Choosing a Confidence Level for Value at Risk

The confidence level of a Value at Risk (VaR) calculation determines how extreme a loss scenario the measure is designed to capture. A 95% value at risk (VaR) describes the maximum expected loss over a holding period with 95% confidence; a 99% VaR is more conservative, capturing a more severe tail loss. The choice between 95% and 99% is not academic—it directly drives capital requirements, buffer sizes, and whether the firm is prepared for truly exceptional market events.

Interpreting the Confidence Level

A 95% VaR of $1 million over a one-day holding period means: “There is a 95% probability that losses will not exceed $1 million tomorrow.” Equivalently, there is a 5% probability that losses will exceed $1 million—a “1 in 20” event.

A 99% VaR of $1.4 million over the same period means: “There is a 99% probability that losses will not exceed $1.4 million tomorrow.” There is a 1% probability that losses exceed $1.4 million—a “1 in 100” event.

The difference is in the tail of the loss distribution. Both measures are estimates of potential loss, but 99% VaR reaches further into the left tail (the extreme loss scenarios) than 95% VaR does. When you increase the confidence level, you are asking: what is the loss threshold that captures an even more rare, more severe scenario?

Why the Confidence Level Matters: Capital and Buffers

The choice of confidence level directly determines capital requirements and risk buffers.

Under Basel III, the international banking standard, banks must hold capital equal to at least their 99% VaR (plus other adjustments) over a 10-day holding period. This regulatory standard ensures that banks can absorb losses up to a more extreme threshold before depleting reserves. The use of 99%, not 95%, reflects the regulatory philosophy that banks are critical infrastructure; they must be prepared for scenario that is rarer and more severe than a typical bad day.

A bank with a 99% 10-day VaR of $100 million must hold at least $100 million in equity capital (plus other capital tiers) to meet the standard. If the bank were instead to use a 95% VaR of, say, $60 million, its capital requirement would be $60 million—a material reduction. But that capital buffer would be inadequate during a 1 in 20 loss event, risking insolvency.

For hedge funds and trading desks, 95% VaR is more common for day-to-day position limits and intra-day risk monitoring. A desk manager might issue a mandate: “No single position may have a 95% 1-day VaR greater than $10 million.” This allows traders flexibility for typical market days while policing excessive tail exposure. The assumption is that traders monitor these limits continuously and can react if limits are breached; the 5% tail is acceptable risk for operational agility.

The Loss Magnitude Gap: Why 99% VaR Is Larger

The gap between 95% and 99% VaR is non-linear. Assuming a normal distribution of returns, the ratio is roughly 1.3x to 1.4x. But real financial returns have “fat tails”—extreme events occur more frequently than a normal distribution predicts—so the actual ratio can be larger, sometimes 1.5x or higher.

Worked example. A portfolio has historical daily returns with mean 0% and standard deviation 1.5%. Using a normal distribution:

• 95% VaR (1.645 standard deviations): 1.645 × 1.5% = 2.47% loss
• 99% VaR (2.326 standard deviations): 2.326 × 1.5% = 3.49% loss

Ratio: 3.49% / 2.47% = 1.41x

For a $100 million portfolio:

• 95% VaR = $2.47 million
• 99% VaR = $3.49 million

The additional $1.02 million in buffer (from 95% to 99%) may seem small in percentage terms, but it reflects the cost of preparing for a loss scenario that is 5 times rarer. Over many positions and a large institution, these buffers compound.

Impact on Capital Adequacy and Crisis Resilience

The choice between 95% and 99% is a fundamental trade-off between operational flexibility and crisis resilience.

95% confidence is pragmatic for daily operations. It acknowledges that markets are often volatile, losses are frequent, and capital cannot be infinite. A firm using 95% VaR for position limits allows traders to deploy more capital per position and capitalize on opportunities, knowing that roughly 5% of days (about 12 days per year in a 250-day trading year) will exceed the VaR threshold. The firm must have mechanisms to react (margin calls, position reductions) when these exceedances occur.

99% confidence is conservative. It assumes the firm wants to be prepared for a loss scenario that is so rare it may only occur a handful of times per decade. The cost is tied-up capital and reduced return on equity, but the benefit is that ordinary market stress is absorbed without drama. If a 1 in 100 day occurs, the firm can survive it.

For a bank with $1 trillion in assets, the difference between 95% and 99% VaR capital requirements can amount to tens of billions of dollars. That is why regulators mandate 99%: they want banks to survive even uncommon crises without triggering taxpayer bailouts. For a prop trading desk with $100 million under management, the difference might be $5–10 million in tied-up capital, a meaningful constraint but not existential.

The Problem of the “Tail Beyond VaR”

A critical limitation of VaR, regardless of confidence level, is that it says nothing about losses beyond the VaR threshold. A 99% VaR of $100 million tells you losses will likely not exceed $100 million, but it does not tell you whether a loss beyond $100 million could be $101 million or $500 million.

This “tail risk” beyond VaR became painfully relevant in the 2008 financial crisis and the 2020 COVID-19 crash. Firms calculated VaR, believed they were safe, and then experienced losses far beyond the VaR estimate. For this reason, regulators and risk managers increasingly complement VaR with Conditional Value at Risk (CVaR), also called Expected Shortfall. CVaR estimates the average loss in scenarios where the loss exceeds VaR—a more complete picture of tail risk.

A 99% VaR of $100 million with a CVaR of $150 million means: losses will likely not exceed $100 million, but if they do exceed $100 million, the average overage is about $50 million. That tells the risk manager: “Plan for buffers beyond $100 million; tail events are severe.”

Practical Trade-Offs: When to Use 95% vs. 99%

Use 95% VaR when:

• Monitoring intra-day trading desk positions (high operational frequency, daily recalibration)
• Setting position limits for a liquid, actively managed portfolio (traders react continuously)
• Running scenario analysis or stress testing (95% captures typical bad days; stress tests cover tail separately)
• Resources are constrained (capital must be deployed for returns; perfect crisis preparedness is not feasible)

Use 99% VaR when:

• Calculating regulatory capital requirements (Basel III mandates 99%)
• Sizing buffers for systemically important institutions (banks, insurance companies)
• Managing concentrated or illiquid positions (reaction time is slow; buffer must be larger)
• Modeling multi-day or multi-week holding periods (longer holding periods compound risk; higher confidence protects against cumulative shock)

Use both when:

• Designing a firm-wide risk governance framework (report 95% for operational insight, 99% for regulatory compliance and board oversight)
• Stress-testing for capital planning (use 99% VaR as baseline, then model scenarios beyond VaR)
• Communicating with stakeholders (95% for managers and risk officers; 99% for regulators and audit committees)

The Holding Period and Confidence Level Interaction

VaR always includes a holding period (1 day, 10 days, 1 quarter, etc.). The confidence level and holding period interact. A 1-day 99% VaR and a 10-day 99% VaR are measuring different things:

• 1-day 99% VaR: The loss threshold for a single day that is exceeded only 1% of the time. Regulators use this for day-to-day trading risk.
• 10-day 99% VaR: The loss threshold for a 10-day period that is exceeded only 1% of the time. This is larger (roughly sqrt(10) ≈ 3.16x) because risk compounds over time.

Basel III specifies 10-day holding periods because banks need capital that is stable over a multi-day period of stress. A 1-day VaR is unsuitable for a 10-day capital requirement, as it would underestimate cumulative risk. Conversely, using a 10-day VaR for daily position limits would be overly conservative.

See also