Introduction In the ever-evolving landscape of data science and machine learning, a constant factor is that data reigns supreme. The ability to generate synthetic data that accurately mirrors real-world datasets has the potential to unlock proprietary data towards research, policy, and the public. Together with Ghent University and the Katholieke Universiteit Leuven, we have been at the forefront of exploring the capabilities of generative adversarial networks (GANs) to create synthetic data. Our recent work delves into the preservation of causal structures within synthetic datasets, a crucial factor when these data are used for decision-making. The Power of Synthetic Data Generative models, particularly GANs, have revolutionized our ability to simulate realistic data. These models learn the distribution of a dataset and generate new samples that maintain the same statistical properties as the original data. The potential applications of synthetic data are vast, ranging from enhancing machine learning models to preserving privacy in sensitive datasets. However, while the utility of synthetic data is clear, its application in decision-making contexts—where understanding causality is paramount—introduces significant challenges. Our research focuses on evaluating the extent to which generative models, specifically GANs, can replicate the causal relationships inherent in real data. Methodology: Evaluating Causality in Synthetic Data To assess the causal replication capabilities of GANs, we designed an experiment using a dataset with a known causal structure. This allowed us to compare the causal inferences drawn from the synthetic data to those from the original dataset. Data Generation: We created a dataset where the data-generating process and the underlying structural causal model were explicitly defined. GAN Training: Using this dataset, we trained various GAN models, including standard GANs, TimeGAN (which focuses on time-series data), and CausalGAN (designed to respect causal graphs). Causal Inference: We applied classic causal inference methods from econometrics to both the original and synthetic datasets to evaluate how well the synthetic data preserved the original causal relationships. Key Findings Our findings highlight both the promise and the limitations of current generative modeling techniques: Correlation vs. Causation: In cases where the assumptions for causal inference are straightforward (i.e., where correlation equals causation), GAN-generated synthetic data performed well. However, as the complexity of the causal relationships increased, the models often defaulted to simpler structures, potentially missing critical causal links. TimeGAN Performance: TimeGAN, which is tailored for time-series data, struggled to maintain accurate causal relationships in more complex settings. While it captured some temporal dynamics, it often oversimplified the underlying causal structures. CausalGAN Insights: CausalGAN, which incorporates causal graphs into the generative process, showed promise in preserving causal structures. However, it requires accurate prior knowledge of the causal graph, which is not always feasible in real-world applications. Implications for the Future The ability to generate synthetic data that accurately reflects the causal relationships in real datasets holds immense potential for fields where data privacy and ethics are paramount. However, our research underscores the need for caution. When using synthetic data for decision-making, it is crucial to ensure that the data not only captures the statistical properties of the original dataset but also maintains the underlying causal structures. Challenges and Recommendations Complex Causal Structures: The current state of GANs is such that they can struggle with complex causal relationships, often simplifying them to fit more straightforward models. This can lead to incorrect inferences if the synthetic data is used without further validation. Augmenting Observational Data: Incorporating additional information, such as environmental contexts or interventional data, can enhance the causal fidelity of synthetic datasets. This approach, however, is not always practical, especially in fields like finance or healthcare. Real-World Applications: Organizations must be aware of the limitations of synthetic data in causal analyses. While synthetic data can be an invaluable tool for privacy-preserving data sharing and preliminary analysis, it should be used cautiously when making decisions that depend on accurate causal inference. Conclusion Our exploration into the causality-preservation capabilities of GANs highlights the exciting potential and the significant challenges of using synthetic data in decision-making contexts. As generative models continue to evolve, enhancing their ability to replicate complex causal relationships will be crucial. For now, the use of synthetic data should be carefully considered, particularly in applications where understanding causality is essential. Together with Ghent University and the Katholieke Universiteit Leuven, we remain committed to advancing this field, ensuring that synthetic data can be a reliable and ethical tool for future research and decision-making. Read the full paper here: LINK

