When used for parameter optimization and/or model selection, traditional mean squared error (MSE)-based measures of forecast accuracy often exhibit a weak or even negative correlation with the economic value of return forecasts measured by, for example, the Sharpe ratios of the resulting portfolios. Deriving a multivariate risk-adjusted error measure, we show that the RMSE is a special case of this measure under quite restrictive simplifying assumptions. We analyze the contribution of each of these simplifications to the reduction in the explanatory power of the forecast accuracy measure for the shortfall in the attainable Sharpe ratio across a range of well-known portfolio strategies. We do so both in a Monte Carlo simulation under the assumption of normal i.i.d. returns and in an empirical application.
Article 70(1) MiCAR (Markets in Crypto-Assets Regulation) regulates the safekeeping of crypto assets of the client. The regulation is based on existing provisions of financial market law. However, while Markets in Financial Instruments Directive II (MiFID II) expressly allows securities to be used for the service provider's own account with the client's consent, the wording of MiCAR is not clear in the case of crypto assets. This article analyses the permissibility of using crypto assets of the customer for their own account under the MiCAR.
Immersive technologies, including virtual reality (VR), augmented reality (AR), and other forms of extended realities (XR) are increasingly adopted to study and improve intergroup relations. However, evidence of their effectiveness is mixed, highlighting the need for stronger theoretical integration and greater methodological rigor. This critical review seeks to address the key theoretical questions and methodological challenges facing the field.Theoretically, research must first clarify the psychological mechanisms through which immersive interventions influence intergroup outcomes. A second question concerns the durability and generalizability of these effects. While some interventions yield lasting improvements and transfer to other contexts, others show rapid decay. A third issue is the potential for backfire, as immersive experiences may at times reinforce rather than reduce bias. Understanding when, why, and for whom immersive interventions succeed or fail is therefore crucial.Methodologically, current evidence is limited by overreliance on self-reports and the lack of methodological frameworks. Comparative studies with traditional interventions show promise but yield inconsistent results, highlighting the importance of assessing the added value of immersive formats. Progress also depends on greater transparency and replicability, as their absence hinders cumulative knowledge building. Moreover, researchers should exploit the full potential of immersive environments for multimodal assessment.By critically addressing current limitations and boundary conditions, as well as emerging trends and underexplored strategies to improve intergroup relations using immersive technologies, we set the stage for future work that is theoretically integrated, methodologically rigorous, and capable of delivering scalable, ethical, and impactful interventions.
Large Language Models (LLMs) are increasingly adopted in the financial domain. Their exceptional capabilities to analyse textual data make them well-suited for inferring the sentiment of finance-related news. Such feedback can be leveraged by algorithmic trading systems (ATS) to guide buy/sell decisions. However, this practice bears the risk that a threat actor may craft "adversarial news" intended to mislead an LLM. In particular, the news headline may include "malicious" content that remains invisible to human readers but which is still ingested by the LLM. Although prior work has studied textual adversarial examples, their system-wide impact on LLM-supported ATS has not yet been quantified in terms of monetary risk. To address this threat, we consider an adversary with no direct access to an ATS but able to alter stock-related news headlines on a single day. We evaluate two human-imperceptible manipulations in a financial context: Unicode homoglyph substitutions that misroute models during stock-name recognition, and hidden-text clauses that alter the sentiment of the news headline. We implement a realistic ATS in Backtrader that fuses an LSTM-based price forecast with LLM-derived sentiment (FinBERT, FinGPT, FinLLaMA, and six general-purpose LLMs), and quantify monetary impact using portfolio metrics. Experiments on real-world data show that manipulating a one-day attack over 14 months can reliably mislead LLMs and reduce annual returns by up to 17.7 percentage points. To assess real-world feasibility, we analyze popular scraping libraries and trading platforms and survey 27 FinTech practitioners, confirming our hypotheses. We notified trading platform owners of this security issue.
Large Language Models (LLMs) are transforming human decision-making by acting as cognitive collaborators. Yet, this promise comes with a paradox: while LLMs can improve accuracy, they may also erode independent reasoning, promote over-reliance and homogenize decisions. In this paper, we investigate how LLMs shape human judgment in security-critical contexts. Through two exploratory focus groups (unaided and LLM-supported), we assess decision accuracy, behavioral resilience and reliance dynamics. Our findings reveal that while LLMs enhance accuracy and consistency in routine decisions, they can inadvertently reduce cognitive diversity and improve automation bias, which is especially the case among users with lower resilience. In contrast, high-resilience individuals leverage LLMs more effectively, suggesting that cognitive traits mediate AI benefit.
