Understanding Value at Risk (VaR) with Basic Probability

What Is Value at Risk (VaR)?

Think of VaR the way you might think about weather forecasting. A meteorologist cannot say exactly how much rain will fall tomorrow. What they can say is: there is a 90% chance of less than two inches. Value at Risk works the same way, but for financial losses instead of rainfall.

Before VaR became standard in the 1990s, risk managers described exposure through simple metrics like the total size of a position or the percentage allocated to a risky asset. These measures told you how much was at stake, but they gave no sense of the probability that any of it would actually be lost. VaR brought probability into the conversation. A $10 million trading book and a $1 million position in a volatile stock both have "exposure," but their VaR figures will look completely different because they have different probability profiles.

J.P. Morgan published its RiskMetrics methodology in 1994, which standardized VaR calculation and made it a global benchmark. Today, VaR appears in bank annual reports, regulatory filings under Basel III, investment fund prospectuses, and CFA exam questions. It is not a perfect measure — its limitations are real and worth understanding — but it remains the starting point for almost every serious conversation about market risk.

Why Probability Is the Foundation of VaR

Markets are uncertain. A stock that gained 2% yesterday might lose 3% tomorrow, gain 0.5% the day after, and then drop 5% the day after that. Over hundreds of trading days, those daily returns form a distribution — a picture of how the market has behaved across a wide range of outcomes.

Probability is what lets us work with that uncertainty in a structured way. Rather than guessing what will happen, probability assigns a number to how likely each outcome is. The distribution of past returns tells us that small daily changes happen frequently, moderate changes happen occasionally, and very large changes are rare but possible.

VaR picks a specific point on that distribution and says: losses will not exceed this level on X% of days. That X% is the confidence level, and it is entirely a probability statement. A 95% confidence level means the probability of staying below the VaR loss is 0.95, and the probability of exceeding it is 0.05.

Three probability concepts do most of the work in VaR: the probability distribution of returns (which describes all possible outcomes), the standard deviation (which measures the spread of that distribution), and the confidence level (which picks the specific percentile of interest). Understanding these three concepts is understanding VaR. The rest is arithmetic.

For a review of the probability concepts that underpin this, the basic probability guide and probability rules guide cover the foundations that VaR builds on.

Key Statistical Concepts Behind VaR

VaR draws on a specific set of statistical ideas. None of them require advanced math. Each one is explained below with an investment example that shows why it matters in practice.

Probability

Probability is a number between 0 and 1 that describes how likely an event is. Zero means impossible. One means certain. A probability of 0.05 means the event happens roughly once in every 20 similar situations.

In VaR, probability answers: how likely is a loss of this size or larger? A probability of 0.05 attached to a daily loss of $2,000 means you should expect that threshold to be crossed about once every 20 trading days, or roughly 13 times a year. That framing helps risk managers plan capital reserves, set stop-loss triggers, and communicate exposure to investment committees. The probability calculator can translate raw frequencies from historical data into the probability inputs that VaR requires.

Expected Value

Expected value is the probability-weighted average outcome across all scenarios. It tells you what the average result would be if the same situation played out many times. For a portfolio with a 10% chance of a $10,000 loss, a 60% chance of a $500 gain, and a 30% chance of a $2,000 gain, the expected value is (0.10 × −$10,000) + (0.60 × $500) + (0.30 × $2,000) = −$1,000 + $300 + $600 = −$100.

VaR does not describe the expected loss — it describes a specific percentile of the loss distribution. But expected value sits underneath VaR as the center of the distribution that VaR samples from. The expected value guide walks through both the formula and the business applications in detail.

Standard Deviation and Variance

Standard deviation measures how spread out a set of observations is around the mean. In investment terms, it measures how much daily returns tend to vary from the average daily return. A stock with daily returns averaging 0.05% and a standard deviation of 1.2% is much more variable than a bond fund with the same 0.05% average return and a standard deviation of 0.3%.

Standard deviation is the single most important input to parametric VaR. Once you know the standard deviation of daily returns and the portfolio value, the only remaining question is which confidence level to use — and the formula gives the VaR directly. Variance is standard deviation squared; it appears in the mathematics of combining portfolio risks but is less intuitive for reporting purposes.

The standard deviation guide and variance guide cover both the population and sample versions, and the standard deviation calculator lets you compute it directly from your return data.

Confidence Level and Percentiles

The confidence level in VaR is a probability: the probability that the loss will not exceed the stated VaR amount. A 95% confidence level means that in 95 out of every 100 scenarios (trading days, in the usual interpretation), the actual loss will be smaller than the VaR figure.

The confidence level corresponds directly to a percentile of the loss distribution. The 95% VaR is the 95th percentile of the loss distribution — the point where 5% of outcomes are worse. The percentiles guide explains how to read and interpret percentiles, which is all VaR does: it reads off a specific percentile of a financial loss distribution.

