Tariffs returned to the spotlight in financial markets once it became clear that self-proclaimed ‘Tariff Man’ Donald Trump was set to return to the White House. Attention increased further following ‘Liberation Day’ on 2 April, when Trump announced a new round of ‘reciprocal’ tariffs.¹

Since then, tariffs have been a recurring topic in both markets and boardrooms. Figure 1 below illustrates this trend. Each bar represents the number of MSCI World Index constituents that mentioned tariffs in their earnings calls, quarter by quarter. ² While the topic was prominent during Trump's first term, the number of companies referencing tariffs has since doubled from the Q3 2018 peak, with 600 out of 862 companies mentioning tariffs in the current (and still-ongoing) quarter.

Beyond the sheer number of companies mentioning tariffs, investors are also interested in the tone of those discussions. Are companies mainly worried, broadly neutral, or might some even be perceiving opportunities?

To extract such insights, we apply modern natural language processing (NLP) techniques to analyze earnings calls.³ In the chart above, the different bar segments indicate the sentiment classification according to our proprietary NLP sentiment pipeline. Notably, most companies spoke neutrally (orange) about tariffs. However, in periods when tariffs are high on the agenda, such as in 2018/2019 or now (April/May 2025), the share of companies expressing concerns increases, as illustrated by the purple line. At the same time, a small group of companies continues to see tariffs as opportunities.⁴

The tariff case shows how NLP can help investors extract sentiment from unstructured data – such as corporate filings, news articles, earnings calls, management interviews, and social media – and spot shifts in market narratives, even when the topic itself ebbs and flows over time. As these tools become more sophisticated, their role in identifying broader themes, tracking their evolution, and measuring associated sentiment continues to grow.

These advances mean the questions investors ask themselves are also evolving: from our tongue-in-cheek, ‘Who’s afraid of the Tariff Man?’ to the more serious, ‘What else can AI tools like NLP help us understand?’ Such next-gen quant capabilities – as well as knowing how to use them – are increasingly relevant, both for quantitative but also for thematic investors aiming to capture both emerging and established trends.

¹ For more details on how the reciprocal tariffs were computed, see Nangle T., April 2025, “The stupidest chart you’ll see today”, Financial Times.
² Earnings call coverage typically ranges from 60% to 85% of MSCI World Index constituents, depending on the quarter.
³ Unlike structured financial or market data, text data from sources such as earnings calls, news articles, or management interviews is unstructured and context-dependent. Early methods like ‘Bag of Words’ have long been used to quantify sentiment, but they lack the ability to capture relationships between words. More advanced NLP techniques can address this by using contextual embeddings – such as FinBERT – or transformer-based models like GPT, which better interpret meaning in context.
⁴ Analyzing sentiment at the sector level highlights the effectiveness of our NLP pipeline. On average, companies in Utilities, Communication Services, and Financials express less concern about tariffs, while those in Information Technology, Consumer Discretionary, and Consumer Staples tend to be more worried.