Build and test a strategy Lesson 9 of 14
Strategies · Mean reversion
Mean reversion vs trend following: two rules on one market
A trend rule and a mean-reversion rule tested side by side on SPY, daily and monthly, with costs: how their trades, win rates, losing periods and bad markets differ.
On this page
Almost every trading rule is one of two bets. A trend-following rule bets that what has been rising will keep rising: it buys strength and sells weakness. A mean-reversion rule bets that what has moved too far will come back: it buys weakness and sells strength. Given the same chart on the same day, the two often want to do opposite things.
Both cannot be right about the same move, and yet both ideas have long histories of evidence behind them. The way out of the puzzle is that they work on different time scales and fail in different markets. This lesson runs one simple rule of each kind on the same SPY data, first on daily bars for two years and then on monthly bars for 33, and compares what comes out: not which one is better, but how differently they behave.
The idea
Think of a price as a long drift with short wobbles on top of it. The two families of rules are aimed at different parts.
- Trend following targets the drift. If prices move in long swings, then being above a moving average today says something about next month. The rule gives up the first part of every move, because it waits for proof, and the last part, because it waits for proof again. In between, it holds.
- Mean reversion targets the wobble. If a two-day drop is mostly noise, the next few days tend to undo part of it. The rule buys the drop, takes the rebound and leaves.
That difference in target produces opposite trade statistics, which is the most useful thing to know about each family before using either.
| Feature | Trend following | Mean reversion |
|---|---|---|
| Buys | After a rise | After a fall |
| Holds for | Weeks to years | Days |
| Win rate | Low | High |
| Size of gains and losses | A few large gains, many small losses | Many small gains, a few large losses |
| Time invested | Most of the time | A small part of the time |
| Worst market | Sideways and choppy | A decline that does not bounce |
| How it feels | Often wrong, occasionally very right | Usually right, occasionally very wrong |
The rules of the two strategies
Both are checked once a day at the close and trade at that close. Both hold SPY or cash.
Trend rule (a 50-day moving average):
- Compute the average of the last 50 closes.
- If the close is above the average, hold the fund.
- If the close is at or below the average, hold cash.
Reversion rule (the two-day RSI):
- Compute the two-day RSI.
- If no position is held and the RSI is below 10, buy.
- If a position is held and the close is above the five-day average, sell.
The reversion rule is the one tested in full in RSI mean reversion: the two-day RSI rule, tested. The trend rule is a daily cousin of the monthly rule in Moving average trend following: the 10-month rule, tested; a 50-day average is used here because the 200-day version would consume 200 of the 500 daily bars available.
The same two days, opposite trades
On February 5, 2026 SPY closed at $677.62, below its 50-day average after two weak sessions. The trend rule sold at $677.62. The two-day RSI was under 10, so the reversion rule bought, at the same price.
The next day SPY closed at $690.62, back above both averages. The reversion rule sold for a gain of 1.7% after costs. The trend rule bought back at $690.62, 1.9% higher than it had sold. One rule's small gain was the other's small loss, on the same two closes.
It does not always go that way. On April 3, 2025 the reversion rule bought at $536.70 and was 7.5% down three sessions later; the trend rule had been in cash since February 24 and stayed there until May 1.
What the daily backtest shows
The test, stated in full so it can be checked:
- Data. SPY daily closes. The 50-day average needs 50 bars, so both rules are measured over the same 452 trading days, from January 31, 2025 to November 17, 2026.
- Trading. Signal and trade at the same close.
- Costs. 0.10% of the amount on every buy and every sell.
- Left out. Dividends, interest on cash and taxes. The reversion rule is in cash 83% of the time, so interest would have helped it most; both rules produce short-term gains.
| Measure | Trend rule | Reversion rule | Buy and hold |
|---|---|---|---|
| Total return | 13.9% | 24.0% | 21.7% |
| Return per year (CAGR) | 7.5% | 12.8% | 11.6% |
| Deepest fall from a peak | −9.1% | −7.6% | −19.0% |
| Longest time below a peak | 134 trading days | 26 trading days | 88 trading days |
| Time invested | 73% | 17% | 100% |
| Trades | 17 | 27 | 1 |
| Trades with a gain | 24% | 78% | n/a |
| Average gain | 7.0% | 1.3% | n/a |
| Average loss | −1.0% | −0.8% | n/a |
The reversion rule came out ahead in this stretch. That ranking is the least reliable line in the table: it is 22 months of one fund. The shape of the two columns is what carries over.
