Infundly Weekly Briefing - 11 September 2026
This week’s Infundly Research Reads focuses on how investment outcomes can be shaped by the processes and assumptions sitting behind the headline proposition. The selected research covers AI bias in investment workflows, benchmark-free portfolio construction, multi-asset allocation, structured-credit opportunities linked to AI infrastructure, and the limits of universal manager-selection signals, all with direct relevance for professional fund selectors and allocators.
CFA Institute — Managing LLM Bias in Investing: From Detection to Mitigation
What changed: Published on 10 September, this is unusually practical research on AI governance in investment workflows. CFA Institute tested nine OpenAI models across 10 investment scenarios and found that changing only the positive or negative framing of identical information could alter the models’ assessments. Simply instructing a model to avoid bias did little; balanced prompts, comparative testing and human review were more effective.
Worth reading because: This has direct implications for any allocator or research team using LLMs for fund analysis. The due-diligence question should move beyond “which model do you use?” towards how prompts, source selection, analytical steps and human review are governed and tested. A polished AI-generated conclusion is not evidence that the underlying process is unbiased.
GMO — 25 Years of Benchmark-Free Investing
What changed: Published on 9 September, Ben Inker reflects on 25 years of benchmark-agnostic asset allocation at GMO and the lessons from allowing valuation and expected returns, rather than benchmark weights, to determine portfolio positioning. GMO argues that today’s unusually concentrated and valuation-dispersed markets may again reward investors prepared to look materially different from conventional portfolios.
Worth reading because: This is useful for assessing genuinely flexible multi-asset managers. Benchmark freedom can create opportunity, but it also increases the importance of valuation discipline, governance, risk limits and patience through periods of uncomfortable relative performance. Selectors should distinguish genuine unconstrained investing from benchmark awareness dressed up as flexibility.
Goldman Sachs Asset Management — Market Pulse: September 2026
What changed: Goldman remains constructive after strong August markets but expects a less straightforward environment around the September Fed meeting. Its base case combines resilient global growth with gradually easing US inflation, while expecting Europe and Japan to run above potential and monetary-policy paths to diverge. Its portfolio response emphasises diversification across fixed income, alternatives and higher-beta equity exposures rather than a single macro bet.
Worth reading because: The useful element is the portfolio-construction framework rather than the forecast itself. It gives selectors a good benchmark against which to test multi-asset managers: are returns genuinely coming from differentiated allocation decisions, or mainly from equity beta while diversification is discussed but contributes little in practice?
Man Group — The Mispriced Debt Powering the AI Boom
What changed: Man highlights the rapid emergence of data-centre asset-backed securities as hyperscalers finance the AI infrastructure build-out. It argues that similarly rated data-centre ABS can offer materially wider spreads than unsecured corporate debt despite having collateral and structural protections, partly reflecting liquidity and complexity premia rather than simply higher expected credit losses.
Worth reading because: It provides a useful example of where thematic exposure and credit underwriting intersect. The attractive spread needs to be tested against tenant concentration, refinancing risk, technological obsolescence, collateral assumptions and liquidity. For selectors reviewing structured-credit managers, the important question is whether apparent relative value reflects genuine underwriting insight or compensation for risks that conventional ratings do not capture well.
Journal of Econometrics — Heterogeneous Predictability on Mutual Fund Alphas
What changed: The September issue tackles an important weakness in fund-selection models: predictors of future alpha do not work equally well across all managers or through all market environments. The paper uses a clustering approach to identify groups of funds for which different characteristics and market variables appear informative, rather than assuming a universal relationship between fund characteristics and subsequent alpha.
Worth reading because: This is relevant to evidence-based manager research. It cautions against building selection frameworks around universal rules such as “high active share is good” or “smaller funds outperform”. The significance of any characteristic may depend on the type of strategy, market environment and combination of other attributes present.
Infundly takeaway
The strongest theme this week is the importance of understanding how investment decisions are produced, not just what the portfolio looks like at the end. CFA Institute’s work on LLM bias shows how framing can influence AI-generated analysis, while GMO and Goldman Sachs highlight how portfolio outcomes depend on valuation discipline, governance and the way diversification is implemented rather than simply described.
The research on data-centre ABS and heterogeneous alpha prediction reinforces the same point from different angles. Attractive opportunities can arise from complexity and structural mispricing, but they also require selectors to understand the assumptions behind underwriting and risk models. More broadly, manager-selection signals should not be treated as universally predictive. The practical implication is to test the decision architecture, evidence base and implementation discipline behind a strategy rather than relying too heavily on headline labels, historical outcomes or single metrics.
Important information
This material is intended for professional advisers, regulated firms, discretionary managers, institutional investors and other professional investment decision-makers. It is not intended for retail clients and should not be relied upon by retail investors.
This material is provided for informational and professional research purposes only. It reflects general market observations and Infundly’s analysis at the time of writing, unless otherwise stated. It does not constitute investment advice, a personal recommendation, investment management, arranging activity, or an invitation or inducement to engage in investment activity.
References to individual funds explain their characteristics and the questions they raise for professional due diligence. Inclusion within the Fund Discovery process does not constitute a recommendation, product endorsement or conclusion that a fund is appropriate for any portfolio. Regulated firms and professional users remain responsible for their own research, due diligence, product approval, suitability assessments and client outcomes.
The views, figures and data included are based on information available at the time of writing and sources believed to be reliable, but their accuracy, completeness and timeliness are not guaranteed. Markets, funds, personnel and circumstances can change without notice. The value of investments may fall as well as rise and capital is at risk. Past performance is not a reliable indicator of future results.
Third-party links and external material are provided for professional reference and convenience only. Their inclusion does not constitute endorsement by Infundly. This material must not be copied, redistributed or made available to retail clients without Infundly’s prior permission and, where appropriate, review and approval by the regulated firm responsible for the communication.
