Quant Portfolio Risk Course: Sizing Correlated Trading Signals

A collection of individually sensible signals is not yet a portfolio. Two trades can have different tickers and the same underlying risk. Ten event strategies can all lose when liquidity vanishes. A backtest can show modest average volatility while hiding the one path that forces a trader to stop. A quant portfolio risk management course should teach students to see those connections before teaching them to optimize weights.

Our existing guide to sizing and concentration for short portfolios focuses on practical limits for the short side. This curriculum is broader. It explains how a course should move from single signals to sleeves, dependency maps, negative-skew stress, and portfolio-level gates across a systematic book.

Begin with risk ownership

Every position needs an owner in the research design. A dilution short belongs to a financing sleeve. A de-SPAC lockup trade belongs to a post-merger supply sleeve. A volatility premium position belongs to an options sleeve even if its ticker overlaps another event strategy. Naming the sleeve matters because capital is allocated to mechanisms, not to strings in a ticker column.

The first exercise should make students write a risk inventory before calculating an optimizer. For each sleeve they identify the expected holding period, gross and net exposure, liquidity requirement, tail direction, event concentration, and the market condition in which the mechanism may stop working. This exposes duplicated bets that a covariance matrix estimated from quiet history can miss.

Teach sizing as a constraint problem

Sizing formulas are useful only after constraints are explicit. Equal weight ignores differences in volatility and liquidity. Inverse volatility can place the largest dollars in an asset immediately before its volatility changes. Kelly-style sizing is extremely sensitive to estimated edge, which is usually the least reliable number in the system. A good course presents these methods as starting points, not answers.

Students should size from the loss they can survive, then check whether execution is plausible. That means estimating adverse movement, borrow or option costs, gap risk, and the fraction of normal volume the position represents. The position is the smallest size implied by the risk budget, liquidity budget, and concentration limit. Expected return does not override any of them.

Correlation is a story about shared failure

Historical return correlation is one clue, but event strategies often have sparse and unstable histories. Two sleeves may look uncorrelated because their events did not overlap in the sample. They can still share a financing, volatility, liquidity, or crowded-short driver. The curriculum should therefore pair estimated correlation with a dependency map built from mechanisms.

A useful workshop gives students several apparently distinct trades and asks what common shock could hurt them together. Small-cap dilution shorts, de-SPAC shorts, and distressed issuers may all squeeze when speculative liquidity returns. Short-volatility positions across unrelated names may all gap after a macro surprise. The exercise replaces a false sense of diversification with explicit hypotheses about joint loss.

• Same issuer. Multiple signals on one ticker remain one concentration.
• Same mechanism. Different issuers can share the same forced buyer or seller.
• Same funding condition. Several sleeves can depend on cheap leverage or abundant borrow.
• Same tail. Small steady gains can hide one common gap-loss regime.

Negative skew requires path-aware controls

Many systematic strategies collect frequent small gains and experience rare large losses. Average volatility and a normal-distribution assumption describe this poorly. Students need to inspect maximum adverse excursions, clustered losses, gap scenarios, and the time required to recover. They should learn that a strategy can have positive expectancy and still be untradeable at the chosen size.

Portfolio gates belong here. A regime filter can reduce gross exposure when volatility, spreads, or signal correlation rises. A drawdown gate can pause new risk while preserving a documented restart rule. A liquidity gate can reject otherwise attractive trades. These controls should be specified before the next crisis, because a rule invented during a loss is usually an emotion with a spreadsheet attached.

Stress tests should be written as narratives before they become numbers. Ask what happens if every hard-to-borrow fee triples, an expected catalyst is delayed, index volatility gaps higher, or several positions cannot be exited near the closing mark. Then translate each narrative into shocked prices, costs, correlations, and liquidity. This order matters. Purely statistical stress often repeats the shape of the historical sample, while mechanism-based stress can represent a failure that has not occurred inside that sample. Students should report which limit binds under each scenario and what action the operating rules require.

The course should distinguish forecasting error from sizing error. A thesis can fail for a sensible reason and still have been sized correctly. A modest edge can also produce an unacceptable loss because the portfolio treated uncertain estimates as facts. Post-trade review therefore asks two separate questions: did the mechanism behave as expected, and did the risk controls keep the surprise inside budget? Keeping those questions separate prevents every loss from becoming a strategy rewrite and every gain from becoming evidence that size should increase.

The final project is a risk constitution

The course should end with a short document that another researcher could operate: sleeve definitions, position limits, shared-driver limits, stress cases, escalation rules, and portfolio gates. The Sizing, Concentration, and Correlation lesson teaches the core mechanics, and the curriculum hub shows where portfolio work sits after signal research.

A risk course succeeds when students stop asking for the perfect weight and start asking what can fail together. The objective is not to make a backtest smooth. It is to build a book whose exposures are legible, whose losses remain survivable, and whose operating rules exist before they are needed.