The trend rule won 4 trades out of 17. Two of those four made almost all of its money: one held from May 1 to November 17, 2025 (19.0%) and one from April 8 to June 26, 2026 (7.6%). The other fifteen were small, mostly losses of about 1%. The reversion rule won 21 of 27, none by more than 2.4%.
The two halves of the period
| Period | Trend rule | Reversion rule | Buy and hold |
|---|---|---|---|
| January 2025 – December 2025 | 18.9% | 10.7% | 14.3% |
| December 2025 – November 2026 | −4.3% | 11.9% | 6.5% |
Split the period and the ranking flips. In 2025 the market fell hard, then rose for six months without a real pause: the trend rule skipped the fall, caught the rise and beat both. In 2026 the market chopped sideways around its 50-day average, and the trend rule made 13 trades, of which three gained. A reader who saw only one half would reach the opposite conclusion from a reader who saw the other.
Month by month
The clearest difference shows in the months when the fund fell:
| Month | SPY | Trend rule | Reversion rule |
|---|---|---|---|
| March 2025 | −5.9% | 0.0% | 1.6% |
| April 2025 | −0.9% | 0.0% | 5.3% |
| May 2025 | 6.3% | 5.4% | 1.2% |
| June 2025 | 4.8% | 4.8% | 0.0% |
| February 2026 | −0.9% | −6.0% | 2.3% |
| March 2026 | −5.2% | 0.0% | 1.8% |
| April 2026 | 10.5% | 6.2% | 0.0% |
| July 2026 | 0.0% | −2.2% | 2.1% |
When the market trended up (May and June 2025, April 2026) the trend rule earned and the reversion rule had nothing to do. When it chopped (February and July 2026) the trend rule lost on false signals and the reversion rule collected. Across the months of the test the two rules' monthly returns had a correlation of −0.31: they tended to do well at different times.
The two rules held the fund together on only 24 of the 452 days. Putting half of the money in each, rebalanced daily, returned 19.3% with a deepest fall of 4.5%, smaller than either rule alone. That is an illustration of why some traders run both kinds, not a tested result: it is one period, and the combination was chosen after seeing it.
The same two rules on 30 stocks
| 30 Dow stocks, daily, same two years | Trend rule | Reversion rule | Holding |
|---|---|---|---|
| Median total return | −3.5% | 8.6% | 24.8% |
| Stocks where the rule gained | 12 of 30 | 22 of 30 | — |
| Stocks where it beat holding | 6 of 30 | 10 of 30 | — |
| Trades with a gain | 20% | 65% | — |
| Best trade | 113.9% | 12.1% | — |
| Worst trade | −16.1% | −30.6% | — |
On single stocks both rules trailed holding, and the last two rows show their characters at full size. The trend rule's best trade more than doubled; its worst lost 16.1%. The reversion rule's best trade made 12.1% and its worst lost 30.6%. One has a capped downside per trade and an open upside. The other has the reverse.
The 30 stocks are today's Dow members, so companies removed after poor results are missing (survivorship bias).
The long view: a monthly backtest, 1993 to 2026
Two years cannot contain a real bear market. Monthly bars can, at the price of much slower rules. On SPY month-end closes from November 1993 to October 2026:
- Trend rule: hold the fund when the close is above its 10-month average (the rule of Moving average trend following: the 10-month rule, tested).
- Reversion rule: hold the fund for one month after any month that closed down; otherwise hold cash. This is a deliberately plain "buy weakness" rule, not a published strategy.
