Insights & Announcements
A weekly Newsletter on technology applications in investment management with an AI / LLM and automation angle. Curated news, announcements, and posts, primarily directly from sources. We apply some of the ROI's Kubro(TM) Engine tools at the backend for production, yet with a human in the loop (for now). It's a start, and it will evolve on a weekly basis.
Disclaimers: content not-fully-human-verified, with AI summaries below. AI/LLMs may hallucinate and provide inaccurate summaries. Select items only, not intended as a comprehensive view. For information purposes only. Please DM with feedback and requests.
Picking a select few (no review) from the most recent submissions, eyeing potentially interesting bits on LLMs and AI in the finance/investment world.
Submitted on 27 Oct 2025, Claire Barraud (UGA UFR FEG) presents arXiv:2510.23175, proposing a complementary behavioural finance framework that interprets financial markets in boom-and-bust episodes as a Le Bonian crowd. Drawing on classical crowd psychology (Le Bon, 1895; Tarde, 1901; Freud, 1921), the paper adopts a macro-psychological perspective, arguing that booms and crashes reflect collective mental states governed by unconscious processes, suggestion, emotional contagion, and impulsive action, clarifying market-wide irrationality and providing a macrobehavioural basis for financial instability. 🔗 Source: arxiv.org | Found on Oct 29, 2025
Submitted on 24 Oct 2025, Dong Yan, Ke Zhou, Zirun Wang, and Xin-Jiang He study portfolio selection with transaction costs under a two-factor stochastic volatility model, where volatility is mean-reverting with a stochastic mean-reversion level. The framework includes proportional exogenous transaction costs and endogenous costs driven by a stochastic liquidity risk process. An option-implied S-shaped utility is extracted and its concave envelope used to address non-concavity, yielding a five-dimensional nonlinear Hamilton-Jacobi-Bellman equation. A deep learning-based policy iteration computes the value function and optimal policy, with numerical experiments assessing cost and volatility impacts on investment decisions. 🔗 Source: arxiv.org | Found on Oct 27, 2025
Submitted on 29 Oct 2025, Mohamad Abou Ali and Fadi Dornaika present a PRISMA-based survey of 90 studies (2018–2025) that introduces a dual-paradigm framework for agentic AI: Symbolic/Classical (algorithmic planning, persistent state) and Neural/Generative (stochastic generation, prompt-driven orchestration). Across theory, applications in healthcare, finance, and robotics, and ethics/governance, they find symbolic systems dominate safety-critical domains, while neural systems prevail in adaptive, data-rich contexts. They identify deficits in governance models for symbolic systems and a need for hybrid neuro-symbolic architectures, advocating intentional integration to build adaptable, reliable agentic AI. 🔗 Source: arxiv.org | Found on Oct 30, 2025
Submitted on 28 Oct 2025, The Economics of AI Training Data by Hamidah Oderinwale and Anna Kazlauskas (18 pages) establishes data economics through three contributions. It identifies data’s properties—nonrivalry, context dependence, and contamination-induced rivalry—traces precedents from oil and grain, and documents 2020–2025 training data deals showing persistent fragmentation, five pricing mechanisms (per-unit licensing to commissioning), and that most exclude original creators from compensation, reaching hundreds of millions of dollars. It proposes a hierarchy (token, record, dataset, corpus, stream), urges production-function representation, and outlines problems on context-dependent value measurement, governance-privacy tradeoffs, production contribution, and mechanisms for heterogeneous compositional goods. 🔗 Source: arxiv.org | Found on Oct 30, 2025
Submitted on 28 Oct 2025, this arXiv cs.AI/cs.SI paper (arXiv:2510.24337) presents a practical guide for gLLM-assisted quantitative content analysis in communication research. It reports that gLLMs can outperform crowd workers and trained coders on coding tasks at a fraction of the time and cost, decode implicit meanings, be instructed via natural language, require minimal annotated data, and be deployed with basic programming skills. The guide addresses seven challenges: codebook development, prompt engineering, model selection, parameter tuning, iterative refinement, reliability validation, and optional performance enhancement, emphasizing validity, reliability, reproducibility, and ethics. 🔗 Source: arxiv.org | Found on Oct 29, 2025
Submitted on 29 Oct 2025, this paper presents an agentic, modular RAG architecture for fintech, addressing domain-specific ontologies, dense terminology, and acronyms. The pipeline enables intelligent query reformulation, iterative sub‑query decomposition via keyphrase extraction, contextual acronym resolution, and cross‑encoder context re‑ranking. Evaluated against a standard RAG baseline on 85 question–answer–reference triples from an enterprise fintech knowledge base, the agentic system achieved higher retrieval precision and relevance, with increased latency. 🔗 Source: arxiv.org | Found on Oct 30, 2025
Actions, news, and announcements by investment firms, in the context of technology applications.
