---
title: Huge If True: Learning from the Size of the Surprise
url: https://www.ml-quant.com/papers/arxiv/2611.15747/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-11-14
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2611.15747
source_url: https://arxiv.org/abs/2611.15747
featured: 2026-11-13
citations: unknown
topic: Asset Pricing & Factors
---


# Huge If True: Learning from the Size of the Surprise

The study shows that investors learning about both news and valuation multipliers from earnings surprises respond less on average but more sharply near zero when multiplier uncertainty is high.

- Source: https://arxiv.org/abs/2611.15747
- Identifier: arXiv:2611.15747
- Released: 2026-11-13
- First featured: Quant Letter No. 139 (2026-11-13): https://www.ml-quant.com/issues/2026-11-13/
- Citations (Semantic Scholar): not tracked
- Published in: not yet
- Topic: Asset Pricing & Factors
- Authors: Thomas Graeber, Christopher Roth, Marco Sammon, Constantin Schesch

## Abstract (arXiv, CC0)

Responses to earnings surprises depend on an unknown multiplier mapping earnings to value. Investors, however, learn about the news and the multiplier from the same number. Large surprises may indicate small multipliers and tiny surprises large ones. We model how greater uncertainty about the multiplier attenuates responses on average but makes them steeper near zero surprises. In an incentivized experiment, we find that greater processing constraints lower average sensitivity but raise it near zero. We confirm this with real-world earnings announcements: stocks with higher valuation uncertainty, our proxy for multiplier uncertainty, respond less on average but more steeply near zero.

## Related

- [Behavioral Finance at 40](https://www.ml-quant.com/papers/repec/nbr-nberwo-35959/): A retrospective reviews behavioral finance's progress over four decades, covering earnings beliefs, return beliefs, prospect theory, and emerging directions in cognitive foundations and AI.
- [Investing in Data Capital](https://www.ml-quant.com/papers/ssrn/7680002/): Measuring data investment from earnings calls, the research shows data capital's share of investment reached about one-eighth in 2025 and commands a market premium of 12 dollars per dollar invested.
- [Not All Beats Are Equal: Anchoring on the Nominal Earnings Surprise](https://www.ml-quant.com/papers/ssrn/7676359/): Shows that stock returns respond more to the same percentage surprise when reported earnings are larger in cents, consistent with anchoring on nominal rather than percentage earnings beats.
- [What Language Models Add to Healthcare-Equity Rankings, and What They Could Have Known](https://www.ml-quant.com/papers/arxiv/2610.38056/): Language models improve one-year stock rankings for healthcare firms by 48%, especially when scientific texts cover trials or regulatory records, but contribute less at other horizons.
- [Linguistic Passiveness in Earnings Calls and Stock Returns](https://www.ml-quant.com/papers/arxiv/2610.19087/): Passive-voice fraction in earnings calls predicts temporary mispricings for good news and longer-run underperformance for bad news, independent of standard sentiment measures.
- [Debt Dictionaries](https://www.ml-quant.com/papers/repec/fip-fedpwp-103784/): Stock and bond investors extract different value signals from earnings calls, with bondholders emphasizing operations and risk while equityholders focus on growth and innovation.