Bucharest University of Economic Studies, Bucharest, Romania

3-5 June 2026

Register now: https://forms.gle/TXVGX2SdvzuAB8LR6

Official Ceremony: June 4, Aula Magna, ASE Bucharest

The conference focuses on advancing the statistical & theoretical understanding of digital assets and energy finance, bringing together leading academics to address key open questions in these fields.

Keynote Speakers:

• Matthias Fengler (University of St. Gallen, Switzerland): From Digital Transactions to Real Spending: Insights from a Novel Consumption Index for Switzerland
• Rafał Weron (Wrocław University of Science and Technology, Poland): Machine learning for electricity price forecasting
• Chung-Ming Kuan (National Taiwan University, Taiwan): Large-Scale Multiple Inequality Testing without Data Snooping Bias

Themes:

• Digital Asset Market Microstructure & Quantitative Modeling
• AI-Driven Energy Finance & Power System Analytics
• Digital Asset Risk, Portfolio Optimization & Systemic Interaction
• Decentralised Finance, Blockchain Systems & Market Design

Read more and discover the full program by clicking here!

Date: 07-09/05/2026

Organization:

• Florian Ziel: Chair of Data Science in Energy and Environment, House of Energy, Climate & Finance,
Universität Duisburg-Essen.

• Wolfgang Karl Härdle: Institute for Digital Assets IDA, Bucharest University of Economic Studies.

• Souhir Ben Amor: House of Energy, Climate & Finance, Universität Duisburg-Essen.

Agenda:

24/04/2026

About:

This work studies how to improve the practical effectiveness of privacy-preserving machine learning in modern AI systems. It argues that the privacy-utility trade-off can be significantly mitigated through two complementary mechanisms: stronger data representations and tighter statistical privacy accounting. First, it shows that curated synthetic data can improve fine-tuning performance by preserving high-quality representations while avoiding model collapse. Second, it analyzes differentially private fine-tuning of pretrained models, demonstrating both theoretically and empirically that strong public representations are key to high private accuracy. Finally, it extends this perspective to societal-scale applications, including the U.S. Census and federated learning, where more accurate privacy accounting yields stronger utility under the same formal guarantees. Together, these results suggest that practical privacy in AI depends not only on adding noise, but on designing better representations, mechanisms, and accounting tools.

06/03/2026

About:

In functional time series, classical change-detection methods often flag shifts that are statistically detectable but too small to matter in practice. This talk introduces a self-normalized, adjusted-range approach for testing whether a change in a curve is large enough to be considered practically meaningful. The procedure works directly with the functional data and avoids common tuning decisions such as bandwidth choices or dimension reduction via functional principal components. We establish large-sample validity, propose a simple simulation-based calibration for the classical “no-change” benchmark case, and provide theory explaining why the adjusted-range normalization can improve power. Simulation studies across a range of dependence settings show a favorable size–power trade-off. We illustrate the method with executed trade-level Bitcoin options from Deribit, constructing daily 30-day constant-maturity implied-volatility smiles on a common log-moneyness grid and applying a rolling-window monitor. Using a relevance threshold of one volatility point (at the 10% level), the monitor highlights a small number of economically meaningful regime shifts (around early October and early December 2025). A simple straddle-based illustration shows how such relevance-gated signals can be translated into low-turnover decision rules.

20/02/2026

About:

We investigate the transaction patterns of Bitcoin users by their geographical origin and design Random Forest architectures to obtain indices for the share of transactions per region. We show that North American and European users transaction flows have a stronger market relationship than global flows. Using the indices, trained on a unique dataset spanning 5 years of blockchain transactions and their associated geographical origin, we construct region-specific blockchain metrics which we show improve market modelling and forecasting as compared to the global metrics. We further show with the indices regional differences in users willingness to pay transaction fees.

31/10/2025

About:

This talk examines the fiscal trade-offs when central bank digital currency (CBDC) competes with bank deposits. Debt finance supports banks’ money creation and intermediation but raises fiscal costs, while CBDC finance reduces costs and provides liquidity directly—at the expense of weakening banks’ role. Debt-financed stimulus can generate strong expansionary effects, while CBDC-financed stimulus delivers even larger effects by bypassing the banking sector. Under a fiscally dominant regime, the magnitude of these effects hinges on how strongly monetary policy responds to inflation.

