VaR or Not VaR: Why Value-at-Risk Models Are Facing an Uncertain Future
"Explore the economic obstacles challenging the reliability of Value-at-Risk (VaR) models and what it means for investors navigating today's volatile markets."
In the late 1960s, the Value-at-Risk (VaR) measure emerged as a response to a pressing question posed by JP Morgan's Chairman, Dennis Weatherstone: "How much can we lose on our trading portfolio by tomorrow's close?" This question sparked the development of VaR models by RiskMetrics Group, quickly becoming a standard tool for risk assessment in the financial industry.
VaR aims to measure the maximum potential change in the value of a portfolio over a specific time horizon, given a certain probability level. It provides a seemingly straightforward way to quantify risk, making it popular among investors, regulators, and financial institutions. However, the reliability and accuracy of VaR models are increasingly being questioned due to economic obstacles that limit their effectiveness.
One critical issue lies in the traditional approach to calculating VaR, which often relies on frequency-based price probabilities. This method determines the probability of price movements based on the number of trades at a specific price point over a given period. However, this approach overlooks the impact of large trade volumes and market dynamics, potentially leading to inaccurate risk assessments.
VaR: A Ubiquitous but Contested Risk Metric
Value at Risk has become one of the most widely adopted risk measurement tools in finance, used by banks, asset managers, and regulators to estimate potential portfolio losses over a given time horizon and confidence level. Its precise current market penetration is difficult to pin down, as adoption varies across jurisdictions and institution types. What is clear is that VaR remains a central pillar of both internal risk management and regulatory frameworks such as the Basel accords, even as debates about its adequacy persist.
Three Methods, One Persistent Shortcoming
VaR is typically calculated using one of three approaches: historical simulation, the variance-covariance (parametric) method, or Monte Carlo simulation, each with distinct assumptions about market behavior. While VaR provides a useful single-number summary of potential loss at a given confidence level, sources agree it can create a false sense of security because it does not measure worst-case loss beyond the confidence threshold. It is also not directly additive across portfolio segments, making firm-wide aggregation cumbersome. For these reasons, practitioners increasingly combine VaR with stress testing and complementary risk measures to obtain a more comprehensive view of exposure.
From Obscurity to Industry Standard
Although the conceptual origins of VaR-style risk measures stretch back further, the term 'value-at-risk' did not enter the financial lexicon until the early 1990s. Multiple sources trace VaR's rapid rise to prominence in the late 1980s and 1990s, driven by growing computing power and regulatory demand for standardized risk metrics. By the time of the Basel II framework, VaR had become the global standard for market risk measurement in banking. The tool's evolution reflects broader shifts in financial risk management, from intuitive judgment toward quantitative, model-driven approaches.
The Flaw in Frequency-Based Price Probability: Why Volume Matters
The conventional frequency-based approach to price probability assumes constant trade volumes, which is rarely the case in real-world markets. Large market transactions can significantly influence price movements, making the randomness of trade volumes a crucial factor in risk assessment. Ignoring this randomness can lead to an underestimation of potential losses and a false sense of security.
- Market-Based Probabilities: Reflect the impact of trade values and volumes on price movements.
- Frequency-Based Probabilities: Assume constant trade volumes and may not accurately capture market dynamics.
Machine Learning Enters the VaR Arena
Recent academic work has begun testing whether neural network architectures — including convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and feed-forward networks — can reduce VaR breaches at the 5% and 1% thresholds in equity and bond portfolios. A systematic literature review reported by Risk.net has also examined which VaR model types remain most popular and whether the 2007–09 financial crisis prompted any shift in model preference among practitioners. Meanwhile, newer research integrates textual and unstructured data into VaR forecasting, reflecting a broader trend toward hybrid quantitative-qualitative risk models. These developments suggest that the VaR framework is not static but is being actively reworked through computational innovation.
