What To Watch On The Risk Screen All Day

At 11:00 you are down $210,000. Is that a normal Tuesday, or is something wrong with your book? Your risk screen will happily tell you forty things — exposures, greeks, VaR, factor loadings, twelve versions of PnL — and none of them answers that question directly. Traders who lose money on days they could have saved usually did not lack data. They lacked a way to decide, in thirty seconds, whether what they were looking at was noise.

Think of a cockpit. A pilot has a hundred instruments and does not scan them equally. There is a fixed pattern built around the handful that predict a stall, and everything else gets looked at when one of those handful moves. A risk screen works the same way. You want five numbers you check constantly, and thirty-five you check when one of the five misbehaves.

The five

1. Gross and net exposure — how much you own in total, and how much of it is a directional bet.
2. Today's PnL measured in normal days — not the dollar number, the dollar number divided by what a typical day looks like.
3. Limit utilisation — how close each risk limit is to being touched, in percent.
4. Concentration — your largest single-name and largest sector loss if that thing has a bad day.
5. Drift since the open — how far the book has quietly moved away from where you set it this morning.

Everything else on the screen is diagnostic. These five are the alarm.

Why "net exposure" lies to you

The number most traders quote is net dollar exposure: longs minus shorts. It is the easiest number on the screen and often the most misleading, because a dollar of a sleepy utility and a dollar of a high-beta semiconductor are not the same dollar. What you want is exposure weighted by how much each position actually moves.

Write $E_i$ for the dollar exposure of position $i$ — positive if you are long it, negative if you are short. Write $\beta_i$ for that position's beta: how many percent it tends to move when the index moves 1 percent. Then the number that matters is

$$\text{beta-adjusted net} = \sum_i \beta_i \, E_i .$$

In plain English: take every position, scale its dollar size by how sensitive it is to the market, and add them all up. The result tells you how many dollars of index you are effectively long or short.

Worked example: the flat book that isn't

Your screen shows a $50m gross book: $26.0m of longs across 22 names and $24.0m of shorts across 19 names.

• Net dollar exposure: $26.0 - 24.0 = +2.0$, so $2m net long on $50m gross. Four percent. Looks flat.
• But the longs are growth names with an average beta of 1.15, and the shorts are defensives with an average beta of 0.85.
• Beta-weighted longs: $26.0 \times 1.15 = 29.9$.
• Beta-weighted shorts: $24.0 \times 0.85 = 20.4$.
• Beta-adjusted net: $29.9 - 20.4 = +9.5$, so $9.5m.

If the index falls 1 percent, the naive number predicts a loss around $20,000. The honest number predicts about $95,000 — nearly five times more. You are not running a market-neutral book. You are running a $9.5m long that is disguised as one.

Reading today's PnL in units of a normal day

The second number is the one that answers the 11:00 question. Your risk model gives you a predicted standard deviation for a full day's PnL; call it $\sigma_{\text{day}}$, "the size of a typical day's swing, up or down". A day's move does not accumulate evenly, though. Risk builds with the square root of elapsed time, so if $t$ is the minutes elapsed and $T$ is the minutes in the whole session,

$$\sigma(t) = \sigma_{\text{day}} \sqrt{t/T} .$$

In plain English: by a quarter of the way through the day you should already have moved about half of a typical full day's amount, not a quarter. Most of a day's variation happens early.

Worked example: is this drawdown normal?

Your risk model says $\sigma_{\text{day}}$ is $180,000. Your daily loss limit is $500,000. It is 11:00, which is 90 minutes into a 390-minute session, and you are down $210,000.

• Expected swing by now: $180{,}000 \times \sqrt{90/390} = 180{,}000 \times \sqrt{0.231} = 180{,}000 \times 0.480 = 86{,}400$, so about $86,000.
• Your actual move in those units: $210{,}000 / 86{,}400 = 2.43$. A 2.4-sigma move.
• Under a normal distribution that is roughly a 0.8 percent one-tailed event; with the fat tails real books have, call it 2 to 3 percent. Either way, this is not a typical morning.
• Now look forward. The remaining 300 minutes carry $180{,}000 \times \sqrt{300/390} = 180{,}000 \times 0.877 = 157{,}900$, about $158,000 of standard deviation.
• So an ordinary one-sigma bad afternoon lands you at $-210 - 158 = -368$, i.e. $368k. A two-sigma bad afternoon lands you at $-210 - 316 = -526$, i.e. $526k — through your $500k limit.

That last line is the decision. You are not merely having a bad morning; you are one ordinary bad afternoon away from being derisked by someone else. Cutting exposure at 11:15 is a choice. Cutting it at 15:30 is an instruction.

The explorer below makes the same point. Leave the mean at 0 and read the horizontal axis in units of a normal day: a $500k limit on a $180k book sits at about 2.8. Now drag the standard deviation up and watch how fast the mass beyond 2.8 stops being negligible — that is what happens when volatility doubles and nobody resizes.

Limit utilisation, at a glance

The third number is the cheapest to read and the most often ignored. Every limit you run against — gross, beta-adjusted net, VaR, single-name — should be displayed as a percentage of itself, not in its native units, so that four incomparable numbers become one comparable row.

What this means in practice

You are not staring at the screen. You run a loop every twenty minutes or so that takes half a minute: beta-adjusted net still where I set it, PnL in sigma terms, any bar past 85 percent, largest name inside its cap, book still resembling this morning's. If all five pass, go back to work. If one fails, that is when the other thirty-five numbers earn their keep.

Two habits break the loop. Watching dollar PnL instead of sigma PnL makes you panic on a quiet day and relax on a violent one. Checking only when you are losing means you never notice the days you made money for reasons unrelated to your thesis.