Standard risk models assume returns are normally distributed. The empirical record has not supported that assumption for decades, and the gap matters more in a sustained high-geopolitical-risk environment than it did in the quieter 2010s. A look at fat tails, tail risk, and the scenario-based analysis the literature describes as the response.
In April 2026, the World Bank reported that global energy prices were on track to rise approximately 24 percent during the year, the largest annual increase since the 2022 invasion of Ukraine, as the conflict in the Middle East transmitted through commodity markets.¹ Brent crude futures moved sharply through the first half of the year, with intraday volatility well outside what historical-data risk models, built on the assumption of normally distributed returns, treat as plausible. The mismatch is not new. The literature on financial returns has documented it for decades.² What is new is the frequency with which the mismatch is being observed in real time. The property behind it is what the literature calls fat tails, and the exposure it creates is tail risk.
What is tail risk?
In financial statistics, the term “tail risk” refers to the probability of returns that fall in the extreme regions of a distribution. The Wikipedia summary, drawing on the academic literature, defines it as the risk of an asset or portfolio moving more than three standard deviations from its mean.³ The Basel Committee on Banking Supervision identifies the supervisory concern with tail events directly: institutions are expected to consider not only historical events but also “hypothetical future events that take into account new information and emerging risks,” particularly where “new or heightened vulnerabilities are identified, or if historical data do not contain a severe crisis episode.”⁴
Two distinct but related ideas often appear in this discussion. The first is statistical: that the actual distribution of financial returns has more probability mass in its tails than a normal distribution predicts. The second is institutional: that risk frameworks calibrated on historical data tend to underestimate the size of losses that can occur in a single period. Both points have been documented across the academic and practitioner literature, and both have direct implications for how portfolios are analysed under stress.
The normal distribution problem
A normal distribution is characterised by its mean and standard deviation. Under that assumption, the probability of a return falling more than three standard deviations from the mean is approximately 0.27 percent, and the probability of a five standard deviation event is approximately one in 3.5 million periods. Many standard risk metrics, including the most common formulations of Value at Risk, rest on this distributional assumption either directly or through historical samples that implicitly inherit it.
The empirical record does not support the assumption. Benoit Mandelbrot, writing in 1963, documented that the distribution of speculative price changes is characterised by what he termed “pathological” properties relative to the normal distribution: infinite variance, scaling, and significantly more frequent extreme observations.⁵ Rama Cont, in a widely cited 2001 survey of empirical properties of asset returns, summarised what subsequent decades of analysis had confirmed: the unconditional distribution of returns displays “heavy tails,” with kurtosis values significantly above those of a normal distribution, across virtually every financial market studied.⁶ The Bank for International Settlements, in its 2018 stress testing principles, explicitly cautioned that frameworks relying solely on historical statistical relationships rest on the assumption that historical relationships are a good basis for forecasting future risks, and that the 2008 crisis is widely regarded as having revealed “serious flaws with relying solely on such an approach.”⁴
Value at risk under the assumption
The methodological implication is structural. A risk model that assumes normality, or that calibrates risk parameters on samples drawn from a recent low-volatility regime, will produce loss estimates that are systematically too low for the extreme region of the distribution. The 99 percent VaR of a normally distributed return at one standard deviation is approximately 2.33 standard deviations. The corresponding VaR for a fat-tailed distribution, depending on parameterisation, can be materially larger. The difference between the two estimates is the practical consequence of the distributional mismatch.
Why geopolitical events are the canonical fat-tail event class
Multiple strands of recent academic research have examined how geopolitical events specifically interact with tail risk in financial markets. The literature documents three structural reasons why geopolitical events tend to produce the kinds of return distributions that standard models struggle to capture.
Compounding transmission across markets
Geopolitical events tend to transmit through multiple markets simultaneously, with the transmission paths reinforcing one another. A 2024 study published in Applied Economics examined the tail-risk connectedness across G7 stock markets during the Russia-Ukraine conflict and documented that the conflict “significantly increased tail risk connectedness among G7 stock markets, with the highest estimated levels observed two- and three-months thereafter,” attributing the increase to “heightened geopolitical and economic uncertainty, increased interconnectivity due to elevated risk and concomitant safe-haven behaviour, financial contagion, disrupted supply chains, and shifts in investor sentiment.”⁷ The methodological observation is direct: when several markets simultaneously experience tail behaviour driven by a common cause, the assumption that asset-class correlations remain stable under stress is no longer reliable.