The conference explores the convergence of AIoT, smart manufacturing, digital finance, and Industry 4.0, bringing together academics, educators, industry experts, and technologists to discuss how education and research can respond to a data-driven, hyperconnected economy. The event focuses on how AIoT is transforming production systems, how fintech is reshaping industrial transactions and supply chains, and how integrated skill development can prepare future-ready talent.

Featured Speakers & Talks

• Wolfgang Karl Härdle — Welcoming; Bird, Plane, Drone: EMD Solves It
• Daniel Traian Pele — AI and Digital Finance
• Xiaorui Zuo — May the Course Be With You
• David Siang-Li Jheng — Crypto Currency Returns
• Wing-Yi Chan — Analyzing Twitter Emoji Sentiment for Ethereum Trading Strategy
• Yo-Ping Huang — AIoT in Industrial Applications
• Owen Chaffard — Hierarchical Loan Forecasting
• Meng-Jou Lu — Which Risk Do Crypto Index Investments Have?
• Brent Cheng — The CRETRIX Project of Robotic Trinity for Combat/Rescue/Explore
• Tsung-Hsun Wu — Real-Time Voice Translation: AISPK

The detailed agenda also includes a second-day session on Modern Econometrics for Business AI with Q2 on Platograph, including Python, ChatGPT, Q2, and Platograph-based learning activities.

Themes

• AIoT, smart manufacturing, and Industry 4.0
• Fintech, digital finance, and industrial transactions
• Cryptocurrency markets, trading strategies, and risk analysis
• AI-assisted learning, modern econometrics, Python, and ChatGPT
• Enterprise digital intelligence, Plato Graph System, and digital twins
• Supply-chain integration, robotics, voice translation, and applied AI
• Future-ready education and workforce transformation

20/06/2025

About:

Speculative markets may be characterized by sharp falls after a slow build up. Sometimes the converse happens. The authors suggest a number of mechanisms that are able to produce this kind of behaviour and they demonstrate their plausibility by simulation. The models are then fitted to daily data on Bitcoin. In constructing these models they show that it is essential to take account of volatility and non-normality. The authors also investigate the possibility of a dynamic tail index.

The conclusion, at least for Bitcoin, is that speculative markets are more likely to behave like balloons than bubbles. In other words, there is rapid inflation followed by a slow decline.

28/02/2025

About:

Existing methods of using financial sentiments relies on equal-weight and static aggregation to manage sentiments from multiple news items. This leads to a critical issue termed “Aggregated Sentiment Homogenization”. This phenomenon occurs when aggregating numerous sentiments, causing representations to converge towards the mean values of sentiment distributions and thereby smoothing out unique and important information.

To address this issue, we introduce a novel method that leverages a dynamic market-news attention mechanism to aggregate news sentiments for market prediction. Our proposed approach learns the relevance of news sentiments to price changes and assigns varying weights to individual news items. By integrating the news aggregation step into the networks for market prediction, it allows for trainable sentiment representations that are optimized directly for prediction.

In addition, the diffusion of financial news into market prices is complex, making it challenging to evaluate the connections between news events and market movements. In this talk we will introduce introduces FININ (Financial Interconnected News Influence Network), a novel market prediction model that captures not only the links between news and prices but also the interactions among news items themselves. FININ effectively integrates multi-modal information from both market price and news articles. We conduct extensive experiments on two datasets, encompassing the S&P 500 and NASDAQ 100 indices over a 15-year period and over 2.7 million news articles. The results demonstrate FININ’s effectiveness, outperforming advanced market prediction models and latest large-language models. Moreover, our results reveal insights into the financial news, including the delayed market pricing of news, the long memory effect of news, and the limitations of financial sentiment analysis in fully extracting predictive power from news.

17/01/2025

About:

We propose to study the interaction effects of social and spatial networks in the presence of a noisy adjacency matrix. First, we provide evidence that existing network datasets exhibit low-rank, sparse, and noisy structures, and we utilize this information to create a de-noised version of the network.

