Historically the first hike of a tightening cycle costs the S&P 500 7.5 percentage points against what the market would otherwise have done over the following quarter. Every hike after it is a non-event. Measured against the market's own base rate, 51 announced hikes since 1994:
| day 0 | +21d | +63d | +252d | |
|---|---|---|---|---|
| turn (n=6) | +0.93 | −2.74 | −4.07 | +5.58 |
| continuation (n=43) | −0.02 | +0.71 | +2.58 | +12.96 |
| base rate | +0.07 | +1.33 | +3.42 | +12.78 |
A turn is a first hike whose previous policy move was a cut. The Fed pauses, then reverses. The turn median over 63 trading days is −4.07% against a base rate of +3.42%. The continuation median is +2.58% against the same base rate. That 0.8pp gap is inside the noise.
The six turns since 1994:
| day 0 | +21d | +63d | +252d | |
|---|---|---|---|---|
| 1994-02-04 | −2.27 | −0.84 | −5.85 | +2.41 |
| 1997-03-25 | −0.23 | −2.27 | +13.59 | +39.65 |
| 1999-06-30 | +1.57 | −3.20 | −7.60 | +5.98 |
| 2004-06-30 | +0.41 | −3.43 | −2.28 | +5.17 |
| 2015-12-16 | +1.45 | −9.25 | −1.13 | +9.11 |
| 2022-03-16 | +2.24 | +0.80 | −13.03 | −10.12 |
Five of six lower at 21 days, five of six lower at 63 days, five of six higher at a year. Extending to every easing-to-tightening turn since 1954 gives 16 events, of which 4 were positive at 63 days against a 67% base rate. That carries a p-value of 0.001, meaning a one-in-a-thousand chance of a split this lopsided if the first hike made no difference at all. The bootstrap puts the median shortfall at −4.4pp with a 95% interval of −8.9 to −0.9. At twelve months the effect is gone.
The announcement day runs the other way. The median turn was a green day, +0.93%. Across all 51 hikes, a strong day-0 reaction is followed by a weak quarter, at a correlation of −0.34. Run the same calculation on every trading day since 1994 and it is −0.02, which is to say no relationship at all. The pattern belongs to decision days. The worst day-0 tercile returned +3.16% over the next 63 days; the best tercile returned −0.69%.
Which turns hurt is not forecastable from anything measurable beforehand. Splitting the 16 turning hikes by inflation, by length of pause, by how extended the market was, or by how much tightening the Fed eventually delivered separates nothing. The usual bar for calling a split real is a p-value under 0.05. The best of those four reached only 0.11.
The two nearest misses make the point. 1997 was one and done. The target went to 5.50% and the next move, eighteen months later, was a cut. The index gained 13.6% over the following quarter. 1999 was the first of six hikes that took the target from 4.75% to 6.50% by May 2000, and the index lost 7.6%. What separated them happened after the first hike.
One thing does sort outcomes, though only after the fact. Taking the last cut of every easing cycle since 1983, fifteen such pauses across eleven cycles, and splitting by what the Fed did next:
| +63d | +252d | max drawdown | |
|---|---|---|---|
| next move was a hike | +3.2% | +21.0% | −8.5% |
| next move was a cut | −1.8% | +5.4% | −19.4% |
Once the Fed has stopped cutting, hiking next is the good outcome. It means growth held.
What's happening now: nine months on hold at 3.50-3.75%, the two-year at 4.65% against a 3.625% midpoint, and the three-month bill above that midpoint since 11 September. Since 1982 the bill has done that outside a tightening cycle twice. The front end is pricing a turn. Historically a hike followed within twelve months only 41% of the time from this state. If this one lands, the six-event turn hike table above is the map, and it points to roughly a quarter of digestion rather than a cycle top.
Method notes
- Equities: daily bars for SPY, QQQ and the nine SPDR sector ETFs from March 1999, plus the S&P 500 index back to 1950. Price returns only on both sides of every comparison, so no dividends anywhere.
- Policy: FRED DFEDTAR (September 1982 to December 2008) spliced with DFEDTARU. 93 hikes and 91 cuts since 1982, of which 51 are announced hikes since 1994. Pre-1983 turns come from a 50bp zigzag filter on the monthly effective funds rate, so those dates are month-accurate only.
- FRED stamps a target change on its effective date. Before 2008 that is the announcement day. In the band era the statement is the preceding business day, and events are rolled back accordingly. Checked against the tape: 1994-02-04 −2.27%, 2015-12-16 +1.45%, 2022-03-16 +2.24%.
- Announcement effects are measurable only from 1994, when the FOMC began announcing its decisions. The 1954 track measures follow-through alone.
- Monthly macro series carry their release lag (CPI 45 days, labour data 35 days), so a regime tag on any date uses only what was published by then. Rates of change are computed on a calendar-daily frame before reindexing to trading days.
- Tests: binomial against the unconditional hit rate, Mann-Whitney for group differences, 20,000-draw bootstrap for the median excess. Turn and continuation hikes separate at p=0.022 over 21 days and p=0.028 over 63. Cycles are treated as independent draws. Overlapping windows inside one cycle are not, which is why continuation hikes are reported apart.
- Four regime splits were run on the same 16 events and none survives multiple testing. The only result carrying weight is the 63-day underperformance after a first hike.
Not investment advice.