---
title: Measuring What is Repeated and Novel in Earnings Disclosures Using Optimal Transport
url: https://www.ml-quant.com/papers/arxiv/2610.04094/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-10-09
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2610.04094
source_url: https://arxiv.org/abs/2610.04094
featured: 2026-10-09
citations: unknown
topic: Asset Pricing & Factors
---


# Measuring What is Repeated and Novel in Earnings Disclosures Using Optimal Transport

Decomposing earnings press releases and conference calls into shared and unique textual components shows that unique call content predicts returns better than repeated material.

- Source: https://arxiv.org/abs/2610.04094
- Identifier: arXiv:2610.04094
- Released: 2026-10-06
- First featured: Quant Letter No. 134 (2026-10-09): https://www.ml-quant.com/issues/2026-10-09/
- Citations (Semantic Scholar): not tracked
- Published in: not yet
- Topic: Asset Pricing & Factors
- Authors: Yuntao Wu, Lynn Tao, Charles Martineau, Vincent Grégoire, Andreas Veneris

## Abstract (arXiv, CC0)

We introduce an optimal transport framework to decompose the textual content of earnings press releases and conference calls into aligned (shared) and unaligned (unique) components, and use these as predictors of stock returns around earnings announcements. Using over 105,000 document pairs from 2008 to 2023, we find that the aligned portions of both disclosures explain announcement-day returns with high statistical significance. The unaligned component of conference calls, which captures forward-looking content absent from press releases, provides substantial incremental explanatory power, shedding light on why conference calls are central to market reactions to earnings news. Extending to post-announcement returns, investors appear to overreact to shared press release content while underreacting to the unique information conveyed in conference calls.

## Related

- [Exogenous Risk, Hedging Pressure, and Risk Premia in Agricultural Commodity Markets](https://www.ml-quant.com/papers/repec/ags-aaea26-404411/): Traders place 15% weight on USDA crop reports relative to private priors when forming price expectations, with this anchoring weight rising when private analyst disagreement increases.
- [Interpretable Deep Learning for Stock Returns: A Consensus-Bottleneck Asset Pricing Model](https://www.ml-quant.com/papers/arxiv/2512.16251/): The Consensus-Bottleneck Asset Pricing Model uses a neural network to mimic analyst reasoning, showing how investor beliefs influence asset prices and enhancing long-term predictions for U.S. stocks.
- [Deep IV Factor Models](https://www.ml-quant.com/papers/ssrn/5283770/): The Deep Implied Volatility Factor Model, combining neural networks and linear regression, is proposed for estimating the daily Implied Volatility surface of individual stock options, improving performance around earnings announcements.
- [Search and Prices](https://www.ml-quant.com/papers/ssrn/4925143/): A new peer momentum strategy, based on firms co-searched by investors on the SEC EDGAR server, yields an annualized return alpha of 17, outperforming the shared-analyst peer momentum.
- [Cross-Bond Momentum Spillovers](https://www.ml-quant.com/papers/ssrn/4880869/): The study finds that bond peer momentums, particularly those with shared analyst linkage, can predict future bond returns in the corporate bond market. This strategy can yield significant profits, especially for bonds with less investor attention and higher arbitrage costs.
- [Owner's Earnings, Cash-based Operating Profits, and Capital Expenditures in the Cross Section of Stock Returns](https://www.ml-quant.com/papers/ssrn/4794183/): Stocks with high owner's earnings tend to predict average stock returns and outperform other factors, providing significant alpha over the FamaFrench 6factor and q5 factor models.