Different users choose different confidence levels depending on how conservative they want to be. A 90% VaR is a relatively permissive benchmark — losses will exceed it on 10% of days. A 99% VaR is demanding — losses will exceed it on only 1% of days, roughly 2.5 trading days per year. The confidence intervals guide covers the related concept of confidence intervals, which is also used in VaR backtesting.

Normal Distribution

The normal distribution — the bell curve — is the mathematical model most commonly used for investment returns in parametric VaR. It is symmetric around the mean, and its shape is completely described by just two numbers: the mean and the standard deviation. Roughly 68% of observations fall within one standard deviation of the mean, 95% within two, and 99.7% within three.

This is why the Z-score appears in VaR formulas. A Z-score tells you how many standard deviations a particular outcome is from the mean. The Z-score for a 95% confidence level is 1.645, meaning the 95th percentile of losses is 1.645 standard deviations away from the average. The normal distribution guide, empirical rule guide, and normal distribution calculator are the tools you need to work with this distribution directly. You can also read the Z-score corresponding to any probability from the Z-table.

Volatility

Volatility and standard deviation mean the same thing in finance. Both describe how much returns vary from their average. Volatility is typically quoted as an annualized figure (e.g., "20% annual volatility") while VaR calculations use daily volatility. The conversion is simple: divide annual volatility by the square root of 252 (the number of trading days in a year).

A stock with 20% annual volatility has a daily volatility of approximately 20% ÷ √252 = 1.26%. That 1.26% daily figure feeds directly into the VaR formula. Higher volatility always produces higher VaR for the same portfolio value and confidence level — which is intuitive: a more volatile asset can lose more in a single day.

Tail Risk and Loss Distribution

Tail risk refers to the probability of extreme, outlying outcomes at either end of a distribution. In the VaR context, tail risk specifically means the probability of losses much larger than the VaR figure itself. VaR at 95% confidence says nothing about what happens in the remaining 5% of scenarios — those are the tail, and they could contain losses of 2x, 5x, or 10x the VaR amount.

A loss distribution maps every possible loss amount to its probability. Building one from historical data requires collecting daily portfolio returns over a long period, treating negative returns as losses, and ranking them from smallest to largest. The 95% VaR is the loss at the 95th percentile of that ranking. Expected Shortfall (CVaR) is the average of all losses beyond that point — a more complete picture of tail risk.

How Value at Risk Is Calculated

The parametric VaR formula is straightforward. You need four inputs: the portfolio value, the daily volatility, the Z-score for the chosen confidence level, and the time horizon.

The Z-scores for the three most common confidence levels are fixed values you can look up once and reuse in every calculation. For 90% confidence, Z = 1.282. For 95% confidence, Z = 1.645. For 99% confidence, Z = 2.326. These come directly from the standard normal distribution, and the Z-score calculator can verify any of them.

The √T term accounts for the time horizon. Risk accumulates over time because more trading days give more opportunities for losses to add up. A 5-day VaR is not simply five times the 1-day VaR — it is √5 ≈ 2.24 times larger. This scaling assumes daily returns are independent, which is a reasonable approximation for most liquid markets over short horizons.

Types of Value at Risk

Three distinct methods exist for calculating VaR. They differ in their assumptions, their data requirements, and how they handle the real-world complications that a pure normal distribution misses.

Historical Simulation VaR

Historical simulation is the most transparent method. It takes the current portfolio weights and applies every day of historical returns — typically the past 250 or 500 trading days — to calculate what the portfolio would have returned on each of those days. The resulting collection of hypothetical daily profits and losses is then sorted, and the VaR is read directly from the percentile that matches the chosen confidence level.

No distribution is assumed. If the historical data contains a crash, a short squeeze, a currency collapse, or a spike in credit spreads, all of those events are represented in the simulation. This is the method's main appeal: it uses actual market behavior rather than a statistical model of it.

The limitation is that the method can only produce scenarios that actually happened in the historical window. If the next market event has no precedent in the dataset — the 2020 COVID crash is a good example — historical simulation VaR will give no warning of it.

Parametric VaR (Variance-Covariance Method)

Parametric VaR assumes returns follow a normal distribution. Given that assumption, the entire distribution is described by just two numbers — the mean daily return and the standard deviation of daily returns — and the VaR for any confidence level can be calculated analytically using the formula shown above.

This method is fast and works well for straightforward portfolios of liquid assets under normal market conditions. The problem is that it systematically underestimates tail risk because actual financial returns have heavier tails than a normal distribution predicts. Large losses happen more often than the normal distribution would imply, which means parametric VaR tends to be too low for risk events at 99% confidence and beyond.

Monte Carlo Simulation VaR

Monte Carlo VaR runs thousands of hypothetical scenarios by drawing random return values from an assumed distribution (or a set of correlated distributions for a multi-asset portfolio), recording the portfolio return each time, and reading the VaR from the percentile of the resulting distribution. See the Markov Chain Monte Carlo guide for the simulation methodology in detail.

