skip to main content

Explore Research Products in the NSF-PAR

Title: Deep empirical risk minimization in finance: Looking into the future
Abstract

Many modern computational approaches to classical problems in quantitative finance are formulated as empirical loss minimization (ERM), allowing direct applications of classical results from statistical machine learning. These methods, designed to directly construct the optimal feedback representation of hedging or investment decisions, are analyzed in this framework demonstrating their effectiveness as well as their susceptibility to generalization error. Use of classical techniques shows that over‐training renders trained investment decisions to become anticipative, and proves overlearning for large hypothesis spaces. On the other hand, nonasymptotic estimates based on Rademacher complexity show the convergence for sufficiently large training sets. These results emphasize the importance of synthetic data generation and the appropriate calibration of complex models to market data. A numerically studied stylized example illustrates these possibilities, including the importance of problem dimension in the degree of overlearning, and the effectiveness of this approach.

  more » « less
NSF-PAR ID:
10442964
Author(s) / Creator(s):
 ;  
Publisher / Repository:
Wiley-Blackwell
Date Published:
Journal Name:
Mathematical Finance
Volume:
33
Issue:
1
ISSN:
0960-1627
Page Range / eLocation ID:
p. 116-145
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ; ; ; ; ; ; ; ; ; et al ( , Journal of the American Medical Informatics Association)
    Abstract Importance

    The study highlights the potential of large language models, specifically GPT-3.5 and GPT-4, in processing complex clinical data and extracting meaningful information with minimal training data. By developing and refining prompt-based strategies, we can significantly enhance the models’ performance, making them viable tools for clinical NER tasks and possibly reducing the reliance on extensive annotated datasets.

     Objectives

    This study quantifies the capabilities of GPT-3.5 and GPT-4 for clinical named entity recognition (NER) tasks and proposes task-specific prompts to improve their performance.

     Materials and Methods

    We evaluated these models on 2 clinical NER tasks: (1) to extract medical problems, treatments, and tests from clinical notes in the MTSamples corpus, following the 2010 i2b2 concept extraction shared task, and (2) to identify nervous system disorder-related adverse events from safety reports in the vaccine adverse event reporting system (VAERS). To improve the GPT models' performance, we developed a clinical task-specific prompt framework that includes (1) baseline prompts with task description and format specification, (2) annotation guideline-based prompts, (3) error analysis-based instructions, and (4) annotated samples for few-shot learning. We assessed each prompt's effectiveness and compared the models to BioClinicalBERT.

     Results

    Using baseline prompts, GPT-3.5 and GPT-4 achieved relaxed F1 scores of 0.634, 0.804 for MTSamples and 0.301, 0.593 for VAERS. Additional prompt components consistently improved model performance. When all 4 components were used, GPT-3.5 and GPT-4 achieved relaxed F1 socres of 0.794, 0.861 for MTSamples and 0.676, 0.736 for VAERS, demonstrating the effectiveness of our prompt framework. Although these results trail BioClinicalBERT (F1 of 0.901 for the MTSamples dataset and 0.802 for the VAERS), it is very promising considering few training samples are needed.

     Discussion

    The study’s findings suggest a promising direction in leveraging LLMs for clinical NER tasks. However, while the performance of GPT models improved with task-specific prompts, there's a need for further development and refinement. LLMs like GPT-4 show potential in achieving close performance to state-of-the-art models like BioClinicalBERT, but they still require careful prompt engineering and understanding of task-specific knowledge. The study also underscores the importance of evaluation schemas that accurately reflect the capabilities and performance of LLMs in clinical settings.

     Conclusion

    While direct application of GPT models to clinical NER tasks falls short of optimal performance, our task-specific prompt framework, incorporating medical knowledge and training samples, significantly enhances GPT models' feasibility for potential clinical applications.

     
    more » « less
  2. ; ( , Computer-Aided Civil and Infrastructure Engineering)
    ABSTRACT

