How the risk numbers are calculated, and how we test them.
Every trading night the model is re-scored against what markets actually did, on a fixed set of portfolios chosen in advance. The results below update automatically, good or bad.
The nightly test
Loading the latest test results…
Daily returnForecast loss for a 1-in-100 day (set the evening before)Day the loss was worse than the forecast
Portfolio
Tested from
Days
Worse than 1-in-100 forecast
Last 250 days
Worse than 1-in-40 forecast
Average loss on bad days
Loading…
How to read the test
1-in-100 forecast. Each evening the model sets the loss that should only be beaten on about 1 trading day in 100. Over 250 days, about 2 or 3 worse days are expected. Many more means the model understates risk; almost none means it overstates it.
Last 250 days. Banks use the same count with a traffic light: 0 to 4 worse days is green, 5 to 9 is yellow, 10 or more is red.
1-in-40 forecast. The same check for a less extreme loss level (about 6 worse days a year expected).
Average loss on bad days. Checks whether losses on the worst days were as large as the model's average-loss forecast. A small p-value (below 0.05) suggests those losses were bigger than forecast.
No hindsight. Each forecast uses only prices known before that day. Model settings are re-estimated once a quarter using past data only.
Chosen in advance. The test portfolios were fixed on 27 Sep 2026, before the first run, and are never swapped for better-looking ones.
Many tests at once. With dozens of statistical tests, a few fail by chance even for a correct model. The count of expected chance failures is shown under the table.
How the risk numbers on your check are calculated
Value. Each holding is valued at the latest closing price times your quantity.
Clean prices. Prices are adjusted for bonuses, splits and demergers. Real crashes are kept in full.
Today's volatility. For each stock a volatility model with fat tails estimates how jumpy it is right now.
Simulation on past data. The last 750 trading days are simulated 10,000 times, picking whole days at a time so stocks keep moving together.
The headline number. From those scenarios we report the loss beaten on only 1 day in 100, and the average loss on those worst days.
Market what-if. Shown only when the Nifty 50 explains at least 40% of your portfolio's daily moves over the past year and at least 80% of it is in Indian stocks.
Technical names: 1-day VaR and Expected Shortfall from filtered historical simulation with GARCH(1,1) Student-t volatility (EWMA fallback).
Honest limits
The model looks one trading day ahead. It is not a forecast of where prices will go.
The past 750 days may not contain a crisis like the next one.
Very large positions can take days to sell; the one-day number does not include that.