Same costs, 0.10% per trade; dividends left out.
| Measure | Trend rule (10-month) | Reversion rule (after a down month) | Buy and hold |
|---|---|---|---|
| Return per year, 1993–2026 | 7.8% | 3.4% | 8.9% |
| Deepest fall from a peak | −23.8% | −40.6% | −52.2% |
| Time invested | 76% | 37% | 100% |
| Round trips | 25 | 90 | 1 |
| Trades with a gain | 64% | 79% | n/a |
| August 2000 – September 2002 | −11.5% | −30.3% | −46.3% |
| October 2007 – February 2009 | −5.6% | −35.5% | −52.2% |
| Return per year, 1993–Feb 2009 | 8.0% | −0.2% | 3.1% |
| Return per year, Feb 2009–2026 | 7.7% | 6.7% | 14.1% |
The monthly reversion rule won 79% of its 90 trades and still earned 3.4% a year, because its losing trades were the months inside the two bear markets: a down month followed by another down month, again and again. A bear market is a mean-reversion rule's worst case and a trend rule's reason for existing. From 2009 on, with no long decline, the reversion rule was nearly level with the trend rule.
The monthly reversion rule is much weaker than the daily one, which fits the research: the bounce after weakness is a matter of days to a few weeks, while momentum, the tendency of winners to keep winning, shows up at 3 to 12 months (Jegadeesh and Titman, 1993).
Where each one fails: the losing periods
The trend rule lost in sideways markets. Its deepest fall, 9.1%, ran from October 29, 2025 to April 8, 2026 and was made of whipsaws: it sold on February 5, 2026 at $677.62, bought on February 6 at $690.62, sold on February 12 at $681.27, bought on February 20 at $689.43, sold on February 23 at $682.39 and bought on February 25 at $693.15. Six trades in three weeks, each buy higher than the sale before it. From December 2025 to November 2026 it lost 4.3% while the fund gained 6.5%.
The reversion rule lost when a fall continued. In the daily test that was brief: 7.6% from April 2 to April 8, 2025. In the monthly test it was 40.6% from August 2000 to February 2009, and on single stocks one trade lost 30.6%. Its smooth-looking record depends on the absence of a long decline in the sample.
Both pay more in costs and taxes than holding, and both trailed holding on most single stocks.
Compared with its neighbours
| Rule | Family | Tested in |
|---|---|---|
| 10-month moving average | Trend | Moving average trend following: the 10-month rule, tested |
| Two-day RSI | Mean reversion | RSI mean reversion: the two-day RSI rule, tested |
| Bollinger Band lower-band buy | Mean reversion | Bollinger Band mean reversion: buying the lower band, tested |
| Pairs trading | Mean reversion | Pairs trading: rules, a backtest and the pair that broke |
| Last week's losers | Mean reversion | Short-term reversal: buying last week’s losers, tested |
How Finabase does it
No Finabase tool chooses between these approaches or tells you when to trade. The tools show each kind of signal and let you replay simple rules.
- Chart. The link above opens SPY with a 50-day moving average on the price and a two-day RSI below it, so the two rules' triggers can be compared day by day.
- Backtest. The Backtest tool trades the signals of the studies you add. For a moving average that is the price crossing the average, close to the trend rule here. For the RSI it is the 30 and 70 levels, not "below 10, exit above the five-day average". Each signal buys a fixed $10,000 and no trading cost is charged, so neither column of this page will be reproduced exactly, and frequent-trading rules look better there than they are.
- Strategy. The Strategy tool's starting rule, "if the 50-day SMA is above the price …", is a trend condition, and it replays two years. As the split above shows, two years can contain one kind of market and not the other.
Frequently asked questions
Which is better, mean reversion or trend following?
Neither, in general. In this test the reversion rule led over 22 months of daily data and the trend rule led by a wide margin over 33 years of monthly data, mainly because of two bear markets. Which one did better depended on the time scale and on the market in the sample.
Why does trend following have such a low win rate?
Because it is built to lose small and often. It buys every move above the average, and most of those moves fade within days, giving a small loss. The few that become long advances pay for the rest. Here 4 winning trades out of 17 averaged 7.0%, and 13 losing trades averaged −1.0%.
Can the two be combined?
Many systematic traders run both, because the two tend to struggle at different times. In the daily test the rules' monthly returns had a correlation of −0.31. That is a property of one 22-month sample, and a combination adds the costs and the complexity of both.
Why do stocks trend and revert at the same time?
They do so at different horizons. Over days, prices overshoot and bounce, which favors reversion. Over months, moves tend to persist, which favors trend rules. Over several years, research has found reversal again. A rule only makes sense together with its holding period.
