4. A practical risk management framework

Risk 6: The limits of the tools

This is the last module of level 6 and a stopping point along the way, not the end: the place to take stock of the toolkit you are carrying so far before going into the three levels that remain.

What you are carrying so far

• Measuring risk (level 1): standard deviation and its limits, why real tails are fatter than the bell's and why the time horizon changes the meaning of any figure.
• Position sizing (level 2): that size is the decision that weighs most, and that the formulas for the optimal size are highly sensitive to an edge estimated from noise.
• Maximum drawdown and ruin (level 3): that the arithmetic of recovery is asymmetric and that the risk that matters is being knocked out before the idea has had time to work.
• Leverage (level 4): that it multiplies the outcome and, above all, that it introduces the possibility of somebody else deciding when you sell.
• Diversification and the risk budget (level 5): that diversifying means spreading risk and not positions, and that dependencies tighten precisely when you needed them not to.
• And this level: that all of the above is measured with tools that inherit the defects of the data they are calibrated on.

Notice what does not appear: any promise of bounding the future loss precisely. That outcome is not a failure of the course, it is its conclusion. What you are carrying is the judgment to know what each figure tells you and, above all, what it does not tell you.

Four questions about your own portfolio

If you cannot answer them in writing and with figures about your real portfolio, you do not have risk management, you have an opinion.

• What am I really exposed to? Not the list of positions, but the common factors sitting behind them. Ten different shares can be the same bet on rates, on a sector or on a currency. Write it down: if rates rise 200 basis points tomorrow, which part of my portfolio feels it, and in which direction?
• How much can I lose in the reasonable worst case? In euros, not as an abstract percentage, and worked out with scenarios rather than only with recent volatility. With the discipline of module 2: the worst case you consider must not coincide with the worst case you have lived through.
• Could I ride out that case without being forced to sell? It depends on your payment commitments, your liquidity, whether margin is involved and your real time horizon, not the one you declare. The same 45% loss is a bad year for someone who can wait and a final outcome for someone who cannot.
• What would have to happen for me to change my mind? Written before, not after. Without this, any loss gets reinterpreted as an opportunity and any gain as confirmation of talent.

A real case: August 2007, the week everyone was the same portfolio

Between 6 and August 9, 2007, dozens of quantitative equity funds, run by independent teams, lost money at the same time and with a violence unprecedented in their track records. The work of Amir Khandani and Andrew Lo documented that a simple strategy of the kind those funds used lost around a quarter of its value over those three sessions. One of Goldman Sachs's large quantitative funds lost around 30% over the month and received a capital injection of some 3 billion dollars.

And then came the revealing detail: on August 10, a good part of what had been lost was recovered. That rules out a change in fundamentals, because fundamentals do not retrace their steps in twenty-four hours. What had happened was that many funds held, without knowing it, the same exposure; somebody had to cut risk urgently for unrelated reasons, their selling moved prices against everyone else, and the risk control systems of the rest ordered selling too. The correlation was not in the historical data: it was created by the very structure of shared positions. It is the summary of these six levels compressed into a single week.

What comes next

The first six levels have dealt with risk as it appears on a price chart: width, tails, size, drawdowns, leverage and the tools all of that is measured with. Three levels remain, and none of them is about that kind of risk.

• Level 7, the risk that is not in the prices: the screen price holds for one share and not for your entire position; where your money physically is when you believe you have it, and what difference it makes whether your broker fails or the issuer of your product does; the concentrations that never show up in the portfolio because they are not assets; and the investor's own operational risk.
• Level 8, time horizon, contributions and withdrawals: this is where the silent assumption of the first six breaks, namely a portfolio nobody puts money into or takes money out of. Sequence risk, how accumulating differs from decumulating, where the four percent rule came from and what regular contributions really protect you against.
• Level 9, a risk system of your own: the limits written down before you need them, what to review every month and what never to look at, how to tell cutting because risk has risen from cutting because it hurts, and the four questions above turned into a single sheet you fill in once a year.

There is one gap, mind you, that none of the nine levels covers: deciding whether a difference in results between two strategies is real or fits comfortably inside chance, separating luck from skill in a track record, or building a historical test that does not lie to you. Here you can reason well and be wrong with judgment; measuring is another matter and it lives on another track, math and statistics, which is the third intermediate one alongside this course and Derivatives. And if you also find yourself missing the asset, the environment or the bill, those are the three easy ones with no code: Market theory, Macro and Costs. The three hard tracks are Technical and Python, Data and production, and Expectations and edge, the last of these without a single line of code, hard by demand rather than by programming.

The complete map, with the recommended order, is at the end of level 9. Here it is enough to know that it exists.