September 2026 Portfolio Review: In a Closed Pool, Averaging Down Means Selling Winners
September was a quiet month for me in terms of activity. I didn’t make a single trade, and no new money went into the portfolio. So whatever happened, I can’t blame timing or a badly placed order. It was simply the portfolio I had built, marked to market.
The market didn’t like it very much. The portfolio fell 4.31% in September, which took away the whole of the year’s gain. At the end of August I was up 4.10% for the year. At the end of September I am down 0.39%.
That second number bothers me more than the first. After nine months of fairly active work, I’d have done better just leaving the money in T-bills.
My numbers
| September | YTD | |
|---|---|---|
| My Portfolio | -4.31% | -0.39% |
| China-related stocks | -5.3% | |
| Non-China stocks | -3.8% | |
| S&P 500 Index | -0.4% | +11.8% |
| Hang Seng Index | -3.7% | -4.0% |
I underperformed the S&P 500 by about four percentage points this month, and by about twelve for the year. I chose not to chase the large technology names that carried the index, and I’m fine with that choice. What I’m less fine with is what I did instead.
Where September went
| Sector | Weight (end-Sep) | Contribution to September |
|---|---|---|
| Healthcare | 35% | +1.3 pts |
| Industrials | 31% | -2.3 pts |
| Consumer Discretionary | 19% | -1.6 pts |
| Materials | 6% | -1.4 pts |
| Technology | 7% | -0.5 pts |
| Financials | 3% | +0.1 pts |
Industrials and Consumer Discretionary look like two different things on paper. In my portfolio they are mostly one thing: robotaxis and the companies around them. Pony AI (PONY), WeRide (WRD), XPeng (XPEV), Grab (GRAB), Uber (UBER) and Hesai (HSAI) together make up a little under 30% of the portfolio. That group fell about 10% in September and accounted for roughly two-thirds of the month’s decline.
Healthcare was the only large sector that went up, but that was mostly one stock. Tempus AI (TEM) rose about 30% in the month and on its own contributed more than the whole healthcare sector did. Without Tempus, healthcare was slightly negative.
The habit I didn’t notice
I haven’t mentioned this in earlier reviews, but this portfolio runs on a fixed pool of capital. I set the amount in 2023 and I don’t top it up. If I want to buy something, I have to sell something else first.
I knew that, of course. What I hadn’t properly thought through is what it means when I average down.
This year I added to XPeng several times as the price slid from around US$20 to around US$11. I added to Pony AI at lower and lower prices through the first half. I bought more WeRide in March after it had nearly doubled since December, then a little more in July after it had come off. None of that money was new. It came from selling energy at the start of the year, from taking profits on Albemarle (ALB) and Ovid (OVID), and from selling Amazon (AMZN) and a few others.
When fresh savings come in every month, averaging down is just buying more of something at a lower price. In a closed pool, it means selling what is working to buy more of what isn’t. I never framed it that way to myself while I was doing it.
The result is that the largest positions ended up in the biggest losers. Pony AI is down about 43% on my cost this year, XPeng about 39%, WeRide about 27%. Together with Alibaba (BABA / 9988.HK), the China and robotaxi names are the whole reason the year is negative. Without them, the rest of the portfolio would be comfortably in the black.
Part of this is temperament. I don’t like crystallising losses, and I’m usually willing to wait for a stock to recover. Sometimes that patience has paid off. This year it mostly meant I held on to the losers and sold the winners.
To be fair to myself on one point, selling winners isn’t always wrong. My best result this year, Albemarle, came from selling most of it into strength in the first half. The problem wasn’t that I trimmed winners. It was where the money went next.
Even Albemarle isn’t a clean win, though. After a strong first-quarter result in May, it went above US$200 and peaked at about US$221. In hindsight, that was the time to sell the rest. I got greedy and thought it could run past US$240. It didn’t. It now trades around US$101, less than half the peak, and in September it fell about 22% after a CEO succession announcement and more caution on lithium. On paper it’s a materials stock, but in practice it rides on EV demand, which is a consumer discretionary story. With inflation still sticky and China sitting on plenty of lithium supply, I can see the price staying under pressure for a while. I’m considering rotating out.
There’s a part of the Albemarle story I’m less comfortable with. Having taken most of my profit before the peak, I bought back in June at around US$135 and again in July at around US$118, on the way down. My holding today is larger than it was at the top. It’s the same habit as with the robotaxi names. The profit I booked earlier is still the best result in the portfolio this year, but the shares I bought back are underwater.
