Italy is the one market with a normal level of risk: a 26% chance of a −2% day, level with its long-run share. The other six are below their own norms, the Nasdaq 100 and the S&P 500 by about half. For a −3% day every market is below normal, between 1% (S&P 500, Australia) and 6% (Italy).
The odds for 21–25 September
Chance that at least one daily close of the week is 2% (or 3%) or more below the previous day’s close, on each market’s own index.
| Market | −2% day | Long-run share | −3% day | Long-run share | Main drivers today |
|---|---|---|---|---|---|
| Italy40 FTSE MIB | 26.1% | 25.8% | 5.8% | 10.5% | Near the bottom of its 3-month range (15%), −1.8% last week, volatility 13% |
| Europe50 Euro Stoxx 50 | 18.7% | 20.4% | 4.4% | 9.0% | Near the bottom of its 3-month range (9%), −1.4% last week |
| NASDAQ100 | 16.2% | 29.6% | 3.2% | 13.6% | Highest volatility of the group (13%), 8.9% above its 200-day average |
| Germany40 DAX | 15.0% | 20.8% | 3.0% | 8.4% | Low in its 3-month range (35%), −1.0% last week |
| UK100 FTSE 100 | 8.9% | 12.6% | 1.6% | 4.4% | Low volatility; the only market where correlation among members is rising fast |
| USA100 S&P 500 | 7.5% | 13.6% | 1.1% | 4.5% | Close to its high (−1.9%), volatility below 10% |
| Australia50 ASX 200 | 7.0% | 9.5% | 1.0% | 2.6% | Lowest volatility, but at the bottom of its 3-month range (10%) |
Long-run share = share of all weeks, 1990–2026, with at least one close that far below the previous close (Europe50 from 2007, Italy40 from 1997, Australia50 from 1992). Daily closes only, not intraday moves. The seven markets tend to fall together, so the figures cannot be combined as if they were separate bets.
What 36 years of data say
Each Friday’s state of the market is compared with what happened in the following Monday–Friday week. Volatility explains most of it; several other factors add something, and several do not.
| Factor | What the data show | Effect this week |
|---|---|---|
| Volatility (21 and 5 days) | By far the strongest driver: calm markets rarely have a −2% day. | Lowers every market, most for Australia, the S&P 500 and the UK. |
| Position in the 3-month range | Near the lows, risk is higher (t = −4.3). | Raises Europe50 (9% of the range), Australia50 (10%), Italy40 (15%), Germany40 (35%); lowers the S&P 500 (69%) and Nasdaq 100 (78%). |
| Position in the 12-month range | Adds nothing once the 3-month range is known (t = −0.8). | None. |
| Distance from the 12-month high | Closer to the high, lower risk (t = −4.9). | Lowers the S&P 500 (−1.9%) and UK (−2.3%). |
| Stretch above the 200-day average | The further above, the higher the risk (t = +5.1). | Raises the Nasdaq 100 (+8.9%), S&P 500 (+6.5%), Italy (+6.2%). |
| Last week’s return | After a down week, risk is higher (t = −3.6). | Raises Italy (−1.8%), Europe50 (−1.4%), Germany (−1.0%). |
| Day of the week | In calm markets Monday and Friday carry the most risk: 25% and 24% of the weekly −2% risk, against 14% on Wednesday. | Timing only; it does not change the weekly figure. |
| Pairwise correlation of the index members | Measured since October 2022 only (1,423 market-weeks). Its level adds nothing once volatility is known (t = −1.1 at −2%, −1.6 at −3%); a fast rise over five days matters slightly (t = +1.4 and +1.9). | Not used in the odds. Correlation is rising fast only in the UK. |
| Weekly calendar | Weeks that contain the first Friday of the month (US jobs report) have 23–42% higher odds (t = 1.8–2.1). Earnings-season weeks show no effect (t = 0.4). | Next week is neither: slightly favourable. |
| Rates and oil | The three-month change in the US 10-year yield and in oil does not move the weekly odds once volatility is known (t = −1.2 and −0.1). | None in the model, although both are elevated. |
| Elections | Weeks in a US midterm year have 36% higher odds of a −2% day (t = 2.5), but this is offset by the September–November election-season factor (−27%). Raw share of weeks: 21.5% in those months of even years, against 18.2% otherwise. | Net effect close to zero, and not robust. There is no US presidential election in 2026; the midterms are on 3 November. |
t-statistics are clustered by calendar week, because the seven markets move together. |t| above 2 is conventionally significant. Distances and ranges are computed on each market’s own index closes.