Summary
Trend following buys strength and holds it; mean reversion buys weakness and sells the bounce. On the same SPY data the trend rule won a quarter of its trades and made its money on two long holds, while the reversion rule won three-quarters of its trades a percent at a time. They earned in different months, and they failed in different markets: the trend rule in the sideways chop of 2026, the reversion rule in the bear markets of 2000 and 2008, where a plain monthly version lost 30% to 36% while the trend rule lost 6% to 12%. Neither is the better strategy. They are different bets, and knowing which one a rule is tells you what kind of market will hurt it.
Check your understanding
5 questions on this lesson. Each answer comes with the reason, and nothing is scored.
A fund closes below its 50-day average after two sharp down days, and its two-day RSI is 5. What do the two rules on this page do?
- Both buy
- The trend rule sells or stays in cash; the reversion rule buys
- The trend rule buys; the reversion rule sells
Show the answer
B. The trend rule sells or stays in cash; the reversion rule buys Weakness is a sell signal for a trend rule and a buy signal for a reversion rule. On February 5, 2026 the two rules traded SPY at the same price in opposite directions.
The trend rule won 24% of its trades and the reversion rule 78%, yet both made money. How?
- The trend rule’s few gains were large and its many losses small; the reversion rule’s gains were small and frequent
- The trend rule paid lower costs
- Win rate is the same thing as return
Show the answer
A. The trend rule’s few gains were large and its many losses small; the reversion rule’s gains were small and frequent Four trend gains averaged 7.0% against thirteen losses averaging −1.0%. Twenty-one reversion gains averaged 1.3%. A win rate means nothing without the size of the gains and losses.
Which market is hardest on a trend-following rule?
- A long, steady decline
- A long, steady advance
- A sideways market that keeps crossing its average
Show the answer
C. A sideways market that keeps crossing its average Each crossing triggers a trade that reverses a few days later. In February 2026 the 50-day rule made six trades in three weeks, each buy higher than the sale before it. A long decline is where a trend rule protects.
From October 2007 to February 2009 the monthly trend rule lost 5.6% and the monthly reversion rule lost 35.5%. Why did the reversion rule do so badly?
- It traded too rarely
- It kept buying after down months, and in a bear market down months are followed by more down months
- It was charged higher costs
Show the answer
B. It kept buying after down months, and in a bear market down months are followed by more down months A mean-reversion rule assumes a fall will be partly undone. In a lasting decline that assumption fails repeatedly, so its losses pile up exactly where a trend rule is in cash.
In the first half of the daily test the trend rule was best; in the second half the reversion rule was. What is the right conclusion?
- Use the trend rule in odd years and the reversion rule in even years
- A short test mostly tells you which kind of market the period was, not which rule is better
- The reversion rule is improving
Show the answer
B. A short test mostly tells you which kind of market the period was, not which rule is better The first half had a fall and a long rise, which suits trend rules; the second was choppy, which suits reversion. Splitting a test shows how much a result depends on the stretch it covers.
Exercise
One task to do in Finabase. Nothing is graded or saved.
Open the Scanner with Indicators and Bullish selected. Open the SMA (50) card and click through two or three of its stocks to their charts. Then do the same for the RSI (14) card. For each stock, look at the week before the signal: had the price been rising or falling? Which card is buying strength and which is buying weakness?
Open the ScannerWhat you should find
A bullish moving-average signal is the price crossing above its average, so those stocks had been rising: that is trend following. A bullish RSI signal comes from the oversold zone, so those stocks had just fallen: that is mean reversion. Both are labelled bullish on the same day for opposite reasons. The Scanner's RSI uses 14 days, not the two-day rule tested in the lesson.
Sources and further reading
- Larry Connors and Cesar Alvarez, Short Term Trading Strategies That Work (TradingMarkets, 2008) — the two-period RSI pullback rule
- Mebane T. Faber, “A Quantitative Approach to Tactical Asset Allocation,” Journal of Wealth Management (Spring 2007) — the 10-month moving-average rule
- Narasimhan Jegadeesh and Sheridan Titman, “Returns to Buying Winners and Selling Losers,” Journal of Finance 48(1), 1993 — momentum at 3 to 12 months
- Finabase chart data — SPY daily bars November 2024 to November 2026 and monthly bars February 1993 to October 2026, plus the 30 Dow stocks daily; rules and results computed by Finabase