We employ the Least Absolute Shrinkage and Selection Operator (LASSO) in conjunction with nuclear norm penalization to simultaneously regularize the sparse and low-rank components. We introduce two procedures:

1. A two-step estimator, where we first de-noise the adjacency matrix before using it in regression analysis.

2. A one-step supervised Generalized Method of Moments (GMM) estimator to estimate the regression parameter and the adjacency matrix.

Our results show that our estimation method performs favorably compared to GMM, especially when dense errors are present, and networks are endogenous to measurement errors. Simulation exercises indicate that our method outperforms GMM in estimating the regression coefficients by 50-70% in root mean squared error (RMSE) terms when noise is present in the network and maintains a significant advantage of approximately 60-80% with endogenous networks.

Additionally, we apply our method to the Besley and Coate (1995) (BC) dataset and find our methods deliver estimated spillover and multiplier effects that differ significantly from those obtained using BC. Furthermore, we show how our decomposition can be used to provide reliable and more detailed guidance for policy targeting under constraints.

16/01/2025

Watch a recording of our hybrid workshop covering topics such as AI-driven solutions for financial markets, blockchain, forecasting using LLMs and more!

Timestamps:

00:00 – Price Market Responses to Ethereum Development Milestones: An Event Study Approach

41:30 – Financial Risk Meters in Taiwan’s High-Cap Sectors

1:10:26 – Energy Price Modelling: A Comparative Evaluation of four Generations of Forecasting Methods

3:51:45 – Uncovering the Factors Behind Market Jumps: Insights from the First 100 Days of the Ukraine War

4:22:00 – LLMs for time series forecasting

4:58:46 – The link between energy prices and stock markets in EU countries

13/12/2024

About:

Based on options and realized returns, we analyze risk premia in the Bitcoin market through the lens of the Pricing Kernel (PK). We identify that: 1) The projected PK into Bitcoin returns is W-shaped and steep in the negative returns region; 2) Negative Bitcoin returns account for 33% of the total Bitcoin index premium (BP) in contrast to 70% of S&P500 equity premium explained by negative returns. Applying a novel clustering algorithm to the collection of estimated Bitcoin risk-neutral densities, we find that risk premia vary over time as a function of two distinct market volatility regimes. In the low-volatility regime, the PK projection is steeper for negative returns and has a more pronounced W-shape than the unconditional one, implying particularly high BP for both extreme positive and negative returns and a high Variance Risk Premium (VRP). In high-volatility states, the BP attributable to positive and negative returns is more balanced, and the VRP is lower. Overall, Bitcoin investors are more worried about variance and downside risk in low-volatility states.

The conference focuses on the role of Artificial Intelligence in security, ethics, and finance, bringing together scholars and industry professionals for keynote talks, interactive sessions, and research discussions on the latest developments in AI, digital finance, risk analytics, cybersecurity, and responsible innovation. The initiative aligns with Taiwan’s Ministry of Education Yushan Scholar Fellow Program, supporting international research cooperation and global academic integration.

Keynote Speakers

• Daniel Traian Pele — Bucharest University of Economic Studies, Romania; Institute for Economic Forecasting, Romanian Academy
  In the Beginning was the Word: LLM Risk Measures
• Stefan Lessmann — Humboldt University of Berlin, Germany
  A Tale of Bias and Missingness in AI-based Scoring Systems: Evidence from the Credit Industry
• Chuan-Ju Wang — Academia Sinica, Taiwan
  Textual Data Analytics in Finance
• Mark L. Shope — National Yang Ming Chiao Tung University, Taiwan
  Co-Synergy: The Power of Combined Human and AI Efforts in Interactive Processes

Themes

• AI, security, and digital identity
• Ethical and legal dimensions of AI
• Machine learning, LLMs, and financial risk analytics
• Digital finance, cryptocurrencies, Web3, and cybercrime
• Financial crime, data security, and regulatory challenges
• Human–AI collaboration and knowledge platforms

29/11/2024

About:

Communication clarity is crucial in central banking, yet its measure and impact remain under-explored. This study introduces a novel text-based model tailored exclusively for FOMC statements, extracting the text-implied policy stance from statements and offering a nuanced measure of communication clarity. The proposed clarity measure reflects to what extent the textual and numeric components of the FOMC communication are consistent. The clarity measure is considered as an information treatment in central banks’ communication strategies. The causal effect of communication clarity on the reduction of market uncertainty is significant; a one standard deviation increase in clarity results in a 22% reduction in VIX.