Published October 27, 2025, at 16:24 by ovaughn, the article recounts Citadel Chief Technology Officer Umesh Subramanian’s discussion with Bloomberg’s Dani Berger at the Bloomberg Enterprise Data & Tech Summit. Subramanian said, “AI helps us move faster and see more clearly—but it’s our people who turn insights into alpha,” highlighting disciplined technology use and people as drivers of competitive edge. He added, “AI is a force multiplier, not a leveler,” asserting that those who harness data and transform insights into action will be most successful. 🔗 Source: citadel.com | Found on Oct 27, 2025
US equities rose 87% in three years as the equity risk premium nears zero; reverting to median implies a 35% fall, while dot‑com‑style euphoria could add 51%. Risks include AI hype (BLS productivity 2.2% vs 1.8% pre‑COVID; OpenAI CAPEX commitments US$1.5 trillion) and stretched valuations (Palantir, Cloudflare, Snowflake from 10x to 26x sales). Rates and fiscal strain could lift the UST10 to 5.5% (+100bp term premium, +50bp inflation expectations), with France’s non‑financial debt at 322%. Private markets are surging (Bain CAGRs up to 52%). Policy volatility raised tariffs from 2% to 19%. 🔗 Source: man.com | Found on Oct 29, 2025
Focused on specific company types in financial information, alternative data, software solutions, and AI / LLM technologies.
Anthropic is expanding Claude for Financial Services with an Excel add-in, new real-time connectors, and Agent Skills, building on Sonnet 4.5’s 55.3% accuracy on Vals AI’s Finance Agent benchmark. Claude for Excel is a beta research preview for Max, Enterprise, and Teams, gathering feedback from 1,000 initial users in an Excel sidebar that reads, modifies, and creates workbooks with transparent tracking. New connectors include Aiera, LSEG, Moody’s (data on 600 million companies), and MT Newswires. Six new Skills cover comps, due diligence packs, teasers/profiles, and earnings analyses; Citi reports 75% of engineers saving 8–10+ hours weekly with “goose.” 🔗 Source: anthropic.com | Found on Oct 27, 2025
LSEG announced a collaboration with Anthropic to grant Claude customers access to licensed data from products like Workspace and Financial Analytics via the expanded Claude for Financial Services, launched today. AI-ready content will roll out through MCP starting with LSEG Financial Analytics, with initial capabilities later this month and more categories to follow. LSEG’s MCP server is live in the Claude MCP Partner Directory from Monday 27 October. The partnership embeds mutual lead generation and offers Workspace licences; LSEG operates in 65 countries and employs over 26,000 people, over half in Asia Pacific. 🔗 Source: lseg.com | Found on Oct 30, 2025
Microsoft and OpenAI signed an agreement supporting formation of OpenAI Group PBC and recapitalization. Microsoft holds approximately $135 billion investment, 27% on an as-converted diluted basis, inclusive of owners; previously held 32.5% in the for‑profit. Azure API exclusivity continues until AGI, with AGI declarations verified by an independent panel. Microsoft’s model/product IP extends through 2032; research IP until AGI verification or 2030. OpenAI will purchase an incremental $250B of Azure, can jointly develop products with third parties (API on Azure; non‑API on any cloud), provide API access to US government national security customers on any cloud, and release open‑weight models. 🔗 Source: blogs.microsoft.com | Found on Oct 28, 2025
LSEG announced an expanded partnership with BlackRock, integrating Preqin private markets data into LSEG’s Workspace and Data & Feeds to give customers deeper insights into private markets and alternative assets. The addition enhances LSEG’s data coverage across public and private markets and brings Preqin’s data to new investors. LSEG also renewed its multi-year partnership with BlackRock’s Aladdin platform, providing access to LSEG’s Pricing and Reference Services data, and BlackRock extended its partnership with FTSE Russell to continue licensing benchmarks for creating investment vehicles. 🔗 Source: lseg.com | Found on Oct 30, 2025