximum drawdown (55%). The classic 30/70 did even worse. The textbook rule, repeated in thousands of courses, lost to simply holding the index on historical data.

Connors' RSI-2 system (on the QQQ ETF): the same indicator family, different rules — and a different world:

The most telling number is the exposure: the system was in the market barely 18% of the time, and its deepest drawdown was almost four times shallower than simply holding QQQ. It only made money when the statistics were on its side, and sat in cash the rest of the time.

The mandatory caveats, without which these numbers are clickbait: sources: QuantifiedStrategies, RSI 14 Debunked and Larry Connors RSI Strategy, as of 09.2026. The tests concern US indexes/ETFs, not crypto — porting this to BTC/ETH requires its own backtest. And as always: past results don't guarantee future ones — a strategy that's published and popular can lose its edge over time.

Why did 30/70 fail while RSI-2 worked? Three structural differences: a trend filter (RSI-2 only buys above SMA200 — it plays dips in a bull market, it doesn't catch falling knives in a bear market), a deeper extreme (a 2-period RSI below 10 is a far rarer, stronger state than a 14-period RSI under 30), and a fast exit (profit is taken on the first bounce, instead of waiting for "overbought").

How to Apply It Step by Step — the RSI-2 System

The classic rules (daily timeframe):

1. Trend filter: the closing price is above the SMA200. Below it — no trades. (An alternative regime filter: the golden cross.)
2. Entry signal: RSI(2) drops below 10 (a more aggressive variant: below 5). You buy on the close of the signal day.
3. Exit: close the position when price closes above the SMA5 (variant: RSI(2) > 70). Win or lose — you get out; this is a quick-bounce system, not a positional one.
4. Risk: size the position with a percent-risk model (the formula is here). Note: in the original tests, the system ran without a classic stop loss — in mean-reversion strategies a stop loss historically tended to worsen results (details and pitfalls) — so risk control comes from a small position size and a time stop instead (e.g. exit after 5 days with no bounce).

A numerical example on ETH (illustrative): a $10,000 account. ETH is in an uptrend (price $2,600 > SMA200 at $2,300). A three-day pullback drags price to $2,450, RSI(2) = 6. You buy at the close for a nominal amount set in advance — assuming an accepted loss of 1% of the account ($100) and an emergency time/catastrophic exit ~4% below entry, position size = 100 / 98 = ~1 ETH. Two days later price closes at $2,570, above the SMA5 — you close with a profit of roughly $120. If the bounce hadn't come within 5 sessions, you'd exit without discussion.

Calibrating RSI(14) to the Market Regime

If you still use classic RSI(14), the thresholds need to fit the regime — a fixed 30/70 really only behaves sensibly in a sideways market:

On top of that, apply the multi-timeframe rule: interpret an H1 RSI reading in the context of the daily. Overbought on the hourly with a neutral daily is usually just a local pause, not a trend-reversal signal.

It's also worth knowing where RSI has its best-documented uses outside of thresholds. Divergences — the gap between new price extremes and a weakening RSI — hit roughly a 58% win rate in cited forex tests, which can be enough for an edge with a good RR. Our own pattern research (a pattern ablation study on crypto data) likewise flagged bearish RSI divergence as one of the few statistically significant short signals. It's the same lesson again: an indicator's value depends on which of its behaviors you use — not on the mere fact that it's on the chart.

RSI signals also gain credibility when combined with levels — support/resistance, Bollinger Bands, or candle structure: oversold RSI right at strong support is a different quality of information than oversold RSI in a vacuum, halfway down a decline with no level underneath.

When It Doesn't Work and Common Pitfalls

• A strong trend kills the 30/70 thresholds. In a bull run, BTC can hold RSI above 70 for weeks — shorting "overbought" in that kind of market is systematically giving money away. Mirror image: in a bear market, an "oversold" altcoin can still be oversold by another -60%.
• RSI-2 without a trend filter is a different (worse) strategy. Buying every RSI(2) extreme, including below the SMA200, means catching falling knives in a bear market — exactly where bounces tend to be shallowest.
• Porting parameters without testing. A system calibrated on QQQ isn't automatically a system for BTC — different volatility, different trading hours (24/7), different costs. Test parameters on your own data, ideally with a training/verification split.
• Fiddling with thresholds after the fact. "30/70 doesn't work, let me check 25/75, 20/80…" — along the way you'll find a combination that looks beautiful on history and doesn't work live. That's overfitting, the main mechanism that produces disappointed traders.
• Ignoring costs on a short horizon. RSI-2 generates fast, short trades; on an expensive exchange, fees and spread can eat a significant chunk of the historical edge.
• Confusing exposure with idleness. 18% time in the market means the system does nothing most days — and that's a feature, not a bug. Traders ruin strategies like this by "manufacturing" signals during quiet periods because they're bored of waiting. If you need daily action, that's a temperament problem no indicator will solve.

One last regime-related note: RSI-2 is a mean-reversion strategy, and markets differ in how strongly they "revert." US stock indexes have historically been among the most mean-reverting markets in the world — and that's exactly what the system was calibrated on. For most of its history, BTC has behaved more momentum-driven (moves once started were more likely to continue than reverse), which means the edge measured on QQQ might simply not exist on crypto, or might require different parameters. That's not a reason to reject the system — it's a reason to test it on BTC/ETH data before you risk your first dollar.

The broader lesson goes beyond a single indicator: "everyone teaches it this way" isn't proof — a measurement is. The classic 30/70 sounds great in a course and doesn't hold up on data; the less-known RSI-2 looked solid on data, though it demands discipline and verification. Before you hand your money to any rule, check whether anyone has actually run the numbers on it. Better yet, run them yourself.

FAQ

Rafał — Krypto Bez Ściemy

Trader and founder of Strefa Tradingu. He’s been breaking crypto down on YouTube for years — no hype, no signals, with a focus on market structure and risk management.

kryptobezsciemy.plInstagramYouTubeKlub Zero