CFA Institute

source

Gregory W. Brown, Wendy Y. Hu, David T. Robinson, and William M. Volckmann II A large-sample analysis shows SBIC funds outperform non-SBIC peers across IRR, MOIC, and PME. Performance is strongest for funds using moderate SBA leverage and larger fund sizes, with equity strategies showing greater variability than debt funds. source

At roughly 128% debt-to-GDP, the United States sits alongside France, Italy, and the United Kingdom — not in isolation. Japan stands out at over 230% debt-to-GDP, yet faces no immediate funding stress. Why? Because foreign dependence — not absolute debt — is the real constraint. China: roughly 102% debt-to-GDP, with about 3% foreign-held Japan: roughly 230% debt-to-GDP, with about 12% foreign-held United States: roughly 128% debt-to-GDP, with about 22% foreign-held The United States is unusual: it carries a large debt load, yet remains overwhelmingly domestically financed. That composition matters far more than the headline number. The foreign debt also reduced in percentage from 2019 to 2025, as seen in the following figure. source

We examine downside protection—or defensive—strategies over more than 220 years of global financial history, covering many years in which traditional equity–bond portfolios suffer and across a wide range of economic scenarios and historical regimes. Traditional defensive equity factors—low-risk, quality, and value—consistently provide effective downside protection, whereas gold and put options prove less drawdown or cost-effective. Our long-run evidence shows that multi-asset defensive strategies, particularly a return-enhanced version of the defensive absolute return (DAR) portfolio introduced by Cavaglia et al. (2022) and trend-following, provide the most effective downside protection. DAR and trend-following are complementary across tests by diversifying each other across stages of drawdowns. Investors can improve the defensive properties and improve total portfolio outcomes of traditional portfolios by considering the deep sample evidence on defensive strategies provided in this paper. source

What is machine learning in commodity futures?It is the application of supervised learning models to forecast returns in commodity futures markets. By grounding features in established theories such as the theory of storage and the hedging pressure hypothesis, ML identifies signals such as momentum, basis, carry, and skewness and translates them into long–short portfolio strategies. Why is ensemble modeling important in commodities?Ensemble modeling combines predictions from multiple horizons (short, medium, and long term) into a single signal. This approach reduces model risk, lowers volatility, and improves drawdown control compared with single-horizon models. Can machine learning generate alpha in commodity markets?Yes. When features are carefully designed and portfolios are constructed cross-sectionally, machine learning can uncover persistent patterns in commodity prices. These patterns align with macroeconomic cycles and provide systematic sources of alpha. Are the results interpretable for institutional investors?Yes. Because the features are drawn from established commodity economics, the models are not “black boxes.” They remain transparent, interpretable, and consistent with fiduciary and governance requirements. source

Allocations by institutional investors, represented by public pensions, have plateaued in recent years. This is unsurprising given the sheer volume of capital already committed, combined with the fact that private equity, the larger of the two allocations, has failed to deliver returns comparable to public markets for many years. The tapering of new institutional commitments, coupled with a clogged exit environment, created pressure across the private-markets ecosystem. Asset managers still had large portfolios to finance, consultants still had asset classes to recommend, and distributors still needed new products to sell. The solution was a structural innovation that allowed the industry to expand its investor base: semi-liquid vehicles designed specifically for individual investors and marketed as the “democratization” of private markets. These structures typically offer periodic liquidity, often through quarterly redemption windows, while investing in assets that may take years to sell at reliable prices. The appeal is obvious. Investors are offered exposure to private markets together with the appearance of stability and the reassurance that they can redeem capital periodically. The problem is that this model violates the previously explained principle of finance. Long-duration, difficult-to-price assets should never be financed with short-term liabilities unless a lender of last resort stands behind the structure. When that rule is ignored, the structure is unstable. As long as inflows continue and redemptions remain manageable, it seems advantageous to both investors and fund managers. But once investors begin to withdraw capital, the mismatch between liquidity promises and underlying assets becomes visible very quickly. History provides many examples of this dynamic. Wildcat banks in the 1800s, trust companies in the early 1900s, and investment bank warehousing facilities in the early 2000s. In each case, when confidence weakened, investors rationally attempted to redeem before others did. It doesn’t take long before investors run, simply in anticipation of other people running – which is the hallmark of a bank or fund run. This risk is substantially amplified when individual investors provide a large percentage of the capital. Taken together, semi-liquid private credit and private equity funds are unusually vulnerable to run mechanisms. Not only are Illiquid assets financed with redeemable capital, but the underlying investments were raised at the tail-end of two aged investment cycles. Financial history suggests that such combinations rarely remain stable for very long. They may function smoothly for several years. But when confidence weakens, the structural mismatch becomes impossible to ignore. That day arrived on February 18, when Blue Owl announced that it had permanently eliminated quarterly liquidity in its OBDC II private credit fund. source

