Capstone: Run the Book Under the Policy

The brief: what you are being asked to run

The capstone is a **30-day paper book under a policy you wrote**. That means a written policy in the shape of the previous lesson — a per-trade risk fraction, a heat cap, a concentration cap and a drawdown trigger with a stated way back — and then a record of the book obeying it. Before any position is opened you should be able to say what each one is risking, where its stop sits, and which cluster it belongs to; if any of those three questions does not have an answer, the position is not ready to be opened. The evidence you submit is an **exposure log**, and it is the point of the whole exercise: for each position, the entry, the stop, the distance between them, the number of shares that makes that distance worth the risk budget, the risk in dollars, the risk as a percent of capital, and the cluster it belongs to. Then a running total — the heat — and the cluster subtotals. That log is what makes the book auditable, because every cap in the policy is a statement about one of its columns, and a book that has kept the log can answer the audit in a minute. Two runs can produce the same return and be graded very differently, because **the grade is process, not profit**. A book that made money by taking a 6% risk on a single name and got away with it has failed the exercise, and a book that lost a small amount while keeping every rule has passed it. That is not a stylistic preference. It is the same argument the expectancy lesson made: a single 30-day outcome is one draw from a distribution, so it tells you almost nothing about the process, whereas the rules can be checked directly — and the whole curriculum exists because the rules are the part that is knowable in advance. What each row of an exposure log holds — Entry and stop: the two prices that define the idea · Stop distance: the percent the risk budget will be divided by · Shares: risk budget ÷ (entry − stop) · Risk in dollars and as a percent: the number the heat and cluster caps are checked against ← · Cluster: which related group it is added to before the audit Sizing and risk are different questions. In the audit table the smallest position by dollars carries the third-largest risk, because a 15% stop on a small position is a large risk, while the second-largest position carries the second-smallest, because a 4% stop on a diversified holding is not the same kind of bet.

Running the 30 days, and the honesty requirement

The book is run for 30 days, and three things happen during that time that the policy has to have anticipated. Positions hit their stops, which is the plan working rather than failing. Correlations do what they do in stress and several names in the same cluster fall together, which is the case the cluster cap exists for. And at some point the account is down enough that the drawdown trigger fires, halving the risk fraction as the policy states — at which point the discipline is to apply the rule that was written, not to re-derive it. Each of those events belongs in the log with a date, and the log is the artefact: a heap of trades with no stops and no clusters cannot be audited at all, so the exercise is ungradeable rather than merely untidy. The honesty requirement is the part worth understanding rather than just obeying. A violation — an oversized position, a breached heat cap, a stop moved because the price approached it — is not an automatic fail. What is being assessed is whether you noticed it, logged it with a date and a reason, and adjusted. **An honestly logged violation is evidence of a working process; an unlogged one is evidence of a broken record**, because a log that only contains compliance cannot be used to learn anything and cannot be trusted at all. The same reasoning governs the review at the end: the question is not whether the month made money but whether the arithmetic you wrote down in advance was the arithmetic that governed what you did. The stress test closes it. Before the run, write down what the book does under two or three specific shocks — a 10% index fall with correlations rising, a rate move that hits the long-duration names, a single cluster falling 20% while everything else is flat — and compute the loss rather than describing it. A book whose caps were set by adding related names together will pass those shocks by design, and the point of the exercise is to see the number before the market supplies it. That is the whole subject in one page: size from the stop, add the clusters, cap the heat, write the trigger, log the result, and judge the process rather than the month. The capstone deliverable — A written policy: per-trade, heat, cluster, and a drawdown trigger with a way back · An exposure log: entry, stop, distance, shares, dollars of risk, percent, cluster · A stress test: two or three named shocks, with the loss computed rather than described ← · A 30-day run: with every violation dated and explained · The final defence: why each rule was followed, argued from the arithmetic Do not set the caps by feel and then look for trades that fit them. The policy is written first and the positions are sized into it; a book designed backwards from the trades will pass every audit and mean nothing.

Limits, overrides and who decides

A risk system is not a set of numbers. It is a set of rules about who may do what, and the characteristic failure of a self-managed book is that the rules are private and revisable in the moment — which is the same escalation pattern this curriculum names elsewhere, wearing the costume of risk management. The document has four parts. The **limits**: per position, per factor or sector, per instrument type, and in aggregate, each expressed in the same units as the rest of the book so that they can be added. The **trigger**: the observation that requires an action, stated so precisely that there is no argument about whether it has occurred. The **escalation ladder**: what happens at half the limit, at three-quarters, at the limit, and beyond it, decided in advance so that the response does not have to be invented while something is happening. And the **override rule**: who may set a limit aside, on what grounds, and where it is recorded. The distinction that decides whether any of this is real is between a limit and a guideline. A limit has a consequence when it is breached. A guideline is a preference. Most private risk rules are guidelines using the vocabulary of limits, which is precisely why they evaporate in the moment they are needed — nothing happens when they are broken, so nothing holds. Overrides are the hard part, because they are genuinely necessary: a limit can be wrong, and a rigid system that forces a sale into a temporary dislocation is a system that manufactures losses. The design that survives contact is not prohibition but friction and record. Overrides are permitted; each one is logged at the time with the reason; and the *frequency* of overrides is reviewed on a schedule. A limit overridden three times in a month is not a limit, it is an aspiration — and the honest response is either to enforce it or to rewrite it deliberately, rather than to keep the language and lose the substance. There is a failure mode in the other direction, and it is common among careful people: limits set so tight that ordinary variance trips them. A position limit that a normal week breaches forces a sale at the bottom and converts noise into behaviour. That is why limits belong *after* the risk arithmetic rather than before it — sized from the expected maximum drawdown and the ordinary range, so that being at the limit means something has actually changed. The practical form is one page: four limits, one escalation ladder, one override rule with a log, and a monthly review of the log rather than of the P&L. It takes an hour to write and it is the difference between a risk policy and a set of intentions. • A limit has a consequence; a guideline has a preference — most private rules are the second. • Write the escalation ladder before it is needed, not during. • Allow overrides, require a logged reason, and review their frequency. • Size limits from the expected drawdown, or ordinary variance will trip them. The audit question for any limit system: name the last time a limit was breached and say what happened. If the answer is “nothing”, the limit is decoration, and the book is running on the trader’s judgement with a document attached.

Sources

• Process over outcome in evaluating a decision makerAnnie Duke, "Thinking in Bets" (2018)
• Risk limits, exposure logs and drawdown reviewProfessional risk-policy practice; managed-futures drawdown controls
• Position sizing and expectancy across a sample of tradesVan Tharp, "Trade Your Way to Financial Freedom"
• Sizing from the stop and the noise it must surviveWilder, "New Concepts in Technical Trading Systems"