Path dependency and persistence
Geopolitical events produce path-dependent outcomes that historical data captures imperfectly. A six-month closure of a strategic chokepoint differs from a three-week closure in ways that cannot be linearly extrapolated. A trade dispute that escalates to comprehensive tariffs produces different dynamics from one that is resolved through negotiation. The Caldara and Iacoviello Geopolitical Risk Index, which has been tracked since 1985 and updated monthly, shows that the post-2022 period has sustained levels above the long-run average for longer than any equivalent stretch since the early 2000s elevated period. The methodological challenge that path dependency creates is that historical samples include outcomes whose realised path is known. Forward-looking risk analysis has to engage with paths that have not yet been observed.
Reflexivity in market structure
Geopolitical events also tend to change the structure of the markets they affect, rather than producing transitory shocks that revert. The Russia-Ukraine conflict’s effect on European energy markets has not unwound. Tariff structures introduced during 2018 to 2019 reshaped supply chains in ways that subsequent policy changes have not fully reversed. The methodological implication is that the post-event distribution of returns may differ from the pre-event distribution in ways that historical-data models, which by construction sample from past distributions, are slow to incorporate.
How the institutional literature describes the response
Two methodological responses to tail risk are widely documented in the institutional finance literature. They are not mutually exclusive, but they have different analytical structures and address different aspects of the problem.
Tail-risk hedging
The first response, widely described in the literature, is direct hedging of tail events through derivative structures: dedicated tail-risk funds, put option overlays, and dynamic volatility-triggered overlays. AQR Capital Management, in its widely cited 2012 paper “Chasing Your Own Tail (Risk),” documented the structural challenge with this approach: hedging strategies generally have negative expected returns because the insurance buyer faces a long-run cost above the long-run payout.⁹ PIMCO has separately published institutional educational material describing how tail-risk hedging is constructed and what trade-offs it presents.¹⁰ The literature consistently describes hedging as one component of an institutional response rather than a standalone solution.
Scenario analysis and scenario-based stress testing
The second response, also widely documented, is forward-looking scenario analysis that examines what specific hypothetical events would do to a portfolio at the position level, rather than relying on historical-data risk parameters to forecast tail behaviour. This is the methodological framework that the Basel Committee, the European Central Bank, and the EY/IIF 2026 Global Bank Risk Management Survey all describe as the institutional direction of travel for risk management in the current environment.⁴ Scenario-based stress testing addresses a different question from tail-risk hedging. It does not protect the portfolio against tail events; it identifies which positions in the portfolio would be most exposed, by what mechanism, under what specific configurations of stress. The output is analytical rather than protective.
The two approaches are typically combined in institutional risk frameworks. Hedging addresses the question of what happens when a tail event materialises. Scenario analysis addresses the question of which scenarios would produce material impairment, and by what transmission mechanism. The current literature, including the Basel Committee’s stress testing principles and recent EY/IIF CRO survey work, suggests that the analytical capacity for forward-looking scenario analysis is the area where institutions are most actively investing in 2026.
What this means for wealth managers, family offices, and asset managers
The methodological points above have direct implications for institutions outside the directly supervised banking sector.
First, the bar for what counts as a credible risk assessment is rising. Sophisticated private clients, family offices, and institutional allocators are increasingly asking scenario-specific questions about specific portfolios under specific configurations of stress. The literature suggests that institutions whose risk frameworks can answer those questions, with documented analytical chains from scenario to factor to position to impact, are differently positioned in client conversations than those that can produce only aggregate VaR estimates.
Second, the limitations of historical-data approaches are particularly material in environments that have no recent historical analogue. The Caldara and Iacoviello geopolitical risk data shows that the post-2022 period is the longest sustained stretch of above-average geopolitical risk in the index’s 40-year history.⁸ The methodological implication is that risk frameworks calibrated on samples from the 2010s, when geopolitical risk readings were materially lower, may underestimate the conditional risk profile of the current environment.
Third, scenario-based analysis enables a different kind of conversation. A client who asks what a sustained closure of a strategic shipping chokepoint would mean for their fixed-income allocation is asking a question that VaR cannot answer. A framework that can decompose that scenario to factor activations, propagate those through the portfolio at the instrument level, and produce a defensible analytical chain is engaging a different conversation than one that can only report historical volatility.
The methodological point
The argument is not that historical-data approaches should be discarded. The empirical record they describe is real and the regulatory frameworks that incorporate them are well-founded. The argument is that frameworks built solely or primarily on historical statistical relationships, calibrated on samples drawn from past distributions, are documented in the literature to underestimate the probability and severity of the extreme region of the return distribution. The current environment, characterised by sustained elevated geopolitical risk by historical standards, is one in which that limitation is more material than in lower-volatility periods. Forward-looking, hypothetical scenario analysis is the methodological response that the Basel Committee, the European Central Bank, and the consultancy literature consistently describe as complementing historical-data approaches. The analytical capability to do that work, propagating configured scenarios through real portfolios at the instrument level, is what the literature describes.
Read the original article here.
