Graveyard / Calendar effects / ETFs
Pre-FOMC drift
Do stocks rise in the day before Federal Reserve announcements?
The idea
A well-known 2015 paper found that US stocks drift upward in the 24 hours before scheduled FOMC statements, and a later paper found the equity premium is earned in the even weeks of the FOMC cycle. Buy before the meeting, sell just before the 14:00 statement.
How we tested it
| Data | Test 1: SPY and QQQ one-minute prices around 82 statement days. Long from the prior close to 13:55 on the day. Test 2: the same on four ETFs (SPY, QQQ, IWM, TLT) over 46 meetings. Test 3: SPY daily over 196 statements, long in weeks 0, 2, 4 and 6 of the FOMC cycle. |
|---|---|
| Period | Test 1: January 2016 to October 2026. Test 2: 2021 to 2026. Test 3: July 2002 to October 2026. |
| Costs | Test 1: 1 bp plus 2 bp, stress 6.5 bp per side. Test 3: 1 bp per side, stress 6.5 bp. |
| Holdout | In-sample and out-of-sample Sharpe. With 8 events a year, the out-of-sample slice holds about 17 to 18 events, below the minimum of 20. |
| Control | Buy and hold SPY in test 3. |
Result
| Test 1: events | 82 |
|---|---|
| Test 1: SPY per event | +16.4 bp (t 2.29) |
| Test 1: QQQ per event | +25.6 bp (t 2.49) |
| Test 1: Sharpe full / in-sample / out-of-sample (18 events) | 0.68 / 0.72 / 0.48 |
| Test 1: Sharpe 2016-19 vs 2020-26 | 0.09 vs 0.87 |
| Test 1: Sharpe at stress cost | 0.27 |
| Test 1: max drawdown / profit factor | −2.7% / 2.19 |
| Test 3: Sharpe full / since 2016 | 0.56 / 0.38 |
| Test 3: buy and hold SPY since 2016 | 0.88 |
Why it failed
The drift is real, with t of about 2.3, but it is about 8 trades a year, so the out-of-sample slice has fewer than 20 events and a daily Sharpe of 0.8 is out of reach. It is concentrated in bear and high-volatility periods, and in the four-ETF test it was mostly 2022. The FOMC-cycle pattern disappeared after the sample of the paper: the even and odd weeks have earned about the same since 2016.
What would make it worth retesting
Only as one leg of a pre-registered book of scheduled events with enough trades. The notes name no partner event window that works yet.
Past results of a backtest say nothing certain about the future.