Erik-Jan Senn

Thumbnail Image
Last Name
Senn
First name
Erik-Jan
Email
erik-jan.senn@unisg.ch
Phone
+49 15255 898827

Settings

Now showing 1 - 3 of 3
  • Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
  • Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    The power of visuals: Using social media images for financial sentiment analysis
    (2024-12-15)
    Erik-Jan Senn  
    ;
    Francesco Audrino  
    Financial sentiment analysis focuses mainly on text data. However, the importance of visual information from images has increased over the last decades, especially on social media. The objective is to investigate whether visual information influences the sentiment of retail investors and improves financial forecasting. The proposed sentiment model is based on visual information for stock-related posts on the social media platform StockTwits. The images are processed by a computer vision model and classified using user-labelled sentiment. The empirical analysis shows how visual sentiment impacts the classification performance of standard text-based models. In a financial forecasting application, the value of visual information is evaluated for financial variables such as realized volatility.
    Type:conference paper
    URI:https://www.alexandria.unisg.ch/handle/20.500.14171/124608
  • Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    LongFinBERT: A Language Model for Very Long Financial Documents
    (2023-12-18)
    Minh Tri Phan  
    ;
    Erik-Jan Senn  
    This paper introduces LongFinBERT, a modern language model specialized for processing long financial documents. Due to an adaptation in model architecture, LongFinBERT demonstrates substantially lower computational requirements for lengthy documents compared to other state-of-the-art language models. This characteristic enables processing of e.g. an entire annual accounting filing at once, which was previously computationally infeasible for LMs. We apply LongFinBERT to two empirical settings: Firstly, we aim to improve the detection of financial misreporting using text from 10-K filings from 1994 to 2018. Misreporting predictions that utilize text-based features from LongFinBERT outperform those based solely on accounting variables or other textual models, namely Latent Dirichlet Allocation, neural document embeddings, and FinBERT. Lastly, we find that market returns respond to year-over-year alterations of accounting disclosures, measured using LongFinBERT.
    Type:working paper
    URI:https://www.alexandria.unisg.ch/handle/20.500.14171/119138