Bank stocks are the hottest thing on SGX right now. DBS closed at S$78.38 on 29 September, up 56% in a year. So I did what a lot of investors are doing. I asked AI whether DBS is still a buy.
Actually, I asked it twice.
First I asked ChatGPT, “Why DBS is not a buy at $78?”
It opened with “The case for not buying DBS at ~S$78 is actually pretty strong.” Then it laid out the numbers. DBS trades at about 3.2 times its book value, against a historical median of around 1.6 times.
Then I asked the opposite. “Why DBS is a buy at $78?”
This time it built me a case. A 4.1% dividend yield plus 5% to 7% earnings growth adds up to a potential 9% to 11% a year. “That’s not unreasonable for a very high-quality bank.”
To be fair to ChatGPT, neither answer was wrong, and the second one still warned me about the lack of margin of safety. But notice what it did. Each time, it led with the case I asked for.
Most of us don’t ask AI neutral questions. We ask the question we already have an answer to. If you like DBS, you ask why it’s a buy. If you missed the rally, you ask why it’s overpriced.
The AI obliges. You come away feeling like you’ve done your homework. What you have actually done is rehearse your own opinion with a very articulate assistant.
That is confirmation bias, and AI makes it faster and more convincing than ever.
Banks are cyclical. When interest rates are high, they earn fat margins on every loan. That is the good times DBS is enjoying now.
People forget what the other half of the cycle looks like. From mid-2007 to the end of 2016, through the financial crisis and the years of near-zero rates that followed, DBS’s share price fell about 12%. Even with dividends, shareholders made about 2.5% a year for almost a decade.
I’m not saying bank stocks are about to crash. Rates are still high and the strong economic growth still favours the banks. But a cyclical stock at the top of its cycle should be judged the way a value investor would. What is it worth across the whole cycle, not just at the best point of it?
And that is not a question you can answer by asking an AI chatbot one leading question.
How I use AI instead
I don’t let the AI take a side until the very end. Instead, I run a workflow where each step has one job.
Step 1. Source documents. The work starts with DBS’s own annual reports and results, checked before anything else. If the inputs are wrong, everything after is wrong.
Steps 2 and 3. Two neutral notes. A valuation note works out what DBS is worth on several measures, price to book, price to earnings and dividend yield. A qualitative note sets out the facts of the business. The AI is not allowed to have an opinion at this stage.
Steps 4 and 5. The case for, and the case against, kept apart. One agent looks for what could push the price higher. Another looks for what could go wrong. Each sees only the neutral notes, never the other’s argument. So neither is tainted by the other’s worldview.
Step 6. A portfolio manager. A final agent reads everything and makes the call, buy, hold or pass. It also has to say what would change its mind.
Then I read it, and I argue with it.
What it decided on DBS
In June, at S$66.22, the portfolio manager passed on DBS. Its reasoning was that DBS is a superb bank at a rich price, and the engine that drove the re-rating, the interest margin, is already shrinking. DBS’s net interest margin has come down from a peak of 2.15% to 1.87%.
I pushed back. I argued that rates are likely to stay higher than in the 2010s, so a five-year average undersells what DBS can earn. The portfolio manager agreed in part and raised its fair range to S$50 to S$60. But it held the line on the call. Even on my more generous view, there was no cushion at that price.
Since then, the price has kept climbing, to S$78. So was the call wrong?
On price, so far, yes. A process does not tell you what a stock will do next quarter. What it tells you is what you are paying for, and what would change your mind.
For DBS, the question it turns on is whether fee income can grow fast enough to replace the interest margin that falling rates take away. The workflow set out exactly what would make the premium look earned. The margin has to stabilise, net interest income has to grow again, fees have to keep growing above 20%, and return on equity has to hold at 16% to 17%, for about two quarters in a row.
After August’s results, three of those four are pointing the right way. Fees grew 22%, return on equity hit 17.9%, and the margin barely moved. Net interest income is still down on a year ago. So the next results matter. I know exactly what I’m looking for.
That is the difference. ChatGPT gave me two confident answers. The workflow gave me a view, the evidence on both sides, and a checklist for changing my mind.
AI is a tool. Use it right
AI is the most powerful research tool retail investors have ever had. Used the wrong way, it’s a mirror that tells you what you want to hear.
In my AI-Powered Investor webinar, I explain why every investor should be using AI, and how to use it properly, so it challenges your thinking instead of confirming it. Register for the free webinar.