Robotaxis, more quietly
I still believe in robotaxis, but more quietly than I did at the start of the year.
The news has been mixed. Waymo keeps expanding, and Tesla (TSLA) held its Cybercab launch in Austin in early September. At the same time, drivers are pushing back. Last week rideshare drivers and unions rallied in Washington DC against a bill that would allow driverless cars there from 2028. None of it seems to have helped the Chinese names at the share-price level.
XPeng is the one I find hardest to judge. It’s ahead of most on humanoid robots, but its robotaxi is only just getting started. XPeng opened an invite-only pilot of its YOYO robotaxi in Guangzhou last week. That’s how every operator starts, Waymo included. You go one city at a time, within approved zones. Even so, XPeng isn’t targeting a service without safety drivers until 2027. So I’m holding XPeng partly for things that haven’t happened yet, and I know that.
The more practical question is whether I need both WeRide and Pony AI. I’ve written before about why I hold them separately. Pony is a robotaxi operator. WeRide is closer to a platform, with ADAS licensing alongside its robotaxis. So the case for keeping one over the other would rest on different reasons, depending on which I keep. The market, though, didn’t make that distinction in September. Both fell about 8%. As I found in my mid-year review, owning different business models inside one theme hasn’t stopped the money moving together. I’m thinking about consolidating into one. I haven’t decided which.
If I had expected quick gains from this theme, I would have bought semiconductors and the large tech names at the start of the year. I didn’t, and I still think physical AI is worth owning. But “not quick” and “too large” are two separate problems, and I’ve been using the first as an excuse for the second. My view that physical AI will be a long-term winner hasn’t changed. What has changed is how much of the portfolio I’m willing to let one corner of it, Chinese robotaxis, take up.
Tempus, and what Ovid taught me
Tempus rose after a Morgan Stanley note in mid-September highlighted new reimbursement pricing for its diagnostic tests. It’s now one of my largest positions, and it got there by going up, not because I bought more.
I’ve been here before with Ovid. At one point I was sitting on a gain of more than 70% on Ovid. Today it’s around 45%. I did trim, but later than I should have. Tempus fell about 13% in a single session on 6 October, and my sense is that it’s the kind of stock investors sell when they want to rotate into something else. I’m likely to trim some.
WuXi AppTec (2359.HK) I’m comfortable holding. I sold my entire position in June, and what I own now is a smaller, fresh tranche bought in July. It rose 13.5% in September. I don’t read too much into that, but the size feels right.
Are we heading into financial repression?
The bigger backdrop in September was interest rates. The Fed raised rates on 16 September, its first hike since 2023, and the US 10-year yield ended the month around 5.3%, the highest in nearly two decades. Small caps and anything valued on earnings far in the future had a hard month. My portfolio is full of exactly those.
Yet there are hints that governments would like to keep their own borrowing costs down. In August the US Treasury stepped up buybacks of long-dated bonds after 30-year yields hit 20-year highs (Treasury calls it liquidity support), and the US and Japan intervened together to support the yen. One Deutsche Bank economist called it “soft-form financial repression”. I’m not sure I’d go that far yet. But with this much government debt around, the old trick of keeping interest rates below inflation and letting savers quietly pay for it is something I’m keeping an eye on.
US Treasuries tell the story better, with a twist. Had I simply parked this portfolio in US T-bills at the start of the year, I’d be up about 2.8%, comfortably ahead of my -0.39%. But T-bills are only paying roughly in line with US inflation, so even the safe option is just treading water. And had I bought longer-dated Treasuries, the textbook safe asset, I’d be worse off: 7–10 year Treasuries are down about 4.7% this year and 20-year-plus bonds about 8.5%, as yields climbed.
I’m not making a big macro call from this. But it does change how I see my -0.39%. If cash is losing value, a flat portfolio is actually going backwards. And inflation usually hits consumer spending hardest, which is where a good part of my China exposure sits.
What I will be watching
In October, I want to decide three things: whether to keep both WeRide and Pony AI, how much of Tempus to trim, and whether Albemarle still deserves a place. I also want to put a ceiling on how much of the portfolio any single theme can take, so I have a rule to check before I average down again.
I don’t know yet whether this year will be remembered as the year I held on too long, or the year I bought early. The portfolio can’t tell me that. What it can tell me is that I put the most money behind the ideas the market liked least, and I paid for them by selling the ones it liked most.
It is not financial advice. The author may hold positions in securities discussed. Readers should conduct their own due diligence and consult with a qualified financial advisor before making investment decisions.