Strategies · Trend and momentum
Golden cross strategy: the 50/200-day moving average crossover, tested
The 50-day and 200-day moving average crossover as a trading rule: exact rules, a 33-year SPY backtest with costs, the whipsaws of 2020 and 2025, and nearby settings.
On this page
A moving average crossover uses two averages of the same price, a fast one and a slow one. When the fast average is above the slow one the rule owns the market; when it is below, the rule holds cash. The famous pair is the 50-day and the 200-day simple moving average. The day the 50-day rises through the 200-day is called a golden cross, and the day it falls through is a death cross. Both make headlines.
This lesson turns the headline into a rule and tests it. The data available here gives 33 years of monthly prices but only two years of daily ones, so the long test uses the closest monthly version of the same rule, and five years of weekly prices show the real 50-day and 200-day lines at work. The result is close to the simpler rule in Moving average trend following: the 10-month rule, tested: a smaller worst fall, a slightly lower return, and a string of costly false alarms since 2009.
The idea
A single price is noisy. One close below a long average may mean a decline has begun, or it may mean nothing. Comparing two averages is a way of asking for more evidence. The fast average has to fall below the slow one, which takes several weeks of weakness, before the rule acts.
That patience has a price. By the time a 50-day average crosses a 200-day average, the market has usually moved a long way from its high. The rule accepts selling late and buying late in exchange for fewer signals. Whether that exchange is a good one is what the test measures.
Like every trend rule, the crossover does not forecast. A death cross does not say that prices will fall further. It says that the last 50 days were, on average, lower than the last 200. What happens next depends on which kind of decline it is, and the rule cannot tell.
The rules
The daily version, as it is usually described:
- After each day's close, compute the average of the last 50 closes and the average of the last 200 closes.
- If the 50-day average is above the 200-day average, hold the fund.
- If it is at or below the 200-day average, hold cash.
- Trade only on the day the answer changes.
The monthly version tested below is the same rule on month-end closes:
- On the last trading day of the month, average the last 2 month-end closes and the last 10 month-end closes.
- Hold the fund for the next month if the 2-month average is above the 10-month average. Otherwise hold cash.
A month has about 21 trading days, so 50 days is about 2.4 months and 200 days about 9.5. The monthly rule looks 12 times a year where the daily rule looks about 252 times, so it can react up to a month later. That difference is real and is named again under what the test leaves out.
The figure shows the weekly equivalent on SPY, the oldest S&P 500 exchange-traded fund. The averages crossed down once in the period, in the week of April 7, 2025, and back up in the week of June 23, 2025.
A worked example
At the end of November 2007 SPY closed at $148.66. Its 2-month average was $151.65 and its 10-month average $148.42, so the rule stayed invested. A month later SPY closed at $146.21. The 2-month average had dropped to $147.43, below the 10-month average of $148.95. The rule moved to cash.
It stayed there for 18 months. At the end of May 2009 the 2-month average was $89.97 and the 10-month average $93.82: still cash. At the end of June 2009 the fast average, $92.24, was above the slow one, $90.13, and the rule bought at $91.95. It had sold at $146.21, so it bought back 37% lower.
What the backtest shows
The test, stated in full:
- Data. SPY month-end closes from February 1993 to October 2026. The first months build the averages, so the test runs from January 1994: 394 months.
- Trading. The signal and the trade both use the month-end close.
- Costs. 0.10% of the amount on every buy and every sell.
- Left out. Dividends, interest on cash and taxes. Buy and hold collects every dividend; the rule collects them about three-quarters of the time and would have earned interest on cash the rest. The daily rule itself is also left out: this is its monthly stand-in.
| Measure | 2/10-month crossover | Buy and hold |
|---|---|---|
| Total return | 1,313.7% | 1,467.2% |
| Return per year (CAGR) | 8.4% | 8.8% |
| Deepest fall from a peak | −20.1% | −52.2% |
| Longest time below a peak | 34 months | 80 months |
| Time invested | 75% | 100% |
| Round-trip trades | 16 | 1 |
| Trades with a gain | 75% | n/a |
The rule traded about once every two years, so costs were small: with no cost at all the return is 8.5% a year, against 8.4% with it. The 12 winning trades gained 29.9% on average and the four losing trades lost 7.2%.
The two halves of the period
| Period | 2/10-month crossover | Buy and hold |
|---|---|---|
| January 1994 – February 2009 | 8.2% a year | 2.9% a year |
| February 2009 – October 2026 | 8.6% a year | 14.1% a year |
The rule earned about the same in both halves. Buying and holding did not. Up to the low of February 2009 the rule was far ahead, because it sat out most of two long bear markets: from August 2000 to September 2002 it lost 6.3% while buy and hold lost 46.3%, and from October 2007 to February 2009 it lost 5.6% against 52.2%. Since then there has been no long decline, and holding has earned five and a half points a year more.
