In short
- Best performing currency: the Japanese yen, +3.24% against the basket of the eight majors, from the London opening on 1 September to the one on 1 October 2026.
- Worst: the New Zealand dollar, -2.84%. The biggest pair move of the 28 was NZD/JPY, -5.89%.
- Central banks: all eight met. Five raised rates and three held.
- Our setups with the mark: 31 calls, 20 the expected way, +3.75% together. The ones without it, in the same lists: 32 calls, 15 the expected way, -3.43%.
- The ranking: the currency we ranked strongest beat the one we ranked weakest on 12 of 21 mornings. In August it was 5 of 18.
Best and worst performing currencies of the month
Split at the 15th, the month reads as two different months. The yen did all of its work in the first half and gave a little back in the second. The dollar did the opposite. The New Zealand dollar lost in both.
| Currency | 1 to 15 Sep | 15 Sep to 1 Oct | September |
|---|---|---|---|
| Japanese yen | +3.50% | -0.25% | +3.24% |
| US dollar | +0.20% | +2.00% | +2.21% |
| British pound | -0.22% | +0.15% | -0.08% |
| Euro | -0.27% | -0.08% | -0.36% |
| Canadian dollar | -0.04% | -0.42% | -0.46% |
| Swiss franc | -0.69% | -0.03% | -0.73% |
| Australian dollar | -0.21% | -0.65% | -0.86% |
| New Zealand dollar | -2.17% | -0.68% | -2.84% |
Against the basket means against the average of all eight, so a currency can rise here while one of its pairs falls. That is exactly what happened to USD/JPY: the dollar was the second best of the eight and the pair still ended 1.00 percent lower, because the yen did better still. The three pairs that moved most all had the yen or the kiwi on one side: NZD/JPY -5.89%, NZD/USD -4.94%, AUD/JPY -3.97%. The quietest was EUR/CAD at +0.10%.
The first half: a war, oil, and the havens
On the weekend of 5 and 6 September the United States and Iran struck each other's tankers in the Strait of Hormuz (NPR). Brent went above $100 a barrel for the first time since July in the days after, and overnight into 15 September Houthi rebels fired missiles and drones at Saudi Arabia (CNBC). At the close on Monday the 14th, WTI stood at $101.8, up 11.3 percent in five days.
Our risk sentiment score, which reads equities, volatility, credit and the commodities that follow industrial demand, called every morning of the first half risk off. The score itself read -15 on 7 September, hovered just above zero for three days, and then sank: -20 on the 11th, -27 on the 15th, -28 on the 17th, the low of the month.
The yen had two reasons to rise and used both. It is one of the currencies people hide in, and the odds of a Bank of Japan hike on 18 September went to about 99 percent on the 7th, the day the yen strengthened to 154.1 per dollar. It gained 3.50 percent in the first half alone.
The kiwi had the opposite month. The RBNZ raised its rate on 2 September exactly as forecast, but signalled a slower path after it, and the kiwi trades on confidence, which was the first thing to go that weekend. It lost 2.17 percent in the first half and finished the month last.
The second half: the Fed
On 16 September the Fed raised rates by a quarter point to 3.75 to 4.00 percent, unanimously, which was expected. What was not expected was the rest: 16 of the 18 members saw at least one more hike this year, and stocks turned from gains to losses during Chair Warsh's first press conference (PBS). The dollar gained 2.00 percent from the 15th to the end of the month, and it stood at the top of our ranking on 13 of the 15 mornings from 10 September, twice level with the Australian dollar, and second on the other two.
The yen got its hike two days later, 1.25 percent on a 7 to 2 vote, and it did nothing for it. The hike was fully priced, and the same morning Japanese core inflation came in at 1.7 percent against 1.8 expected. In the last full week the yen slid towards 160 per dollar until Japan's finance minister spoke on the 25th, which is the story of our week 39 review.
Risk sentiment turned two days after the Fed. The score jumped to +20 on 18 September, reached +40 on the 22nd and eased to +13 by the 28th. Then it fell back to -27 on the 29th and -28 on the 30th, with credit weighing heaviest as high yield bonds slipped. The month ended where it started, in risk off.
All eight central banks, one table
Every one of the eight met in September, which makes the month a clean test of how much a rate decision is worth to a currency on its own.
| Date | Central bank | Decision | Rate after | Source |
|---|---|---|---|---|
| 2 Sep | Reserve Bank of New Zealand | Raised 0.25 | 2.75% | RBNZ |
| 2 Sep | Bank of Canada | Held | 2.25% | Bank of Canada |
| 10 Sep | European Central Bank | Raised 0.25 | 2.50% | Euronews |
| 16 Sep | Federal Reserve | Raised 0.25, 12 to 0 | 3.75 to 4.00% | CNBC |
| 17 Sep | Bank of England | Held, 6 to 3 | 3.75% | Euronews |
| 18 Sep | Bank of Japan | Raised 0.25, 7 to 2 | 1.25% | Bank of Japan |
| 24 Sep | Swiss National Bank | Held | 0% | SNB |
| 29 Sep | Reserve Bank of Australia | Raised 0.25, unanimous | 4.60% | RBA |
Not much, it turns out. Of the five that raised rates, two currencies ended the month higher (the yen and the dollar) and three lower (the euro, the Australian dollar and the kiwi, which was the worst of all eight). A hike everyone expects is in the price before it comes: the market barely moved on the ECB, and the reaction to the RBA was subdued. What moved currencies in September was what came with a decision, like the Fed's dot plot, and what came from outside, like the war. The rate of every bank, with the date of its next meeting, is on the central banks page.