To address these gaps, leading banks are adopting holistic AI risk and control approaches that treat AI as an enterprise-wide risk rather than a technical tool. Effective frameworks embed accountability, transparency, and resilience across the AI lifecycle and are typically built around five core pillars. 1. Board-Level Oversight of AI RiskAI oversight begins at the top. Boards and executive committees must have clear visibility into where AI is used in critical decisions, the associated financial, regulatory, and ethical risks, and the institution’s tolerance for model error or bias. Some banks have established AI or digital ethics committees to ensure alignment between strategic intent, risk appetite, and societal expectations. Board-level engagement ensures accountability, reduces ambiguity in decision rights, and signals to regulators that AI governance is treated as a core risk discipline. 2. Model Transparency and ValidationExplainability must be embedded in AI system design rather than retrofitted after deployment. Leading banks prefer interpretable models for high-impact decisions such as credit or lending limits and conduct independent validation, stress testing, and bias detection. They maintain “human-readable” model documentation to support audits, regulatory reviews, and internal oversight. Model validation teams now require cross-disciplinary expertise in data science, behavioral statistics, ethics, and finance to ensure decisions are accurate, fair, and defensible. For example, during the deployment of an AI-driven credit scoring system, a bank may establish a validation team comprising data scientists, risk managers, and legal advisors. The team continuously tests the model for bias against protected groups, validates output accuracy, and ensures that decision rules can be explained to regulators. 3. Data Governance as a Strategic ControlData is the lifeblood of AI, and robust oversight is essential. Banks must establish: Clear ownership of data sources, features, and transformations Continuous monitoring for data drift, bias, or quality degradation Strong privacy, consent, and cybersecurity safeguards Without disciplined data governance, even the most sophisticated AI models will eventually fail, undermining operational resilience and regulatory compliance. Consider the example of transaction monitoring AI for AML compliance. If input data contains errors, duplicates, or gaps, the system may fail to detect suspicious behavior. Conversely, overly sensitive data processing could generate a flood of false positives, overwhelming compliance teams and creating inefficiencies. 4. Human-in-the-Loop Decision Making Automation should not mean abdication of judgment. High-risk decisions—such as large credit approvals, fraud escalations, trading limits, or customer complaints—require human oversight, particularly for edge cases or anomalies. These instances help train employees to understand the strengths and limitations of AI systems and empower staff to override AI outputs with clear accountability. A recent survey of global banks found that firms with structured human-in-the-loop processes reduced model-related incidents by nearly 40% compared to fully automated systems. This hybrid model ensures efficiency without sacrificing control, transparency, or ethical decision-making. 5. Continuous Monitoring, Scenario Testing, and Stress SimulationsAI risk is dynamic, requiring proactive monitoring to identify emerging vulnerabilities before they escalate into crises. Leading banks use real-time dashboards to track AI performance and early-warning indicators, conduct scenario analyses for extreme but plausible events, including adversarial attacks or sudden market shocks, and continuously update controls, policies, and escalation protocols as models and data evolve. For instance, a bank running scenario tests may simulate a sudden drop in macroeconomic indicators, observing how its AI-driven credit portfolio responds. Any signs of systematic misclassification can be remediated before impacting customers or regulators. source