Intelligent Systems in Accounting, Finance and Management最新文献

The development of an in-house accounting bot—an artificial intelligence (AI) assistant capable of generating internally structured bookkeeping double-entry posting schemes—is explored in this paper. The processes of curating a suitable dataset, selecting, and fine-tuning a seven-billion-parameter language model, categorized as a small language model (SLM) (SLMs typically refer to models with fewer than 10 billion parameters, whereas medium-sized models often have 14B parameters, and large-scale models exceed 70B), are described. A human-evaluated benchmark is also presented to assess model performance. To achieve efficient supervised fine-tuning (SFT), low-rank adaptation (LoRA) was employed, significantly reducing memory requirements by using a small set of trainable parameters while maintaining model expressiveness. The process of backpropagation was further optimized using Unsloth, a high-performance training framework designed for efficient video memory usage and flash attention mechanisms, which accelerates adaptation and reduces memory overhead. The model whose layers were updated is called QwenCoder2.5. It was selected with the presumption that it would be able to learn how to generate and examine bookkeeping patterns generated by accounting information system (AIS) over a 17-year history. This proof of concept aims to support researchers and practitioners exploring the integration of generative AI in accounting by providing insights into both the benefits and challenges of AI-driven automation in bookkeeping tasks. The study demonstrates how an SLM can be fine-tuned on a proprietary dataset of journal posting schemes to assist accountants, auditors, and financial analysts while also facilitating synthetic data generation. Challenges related to AI, data preprocessing, fine-tuning optimization, and evaluation methodology are introduced and examined.

Increasing computing power and access to the internet have amplified the role of social media and online news media on financial market outcomes. However, these two sources of information are intertwined in such a way that information flows between them. As a result, sentiment expressed in one source can affect stock market outcomes through the other source. This study examines this interplay between news media sentiment, social media sentiment and stock returns within the Dow Jones constituent companies from 2016 to 2023. Leveraging an extensive dataset, we adopt an approach that combines causal mediation models with robust statistical techniques to establish the mediation effects of one sentiment proxy on the relationship between the other proxy and stock returns. We also use a range of other methods like path analysis, panel vector autoregression and causal forests for robustness. The study finds that news sentiment is more influential in directly affecting stock returns than Twitter sentiment while the latter is more influential indirectly when mediated by news sentiment.

This paper examines the prediction of IPO withdrawal using machine learning methods (lasso and random forest) and conventional regression (logit). The dataset comprises 2444 US first-time IPOs from 1997 to 2014. Results show that random forest outperforms both logit and lasso in in-sample and cross-sectional out-of-sample predictions when the training and test sets are drawn from the same time period. However, when models are trained on past data and tested on future observations, all models fail to accurately predict IPO withdrawal. This failure is attributed to concept drift—a change in the relationship between predictors and IPO withdrawal over time. I show that concept drift occurs at multiple points in time, affects various predictors, and persists even when accounting for economic shocks, institutional changes, or different prediction horizons. These findings suggest that the generalizability of previous results on IPO withdrawal is limited, as the relationship between various predictors and IPO withdrawal seems to vary across time periods.

This study explores the wealth of information inherent in online employee reviews as an emerging resource in academic research. The focus is on the fields of finance, accounting, economics, and management, with an emphasis on how employee reviews contribute to our understanding of these areas. A systematic literature review (SLR) of 70 high-quality articles highlights the insights gleaned from employee reviews. Their data points, such as employee satisfaction, employee outlook, evaluation of culture, management, and colleagues, and text comments are mainly used in (1) explaining and predicting firm performance, (2) predicting and understanding performance and satisfaction of specific job groups, and (3) CSR- and ESG-related research. This SLR is important because the three main topics mentioned in which employee reviews are mainly used are spread across the fields of finance, accounting, economics, and management. This SLR therefore provides researchers with an important and necessary overview of the research already addressed across these fields. Furthermore, the SLR provides an overview of employer rating platforms utilized for academic research and methods used to harness employee reviews for research purposes. Here, a significant finding of this SLR is the predominant use of Glassdoor as a data source and the focus on US markets. The SLR concludes by proposing five potential avenues for future research, paving the way for a deeper understanding of the interplay between employee reviews (information) and organizational dynamics.

In recent years, research studies have been conducted on analyzing journal-entry data using advanced visualization techniques and machine learning models. However, because of their highly confidential nature, these data are not disclosed externally, which can limit research and business opportunities to analyze the rich organizational information they contain. To address these problems, this study utilized a variational autoencoder to generate synthetic journal-entry data with statistical properties similar to those of actual data. The synthetic journal-entry data we created adhered to the fundamental structure of double-entry bookkeeping and were quantitatively evaluated for quality.

This paper explores the application of causal discovery algorithms to factor investing, addressing recent criticisms of correlation-based models. We create novel causal network representations of the S&P 500 universe and apply them to three investment scenarios. Our findings suggest that causal approaches can complement traditional methods in areas such as stock peer group identification, factor construction, and market timing. While causal networks offer new insights and sometimes outperform correlation-based methods in terms of risk-adjusted returns, they do not consistently surpass traditional approaches. The causal method though shows promise in identifying unique market relationships and potential hedging opportunities. However, its practical implementation presents challenges due to computational complexity and interpretation difficulties. Our study demonstrates the potential value of causal discovery in factor investing, while also identifying areas for further research and refinement.