Understanding VaR in Energy Trading

Executive summary

Value at Risk, universally shortened to VaR, is the single most widely used measure of market risk in trading, and for good reason: it condenses the risk of a whole portfolio into one number that answers a concrete question, how much could we lose. That simplicity is what made VaR the industry standard, and it is why every energy trading desk, risk manager, and regulator speaks the language of VaR.

But VaR is also widely misunderstood, and energy markets stress it in specific ways. Energy prices are volatile, fat-tailed, and prone to spikes that a naive VaR understates, so using VaR well in energy trading means understanding both what the number tells you and what it hides. This guide explains VaR clearly, including the three main ways to calculate it, its limitations, and the complementary measures, notably Expected Shortfall, that address them.

It covers what VaR is, the three calculation methods, Expected Shortfall, the energy-specific challenges, how VaR drives limits and controls, and backtesting. It is a core pillar of the risk-analytics cluster, connecting to stress testing, scenario analysis, real-time position management, and the technical VaR reference.

What is Value at Risk?

Value at Risk is a statistical estimate of the maximum loss a portfolio is likely to suffer over a given time horizon, at a given confidence level, under normal market conditions. It is defined by three things: a time horizon (say, one day), a confidence level (say, 95% or 99%), and an amount (the potential loss).

Put concretely, a one-day 95% VaR of one million means: under normal conditions, there is a 95% chance the portfolio will not lose more than one million over the next day, or equivalently, a 5% chance it will lose more. That single sentence captures the appeal and the trap of VaR. The appeal is that it summarises the risk of an entire, complex portfolio in one interpretable number. The trap is in the phrase "not more than": VaR tells you the threshold that will be breached only 5% of the time, but it says nothing about how bad the loss is on those 5% of days, which turns out to matter enormously in energy markets.

How VaR is interpreted

Because VaR is so often misread, it is worth being precise about what it does and does not say. Reading it correctly is the difference between a useful risk measure and a false sense of security.

The single most important point is that VaR is a threshold, not a maximum. A 99% VaR is the loss that will be exceeded on 1% of days, and on those days the loss can be far larger than the VaR number, VaR is silent about how much larger. Treating VaR as a worst case is the classic and dangerous error, because it is precisely the losses beyond VaR, in the tail, that cause real damage. This is why VaR must always be read as "the loss we expect to exceed occasionally," and why it needs to be complemented by measures that describe the tail, which the sections below address.

The three ways to calculate VaR

There are three standard methods for calculating VaR, and they differ in how they model the distribution of possible losses. Each has strengths and weaknesses, and the choice matters, especially in energy.

The three methods can give materially different VaR numbers for the same portfolio, which is itself a cautionary lesson: VaR is a model output, not a fact. The next sections look at each in turn, but the headline is that the parametric method, though fast and common, assumes a normal distribution that badly misrepresents energy price behaviour, while historical simulation and Monte Carlo can capture the fat tails that energy markets exhibit. Choosing and understanding the method is part of using VaR responsibly.

Historical simulation

Historical simulation calculates VaR by taking the portfolio’s current positions and repricing them against a set of actual historical market moves, then reading the loss at the chosen confidence level from the resulting distribution. If you use the last 500 days of price moves, you get 500 hypothetical P&L outcomes, and the 99% VaR is roughly the 5th-worst of them.

Its great strength is that it makes no assumption about the shape of the distribution, it uses the actual historical moves, including whatever spikes and fat tails really occurred, which suits energy markets well. Its weakness is that it assumes the future will resemble the historical window: if the past period was unusually calm, the VaR will understate risk, and if a new kind of event has never occurred in the window, it will not be captured. Historical simulation is a sound, widely-used default for energy, precisely because it respects the real, non-normal behaviour of energy prices, but it must be paired with an awareness that the past is an imperfect guide to the future.

Parametric (variance-covariance) VaR

The parametric method, also called variance-covariance VaR, assumes returns follow a normal distribution and calculates VaR from the portfolio’s volatility and the correlations between its positions. It is fast and simple, which made it popular, and for portfolios that genuinely behave normally it works reasonably well.

The problem, and it is a serious one in energy, is the normality assumption. Energy prices are emphatically not normally distributed: they have fat tails (extreme moves happen far more often than a normal distribution predicts) and they spike. A normal-distribution model systematically understates the probability of large losses, so parametric VaR tends to be dangerously optimistic in energy markets, reporting a comfortable number while the real tail risk is much larger. For this reason, parametric VaR should be used with great caution in energy trading, if at all, because the very behaviour that makes energy risk dangerous, the fat tails, is exactly what this method assumes away.

Monte Carlo simulation

Monte Carlo VaR calculates risk by simulating a large number of possible future scenarios, generated from a model of how prices and other risk factors behave, repricing the portfolio in each, and reading VaR from the resulting loss distribution. Because the model can be chosen to reflect fat tails, spikes, and complex dependencies, Monte Carlo is the most flexible method.

Its strength is exactly that flexibility: it can model non-normal, fat-tailed price behaviour, complex instruments, and intricate correlations that the other methods handle poorly, making it powerful for the options and structured products common in energy. Its weaknesses are that it is computationally heavy (many thousands of scenarios repriced) and, crucially, only as good as the model behind it, a Monte Carlo VaR built on a normal model is no better than parametric VaR. Used with a model that respects energy’s fat tails, Monte Carlo is the most capable method, but it demands both computational power and modelling care, which is why it rewards a platform with a serious quant engine behind it.

Expected Shortfall: beyond VaR

Because VaR is silent about the size of tail losses, the risk community increasingly complements it with Expected Shortfall (also called Conditional VaR or CVaR), which answers the question VaR cannot: given that we breach VaR, how bad is the loss on average? Expected Shortfall is the average loss on the days worse than VaR.