Introduction In times of economic uncertainty, timely and accurate data is crucial for decision-makers. Traditional methods of estimating Gross Domestic Product (GDP) often lag behind real-world events, leaving policymakers with outdated information. Together with Prof. Kris Boudt, Prof. Koen Schoors, Feliciaan De Palmenaer, and Arno De Block from Ghent University, we have developed a novel approach to nowcasting GDP using bank transaction data, offering near-real-time insights into economic activity. How It Works Our approach utilises anonymised micro-level bank transaction data from both firms and consumers. By aggregating these transactions and mapping them to GDP components—output, expenditures, and income—we have developed indicators that provide a real-time snapshot of the economy. Data and Methodology The bank transaction data includes all types of financial transactions, such as payments for goods and services, wages, and investments. These transactions are categorized and aggregated to reflect the economic activities they represent. Our methodology involves: Labeling Transactions: We tag each transaction with a specific label indicating its economic purpose, such as consumption or investment. Aggregation: Transactions are aggregated on a weekly basis, creating high-frequency indicators that reflect economic activity. Indicator Construction: These weekly indicators are used to construct quarterly growth rates, which are then compared to traditional GDP estimates that are released weeks later. Results Our innovative nowcasting model has shown impressive results, particularly during periods of economic shocks. Here are some key findings: Enhanced Accuracy During Shocks: Our model performed exceptionally well during economic crises, such as the COVID-19 pandemic, by quickly reflecting changes in economic activity. This allowed for more responsive policy decisions compared to traditional GDP estimates. Close Alignment with National Bank Estimates: Our nowcasting indicators closely tracked the estimates provided by the National Bank of Belgium (NBB), validating the model's effectiveness. The correlations between the BNP Paribas Fortis transaction-based indicators and official GDP figures were high, demonstrating the model’s reliability. (See Fig 1. and Fig 2. below) Timeliness: The real-time nature of the bank transaction data provided more timely insights than the traditional quarterly GDP reports. This is particularly valuable in the early stages of economic analysis when data from other sources is sparse. Fig. 1 Aggregated weekly costs- and revenues-based BNP Paribas Fortis-UGent indicators vs the quarterly official GDP growth numbers. Note that the BNP Paribas Fortis-UGent indicators are produced weeks before the official numbers and at a higher frequency (weekly vs quarterly). Fig. 2 BNP Paribas Fortis-UGent nowcast of GDP (red) vs the nowcast of the national bank (green and blue which are different models). In black the official GDP number. Note that the BNP Paribas-UGent indicator is produced weekly and thus comes out first out of all indicators shown, the final GDP number is published weeks after all others. Implications and Future Work The success of our nowcasting model has significant implications for economic policy and business decision-making. By providing near-real-time insights, policymakers can respond more swiftly to economic changes, potentially mitigating the impact of economic shocks. Businesses can also benefit from more accurate and timely data, aiding in strategic planning and risk management. Limitations and Recommendations While our model has shown great promise, it also has some limitations: Data Coverage: Our model relies on data from bank customers, which may not represent the entire economy. We are working to expand data sources and improve representativeness. Transaction Labels: Accurate labeling of transactions is crucial. We are continually improving the algorithms for categorising transactions to enhance the model’s accuracy. Seasonal Adjustments: Adjusting for seasonal variations remains a challenge. Further refinement of these adjustments will improve the precision of the nowcasting indicators. Conclusion Our nowcasting model represents a significant advancement in economic forecasting. By leveraging real-time bank transaction data, we provide timely and accurate estimates of GDP growth, closely aligning with traditional estimates while offering the advantage of immediacy. As our model continues to evolve, it holds the potential to transform economic analysis and policy-making, ensuring that decision-makers have the most current data at their fingertips.