So the whole-period result, nearly level, is the sum of a large lead and a large deficit. A test that ended in 2009 would have made the crossover look brilliant, and one that began in 2009 would have made it look useless.
Where it fails: the losing periods
The crossover fails when a fall is sharp and short. The averages need weeks to cross, so the sell comes after most of the damage, and they need weeks to cross back, so the buy comes after most of the recovery.
| Sold | Bought back | Result |
|---|---|---|
| January 2016, $193.72 | April 2016, $206.33 | Bought back 6.5% higher |
| November 2018, $275.65 | March 2019, $282.48 | Missed the December fall; back 2.5% higher |
| March 2020, $257.75 | June 2020, $308.36 | Sold after the crash; back 19.6% higher |
| February 2022, $436.63 | February 2023, $396.26 | Bought back 9.2% lower: the rule's good case |
| October 2023, $418.20 | November 2023, $456.40 | Bought back 9.1% higher |
| April 2025, $554.54 | June 2025, $617.85 | Bought back 11.4% higher |
March 2020 is the clearest failure. SPY's month-end close fell from $296.26 in February to $257.75 in March. The rule held through that fall, sold at the March close, and then watched the recovery from cash until June. Its deepest fall in the whole test, 20.1% from December 2019 to June 2020, is that one episode. From December 2019 to October 2026 the rule gained 76.4% and buy and hold gained 134.8%.
The weekly lines in the first figure tell the 2025 story at finer grain: the averages crossed down in the week of April 7, 2025, with SPY at $533.94, and crossed back up in the week of June 23 at $614.91, 15.2% higher.
Other ways it fails or misleads:
- Fewer signals is not better signals. Five of the six sells since 2016 were followed by a higher buy. Waiting for two averages to cross removed some false alarms and made each remaining one more expensive.
- Taxes. In a US taxable account a sale at a gain is taxed in that year (IRS Topic 409). Buy and hold defers the tax. The test ignores this, and it counts against the rule.
- One market, one history. Brock, Lakonishok and LeBaron found moving-average rules profitable on the Dow from 1897 to 1986. Sullivan, Timmermann and White tested the same rules on the next ten years and found the advantage gone. A rule that worked for 90 years can stop.
- Two years of daily data cannot judge it. In the daily bars available here, both averages exist only from September 2025, and they did not cross once after that.
Do the settings matter?
| Fast and slow averages | Return per year | Deepest fall | Round trips |
|---|---|---|---|
| 1 and 10 months | 7.8% | −23.8% | 25 |
| 2 and 8 months | 7.4% | −20.1% | 24 |
| 2 and 10 months | 8.4% | −20.1% | 16 |
| 3 and 10 months | 8.0% | −21.9% | 15 |
| 2 and 12 months | 8.6% | −20.1% | 14 |
| 3 and 12 months | 7.7% | −21.9% | 14 |
All six trailed buy and hold's 8.8% and all six cut the deepest fall to less than half. The "1 and 10" row is the plain price-against-average rule of Moving average trend following: the 10-month rule, tested. The spread between the rows, about one point a year, is the size of difference that comes from luck in where a few trades fell. Choosing the best row and calling it the right setting would be overfitting.
The same weekly rule on other funds from August 2022 to November 2026 shows how much the outcome depends on the instrument: 11.9% a year on SPY against 14.6% held, 16.4% on QQQ against 19.2%, and 2.0% on IWM, a small-company fund, against 8.7%. IWM moved sideways in a wide range, which is the worst condition for any crossover.
Compared with its neighbours
| Rule | Checks | Round trips in this test | Return per year | Deepest fall |
|---|---|---|---|---|
| Buy and hold | Never | 1 | 8.8% | −52.2% |
| Price above its 10-month SMA | Monthly | 25 | 7.8% | −23.8% |
| 2-month above 10-month average | Monthly | 16 | 8.4% | −20.1% |
Smoothing the price with a second average cut the number of trades by a third. It did not change the character of the strategy. Both rules are in cash through long declines and both lose in quick ones.
How Finabase does it
No Finabase tool tells you when to buy or sell, and none runs this monthly rule for you. Three tools show the pieces.
- Chart. The chart draws any number of simple moving averages. The link above opens SPY with the 50-day and 200-day lines, so the last cross is visible directly. The Moving averages and trend: filters, the 50/200 and golden crosses lesson explains how to read them.