Monte Carlo is the most flexible method. Unlike historical simulation, it is not limited to scenarios that actually occurred. Unlike parametric VaR, it can use non-normal distributions that better match the fat tails of real financial data. It can also handle complex instruments like options, where the relationship between underlying asset returns and portfolio value is non-linear.

The trade-off is computation time and model risk — the results are only as good as the assumed distributions and the correlations between assets.

Real Example #1: Stock Portfolio VaR

An investor holds $50,000 in a technology stock. Daily returns over the past year have had a standard deviation of 2.0%. The investor wants to know the 1-day VaR at 95% confidence — in other words, how bad could a single trading day get, 95% of the time?

Real Example #2: Mutual Fund Risk Assessment

A client holds $200,000 in a diversified mutual fund. The fund's daily return volatility over the past year was 0.8%. The fund manager reports both a 1-day and a 10-day VaR at 99% confidence for regulatory purposes.

Real Example #3: Bank Risk Management

Banks use VaR differently from individual investors. Under the Basel III framework from the Bank for International Settlements, banks must calculate a 10-day 99% VaR on their trading book daily and hold sufficient capital to cover potential losses. This VaR number directly determines how much regulatory capital a bank must set aside.

Consider a bank with a trading book valued at $100 million. The daily volatility of the trading book — estimated from the previous year of returns — is 0.5%. The bank calculates its regulatory VaR as follows:

For a broader view of how statistical tools fit into the full risk management process — including credit risk, operational risk, and Monte Carlo simulation — the statistics in risk management guide covers each topic in depth.

Interactive VaR Calculator

Enter your portfolio value, daily volatility, confidence level, and time horizon. The calculator uses the parametric VaR formula and gives an interpretation of what the result means in practice.

Confidence Level Reference Table

The choice of confidence level is a deliberate risk management decision, not a technical detail. It sets how conservative your VaR estimate is and how you interpret exceptions when they occur. The table below shows what each of the three standard levels means for investors and institutions.

A 99% VaR is not "better" than a 95% VaR — it is simply more conservative. The 99% figure will always be larger than the 95% figure for the same portfolio, because it is setting a higher bar for what counts as a normal day. Choosing a confidence level means choosing how much tail exposure you are willing to ignore. No confidence level removes tail risk; it only changes where you draw the boundary before acknowledging it.

VaR Decision Framework

Calculating VaR is not the end point — it is part of a workflow that connects data collection to a risk management decision. The six steps below cover the full process from raw return data to actionable output.

Value at Risk vs Other Risk Metrics

VaR does not work in isolation. Risk professionals use it alongside other metrics that capture what VaR misses — primarily the size and shape of the loss distribution beyond the VaR threshold.

VaR vs Expected Shortfall (CVaR)

Expected Shortfall — also called Conditional VaR or CVaR — asks what the average loss is when losses do exceed the VaR threshold. While a 95% VaR of $5,000 tells you where the bad 5% starts, Expected Shortfall at the same level tells you that in those bad scenarios, the average loss is, say, $8,200. That additional information about the severity of tail losses is what makes Expected Shortfall preferable to VaR for capturing true tail risk. Basel IV is moving toward Expected Shortfall as the primary regulatory risk measure for this reason.

VaR vs Standard Deviation

Standard deviation measures the overall spread of returns — both the good days and the bad ones. VaR focuses specifically on the loss side. For symmetric return distributions, the two metrics contain similar information and can be converted into each other using the Z-score. For skewed distributions — where losses are larger or more frequent than gains — VaR and standard deviation diverge, and VaR gives a clearer picture of the downside alone.

VaR vs Stress Testing

VaR is a probabilistic measure based on historical return patterns. Stress testing asks: what happens if an extreme but specific scenario occurs, such as a 30% equity market crash, a 200 basis point rate spike, or a sudden liquidity freeze? Stress tests are not tied to historical probabilities — they model deliberate, hypothetical worst-case scenarios. A sound risk framework uses both: VaR to monitor day-to-day risk levels, and stress testing to prepare for events that historical data cannot fully anticipate.

Common Mistakes When Using VaR

VaR is useful precisely because it is a clear, testable number. That clarity can also mislead if the number is misread or the assumptions behind it are forgotten.

Industries That Use Value at Risk

VaR originated in banking but spread to any industry that holds financial instruments and needs to report market risk in a standardized way.

Best Tools for VaR Analysis

For students working through the VaR concepts in this guide, the full calculator library covers probability, standard deviation, normal distribution, Z-score, and confidence intervals — all the building blocks of the parametric VaR formula — in an accessible, browser-based format.

VaR Calculation Cheat Sheet

Financial Risk Glossary

Frequently Asked Questions

Key sources and further reading: Bank for International Settlements — Fundamental Review of the Trading Book (Basel III market risk framework) · OpenIntro Statistics — Open-access textbook covering probability distributions and standard deviation · Khan Academy — Statistics and Probability (free foundational course) · CFA Institute — CFA Program curriculum covering VaR in quantitative finance · ISO 31000:2018 — International standard for risk management principles and guidelines · Risk.net — Professional risk management coverage including regulatory VaR developments