Ghent, Belgium – 6th of June, 2024 – Today marks the official launch of Prometis Lab, a groundbreaking not-for-profit research organisation founded by BNP Paribas Fortis (BNPPF) and Ghent University. Prometis Lab is dedicated to leveraging real-world anonymised financial data and scientific research through an interdisciplinary approach to generate insights that will aid individuals, businesses, and policymakers in making better, more informed decisions. Our Mission Prometis Lab's mission is to empower socioeconomic decision-making by harnessing the immense power of anonymised administrative financial data. Our unique approach generates value far beyond conventional financial services, offering innovative solutions to pressing economic and social challenges. Our tagline, "Research for a changing world," encapsulates our commitment to dynamic and impactful research. Our Story The journey of Prometis Lab began in 2014 with a collaborative project between Ghent University and BNP Paribas Fortis, focusing on the influence of social networks on banking loyalty. This initial project laid the foundation for a vision to combine anonymised financial micro-data with socioeconomic research. Over the years, this collaboration deepened, marked by significant milestones such as the 2015 proof of concept on the impact of zero-interest rates on account balances, a project on the social mobility of early-career job-starters, and the 2020 research agreement facilitating broader collaboration. In 2021, BNP Paribas Fortis reinforced its commitment by endowing Ghent University with the “Research for a changing world” chair. Today, this evolving partnership culminates in the establishment of Prometis Lab, which will be led by CEO and Research Director Prof. dr. Milan van den Heuvel and aims at addressing socio-economic challenges on a structural level and conducting research to help navigate a rapidly evolving world. Innovative Collaboration Prometis Lab stands out as a unique collaboration model between the financial industry and academia. Unlike traditional project-based research initiatives, Prometis Lab creates a new platform for independent and socially relevant research. This flexibility allows us to quickly respond to both academic and practical opportunities, strengthening our position as a leading knowledge center in applied economic research. Our leadership structure ensures equal representation and decision-making power between our founders, fostering a balanced and innovative research environment. By working with de-identified financial data in a secure, GDPR-compliant framework within BNP Paribas Fortis infrastructure, Prometis Lab guarantees the safe, ethical, and responsible use of data. Research Streams and Impact Prometis Lab's research agenda focuses on several key research streams, including: Sustainability and ESG (Environmental, Social, and Governance) Metrics: Investigating the impact of ESG ratings on financial performance and green investments. Exploring consumer behaviour in response to climate change and sustainable practices. Nowcasting and Economic Indicators: Developing granular economic indicators to enhance understanding of GDP and inflation. Providing real-time insights into economic dynamics to inform better decision-making. AI for Decision-Making: Leveraging machine learning to improve business and policy decisions. Enhancing human decision-making through AI-augmented tools. Data Governance and Privacy: Ensuring data privacy and integrity in financial data usage. Evaluating and designing frameworks surrounding (EU) open data directives in the financial industry. Societal Challenges: Addressing issues such as social mobility, inequality, and the green transition. Informing public policies to create equitable and sustainable economic systems. We will continually evolve our research agenda to maximize our impact, ensuring that we stay attuned to and address the evolving needs of society. By doing so, we remain responsive and proactive in tackling emerging challenges. Our Vision for the Future Prometis Lab aspires to become an internationally recognized knowledge center, fostering research that has both academic rigor and practical relevance. By continuously innovating and expanding our research horizons, we aim to make significant contributions to the global economic landscape. Our goal is to provide insights that lead to smarter policies and business strategies, ultimately benefiting society as a whole. Founders and Independence Prometis Lab was jointly founded by BNP Paribas Fortis and Ghent University, with BNP Paribas Fortis providing initial funding and secure data infrastructure, and Ghent University contributing its academic expertise and research capabilities. This strategic partnership aims to create a lasting impact by supporting research that addresses the socio-economic challenges of our time. Prometis Lab is however an independent entity and as such the claims and opinions of Prometis Lab do not represent those of our founding members. For more information, visit our website at www.prometislab.org or contact us at info@prometislab.org

