Corporate Narratives Shape Analyst Expectations

Giuseppe Matera, author of the December 2025 study “Corporate Earnings Calls and Analyst Beliefs,” tackled a question that goes to the heart of how financial markets work: Do the stories companies tell matter as much as the numbers they report?

Using advanced machine learning and large language models (LLMs), Matera analyzed over 35,000 earnings call transcripts from 2008-2023 to understand how the way corporate executives frame information—not just the quantitative data itself—influences financial analysts’ forecasts and, ultimately, market expectations.

The study goes beyond traditional textual analysis by introducing a novel “text-morphing” methodology. Think of it as creating alternate realities: Matera used AI to generate counterfactual versions of CEO presentations that systematically vary the narrative emphasis (more confident, more uncertain, more optimistic, etc.) while keeping all the numbers identical. He then input the morphed transcripts into the machine learning models trained on analysts’ expectations and realized earnings. The difference between the new predicted outcome and the original prediction represents the predicted treatment effect of the specific narrative. This methodology allowed him to precisely compute analysts’ over- and underreaction to a given narrative

Key Findings: The Power of Narrative

1. Narratives Contain Real Information

The research definitively shows that earnings call language provides genuine incremental information beyond traditional financial metrics. Even after controlling for over 300 numerical firm characteristics, financial statements, and market data, textual features from earnings calls significantly improved the prediction of:

• Analyst earnings expectations (25-40% explanatory power from text alone).

• Forecast accuracy at longer horizons (27-32% improvement at 2-3 year horizons).

• Realized future earnings.

Bottom line: Analysts are rational to pay attention to earnings calls—they contain information not captured in the numbers alone.

2. Analysts Have Systematic Biases

By comparing how narratives affect analyst forecasts versus how they predict actual future earnings, Matera uncovered systematic biases:

Analysts UNDER-react to:

• Uncertainty narratives (−9 basis points in forecasts vs. −41 bps in realized earnings).

• Confidence signals (+6 bps vs. +26 bps).

• Forward guidance (+11 bps vs. +19 bps).

Analysts OVER-react to:

• Sentiment/optimism (+35 bps vs. +26 bps in realized earnings).

• Global macro framing (+11 bps vs. −19 bps—actually negatively predictive!).

3. Uncertainty Drives Disagreement

The research found that narratives of uncertainty have the strongest effect on forecast dispersion among analysts. When management emphasizes risks and unknowns, analysts interpret the information differently, leading to wider disagreement about future earnings.

4. Long-Term vs. Short-Term Effects

While textual information provides modest improvements over analyst forecasts at the one-year horizon (8% improvement), the gains become substantial at longer horizons—27% at two years and 32% at three years—suggesting that the qualitative dimensions of corporate communication capture aspects of firms’ trajectories that unfold gradually.

Investor Takeaways

1. Build better models: Incorporating sophisticated textual analysis (using modern NLP techniques, not just sentiment dictionaries) can meaningfully improve forecast accuracy, especially at longer horizons.

2. Be wary of optimism: Analysts systematically over-react to positive sentiment and optimistic framing. When management is upbeat, dial back your enthusiasm—the numbers may not live up to the narrative.

3. Pay attention to uncertainty: The research suggests analysts should weight uncertainty and risk narratives more heavily and be more skeptical of pure sentiment and global macro framing that may be strategically deployed.

The Bigger Picture

This research sits at the intersection of behavioral finance and artificial intelligence, demonstrating how modern computational tools can help us understand how markets process information. As Nobel laureate Robert Shiller has argued, narratives spread through economies like contagions, influencing behavior in ways traditional rational models overlook.

What makes this study interesting is its methodology. By using AI to create controlled “alternate versions” of the same earnings call, the researcher could isolate the causal effect of language in a way that would be impossible through traditional experiments or observational studies.

Summary

The key insight is both simple and profound: in financial markets, how you say something can matter as much as what you say. Numbers don’t speak for themselves—they’re always embedded in narratives that shape interpretation. The market may not be perfectly efficient, but research like this helps us understand exactly where and why—and how to potentially benefit from that understanding.

Larry Swedroe is the author or co-author of 18 books on investing, including his latest Enrich Your Future.