Risk
What "Value at Risk" means
One number, three ways to get it, and the one thing it cannot tell you.
Value at Risk, usually shortened to VaR, answers one question: on a normal bad day, how much could this portfolio lose? Sentinel states it as a money amount at a confidence level. "One-day VaR at 95% is €1,200" means that on 19 days out of 20 you should lose less than €1,200, and on the twentieth you may lose more.
That last clause matters. VaR is not the worst case. It is the line that normal days stay inside. The days that cross it are exactly the ones that hurt, which is why Sentinel also shows expected shortfall: the average loss on the days that go past the line.
Three ways to get the number
There is no single correct way to estimate VaR, so Sentinel calculates it three ways and shows them side by side.
- From volatility (the parametric method). Assumes returns follow a bell curve and reads the loss off how volatile your holdings have been. Fast and smooth, and it underestimates the wild days because real markets have fatter tails than a bell curve.
- From history (the historical method). Takes your holdings' actual past daily moves and asks what the worst 5% of them would have done to today's portfolio. No assumptions about the shape of returns, but it only knows what has already happened.
- From simulation (Monte Carlo). Generates thousands of possible days from the statistics of your holdings and reads the loss off that crowd of outcomes. It can explore days history has not shown you yet, at the cost of depending on the assumptions you feed it.
When the three agree, you can trust the number more. When they disagree, the gap itself is information: usually it means your holdings have had a few violent days that the bell curve is smoothing over.
What it cannot tell you
VaR describes normal bad days. It says nothing about how bad the abnormal ones get, and it is built from the past, so it is quietest right before something new happens. That is why Sentinel pairs it with stress tests, which ask about specific crises, and with a backtest, which checks whether the VaR line has been holding up. Treat VaR as a speedometer, not a crash test.