In a comprehensive study, in collaboration with Ghent University, examining the early career income trajectories of young professionals in Belgium, we have unveiled critical insights into the relationship between financial wealth and income growth. Published in Humanities and Social Sciences Communications, the study leverages extensive banking data to explore how pre-career financial wealth influences income mobility within the first seven years of employment. Key Findings The research highlights a significant correlation between higher financial wealth at career start and greater income growth. Using data from millions of Belgian banking clients, the study finds that individuals with higher initial financial wealth experience a noticeable boost in income growth as early as three years into their careers. For instance, a €5500 difference in starting wealth between the 25th and 75th percentiles translates to a 4% difference in income growth. Interestingly, the study controls for various factors such as demographic and geographical variables, which include age, gender, education level, and neighborhood characteristics. Even after accounting for these variables, the positive impact of financial wealth on income growth remains robust. Mechanisms and Channels The researchers explored several mechanisms to understand this relationship better. They assessed the role of social capital, innate abilities, and job search conditions. Their findings suggest that while social capital and innate abilities play a role, the capacity to sustain oneself financially during job search periods without immediate income is a crucial factor. Higher financial wealth provides the flexibility to find jobs that better match an individual's human capital, thereby enhancing productivity and income growth. The study introduces the concept of "thriftiness," or the proportion of financial wealth not consumed before starting a career. Higher thriftiness is associated with greater income growth, highlighting that the ability to save and manage financial resources effectively contributes significantly to early career success. Policy Implications These insights have profound policy implications. The researchers argue that policies aimed at reducing financial pressures for first-time job seekers could substantially improve economic mobility. By providing financial support to young professionals, governments could enable better job matches and enhance productivity, leading to higher income growth and reduced long-term income inequality. Conclusion In conclusion, the study provides compelling evidence that financial wealth at the start of a career significantly influences income mobility. It underscores the importance of financial support in enabling young professionals to secure better job opportunities and achieve higher income growth. These findings advocate for targeted policy interventions to alleviate financial constraints for career starters, fostering a more equitable and dynamic labor market. Read the entire study: https://www.nature.com/articles/s41599-022-01064-0 Recording of the "Universiteit van Vlaanderen"-talk (in dutch): https://www.youtube.com/watch?v=BjNCte7tRlw&t=4s

Introduction Together with economists at Ghent University, we have extensively studied the evolution of purchasing power among Belgian families during the recent government period. Using anonymised bank transaction data from 900,000 families, we have uncovered critical insights into how automatic wage indexation—designed to stabilize purchasing power—has instead created significant fluctuations along the income distribution. Our findings highlight the complexities and unintended consequences of this well-intentioned policy, particularly during periods of economic volatility. Key Insights Our analysis focuses on the real disposable income of different income groups, accounting for the composition of household expenditures and actual energy bills paid. Here are some key insights from our study: Overall Increase in Purchasing Power: On average, purchasing power increased by 1.4% during the "VIVALDI" government period, with a cumulative rise of 3.2% in 2021 and 2022. However, there has been a decline of 1.8 percentage points since February 2023. Heterogeneity in Volatility Across Income Quintiles: The highest income quintiles saw a substantial increase in purchasing power, with the top quintile experiencing a 4.5% rise. Conversely, the lowest income quintiles faced significant volatility, with the first quintile initially gaining 4.8% by the end of 2022 but then losing 7.5 percentage points, resulting in a net loss of 2.7% over the entire period. Three main causes are identified: the indexing of wages and benefits based on a distorted inflation rate (largely due to the energy component calculation), leading to an underestimation of inflation in 2021-2022 followed by underestimation in 2023-2024 due to rising social energy rates, and a recent decline in nominal income starting in 2024, reflecting the impact of recent economic setbacks on low-income groups. The Rollercoaster Effect of Wage Indexation The automatic indexation of wages, pensions, and benefits is intended to protect purchasing power against inflation. However, our research shows that this system has led to significant fluctuations, rather than stability, in purchasing power: Overestimation and Underestimation of Inflation: During 2021 and 2022, wages and benefits were indexed to a consumer price index (CPI) that significantly overestimated actual energy costs, artificially boosting purchasing power. However, from early 2023, the opposite occurred as the CPI underestimated inflation, leading to inadequate wage adjustments. Energy Prices and Social Tariffs: The volatility in energy prices and the differential impact of social tariffs on lower-income households further exacerbated these fluctuations. While higher-income families benefitted from declining commercial energy rates, lower-income families faced rising social tariffs. Sectoral Differences in Indexation: The timing and frequency of wage adjustments varied significantly across sectors, creating disparities in how different groups experienced changes in purchasing power. Implications for Policy Our findings suggest that the current system of automatic indexation, while well-intentioned, can inadvertently cause significant fluctuations in purchasing power, particularly during periods of economic instability. To address these issues, we propose several recommendations: Improving CPI Accuracy: To avoid misalignments in purchasing power adjustments, it is crucial to base the CPI on actual