The calendar for 21–25 September
- No central-bank decision in the US and no US inflation or jobs release. The Fed raised rates by 25 basis points on 16 September in a 12–0 vote; the next meeting is 27–28 October and the next jobs report is due in early October. After the hike, futures pricing reported by the media put about a 49% chance on another rise in October and 87% on at least one more by year-end.
- Data: flash PMIs for Germany, the euro area, the UK and the US on Wednesday; jobless claims and new home sales on Thursday; durable goods orders and final University of Michigan sentiment on Friday (the preliminary September reading was 47.8, the second-lowest on record). China’s central bank decides on Monday, Australia’s labour data and the Swiss National Bank on Thursday.
- Treasury auctions: $78 billion of 2-year notes on Tuesday, $70 billion of 5-year on Wednesday and $44 billion of 7-year on Thursday, with the 10-year yield around 5%.
- Earnings are light. Costco reports on Thursday after the close, with General Mills, Cintas, Paychex and Darden earlier in the week; no mega-cap technology name. Micron reports on 30 September, Nike on 1 October and FedEx is expected on 28 October.
- Politics: Congress has passed a stopgap funding bill running to 11 December, which removes the 30 September shutdown deadline. Berlin held a state election on 20 September, with the result due on Monday.
- Energy: the US–Iran conflict is still weighing on oil, with WTI near $96 at the 18 September close.
Method and limits
- The model: a weekly logistic regression on the seven markets pooled, with a separate level for each, over 11,730 market-weeks from 1990 to 2026. The inputs are the state of the market at Friday’s close and calendar markers; the outcome is whether any close in the following Monday–Friday week was at least 2% (or 3%) below the previous close. Only inputs with a statistically visible effect are kept.
- Out-of-sample test: fitted on data to 2012 and tested on 2013–2026, the model scores 0.364 against 0.432 for the long-run frequency at −2%, and 0.172 against 0.219 at −3% (log-loss, lower is better).
- Cross-check: in past weeks with similar volatility and the same side of the 200-day average, a −2% day occurred in 5% of weeks for the S&P 500, 14% Nasdaq 100, 4% FTSE 100, 7% DAX, 6% Euro Stoxx 50, 16% FTSE MIB and 8% ASX 200. The model is higher for the three continental markets because they sit near the bottom of their 3-month range and fell last week.
- Event adjustment: the figures above are the model’s multiplied by 1.15. That factor is a judgement, not an estimate, for risks history cannot measure: a Fed in a hiking cycle, the US–Iran conflict with oil near $100, and very weak consumer sentiment, partly offset by a light calendar. Before it, the −2% figures are 22.7% for Italy, 16.2% Europe50, 14.1% Nasdaq 100, 13.0% Germany, 7.7% UK, 6.5% S&P 500 and 6.1% Australia. A range of 1.0 to 1.4 times the model is reasonable.
- Limits: 36 years and highly correlated markets carry less information than the number of weeks suggests. Indices are in local currency, not in the USD ETFs used by Hedge Gauge. Fed meeting weeks and inflation releases could not be tested, for lack of a dated series. Calendar items and pricing come from news and data-provider pages and were not checked against primary releases.
Sources: index closes from Yahoo Finance; calendars from Investrade (18 Sep) and LiteFinance; Michigan sentiment via Kiplinger; the Fed decision via Charles Schwab; the funding bill via Nextgov and NPR.
Next week has few scheduled shocks: no Fed meeting, no jobs or inflation report, no big earnings. The risk is concentrated in Italy and the euro-area indices, which sit near the bottom of their 3-month ranges after a down week, while the US indices are calm and close to their highs. For all seven, a −3% day is less likely than in a typical week.
Research material only — not personalized financial advice. Not a recommendation to buy or sell any security. Probabilities are historical estimates, not forecasts. Contact: domderrico@gmail.com.