arXiv subjects

Christophe Pérignon

Publications and source records attributed to Christophe Pérignon.

4 recordsLinked to original sources

arXiv2026

Randomness in large language models: What researchers need to know (and report)

Large language models (LLMs) are increasingly used to generate data for research. Typical use cases are classifications, annotations, information extraction, and generation of numerical scores. Unlike conventional measurements, LLM outputs can vary across repeated requests even when the prompt and apparent model settings remain unchanged. This variation arises from deliberate sampling, silent model updates, numerical rounding, or expert routing. Setting a dedicated temperature parameter to zero removes deliberate sampling when that option is available, but it does not eliminate the other sources of randomness. Exact reproduction is therefore generally not possible when using proprietary application programming interfaces. Local execution of open-weight models offers greater control, but reproducibility still depends on the complete hardware and software stack. We illustrate these issues through sentiment classifications of corporate filings and examine their consequences for downstream regression results. We then propose a reporting standard for articles and replication packages, as well as guidance for data editors and authors. Together, these findings and recommendations establish that LLM outputs should be treated as draws from a distribution rather than as fixed measurements.

econ.GN↗

arXiv2026

The Challenger: When Do New Data Sources Justify Switching Machine Learning Models?

Organizations often have an incumbent predictive model in production when new data sources become available. Because historical training data lack the new features, a challenger model must be trained on a small but growing full-feature dataset. We study whether, and when, the organization should switch to the challenger. The decision is statistical and economic: the challenger's predictive performance improves as full-feature data accumulate, but repeated retraining is costly and delays benefits from deployment. We develop a framework linking learning-curve dynamics to model-switching economics. Under a standard power-law learning curve and finite data-collection horizon $T$, the optimal time to train and evaluate the challenger scales as $T^{1/(1+α)}$: learning-curve shape (through its learning speed $α$) is the primary theoretical determinant of when to stop experimenting; costs determine switching profitability. Even without knowing the learning curve, the operational problem is tractable: we show that any algorithm stopping on the $T^{2/3}$ scale and making reliable switch/discard decisions achieves $O(T^{2/3}\sqrt{\log T})$ regret relative to a full-foresight oracle. We propose a sequential evaluation algorithm that uses local learning-curve trends to anticipate improvement, and test it in a real-world credit-scoring study. Even with this local approximation, the algorithm theoretically and empirically achieves near-oracle performance. It is also more stable than greedy sequential evaluation algorithms, where noisy early estimates trigger premature discarding, or simple one-shot evaluation algorithms, which work only when their fixed evaluation time matches the (unknown in practice) theoretical timing scale. Our framework offers a step toward principled model governance when new data sources require costly collection, validation, and deployment.

cs.LG↗

arXiv2025

Global p-Values in Multi-Design Studies

Replicability issues -- referring to the difficulty or failure of independent researchers to corroborate the results of published studies -- have hindered the meaningful progression of science and eroded public trust in scientific findings. In response to the replicability crisis, one approach is the use of multi-design studies, which incorporate multiple analysis strategies to address a single research question. However, there remains a lack of methods for effectively combining outcomes in multi-design studies. In this paper, we propose a unified framework based on the g-value, for global p-value, which enables meaningful aggregation of outcomes from all the considered analysis strategies in multi-design studies. Our framework mitigates the risk of selective reporting while rigorously controlling type I error rates. At the same time, it maintains statistical power and reduces the likelihood of overlooking true positive effects. Importantly, our method is flexible and broadly applicable across various scientific domains and outcome results.

stat.ME↗

arXiv2024

The Fairness of Credit Scoring Models

In credit markets, screening algorithms aim to discriminate between good-type and bad-type borrowers. However, when doing so, they can also discriminate between individuals sharing a protected attribute (e.g. gender, age, racial origin) and the rest of the population. This can be unintentional and originate from the training dataset or from the model itself. We show how to formally test the algorithmic fairness of scoring models and how to identify the variables responsible for any lack of fairness. We then use these variables to optimize the fairness-performance trade-off. Our framework provides guidance on how algorithmic fairness can be monitored by lenders, controlled by their regulators, improved for the benefit of protected groups, while still maintaining a high level of forecasting accuracy.

stat.ML↗