Calculator
CVaR at 95% (5% Tail)
−$6,200
Tail Risk Analysis
At the 95% confidence level, historical VaR is a 6.20% loss ($6,200), and the average of the worse tail losses is CVaR 6.20% — about $6,200 on a $100,000 portfolio. In the worst 1 of 12 observed periods, losses averaged that depth; plan portfolio size and position risk around it.
*Historical simulation only. Past return distributions do not guarantee future tails. Educational only, not investment advice.
Value at Risk answers one question: how much can I lose on a bad day, at a given confidence level. But a 95% VaR of five percent says nothing about the other five percent — the days that are worse than bad. Conditional VaR, also called expected shortfall, asks the harder question: on the days that breach that threshold, how much do losses average? If the worst four percent of months lose eight, ten, and fifteen percent, CVaR captures all of it; VaR stops counting at the threshold and ignores everything beyond. That distinction is why bank regulators moved from VaR to CVaR as their market-risk standard, and why serious portfolio managers run both. For the American investor, tail risk is where financial plans go to die. Most households concentrate their wealth in retirement accounts that can drop thirty to fifty percent in a real crisis, and the dollar loss on a tail event dwarfs years of careful saving. Standard deviation treats an upside surprise the same as a crash, so it understates the danger that actually matters. CVaR isolates the downside alone and prices it in dollars. It is the number to plan around when deciding how large a stock allocation to carry — because knowing what the average crash costs is more useful than knowing the average month's swing.
The historical method needs no distributional assumptions. Sort all observed period returns from worst to best. At a given confidence level, take the tail: for n observations at 95% confidence, roughly five percent of the data falls in the tail — mathematically, the number of tail observations is at least one and equals the largest whole count that does not exceed (one minus confidence) times n. VaR is the loss at the boundary — the largest loss still inside the ordinary range, negated so it reads as a positive loss. CVaR averages every observation at or beyond that boundary and negates the mean. A small example shows why CVaR exceeds VaR. Suppose twelve monthly returns include a worst month of minus 6.2%, and the 95% tail covers the single worst observation: VaR is 6.2% and CVaR equals it, since only one point breaches. But with thirty observations and a 95% level, the tail holds the three worst months — say minus 4.1, minus 5.8, and minus 8.9 percent. VaR reads 4.1%, the boundary, while CVaR reads 6.3%, the average of all three. The gap between the two numbers quantifies how much worse than the threshold the tail truly gets. Multiply the CVaR percentage by portfolio value for the dollar figure the plan should be built around.
The 2008 crisis was the textbook demonstration: VaR models said daily losses would rarely exceed a few percent, and they were technically right about the threshold — but on the days that did breach it, losses ran several times deeper than VaR implied. Every portfolio that sized positions to the VaR number held too much risk exactly when correlations spiked. CVaR forces the full tail into the number. If your risk process can include only one measure, CVaR is the safer choice because it is conservative by construction — it can only overstate tail severity relative to VaR, never understate it.
Historical CVaR inherits what the sample experienced — and nothing it did not. Thirty-six months of calm data will produce an alarmingly small tail number, because the sample simply has no crisis in it. Professional risk desks deliberately include stress windows (2008, 2020, 2022) in their backtests for this reason. When you read your own CVaR, first check what the worst observation actually was and whether your sample includes a real drawdown. The honest question is whether that worst month in the data is plausible to repeat — and in equity portfolios, the answer almost always is.
The practical use is allocation: decide the maximum dollar loss the portfolio can absorb from a tail event without forcing life-changing decisions — selling a home, abandoning a retirement date — and back into the allocation. If the 95% CVaR of the equity sleeve is 12% and the plan can tolerate a 25% drop in that sleeve, the equity allocation ceiling is roughly half. The math is simple; the discipline of running it before the drawdown instead of during one is what separates durable portfolios from emotional ones.
When testing a portfolio's tail, deliberately include monthly returns from 2008, March 2020, and 2022 in the input series. A CVaR computed on tranquil data underestimates the true loss potential, sometimes by half or more. If your own brokerage history is short, use index or fund data covering at least one full crisis as the sample, then adjust the dollar figure to your account size. The point is not pessimism — it is calibration. A tail measure calibrated on calm markets is a false sense of security in numerical form.
Run the same returns at 90%, 95%, and 99% confidence. In healthy distributions the three numbers climb gently. When the 99% CVaR explodes relative to the 95%, the tail is fat — rare but extreme losses dominate, which is exactly the signature of leveraged or concentrated portfolios. Fat tails also mean diversification benefits less than expected. The spread across confidence levels tells you more about the portfolio's fragility than any single number does, and it takes thirty seconds to compute all three.
Convert the CVaR percentage into dollars immediately — the plan lives in dollars, not basis points. Write down: if the 95% CVaR of the portfolio exceeds this dollar amount, the allocation changes. Putting the trigger in writing before the drawdown happens is what turns a risk measure into risk management. Most investors never act on tail statistics because they encounter them as abstract percentages during a crisis, when the emotional response — hold, add, panic — crowds out arithmetic. The pre-committed dollar threshold is what overrides that response.
Rachel held a concentrated growth portfolio she called long-term and never measured. When this calculator ran her three years of monthly returns at 95% confidence, the CVaR came out 14.2% — on a 400,000 dollar portfolio, that is a 57,000 dollar expected tail loss, roughly her annual income. The number reframed her allocation decision overnight: the question was no longer which fund to add but whether a 57,000 dollar crash scenario was survivable without selling. She trimmed to a diversified 70/30 mix, bringing the CVaR to 8.1% while keeping most of the upside.
A small institutional committee ran both metrics on their target allocation: VaR said the 5% worst month cost 4.8%, and most members found that acceptable. But CVaR showed that when the threshold was breached, the average loss was 7.9% — the distribution's tail was much fatter than VaR's single number suggested. They reduced the allocation to high-yield credit, the fattest tail in the mix. When markets stressed eighteen months later, their actual drawdown landed near the CVaR estimate rather than the VaR one. The committee's files show the decision was made calmly, before the stress, which is exactly when it could still be made calmly.
Devon believed his twelve holdings were diversified until he ran the combined portfolio's CVaR and compared it against an index blend. The numbers were nearly identical — his holdings had averaged correlations of 0.8, so they moved as one in every down month and the tail was no smaller than a pure equity portfolio. The CVaR forced an honest question: if the downside is identical to an index fund, what are all these positions adding? He consolidated into three low-cost funds and added bonds, cutting the CVaR by a third. Measurement preceded conviction; conviction preceded action.
Everything you need to know about this topic.

Financial Chaos Analyst
Ivy Sinclair-Wren is a Financial Chaos Analyst covering investing, AI, wealth psychology, and the emotional consequences of opening finance apps during market crashes. Based in Melbourne, she specializes in demystifying the US tax code and helping users navigate the intersection of spreadsheet logic and human irrationality.