This matters enormously in energy. Where VaR tells you the threshold breached 1% of the time, Expected Shortfall tells you the average severity of those 1% of days, capturing exactly the tail risk that VaR hides. Because energy markets have fat tails, the gap between VaR and Expected Shortfall can be large, and a portfolio with a comfortable VaR can have an alarming Expected Shortfall, a warning VaR alone would never give. This is why Expected Shortfall has become the preferred tail-risk measure in modern risk management (and in regulatory frameworks), and why a serious energy risk platform computes both: VaR for the familiar threshold, Expected Shortfall for the severity of what lies beyond it. Together they give a fuller, more honest picture of risk than VaR alone.

Why energy markets challenge VaR

Energy markets stress VaR in ways that make naive application dangerous. Understanding these challenges is what separates using VaR well from being misled by it.

The common thread is that energy returns are not normal and not stable, which is precisely what basic VaR assumes. Fat tails and spikes mean the losses beyond VaR are both more likely and more severe than a naive model suggests; non-linearity from options means the relationship between price moves and P&L is not straightforward; and the hourly, locational nature of power adds dimensions that a simple VaR flattens. This is why energy trading demands methods that respect non-normality (historical simulation, well-modelled Monte Carlo), tail measures (Expected Shortfall), and, crucially, the complementary techniques of stress testing and scenario analysis that deliberately probe the extremes VaR summarises but does not describe.

VaR, limits, and controls

VaR is not just a report; it is a control. Firms set VaR limits, caps on how much VaR a desk or the firm may run, and monitor them continuously, so that risk is kept within the firm’s appetite. This turns VaR from a passive measure into an active constraint on trading.

For limits to be meaningful, VaR has to be computed on the live, current book, not an overnight snapshot, because in a fast market the risk you measured last night is not the risk you hold now. This ties VaR directly to real-time position management: VaR is a function of the live positions, so it must update as they change. A modern platform computes VaR continuously on the governed book and enforces limits against it, ideally flagging as a desk approaches a limit rather than only reporting a breach after the fact. Used this way, alongside the tail and scenario measures that cover its blind spots, VaR becomes a genuine control that keeps risk within appetite rather than a number reviewed the next morning.

Backtesting VaR

A VaR model is a prediction, and predictions must be checked against reality, which is what backtesting does. Backtesting compares the VaR the model predicted against the losses that actually occurred, counting how often losses exceeded VaR to see whether the model is accurate.

The logic is straightforward: a 99% VaR should be exceeded on about 1% of days, so if actual losses breach it far more often, the model understates risk, and if far less often, it overstates it. Regular backtesting is what keeps a VaR model honest, catching a model that has drifted out of line with reality, which matters especially in energy, where regime shifts can quietly degrade a model that once fit. It is also expected by regulators as evidence that a firm’s risk model is sound. A platform that backtests VaR continuously, on governed data with clear lineage, gives a firm confidence that its headline risk number actually reflects the risk it runs, rather than a comforting figure that quietly stopped being accurate.

A worked interpretation

To bring the pieces together, consider an illustrative desk with a one-day 99% VaR of two million and an Expected Shortfall of three and a half million. (This is a representative example, not real figures.)

The VaR says: on a typical day, the desk is very unlikely (1% chance) to lose more than two million. That is the reassuring headline. But the Expected Shortfall says: on the bad 1% of days when it does breach VaR, the average loss is three and a half million, meaningfully worse than the VaR figure. The gap between two million and three and a half million is the fat tail that VaR alone would hide, and in energy that gap can be large. A risk manager reading both numbers understands not just the threshold but the severity beyond it, and would further probe the extremes with stress tests asking what a specific spike or cold snap would do. This is VaR used well: as one honest number among several, read with full awareness of what it summarises and what it leaves to the tail measures and scenarios around it.

Why the Gravitas VaR engine is different

Gravitas computes VaR and Expected Shortfall on the live, governed book, with methods that respect energy’s fat tails.

Because VaR and Expected Shortfall are computed live on the governed book with fat-tail-aware methods and kept honest by backtesting, the desk gets a risk number it can trust and act on, complemented by the tail and scenario measures that cover VaR’s blind spots. And it is delivered at economics that suit desks the incumbents priced out. See the quant engine, the VaR reference, or request a demo.

Best practices

Using VaR well in energy trading rests on a few principles. Read VaR correctly, as a threshold breached occasionally, not a worst case. Prefer methods that respect energy’s fat tails, historical simulation or well-modelled Monte Carlo, over parametric VaR, which assumes away the very tails that matter. Always compute Expected Shortfall alongside VaR, so the severity of tail losses is visible, not just the threshold. Compute VaR live on the governed book and enforce it as a limit. Backtest continuously to keep the model honest. And complement VaR with stress testing and scenario analysis that deliberately probe the extremes.

The through-line is that VaR is an indispensable but incomplete measure: it summarises portfolio risk in one interpretable number, but it is silent about the tail, assumes normality when the standard method is used, and must be read with care. Used honestly, as one number among several, computed with fat-tail-aware methods on the live book and paired with Expected Shortfall, stress testing, and scenario analysis, VaR is a powerful foundation of energy risk management. Treated as a worst case or a single sufficient number, it is a false comfort, which in fat-tailed energy markets is exactly the wrong thing to have.

Risk KPIs

A VaR-based risk capability can be measured across accuracy, coverage, and control.

Backtest exceptions and method suitability measure whether the VaR is accurate; Expected Shortfall and scenario coverage measure whether the tail is captured; freshness and limit enforcement measure whether VaR functions as a live control. Together they describe VaR used as an honest, active risk measure rather than a comforting but incomplete number.