In 2018, four finance professors published a paper saying SPY's first half hour predicts its last half hour. If SPY is up from yesterday's close by 10:00 AM, it tends to keep going up into the close. If it's down, it tends to keep going down. They found it in SPY data from 1993 to 2013.
The rule is simple. At 10:00 AM, see if SPY is above or below yesterday's close. At 3:30 PM, buy if it's above or short it if it's below, then get out at the 4:00 PM close. So it's one trade a day and you're only in it for half an hour.
We backtested it on SPY and it lost a little money. Most days were a small win or a small loss, and the average trade came out to -0.01%.
Just buying SPY and holding it did a lot better.
| Trades | Winners | Average trade | Ended with | Biggest drop | Backtest |
|---|---|---|---|---|---|
| 871 | 49% | -0.01% | $9,379 (-6.2%) | 10.1% | See the backtest |
Year by year
Our data starts in 2023, five years after the paper came out, so none of these days were in their study. If the pattern still worked, you'd expect it to make money most years.
It lost money in 2023, 2024 and 2025, and it's about flat so far in 2026. Maybe markets changed, or maybe enough people started trading it once it was public. Our data can't tell you which.
| Year | Return | Balance at year end | Biggest drop |
|---|---|---|---|
| 2023, from March | -1.7% | $9,831 | 4.4% |
| 2024 | -2.1% | $9,626 | 3.1% |
| 2025 | -2.6% | $9,372 | 6.2% |
| 2026, to September | +0.1% | $9,379 | 2.7% |
How we tested
Lei Gao, Yufeng Han, Sophia Zhengzi Li and Guofu Zhou, "Market intraday momentum," Journal of Financial Economics 129(2), 394-414 (2018). https://doi.org/10.1016/j.jfineco.2018.05.009
We ran it on one-minute candles of SPY from March 28, 2023 to September 30, 2026 in the same Backtest tab you can use. That's all the history we have.
The paper holds every trade to the 4:00 PM close. Our tester fills at the open of the next candle, so the last fill it can get is the 3:59 PM open, and every trade here gets out a minute early. On half days the market closes early, so there's no afternoon trade.
Every test starts with $10,000 and puts the whole balance into each trade, no margin. Costs are the tester's defaults for stocks: half a cent a share in commission and a cent of slippage per fill.
"Just holding" means buying SPY on the first day and holding it to the end.
The paper also looked at other ETFs and found the effect was stronger on volatile days. We only tested the main rule on SPY.
Questions
Did the paper get it wrong?
Not necessarily. Their data ended in 2013 and ours starts in 2023. All we can say is the rule as written hasn't made money on SPY since.
Can I test it on a different stock or dates?
Yes. Hit "Run it yourself" and the Backtest tab opens with the rule loaded on SPY. Change the stock or the dates from there. Free accounts can't go back as far as we did, so your numbers might be different.
Can I change the rule?
Yes. Tell Spyglass, the AI quant on your desk, what you want to change. It rewrites the rule on your chart and runs the backtest again. Spyglass comes with the Trader and Quant plans.