Known Weaknesses, Ongoing Debate
Critics of VaR have long argued that its reliance on historical correlations and normal-distribution assumptions makes it poorly suited to capturing tail risk and regime changes. The 2008 global financial crisis amplified these concerns, as many institutions discovered that their VaR models had significantly underestimated the potential for extreme, correlated losses across asset classes. The model's inability to quantify what happens beyond the confidence threshold — the so-called 'beyond VaR' problem — means it can understate systemic interconnectedness. While no single alternative has fully replaced VaR, these limitations continue to fuel calls for more robust risk frameworks.
VaR, CVaR, and the Search for Better Measures
Before 2008, minimum capital requirements for market risk were calculated primarily using VaR-based models, as noted by FORRS. The ECB has documented how VaR enabled institutions to obtain a firm-wide view of risk and allocate capital more efficiently across business lines. However, risk practitioners operating in liquid markets often use relatively short time horizons in VaR calculations, which can miss longer-duration risks. This has driven interest in complementary and alternative measures such as Conditional Value at Risk (CVaR), which attempts to capture the severity of losses beyond the VaR threshold rather than just the probability of exceeding it.
The Future of VaR: Adapting to Economic Complexity
As markets continue to evolve and become more complex, the limitations of traditional VaR models will become increasingly apparent. The accuracy of price probability predictions is likely to be constrained by the availability and reliability of data on market trade values and volumes. Furthermore, predicting market-based price volatility requires forecasting averages, volatilities, and correlations of trade values and volumes, which is a challenging task given the inherent uncertainty of economic activity.
Blending Quantitative Models with Expert Judgment
Research published in the International Journal of Business Continuity and Risk Management proposes incorporating expert opinions into traditional VaR calculations through techniques such as Bayesian aggregation, fuzzy logic, and Delphi dynamics. The argument is that purely quantitative models overlook qualitative knowledge — subjective assessments and mathematically imprecise insights — that experienced practitioners possess. Studies comparing Bayesian and axiomatic approaches to aggregating expert judgments suggest that structured elicitation can improve the accuracy and robustness of risk estimates. This hybrid approach reflects a growing recognition that no single model, however sophisticated, captures the full picture of financial risk.
AI, Disclosure, and Evolving Regulations
McKinsey's analysis of global risk trends highlights how artificial intelligence and advanced analytics are reshaping operating models and best practices for financial risk management. At the same time, increased availability of risk disclosure data — in areas previously considered secure — has expanded risk management responsibilities, including for liquidity risks. Growing concerns about privacy and systematic surveillance among employees, consumers, and business associates add new dimensions to the risk landscape that traditional VaR models were not designed to address. The convergence of these forces suggests that future risk frameworks will need to be both more data-rich and more attuned to non-financial risks.
Systemic Risk and the Limits of Individual-Portfolio Models
Since their emergence in the late 1980s, VaR models quickly became the global standard in risk management and forecasting. However, their role during international financial crises has raised serious questions about whether such tools can effectively manage risk during extreme, systemic events — with some researchers even attributing aspects of crisis contagion to the models themselves. Sources note that during the 2008 crisis, many institutions found their VaR models underestimated the potential for extreme losses across interconnected markets. Looking ahead, analysts suggest that future risk frameworks may need to integrate ESG metrics and consider the broader financial ecosystem rather than focusing solely on individual-portfolio risk.
People, Judgment, and Organizational Reality
Behind every risk model sits a chain of human decisions — about data selection, model calibration, confidence thresholds, and how results are interpreted and acted upon. VaR's elegant single-number output can obscure the extent to which these judgments shape the final figure, potentially fostering overconfidence among non-specialist users. In practice, the effectiveness of any risk framework depends not only on its mathematical rigor but on the organizational culture that surrounds it: whether dissenting views are heard, whether model limitations are communicated honestly, and whether stress scenarios are taken seriously. These human factors, while harder to quantify than market volatility, may ultimately matter more for preventing the next crisis.