This study will proceed as follows: We will briefly introduce how ChatGPT works and summarize the related research literature. The next section introduces the research method. Subsequently, we will delve into the practical application of the technology in financial services and explore the management of AI applications in financial services organizations. Besides, the study will also discuss the ethical considerations involved in applying ChatGPT. Finally, it will present an agenda for future research to promote the further development of ChatGPT in financial services.
ChatGPT is a deep learning-based language model that uses the transformers model architecture for powerful language generation and comprehension (Kasneci et al. 2023). Learning takes place through two phases: pre-training, which uses a large amount of unsupervised data to learn the structure and semantics of the language, and fine-tuning, which uses a small amount of supervised data to optimize the model for a specific task (Cheng et al. 2023). ChatGPT generates humanized responses, is iterative and extensible, but may have comprehension limitations for domain-specific knowledge. In addition, it may encounter difficulties in handling erroneous or inappropriate requests. ChatGPT can be applied to personal assistants and online customer service content generation.
GPT-3.5 is a transitional version of the GPT series of models between GPT-3 and GPT-4. It improved and optimized GPT-3 and paved the way for the release of GPT-4 (Stahl and Eke 2024). The most recent and sophisticated version, GPT-4, offers more outstanding performance and quality in language creation and understanding than GPT-3.5. Similar architecture and training methods are used, but the model's quality and effectiveness are increased.
To better understand ChatGPT's unique role, it is essential to compare it with other AI tools used in financial services. Traditional AI tools, such as rule-based chatbots and predictive models, often rely on pre-programmed rules or structured data to perform specific tasks. These tools, while effective for well-defined problems, lack the flexibility to handle nuanced, unstructured, and context-rich inputs (Kasneci et al. 2023).
In contrast, ChatGPT, built on the GPT architecture, leverages deep learning and large-scale natural language processing (NLP) capabilities to process vast amounts of unstructured data, such as financial news, reports, and user queries. Unlike traditional AI tools, ChatGPT can generate human-like, dynamic responses and adapt to complex conversational contexts, enabling it to support tasks like financial planning, portfolio analysis, and fraud detection more effectively (Lo and Singh 2023).
Moreover, ChatGPT's iterative learning and real-time adaptability distinguish it from specialized AI tools that are often task-specific. For example, traditional fraud detection systems rely on rigid pattern recognition, whereas ChatGPT can combine historical analysis with natural language inputs to provide real-time insights into suspicious activities ((Zaremba and Demir 2023).
However, it is important to acknowledge that while ChatGPT exhibits significant advantages in flexibility and conversational ability, it still faces challenges such as model bias and the need for human oversight, which are shared with other AI tools (Borji 2023). This comparison highlights ChatGPT's potential to enhance AI applications in finance by addressing unstructured data and providing more human-like interactions.
Theoretical perspective
The Efficient Market Hypothesis (EMH) posits that financial markets are "informationally efficient," meaning that asset prices fully reflect all available information at any given time (Fama 2017). This implies that, in an efficient market, it is impossible to consistently achieve returns higher than the overall market without assuming additional risk. Prices should adjust immediately to new information, which is reflected in the asset prices.
ChatGPT, as an advanced AI model with sophisticated natural language processing capabilities, has the potential to enhance market efficiency by improving information dissemination and accessibility. ChatGPT can analyze vast amounts of unstructured financial data, including news articles, earnings reports, and market analyses, and provide summarized, real-time insights (Lo and Singh 2023). This reduces information asymmetry -- a key factor in market inefficiency -- and enhances the speed at which market participants can access and process relevant data. ChatGPT's real-time capabilities allow for faster decision-making and better market reactions to new information, which theoretically aligns with the principles of EMH. However, the effectiveness of ChatGPT's contributions is highly dependent on the quality and accuracy of the data it processes. Errors or biases in the data could be propagated, leading to skewed insights and potentially affecting the accuracy of market predictions (Zaremba and Demir 2023).
Investor Behavior and Decision-Making. While EMH assumes that investors act rationally, behavioral finance provides a more nuanced understanding by recognizing that investor behavior is often influenced by cognitive biases, such as overconfidence, loss aversion, and herd behavior (Kahneman and Tversky 1979). ChatGPT can mitigate some of these biases by providing objective and data-driven analyses, thereby offering investors a more balanced perspective (Ullah et al. 2024). For instance, ChatGPT could help investors make more informed decisions by analyzing market sentiment, financial trends, and individual preferences, without being swayed by emotional biases.
However, ChatGPT may also introduce new challenges. For instance, if investors rely heavily on AI-driven outputs without critical evaluation, this could result in over-reliance on automated predictions, leading to homogenized strategies across market participants. This aligns with the notion in behavioral finance that, while rational decision-making is ideal, in practice, human biases often distort judgments. In this context, ChatGPT's role is to support investors in making rational decisions, but it cannot eliminate the risks associated with human behavior (George and George 2023).
Market Anomalies and Predictive Analysis. Despite the EMH, various market anomalies have been documented, such as momentum effects and mean reversion (Jegadeesh and Titman 1993). These anomalies challenge the idea that market prices always reflect all available information. ChatGPT's advanced analytical capabilities, particularly its ability to process big data and detect patterns in historical market behavior, may help identify and explain these anomalies. For instance, ChatGPT can assist in uncovering hidden correlations in financial data that are difficult for human analysts to detect.
However, while the integration of AI tools like ChatGPT into market analysis can uncover these anomalies, it also introduces the risk of homogenizing trading strategies. As more investors adopt AI-based tools, there is a risk that AI-generated strategies will converge, potentially reducing the diversity of market strategies. This could, in turn, dampen the effects of traditional market anomalies, such as momentum or mean reversion.
Empirical Evidence and Practical Implications. Recent empirical studies have explored the intersection of AI and market efficiency. Research indicates that AI-driven analyses can uncover value in smaller stocks and undervalued assets, potentially challenging traditional EMH assumptions that all available information is reflected in asset prices. For instance, AI tools like ChatGPT have been shown to improve decision-making accuracy and identify investment opportunities that might be overlooked by human analysts, especially in under-researched markets (Kim et al. 2023).
However, the increasing use of AI-powered tools across the financial sector presents a dual-edged sword: while they improve efficiency and accessibility, they also increase the risk of algorithmic convergence -- the idea that similar tools could lead to identical decision-making patterns, thus reducing market diversity and increasing market volatility (Zaremba and Demir 2023). Therefore, while ChatGPT has the potential to improve market efficiency, it also introduces new complexities to market behavior that require further empirical study.
Technology integration
The efficiency and competitiveness of corporate finance plans may be further improved by integrating ChatGPT with technologies like blockchain and big data. Here are a few potential integration strategies:
Smart contracts and distributed ledgers are two features that blockchain technology can offer. Combining ChatGPT with blockchain enhances data security and traceability (Wang et al. 2023). Companies can use ChatGPT to query and interpret financial data stored on the blockchain, providing accurate information retrieval and analysis. Further, ChatGPT can interact with smart contracts to automate contract execution and financial transactions.
Combining ChatGPT with big data technology can provide diverse data sources to enhance the accuracy and personalization of their answers (Zaremba and Demir 2023). Organizations can use big data analytics to predict market trends, risk assessments, and financial advice. Integrating these analytics into ChatGPT can yield more targeted services and customer advice.