- Backtest. The Backtest tool trades the signals of the indicators you choose on one symbol. Two differences from this page's test change the numbers: every signal buys a fixed $10,000 instead of reinvesting the whole account, and no trading cost is charged.
- Strategy. The Strategy tool takes if/then rules, such as "if SPY's 50-day average is above its 200-day average, hold SPY", and replays them over two years. A crossover trades about once every two years, so a two-year replay may contain one trade or none. It shows how the rule behaves, not whether it works.
Frequently asked questions
Is a golden cross a buy signal?
It is a description of the past, not a forecast. A golden cross says the last 50 days averaged higher than the last 200. In this test, buying on the cross and selling on the opposite cross earned a little less than holding throughout, with a much smaller worst fall.
Why not use the daily 50-day and 200-day averages in the test?
The price history available for these lessons has daily bars for two years only, and a 200-day average needs most of a year before it exists. Two years would hold one trade at most. The monthly version covers 33 years and trades on the same idea, at the price of reacting up to a month later.
Is the crossover better than price above the 200-day average?
It trades less: 16 round trips against 25 in this test. Its return was 0.6 points a year higher here, which is within the range that neighbouring settings produce by chance. The two rules agree most of the time and fail in the same markets.
Does a death cross mean a crash is coming?
No. Death crosses have come before long declines, as in December 2007 in the monthly test, and after short ones that were already over, as in March 2020 and April 2025. The cross cannot tell the two apart when it happens.
Summary
A moving average crossover holds a fund while a fast average is above a slow one and holds cash otherwise. On 33 years of SPY, the monthly version of the 50-day and 200-day rule returned 8.4% a year against 8.8% for buying and holding and cut the deepest fall from 52.2% to 20.1%. All of its advantage came before 2009, from two long bear markets. Since then eight of its nine sales were followed by a purchase at a higher price. The second average reduces the number of trades, not the kind of market that hurts the rule.
Check your understanding
4 questions on this lesson. Each answer comes with the reason, and nothing is scored.
A fund’s 50-day average is $101 and its 200-day average is $100. The price falls 4% today. What does the crossover rule do?
- Sells, because the price fell sharply
- Nothing, unless today’s fall pulls the 50-day average to or below the 200-day average
- Buys more, because the price is cheaper
Show the answer
B. Nothing, unless today’s fall pulls the 50-day average to or below the 200-day average The rule compares the two averages, not the price. One bad day moves a 50-day average only a little, which is why a crossover reacts slowly in both directions.
In March 2020 the monthly crossover sold SPY at $257.75 and bought back in June at $308.36. What does this show?
- The rule was set up wrongly
- A sharp fall that reverses makes the rule sell after the damage and buy back after the recovery
- Trading costs were too high
Show the answer
B. A sharp fall that reverses makes the rule sell after the damage and buy back after the recovery The averages need weeks to cross, so the sell comes late and the buy comes late. In a short fall that turns into a round trip at a loss, here 19.6%. Costs of 0.10% per trade were a small part of it.
The rule returned 8.2% a year before February 2009 and 8.6% a year after, while buy and hold returned 2.9% and 14.1%. What is the fair summary?
- The rule is steadier, and whether it beats holding depends on whether the period contains a long decline
- The rule beats the market
- The rule stopped working in 2009
Show the answer
A. The rule is steadier, and whether it beats holding depends on whether the period contains a long decline The rule’s own return hardly changed. What changed was the market: two long bear markets before 2009, none after. A trend rule earns its keep only in long declines.
Six pairs of averages were tested and returned between 7.4% and 8.6% a year. Which conclusion is sound?
- Use the pair with the highest return
- The idea is not sensitive to the exact pair, and the best row is probably luck
- The test is broken, because the rows differ
Show the answer
B. The idea is not sensitive to the exact pair, and the best row is probably luck A one-point spread among close settings is what a few differently timed trades produce. The stable part, a lower return than holding and less than half the worst fall, is the finding.
Sources and further reading
- William Brock, Josef Lakonishok and Blake LeBaron, “Simple Technical Trading Rules and the Stochastic Properties of Stock Returns,” Journal of Finance 47(5), 1992 — moving-average rules on the Dow, 1897 to 1986
- Ryan Sullivan, Allan Timmermann and Halbert White, “Data-Snooping, Technical Trading Rule Performance, and the Bootstrap,” Journal of Finance 54(5), 1999 — the same rules after 1986
- IRS — Topic no. 409, Capital gains and losses
- Finabase chart data — SPY monthly bars, February 1993 to October 2026, and weekly bars, November 2021 to November 2026; rules and results computed by Finabase