    The presented methodology results in an optimal portfolio of resilience‐oriented resource allocation under weather‐related risks. The pre‐event mitigations improve the capacity of the transportation system to absorb shocks from future natural hazards, contributing to risk reduction. The post‐event recovery planning results in enhancing the system's ability to bounce back rapidly, promoting network resilience. Considering the complex nature of the problem due to uncertainty of hazards, and the impact of the pre‐event decisions on post‐event planning, this study formulates a nonlinear two‐stage stochastic programming (NTSSP) model, with the objective of minimizing the direct construction investment and indirect costs in both pre‐event mitigation and post‐event recovery stages. In the model, the first stage prioritizes a bridge group that will be retrofitted or repaired to improve the system's robustness and redundancy. The second stage elaborates the uncertain occurrence of a type of natural hazard with any potential intensity at any possible network location. The damaged state of the network is dependent on decisions made on first‐stage mitigation efforts. While there has been research addressing the optimization of pre‐event or post‐event efforts, the number of studies addressing two stages in the same framework is limited. Even such studies are limited in their application due to the consideration of small networks with a limited number of assets. The NTSSP model addresses this gap and builds a large‐scale data‐driven simulation environment. To effectively solve the NTSSP model, a hybrid heuristic method of evolution strategy with high‐performance parallel computing is applied, through which the evolutionary process is accelerated, and the computing time is reduced as a result. The NTSSP model is implemented in a test‐bed transportation network in Iowa under flood hazards. The results show that the NTSSP model balances the economy and efficiency on risk mitigation within the budgetary investment while constantly providing a resilient system during the full two‐stage course.

     
    more » « less
  3. ; ; ; ; ( , American Journal of Physical Anthropology)
    Abstract Objectives

    Evolution of human maternal investment strategies is hypothesized to be tied to biological constraints and environmental cues. It is likely, however, that the socioecological context in which mothers' decisions are made is equally important. Yet, a lack of studies examining maternal investment from a cross‐cultural, holistic approach has hindered our ability to investigate the evolution of maternal investment strategies. Here, we take a systems‐level approach to study how human life history characteristics, environments, and socioecology influence maternal investment in their children.

     Materials and methods

    We test how infant age and sex, maternal age, parity, and child loss, and the composition of a child's cooperative breeding network are associated with maternal investment across three small‐scale (hunter‐gatherer, horticultural, and agropastoral), sub‐Saharan populations (N= 212). Naturalistic behavioral observations also enable us to illustrate the breadth and depth of the human cooperative breeding system.

     Results

    Results indicate that infant age, maternal age and parity, and an infant's cooperative childcare network are significantly associated with maternal investment, controlling for population. We also find that human allomaternal care is conducted by a range of caregivers, occupying different relational, sex, and age categories. Moreover, investment by allomothers is widely distributed.

     Discussion

    Our findings illustrate the social context in which children are reared in contemporary small‐scale populations, and in which they were likely reared throughout our evolutionary history. The diversity of the caregiving network, coupled with life history characteristics, is predictive of maternal investment strategies, demonstrating the importance of cooperation in the evolution of human ontogeny.

     
    more » « less
  4. ; ; ( , Evolution)
    Abstract

    A longstanding goal of evolutionary biology is to understand among-individual variation in resource allocation decisions and the timing of these decisions. Recent studies have shown that investment in elaborate and costly weapons can result in trade-offs with investment in testes. In this study, we ask the following questions: At what point plasticity in resource allocation to these different structures ceases during development, if at all? Furthermore, can individuals tailor their reproductive behavior to accompany structural changes? We experimentally addressed these questions in the insect Narnia femorata, quantifying resource reallocation across development for the first time, using a phenotypic engineering approach. To investigate whether allocation plasticity diminishes throughout ontogeny, we induced weapon loss at a range of different developmental stages and examined subsequent testes mass and reproductive behavior. We found that relative testes mass increased as weapon investment decreased, implying a direct trade-off between testes and weapon investment. However, autotomy postadulthood ceased to induce larger testes mass. Intriguingly, losing a weapon while young was associated with extended adult mating duration, potentially enabling compensation for reduced fighting ability. Our results highlight the importance of examining the ontogeny of trade-offs between reproductive traits and the flexibility of the relationship between reproductive morphology and behavior.

     
    more » « less
  5. ; ; ( , APSIPA Transactions on Signal and Information Processing)
    null (Ed.)
    For the past few years, deep learning (DL) robustness (i.e. the ability to maintain the same decision when inputs are subject to perturbations) has become a question of paramount importance, in particular in settings where misclassification can have dramatic consequences. To address this question, authors have proposed different approaches, such as adding regularizers or training using noisy examples. In this paper we introduce a regularizer based on the Laplacian of similarity graphs obtained from the representation of training data at each layer of the DL architecture. This regularizer penalizes large changes (across consecutive layers in the architecture) in the distance between examples of different classes, and as such enforces smooth variations of the class boundaries. We provide theoretical justification for this regularizer and demonstrate its effectiveness to improve robustness on classical supervised learning vision datasets for various types of perturbations. We also show it can be combined with existing methods to increase overall robustness. 
    more » « less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email