What we recommended, and how it went
22 lists went out in September and 21 are measured. The record keeps one list per day, and the one it has for 7 September is from that afternoon, after the opening, so it shares its starting price with the list of the 8th and the record counts the later of the two. Over those 21 mornings the top 3 held 63 setups. 35 went the expected way and together they came to +0.32 percent: about flat.
That flat number is two very different halves of the list. The setups with our mark made +3.75 percent and the ones without it lost 3.43 percent, and the marked ones did better in every week that had any:
| Week of | With the mark | Result | Without | Result |
|---|---|---|---|---|
| 1 Sep | not applicable | 7 of 12 | -0.93% | |
| 7 Sep | 2 of 4 | +0.94% | 4 of 8 | -0.32% |
| 14 Sep | 7 of 10 | +1.13% | 2 of 5 | -0.27% |
| 21 Sep | 6 of 11 | +0.33% | 0 of 4 | -1.77% |
| 28 Sep | 5 of 6 | +1.35% | 2 of 3 | -0.14% |
The dollar was read right. Short NZD/USD stood in five lists and gained in all five, +1.91 percent, the best setup of the month. Short EUR/USD was right seven times out of seven. Long USD/CAD added three of four. All three are the same call: the dollar up, and the currency on the other side not.
The yen was read wrong in week 39. Long AUD/JPY was marked on three mornings in week 39 and lost on all three. Long USD/JPY got away with it, two of three, because the dollar was rising faster than the yen.
Three things in these numbers deserve less trust than the total suggests.
- Most of the mark is applied afterwards. It went on the site on the afternoon of 25 September, so the first list measured from an opening after it was there is the one of 28 September. From then to the end of the month: 6 calls, 5 the expected way, +1.35 percent. The other 25 calls are the same bar applied afterwards to the lists exactly as they were published.
- One setup carries half of it. Without short NZD/USD, the marked calls come to 26, of which 15 went the expected way, for +1.84 percent.
- A month is a small sample. 31 calls is about three times what one week gives, and still not many. The four lists of 1 to 4 September came from the old model in its first form, which the bar cannot be applied to, so they are in the 63 but none of them is in the 31.
The track record counts the same setups another way: each one once per week, from its first opening to the close of the last day it stood in the list. On that count the marked setups of the weeks of 7, 14 and 21 September come to 7 of 12 and +2.66 percent, against 15 of 25 and +2.40 percent here, and week 39 alone is +0.78 percent there against +0.33 percent here. The week of 28 September runs into October on that page, so it is left out of the comparison. Both counts are right; they answer different questions. This one is how the backtests measure, it stops cleanly at the end of the month, and it only counts a day when the setup was actually in that morning's list.
Did the ranking itself work?
The setups are only the far ends of something bigger: a ranking of all eight currencies, every trading day. So the plainest test of the model is not a setup at all. It is whether the currency we put first did better than the one we put last.
Measured from the opening after each list to the next, as the pair between the two: in September the first beat the last on 12 of 21 mornings, by 0.94 points added up. From 15 September, when the current model took over, that was 7 of 12 and 0.98 points. Under the old model in the first half it was 5 of 9 and -0.04, which is nothing.
The other way to look at it is to take the first ranking of the current model, on the morning of 15 September, and hold it to the end of the month without changing a thing:
| Ranked on 15 Sep | Score | 15 Sep to 1 Oct | Finished |
|---|---|---|---|
| 1. US dollar | 72 | +2.00% | 1st |
| 2. Japanese yen | 69 | -0.25% | 5th |
| 3. Swiss franc | 63 | -0.03% | 3rd |
| 4. Euro | 47 | -0.08% | 4th |
| 5. British pound | 46 | +0.15% | 2nd |
| 6. Canadian dollar | 46 | -0.42% | 6th |
| 7. Australian dollar | 41 | -0.65% | 7th |
| 8. New Zealand dollar | 38 | -0.68% | 8th |
Six of the eight finished exactly where they were ranked. The yen and the pound swapped: the yen we had second finished fifth, the pound we had fifth finished second. That is one ranking over twelve trading days, and the scores moved during them, so it is a good sign and nothing more.
The old model deserves the same honesty. On 1 September it ranked the New Zealand dollar first, the currency that went on to lose the most of the eight, and long NZD/CAD stood in all four lists of that first week: -1.36 percent together. That is the reason there is a current model.
Month against month
August is the only month before this one with a full record. It ran entirely on the old model, from 3 August, its first trading day and the day the record starts, and September ran on the old model for its first two weeks, so this is a comparison of two mixtures and not of two models.
| August | September | |
|---|---|---|
| Mornings measured | 18 | 21 |
| Top 3 setups the expected way | 24 of 54 | 35 of 63 |
| Top 3 together | -3.10% | +0.32% |
| Setups with the mark | Cannot be applied | 20 of 31, +3.75% |
| First ranked beat last ranked | 5 of 18 mornings | 12 of 21 mornings |
Better on every line that can be compared, and still not enough months to call it anything. October is the first month the current model runs from the first day to the last, with the mark on the site from the start. That is the one to judge it on.
October
Five of the eight central banks decide in the last three days of the month: the Bank of Canada, the Fed and the RBNZ on 28 October, the ECB on the 29th and the Bank of Japan on the 30th. The RBA follows on 3 November and the Bank of England on 5 November. The SNB does not meet again until 10 December.
So most of October is data, with a pile of decisions at the end of it. The dollar ended September at the top of our ranking, with a Fed that has told everyone more is coming, and the Canadian dollar at the bottom. Every date and time is on the calendar, and what the eight look like this morning is on the strongest currency today.