Advanced portfolio risk — course checkpoint

You began this course able to say a portfolio was "risky." You finish able to say how risky, under which assumptions, and — the part most courses skip — where those assumptions stop working.

The measures, in one breath

Value at Risk — the loss threshold at a chosen percentile. Always quoted with three parameters: horizon, confidence level, and unit, plus the unlabelled fourth dial of window and method. Computed by historical simulation (sort and count), parametrically (z × σ × value), or by Monte Carlo. The methods disagree, and the disagreement is information.

Expected Shortfall — the average loss beyond the VaR threshold. It answers VaR's central silence, it is sub-additive (safe to aggregate), and it is what the Basel Committee moved to at 97.5% in the Fundamental Review of the Trading Book. It is also harder to estimate and harder to backtest.

Backtesting — the discipline that turns a risk number into a claim you can check. Count exceptions against expectation, then look at whether they clustered. Clustered breaches mean the model didn't react to a regime change.

The four things the numbers cannot see

Fat tails. Real markets produce extreme days at rates the normal distribution rules out — 1987's −22.6% Dow session sits around 20 standard deviations under a fitted bell curve. Count your own 3-sigma days and you'll find more than the model allows.

Correlation breakdown. Diversification depends on a statistic that rises under stress. In 2008 government bonds cushioned equity losses; in 2022 they amplified them. Correlations are regime-dependent, and the coming regime is not knowable.

Liquidity. Every model marks positions at screen prices. Days-to-liquidate and widening spreads tell you what an exit would really cost. A twelve-day exit horizon makes a one-day VaR a category error.

Reflexivity. Low measured risk permits leverage; leverage produces forced selling; forced selling produces the volatility the model then reports. The measurement participates in the outcome.

The two tools that don't depend on a distribution

Historical replay — run today's portfolio through a real crisis window. Concrete and undismissable, but the next crisis won't be a copy.

Hypothetical scenarios — construct internally consistent factor shocks with a causal story attached, including a liquidity leg. Limited by imagination rather than by data, which is a different limitation, not a smaller one.

And reverse stress testing: start from the loss that would break the portfolio and work backwards to what would cause it.

Where the risk actually is

Weights show where the money sits; risk contributions show where the risk sits, and they rarely match. A 60/40 portfolio with equity volatility near 16% and bond volatility near 6% typically puts around 92% of the risk in the 60% equity sleeve. Contributions sum exactly to portfolio volatility, which is what makes risk budgeting — allocating risk deliberately rather than accidentally — possible as both a construction method and a governance tool.

The one sentence to carry out

Every measure in this course is built from a sample of the past, and every one of them is least reliable precisely when it is most urgently consulted.

That is not an argument against measuring. Unmeasured risk is not smaller risk. It is an argument for holding two things at once: compute the numbers rigorously, and never mistake them for the territory. The professionals who came through 2008 intact were not the ones with the best models. They were the ones who knew what their models were assuming, and asked a second question the model could not answer.

Before you sit it

Each of these is a minute at your desk. Any one that is not names the lesson to reopen first.

• Name the three numbers that define any VaR, and say what changes when confidence moves from 95% to 99% — Which three numbers define any VaR?
• Say what Expected Shortfall reports about the same tail that VaR refuses to — What is Expected Shortfall and why did regulators switch to it?
• Say which way correlations move in a crisis and what that does to your loss — Why does diversification fail exactly when you need it?
• Say which of two holdings contributes more risk when one carries twice the volatility — Which holding is actually contributing your risk?

Try it now

1. For one portfolio you can describe in three lines, produce a one-page risk summary from the two sleeves below, on one pinned window: 95% VaR, 95% Expected Shortfall, the ES/VaR ratio, risk contributions by holding, days-to-liquidate for the largest position, and the loss under one historical replay. The average daily turnover of each sleeve, which the liquidity figure needs, is in the two tables under the charts.

2. Next to each figure, write the assumption it depends on and the window you used. Then move the window — press MAX, pick a different decade — and note which figures moved most. That second column is the actual skill this course taught.
3. Finish with one sentence naming the risk your summary does not capture. Every honest risk report has that sentence.

Checkpoint quiz next. Nothing in this course states what level of risk is appropriate for anyone, recommends any allocation, or predicts any market outcome — it teaches how the measurements are built and where they break, which is a literacy, not a signal.