Market sentiment analysis in reserve management

The global financial landscape is undergoing a profound transformation, driven by rapid digitalisation and the emergence of advanced technologies such as artificial intelligence and machine learning (ML). For central banks, these developments present both challenges and opportunities, and this chapter explores the potential opportunities of AI in enhancing reserve management strategies at the Bank of Portugal.

According to Concetto and Ravazzolo (2019),¹ optimism – also referred to in the paper as market sentiment – reveals “the movements in the financial markets dictated by the psychological perception of determined operations or trades”. In the US example in this paper, investor sentiment has a negative impact on stock market returns. While no relationship was found in the European example, the authors found a spillover effect from the investor sentiment in the US market to Europe. This finding shows the importance of tracking the evolution of market sentiment to better inform asset management activities. It is of particular importance to monitor sentiment across different market topics and segments, and the respective change over time.

In the Bank of Portugal’s case, market sentiment analysis (MSA) is performed by different teams – namely, asset managers and market analysts, who face several challenges in their daily activities. First, the difficulty of gathering and analysing incoming information, much of it unstructured, given its sheer volume and real-time updates. Second, the dissemination of information across multiple individuals often leads to inefficiencies, redundancies and gaps in critical knowledge. Third, the overwhelming inflow of data makes prioritisation challenging, potentially resulting in overlooked insights and suboptimal allocation of time and resources.

Addressing these challenges requires a structured and centralised approach to information management, leveraging AI-driven tools to enhance efficient filtering, prioritisation and real-time insights. This has led to the development of the Bank of Portugal’s MSA tool for reserve management activities.

Market sentiment analysis tool

In a 2024 Central Banking FinTech and RegTech Global Awards article – ‘Artificial Intelligence Initiative: Bank of Portugal’² – it was reported that this project was built on Alya, an AI platform developed by the Bank of Portugal. Alya was given the Best Government Project award in the Portugal Digital Awards 2023, in the digital transformation projects in public administration category. The platform enhances operational efficiency by automating workflows and providing real-time analysis of documents and data for different operational activities.

The MSA project integrates AI and natural language processing (NLP) techniques for text processing and extraction of entities and topics. As for market sentiment classification, customised ML coding was developed, using NLP and classification algorithms.

During the early development of the MSA project, it was essential to establish a clear definition of market sentiment. For this purpose, it was considered that sentiment comprises three categories: positive; negative; and neutral. Sentences were classified with a positive sentiment whenever they signalled an increase in an asset valuation or higher economic growth, and were labelled with negative sentiment if they signalled a decrease in an asset valuation or lower economic growth. Neutral statements were attributed to sentences that were purely informational or that contained both positive and negative elements, ultimately balancing their overall sentiment classification. Sentences with outcomes that matched measures of market consensus were also classified as neutral. As such, the inclusion of a neutral sentiment was critical to enable a better distinction from positive and negative statements.

Additionally, six market segments were considered – bonds, macroeconomic, geopolitical, foreign exchange, commodities and stocks – as key areas for the segmentation of sentiment analysis, ensuring a comprehensive and structured approach to market interpretation.

Once all requirements were mapped and defined, the MSA project was developed, in four sequential executional steps (see figure 1).

The first one corresponds to data extraction, during which the system automatically retrieves data, in real time, from a variety of structured and unstructured sources. Once integrated into the system, AI and NLP techniques are applied to process, clean and refine the text, ensuring data quality while extracting key entities and topics.

Subsequently, the system categorises sentences based on relevant topics and market segments, before automatically inferring the sentiment per each statement. Lastly, the system also identifies and highlights emerging topics, offering valuable foresight into potential future market trends.

To facilitate data exploration, an interactive dashboard was developed that enables both asset managers and market analysis teams to explore ‘news sentiment’ across multiple market segments. The goal of this dashboard is to showcase the sentiment of a given market segment on a chosen date, as well as the respective sources. This dashboard also contains insights into key topics and emerging trends prevalent across all documents. Through this tool, users are able to examine the market sentiment label of each sentence, which is complemented by daily analysis performed by asset managers and market analysts.

Other potential usage of AI-powered tools in reserve management and its implications

The Bank of Portugal is assessing the potential applications of AI in reserve management activities. The increasing complexity of financial markets, heightened volatility and the massive volume of real-time data all make traditional reserve management approaches less efficient. On the other hand, AI offers powerful tools to enhance asset managers’ decision-making processes.

This may include the development of AI and ML models for advanced data analysis and predictive modelling, enabling deeper insights and more accurate forecasts of financial asset movements. Additionally, the integration of generative AI holds significant promise for enhancing back-office operations and streamlining analysis of legal documents, ultimately improving operational efficiency and mitigating operational risk.

Nonetheless, and although AI systems offer significant advantages, it also presents inherent challenges.

One key consideration is the continuous training and adaptation that the AI/ML models require. Unlike traditional IT systems, where maintenance was primarily the responsibility of IT teams, the ongoing refinement of AI models also needs the intervention of business units, requiring new skills and resources. IT and business units need to work closely to ensure that the model maintains an acceptable level of accuracy and to guarantee that the results lead to correct and informed insights. Another challenge is the ongoing need to monitor data sources to ensure the adequacy of the inputs.

Additionally, there is the balance between expectations and actual benefits. Evaluating and measuring the value of the model – assessing whether the performance of the model justifies its costs – is essential in determining its long-term viability.

Looking ahead

The Bank of Portugal’s commitment to innovation and digital transformation remains a priority. However, the adoption of AI and other emerging technologies must be approached with a strategic and sustainable mindset to ensure long-term success.

A thoughtful and strategic integration process is essential to mitigate potential risks, including data-security concerns, ethical considerations and the impact on workforce dynamics. By fostering a culture of responsible innovation, one can maximise the benefits of AI while ensuring alignment with regulatory standards, business objectives and organisational values.

Notes

1. Concetto, CL, and F Ravazzolo, in ‘Optimism in financial markets: stock market returns and investor sentiments’, Journal of Risk and Financial Management, May 2019.

2. Chow, T, ‘Artificial Intelligence Initiative: Bank of Portugal’, November 13, 2024.