MA crossover trailed buy and hold on SPY, 2000-2026
Kaspar Dolan, Writes about testing trading ideas on tick data (AI-assisted) · · updated
The crossover lost in this window, with and without costs. The gap was small for the slow pair and large for the fast one. This is the site's own calculation on data downloaded on 21 November 2026 [1], so it shows what happened on one instrument over one window and nothing beyond that.
What we tested
- Instrument: SPY, dividend-adjusted daily closes from Yahoo Finance chart data, 3 January 2000 to 20 November 2026, 6,763 days [1].
- Rule: hold SPY when the fast simple moving average of closes is above the slow one; otherwise hold cash earning nothing. Positions take the next day's return [1].
- Costs: 0, 5 and 10 basis points per switch (a basis point is one hundredth of a percent) [1].
- Trials: two rule pairs, 50/200 and 10/50, both reported [1].
- Not tested: drawdowns, interest on cash, taxes, or any other instrument [1].
The slow pair lost by a little, even with free trading
A crossover simply moves you out of the market when the shorter average falls below the longer one [2]. Buy and hold, by contrast, makes the fewest transactions [3]. The table lists the 50/200 results from [1].
| 50/200 test | Final value of $1 | Annual return |
|---|---|---|
| Buy and hold | $8.13 | 8.38% |
| Crossover, no cost | $7.90 | 8.26% |
| Crossover, 10 bp per switch | $7.69 | 8.15% |
The crossover switched 27 times and was invested on 73.2% of days [1]. Without costs it already trailed buy and hold by about 0.1 percentage points a year, and at 10 basis points the gap was about 0.2 points a year. Buy and hold ended at $8.13 for each dollar, against $7.69 for the crossover with costs [1].
Costs were not the main problem. The 27 switches cost about $0.0213 in total at 10 basis points, computed from the switch count, the cost rate and the no-cost multiple. Most of the shortfall appeared before any cost was charged: the rule sat in cash on days when SPY rose.
The faster pair lost by a lot
The 10/50 pair reacts sooner, so it switched 174 times [1]. Buy and hold returned 8.07% a year over its window; the crossover with no costs returned 4.76% a year and reached $3.45 [1]. At 10 basis points it reached $2.90, or 4.08% a year [1].
That is the cleaner lesson for anyone tuning an Expert Advisor or a spreadsheet rule. Faster signals multiply the number of switches, and each switch is a chance to exit just before a rebound. Here the whipsaw cost far more than the fee schedule did.
What this does and does not show
The result is narrow. It covers one fund, one 26-year window, two rule pairs and a cash return of zero [1]. Trying many pairs and reporting the best would overstate any edge, which is why both pairs tried are shown, including the one that did badly [1].
Crossovers are often defended for smaller drawdowns, and this calculation did not measure them [1]. A rule that gives up about 0.2 points a year might still suit someone who values a shallower decline. The data here cannot say either way.
For comparison on the same asset, the site's [opening range breakout test on SPY](/blog/opening-range-breakout-spy-41-win-rate-no-edge-after-costs) also found no edge after costs, though that was an intraday rule rather than a trend filter.
Questions readers ask
Would the crossover win if cash earned interest?
Possibly for the 50/200 pair, since the gap to buy and hold was small. The test paid nothing on cash, so it cannot say by how much [1].
Does a golden or death cross predict market direction?
The 50/200 crossover is called a golden cross when the 50-day average crosses above the 200-day, and a death cross when it crosses below. One source describes the death cross as not a reliable indicator of future declines [2]. This test only compared returns, not signal accuracy.
Is a different pair like 20/100 worth trying?
This test did not try other pairs, so it has no result for them. Each extra pair raises the risk of reporting a lucky fit [1]. If you try more, count every trial and report them all.
How the numbers work
Computed from the numbered sources and the assumptions named below; change an assumption and the results change with it.
| Figure | Value | Where it comes from |
|---|---|---|
| Initial investment amount | 1 | given: the return is expressed as a multiple of 1 dollar; see source [1] |
| Test period (years) | 26 years | source [1] |
| Buy and hold total return, 50/200 window | 8.13 | source [1] |
| 50/200 crossover total return, no cost | 7.9 $ | source [1] |
| 50/200 crossover total return, 10 bp per switch | 7.69 | source [1] |
| 50/200 crossover total switches | 27 | source [1] |
| 50/200 crossover: % of days invested | 73.2 % | source [1] |
| Buy and hold annualized return, 50/200 window | 8.38 % | source [1] |
| 50/200 crossover annualized return, no cost | 8.26 % | source [1] |
| 50/200 crossover annualized return, 10 bp per switch | 8.15 % | source [1] |
| Total cost from 10 bp per switch (50/200 crossover) | 0.0213 $ | computed: 50 ÷ 200 crossover total switches × 0.0001 × Initial investment amount × 50 ÷ 200 crossover total return, no cost |
| 50/200 crossover excess return vs buy and hold (annualized, no cost) | -0.1 % | computed: 50 ÷ 200 crossover annualized return, no cost - Buy and hold annualized return, 50 ÷ 200 window |
| 50/200 crossover excess return vs buy and hold (annualized, 10 bp per switch) | -0.2 % | computed: 50 ÷ 200 crossover annualized return, 10 bp per switch - Buy and hold annualized return, 50 ÷ 200 window |
| Difference in final value: 50/200 crossover (10 bp) vs buy and hold | -0.44 | computed: 50 ÷ 200 crossover total return, 10 bp per switch - Buy and hold total return, 50 ÷ 200 window |
- Buy and hold annualized return, 50/200 window: 8.38 %
- 50/200 crossover annualized return, no cost: 8.26 %
- 50/200 crossover annualized return, 10 bp per switch: 8.15 %
Works out: 50/200 crossover excess return vs buy and hold (annualized, no cost); 50/200 crossover excess return vs buy and hold (annualized, 10 bp per switch)