The RSI(2) strategy won 76% of its trades on the S&P 500. Random entries with the same exit won 67%.

Backtested on S&P 500 index daily closes, January 2015 to September 2026 (2,936 days). Updated 24 Sep 2026.

Short answer: not reliably. The RSI(2) strategy looks excellent at first: 92 trades, 76% won, +0.47% a trade. But buying on random days in an uptrend, with the same exit, also won 67% of the time. Most of RSI(2)'s result came from simply being in a rising market for a few days.

Measured fairly, per day held against random entries, RSI(2) was ahead but not by a statistically meaningful amount, and all of that came after 2020. The calendar effects (turn of the month, FOMC days, options expiration week) were no different from an ordinary day.

76% vs 67%RSI(2) win rate vs random uptrend entries, same exit

3.5 vs 2.1days held per trade: RSI(2) vs random entries

0.92p-value for the turn-of-the-month effect (no effect)

5daily-bar ideas tested, none established

What are RSI(2) and calendar effects?

RSI(2) is a short-term version of the Relative Strength Index, calculated over just two days. It was made popular by Larry Connors as a way to buy sharp pullbacks in a market that's otherwise rising, and it's one of the most widely quoted mean-reversion rules for the S&P 500.

Calendar effects are patterns tied to dates rather than prices: the market supposedly rises around the turn of each month, before Federal Reserve (FOMC) announcements, or behaves differently in the week options expire. IBS (internal bar strength) measures where the day closed within its range; a close near the low is said to predict a bounce.

The exact rules I tested

All rules were written down before I looked at any results.

RSI(2): buy at the close when the 2-day RSI is below 10 and the index is above its 200-day moving average. Sell at the first close above the 5-day moving average.
IBS: IBS = (close − low) ÷ (high − low). When it's below 0.2, buy at the close and sell at the next close. The mirror (IBS above 0.8) should then do worse than average.
Turn of the month: hold from the close before the last trading day of the month through the third trading day of the next.
FOMC: hold on scheduled announcement days (92 of them) and, separately, on the day before.
Options expiration week: Monday to Friday of the week containing the third Friday, and the week after.

The fair comparison: each idea's average daily return was compared with the average of all days, using 20,000 random samples of the same size. RSI(2) was also compared with random entry days that met the same 200-day trend filter and used the same exit.

Results

Idea	Days	Average day	All days	Chance it's luck (p)
IBS below 0.2, next day	562	+0.130%	+0.051%	0.06
IBS above 0.8, next day	913	+0.019%	+0.051%	0.29
RSI(2), days in the trade	318	+0.139%	+0.051%	0.14
Turn of the month	563	+0.056%	+0.051%	0.92
FOMC day	92	+0.060%	+0.051%	0.94
Day before FOMC	92	+0.105%	+0.051%	0.64
Options expiration week	683	+0.003%	+0.051%	0.20

A p-value near 0 would mean the result is unlikely to be luck; anything above 0.05 is usually treated as not established.

RSI(2): 92 trades, +0.47% a trade, 76% won. Random uptrend entries with the same exit made +0.16% a trade and won 67%. Per trade, RSI(2) looks far better. But RSI(2) trades lasted 3.5 days against 2.1, and in a rising index more days means more return. Per day held: +0.136% against +0.077%, with a p-value of 0.18. Split in two: 2015 to 2020, RSI(2) did worse than random (+0.031% against +0.048% a day); 2021 to 2026, better (+0.223% against +0.106%), p = 0.067.
IBS: the closest to an effect, but almost all of it came from 2020 to 2022 (+0.24% the next day). In 2015 to 2019 it was +0.005%; in 2023 to 2026, +0.014%. A crash-era pattern, not a lasting one.
Turn of the month: +0.056% a day against +0.051% for all days. No effect.
FOMC: announcement days and the day before were indistinguishable from any other day.
Options expiration week: slightly below average, and all of that from 2020 to 2022.

I also ran a version of RSI(2) (exit when RSI(2) rises above 70 or after 10 days) on 69 US stocks and ETFs since 2010. It beat its own exposure on 48 of them, but only 24 were positive in both halves of the period, and only 4 had a statistically strong result, about what chance alone produces across 69 tries.

Why it failed

A high win rate can come from the exit, not the entry. "Sell on the first close above the 5-day average" means selling into strength, which wins often in a rising market whatever the entry. Random entries with the same exit won two times in three.
Longer holds earn more in a rising market. The S&P 500 drifts up. A rule that stays in the trade longer will earn more per trade even with no skill at all. The fair test is return per day held, and on that measure RSI(2)'s lead shrank and stopped being meaningful.
Effects found in older data fade once published. Turn-of-the-month and pre-FOMC gains were documented in earlier decades. In 2015 to 2026 they weren't there.
One violent period can carry a whole decade. IBS and options expiration week both owed nearly all of their result to 2020 to 2022. Split by period, they disappear.

Did I test it wrong?

Quite possibly. I tested the version of RSI(2) and calendar effects described above, with settings I thought were fair, and it didn't work. That doesn't prove every version fails. I may have picked the wrong settings, the wrong market or timeframe, or missed a filter that experienced traders take for granted. Many strategies also depend on judgement that a set of fixed rules can't capture.

If you trade RSI(2) and calendar effects and it works for you, I'd genuinely like to hear how. Tell me the chart and timeframe, what counts as a setup, any filters, where you get in, where your stop and target go, and when you get out. I'll test your version on the same data, the same way, and publish the result here, whichever way it comes out.

Send me your version

What would change my mind

RSI(2) is the one I'd test again: its 2021 to 2026 result was close to meaningful. If it keeps beating random entries per day held over the next few years, with the same rules, I'd revisit it. For the calendar effects, I'd need to see them return in data after 2026, not in older samples.

Questions

What is the RSI(2) strategy?

A short-term mean-reversion rule popularised by Larry Connors: buy when the 2-day RSI drops below 10 while the index is above its 200-day average, and sell on the first close above the 5-day average.

Does the RSI(2) strategy work?

It made money on the S&P 500, but so did random entries with the same exit, because the index rises over time and the rule holds for a few days. Per day held, its advantage over random entries wasn't statistically meaningful, and it only appeared after 2020.

Is the turn-of-the-month effect real?

Not in 2015 to 2026. The last day of the month and the first three days of the next returned about the same as any other day.

Does the market rise before FOMC announcements?

Research on earlier decades found a rise before FOMC announcements. From 2015 to 2026, in my data, the day before and the day of the announcement were no different from an average day.

How it was tested

Data: S&P 500 index daily bars, 2 January 2015 to 4 September 2026 (2,936 days). Close-to-close returns, before trading costs. The index has no dividends, which affects every idea equally. Seven weeks in early 2016 with damaged high/low data were left out of the range-based tests.
Stocks and ETFs check: daily bars for 69 US stocks and ETFs from 2010, compared against their own time in the market.
Backtest, not live trading: these results come from historical prices with trading costs included. Backtests usually flatter a strategy, so a failed backtest is a strong "no".

This page records my own testing. It isn't financial advice.

Disclosures · Privacy · [email protected]