25/10/2024

About:

The increasing trading volume and regulatory scrutiny of cryptocurrencies, especially Bitcoin, have solidified their pivotal role in today’s financial markets.

This study examines various factors influencing future cryptocurrency returns. To overcome multi-collinearity problem and enhance prediction, we employ powershap for variable selection, which is a based on model interpretation (Verhaeghe et al, 2022). Our investigations focus on various periods before and after the Covid-19 pandemic, revealing the importance of technical indicators, oil prices, and exchange rates. To showcase the practical application of our method in predicting BTC returns, we present a trading strategy that demonstrates its potential for generating higher returns and Sharpe ratios.

Watch an exclusive and insightful interview with professor Härdle where he answers questions and offers personal insights on topics such as Machine Learning, FinTech, Big Data and the applications of these in economic research. Moreover, he also offers recommendations for students interested in Machine Learning and Econometrics and he also discusses personal research.

Click here to read the transcript!

29/07/2024

About:

This seminar introduces the Financial Risk Meter (FRM), rigorously derived from advanced statistical techniques, and capable of predicting market risks effectively.

The Financial Risk Meter (FRM) is designed to effectively predict future market risks. It identifies systemic network risks and dependencies among extreme events across different asset classes and regions. FRM connects asset pricing kernels, the highest Sharpe ratios, and overall market volatility. On all the presented FRM channels on theIDA.net it demonstrates its strength in detecting systemic risks and reveals network interconnectedness in tail event situations. The FRM predicts recessions and highlights peak risks during crises. Overall, the FRM provides valuable insights into systemic risks across various markets, helping policymakers and investors make informed decisions.

The slides are available on Quantinar, where you can learn more about FRM’s and its usage in different channels.

16/07/2024

About:

Topic modeling and LDA (Latent Dirichlet Allocation) have proven valuable in various fields as an innovative approach to studying areas of interest and identifying topics in a dynamic content. The underlying assumption is that techniques like LDA can swiftly capture emerging topics in textual documents compared to other categorization tools. These unsupervised approaches have been used to identify new industries and technological domains. However, our study on the nascent topic of “sustainability” within the corpora of SGX-listed companies highlights clear limitations in employing techniques like LDA on sparse data. The dynamic LDA approach, also called DTM (Dynamic Topic Modelling), based on an 11-year database of annual reports from publicly listed companies in Singapore, could not detect sustainability’s rise as a critical topic in corporate practice following policy changes. Moreover, despite sustainability reporting becoming mandatory, sustainability-related topics may still not receive significant attention.

The slides are available on Quantinar.

07/06/2024

About:

Exchanges operate limit-order markets where non-synchronous investors can trade. An early investor can leave a price quote for a late investor to consume. High-tech, informed middlemen entered these markets in the last two decades. Naturally, one expects better informed middlemen to hurt gains from trade (i.e., welfare), because they adversely select investor quotes. But, might they raise welfare as market makers who quickly refresh quotes on incoming information? And, as market makers, they offer investors the option to temporarily park their asset. We offer a model with all these features and calibrate it to study how middlemen affect welfare.

The slides are available here.

12/04/2024

About:

Investors face the risk of cryptocurrencies disappearing from the market and becoming zombies. Our study aims to predict which cryptocurrencies will become untradable using predictors based on descriptive statistics of yield, volume and market capitalization. The sample includes crypto assets that have been listed on the markets for at least 210 days in the period from January 2015 to December 2022. We apply various machine learning algorithms and novel XAI tools, namely permutation-based feature importance and PDPs, to identify the main factors explaining the disappearance of cryptos and to understand the shape of the relationships. Our study shows that machine learning models allow us to predict that cryptocurrencies will become zombies within the next 28 days with 84% out-of-time accuracy. The tree-based models, especially random forests, outperformed traditional econometric approaches. The variables with the greatest explanatory power are related to volumes and returns calculated in previous periods.

The slides are available on Quantinar.