ECB decomposes the AI-era rise in US valuations into earnings, discount rates and risk premia
Victoria Shinder2 min readA box in the ECB Economic Bulletin, Issue 6/2026, prepared by Magdalena Grothe, Bruno Lopes Mendes, Ana-Simona Manu and Luca Tondo, examines what has been holding up US equity valuations through what it calls the AI boom — the period of fast innovation and growing use of AI technologies since the public release of ChatGPT in November 2022.
The analytical frame is a dividend discount model, and the authors are explicit that it leaves exactly three channels through which valuations can rise: higher earnings expectations, lower discount rates, and lower risk premia. They then note that this year's earnings seasons saw many US companies report strong results, with overall earnings growth for S&P 500 firms again outperforming earlier expectations, and that strong earnings have fed into expectations of further growth.
What it means
The value of this kind of box is the discipline of the decomposition rather than any conclusion. Only three things can lift a valuation, and they carry very different implications for a rates desk. If the rise is earnings expectations, it is a claim about the real economy that either materialises or does not. If it is lower discount rates, it is a claim about the path of policy, and it unwinds when that path changes. If it is compressed risk premia, it is a claim about how much compensation investors demand for being wrong — and that is the component that moves fastest and with least warning.
Why a euro area central bank is publishing it is the second question, and the answer is straightforward. US equity valuations are a euro area financial stability input, transmitted through portfolio holdings, through the wealth effect on euro area investors, and through the dollar. A repricing that originates in one of the three channels above does not stay in one market.
The honest caveat is in the method itself. A dividend discount model tells you how a move must be attributed given its assumptions; it does not tell you which attribution is true. The useful thing to take from the box is not a verdict on whether US equities are expensive, but the list of three things to watch — and the knowledge that only one of them is about AI at all.