prices paid by households, particularly for essential expenses like energy. Revising Social Tariffs: The social tariff mechanism should be adjusted to better reflect the true cost of living for lower-income families, potentially linking these tariffs more closely to average market prices. Alternative Support Mechanisms: Instead of relying on energy bill adjustments, direct financial support (such as income cheques) could provide more stable and predictable assistance to households, reducing the volatility in the CPI and subsequent wage adjustments. Conclusion Our study reveals that the automatic indexation of wages and benefits in Belgium has led to unintended fluctuations in purchasing power, contrary to its goal of providing stability. By understanding and addressing the underlying causes of these fluctuations, policymakers can better design systems that truly protect households from inflationary pressures without introducing additional volatility. As researchers, we remain committed to exploring solutions that enhance economic stability and improve the well-being of all income groups. By making these adjustments, we can move closer to achieving the intended goal of stabilising purchasing power, ensuring that all families, especially those in lower-income brackets, are better protected against economic uncertainties. Read the full report here (dutch): https://www.ugent.be/eb/economics/en/research/gei/gei14 English version (machine translated):

Introduction In recent years, the term "energy crisis" has become a household term across the globe. The dramatic rise in energy prices has been cited as a key driver of inflation, prompting governments to intervene with various support measures. However, a report by Gert Peersman, Koen Schoors, and Milan van den Heuvel from the Department of Economics at Ghent University presents a compelling case that challenges the prevailing narrative around the energy crisis in Belgium. Key Findings The report analyzes the energy bills of 930,000 Belgian families, providing a detailed and nuanced picture of the real impact of the energy crisis. Here are some of the standout findings: 1. Actual Energy Cost Increase: While official statistics suggest an average increase of 81% in energy costs from 2018 to 2022, the reality for most families was much less severe. Bank transaction data revealed that the median family saw only a 1% increase, with the average increase being 17%. Remarkably, 47% of families paid less for energy in 2022 than in 2021 when taking into account all the government support that was issued. 2. Discrepancies Explained: Three primary factors contribute to the difference between official statistics and actual costs: - Contract Type: The consumer price index (CPI) in Belgium considers only new energy contracts, while many households maintained fixed-rate contracts from before the crisis. - Consumption Changes: Official figures are based on pre-pandemic consumption levels. In reality, many families reduced their energy consumption during the crisis through behavioral changes and energy efficiency investments. - Government Support: The spread of government support over time in statistics differs from the immediate relief experienced by families, leading to an overestimation in official figures. 3. Overestimation of Inflation: The misalignment between actual costs and the CPI resulted in an overestimation of inflation by 3.3%-points on average in 2022. This miscalculation led to an increase in wages and benefits higher than the actual loss of purchasing power due to the automatic indexation of wages in Belgium**. Implications and Recommendations The findings have significant implications for policy and economic planning. The overestimation of inflation has caused wage and benefit increases that do not align with the real economic impact on families. As energy prices stabilize and government support recedes, the CPI will likely show a lower increase than the actual rise in energy costs, potentially leading to a period of higher inflation in real terms for families. [UPDATE: this is exactly what was found in follow-up research here] Policy Recommendations: - Adjusting the CPI Calculation: The report recommends using actual energy contracts that can be provided by energy suppliers (and which is done in other European countries already) to calculate the CPI to avoid unwanted fluctuations in purchasing power and competitive positioning of companies. This adjustment would provide a more accurate reflection of economic conditions and help stabilize wages and benefits. Conclusion The research by Ghent University economists underscores the complexity of the energy crisis and its varied impact on households. By highlighting the discrepancies between official statistics and real-world data, the report calls for more accurate measurement tools and thoughtful policy adjustments to better support families and stabilize the economy. As we navigate the post-crisis period, these insights will be crucial in shaping effective and equitable economic policies. **Clarification on Automatic Indexation in Belgium In Belgium, automatic indexation is a mechanism that links wages and social benefits to the consumer price index (CPI) to protect purchasing power against inflation. When the CPI rises by 2%, wages and benefits are automatically adjusted through a spillover index to match the increase in the cost of living. This system aims to ensure that employees and benefit recipients maintain their purchasing power despite inflationary pressures. However, as this report highlights, inaccuracies in CPI calculations can lead to misaligned adjustments, either overcompensating or undercompensating households for changes in their cost of living. Read the full report here (machine translated to english): Full original report (in dutch): https://www.ugent.be/eb/economics/en/research/gei/gei9