The global financial system is experiencing one of the most intense periods of simultaneous and cross-domain change in decades. After the COVID‑19 pandemic, geopolitical shocks, supply‑chain realignments, de‑globalization pressures, persistent fiscal deficits, and rapid technological disruption have increasingly overlapped.
The defining feature of today’s environment is not any single event, but the speed at which these forces interact – and the strain that puts places on traditional approaches to risk assessment and management.
Many of the drivers reshaping the economy did not originate in the financial system, but they are transmitted through it. Non‑financial shocks such as pandemics, geopolitical conflict, climate transition, and demographic shifts tend to build gradually before accelerating faster than expected. Most legacy risk frameworks were not designed for that pattern.
Inflation dynamics offer a clear example. In many capital‑intensive sectors, the transition toward new technologies effectively requires running parallel production systems – raising costs and embedding inflationary pressure even when short‑term data may appear benign.
Similar dynamics are present in re‑militarization, onshoring, friend‑shoring (re‑routing trade toward aligned partners), and the broader retreat from decades of globalization. These shifts are expensive, and their cost structures often prove sticky even when a few months of inflation data look better.
This divergence between calm surfaces and underlying stress is not an abstract concern – it is how risk now builds in real time. Market‑level indicators reward scale, liquidity and concentration, while issuer‑level fundamentals increasingly diverge beneath the surface.
When averages dominate the narrative, they suppress early warning signals. Not because risk is absent, but because it is unevenly distributed.
Stas Melnikov
A great illustration of that point can be seen in recent U.S. macroeconomic data. Consider March 2026 data from the Bureau of Labor Statistics. Payrolls rose 178K, but health care again did most of the lifting (+76K). BLS noted ambulatory health care employment jumped, including +35K in physician offices as workers returned from a strike.
The household survey was much softer: Employment fell (-64K), and the labor force shrank (-396K), pushing participation down to 61.9%, a multiyear low. Wage growth continued to cool; while average hourly earnings were up 3.5% YoY, hours worked edged down to 34.2 weekly.
The bottom line? Beyond the headline, the report showed a low-hire equilibrium and reinforces the credit signal from SAS’ April 1 credit conditions analysis.
The global economic system is more fragile, particularly in terms of fiscal capacity and shock absorption, than headlines and leading indicators suggest. Sovereign debt levels are historically high, and fiscal deficits persist. In many advanced economies, government balance sheets are stretched, private leverage remains elevated, and policymakers have limited room to maneuver. Yet credit markets continue to behave as if conditions are relatively calm.
This disconnect becomes most visible when looking beyond averages. Equity benchmarks such as the S&P 500 are capitalization‑weighted, allowing a small number of large, strong performers to mask growing weakness across the rest of the market. At the same time, the gap between winners and losers continues to grow.
AI is accelerating that dispersion across industries – not only in technology, but in consumer discretionary, media, materials, utilities and beyond. The divide is increasingly visible between companies that can adapt quickly and those that struggle to keep pace.
Meanwhile, such other concurrent forces as supply‑chain realignment, tariff regimes and shifting cost structures can compound pressure on firms that already lack resilience. Index‑level performance often obscures how wide and quickly these gaps can move.
From a credit standpoint, this matters. In our KRIS daily corporate credit models, if we narrow the universe to the top 3,000 publicly listed companies in the U.S., the median three‑year default probability is currently in the top quintile of its range going back to 1999. And yet credit spreads remain tight.
This gap raises uncomfortable questions: Are risk models missing something? Are markets underpricing risk? Is it some combination of both?
The growing role of passive investment vehicles in credit markets adds another layer of concern. When investors allocate capital based on rules and benchmarks rather than issuer-level investment research and analysis, price signals can weaken, particularly during periods of stress. That makes it easier for risk to build quietly, and it increases the value of disciplined analysis that looks beneath the index.
Again, the pace of change is crucial in today’s environment. Changes that once played out over decades are now unfolding over years, sometimes months. Recalling what happened in 2025 compared to 2023-2024, it feels like we’ve lived a decade, not a year. So far in 2026 there is little sign this acceleration is slowing.
Risk management works best when it rests on foundations that hold up under stress. Data quality and governance matter more than ever, as do disciplined processes and the ability to automate analysis without losing transparency or control. As systems become more complex, the fundamentals matter more than ever.
Longstanding principles of model risk management carry even more weight in the AI age. As models become more sophisticated, organizations struggle to ensure data quality, define appropriate use, maintain AI governance and monitor performance over time. Yet these steps are essential to deploying advanced models responsibly.
No model fully captures reality. Each simplifies an infinitely complex world through a series of assumptions. Some assumptions work in certain environments, while others break as conditions change. That makes explainability critical.
When model outputs change, decision-makers must understand why to separate signal from noise. Models are not crystal balls, and they require fit-for-purpose, decision-support ecosystems to appropriately interpret their output and extract useful signals.
Explainability becomes more challenging with advanced machine learning models, where even model developers may struggle to trace outcomes back to drivers. Without that clarity, organizations risk responding to noise instead of signals – and false confidence can be far more dangerous than acknowledging uncertainty.
Effective risk management is not about perfect forecasts (as such a goal is not achievable). It is about creating frameworks that allow institutions to test assumptions, explore scenarios, deploy models responsibly and adapt as conditions evolve (and not after being prompted by model failures).
Technologies such as AI-driven automation should support human judgment, not replace it. These tools are increasingly necessary given the scale and speed of modern data – but they must support risk management efforts with AI governance and model explainability at their foundation. Otherwise, implementation of sophisticated approaches becomes a weakness rather than an advantage.
Over time, data and AI capabilities have evolved from basic statistical tools to large‑scale analytics, machine learning and domain‑specific risk applications. That evolution reflects a simple reality: Risk does not stand still, and neither can the methods used to measure and manage it.
Today’s fast‑moving and fragile environment calls for clearer thinking and fewer shortcuts. Less reliance on averages. More attention to what is happening beneath the surface. Less confidence in static models, and more awareness of how quickly conditions can change.
Above all, it requires organizations to recognize that calm markets do not always signal low risk.
Stas Melnikov is Head of Quantitative Research and Risk Data Solutions at data and AI provider SAS.