Portfolio and Option Models: Useful Benchmarks, Dangerous Certainty

Portfolio theory and option-pricing formulas are useful validation tools because they make assumptions explicit. They provide relationships that a risk system should reproduce in controlled conditions. Their value weakens when a benchmark assumption is mistaken for a fact about the market or when an expected-return relation is used as if it guaranteed an arbitrage.

REFERENCE · NO CALL

This report consolidates portfolio-risk teaching and handwritten option-model notes dated May and June 2024. Numerical examples are historical teaching constructions with generic asset labels. The focus is validation and interpretation, not investment recommendations or current forecasts.

Risk and return need a joint representation

For a portfolio with weights $\mathbf{w}$, expected asset returns $\boldsymbol\mu$ and covariance matrix $\Sigma$, expected portfolio return and variance are

$$E[R_p]=\mathbf{w}^{\mathsf T}\boldsymbol\mu, \qquad \sigma_p^2=\mathbf{w}^{\mathsf T}\Sigma\mathbf{w}.$$

In words: Portfolio expected return combines weighted expected returns, while portfolio variance also includes how the components move together.

The covariance terms determine how component risks combine. Adding individual volatilities ignores diversification; assuming zero correlation without evidence invents it. A covariance matrix must also be internally valid, including positive semidefiniteness, to avoid impossible negative portfolio variances.

The teaching source compares combinations of two hypothetical assets. Its reported minimum-variance combination is approximately 70% in Asset A and 30% in Asset B, with expected return 15.3% and volatility 13.6%. These numbers are source-example outputs, not empirical estimates offered for a live portfolio.

A validator should be able to reproduce the optimum from the underlying return, variance and covariance inputs. The displayed optimum alone is insufficient: constraints on shorting, borrowing and total weights can change the solution. An optimiser’s successful termination does not establish that the intended problem was specified.

Efficient-frontier results are conditional

An efficient portfolio offers the highest expected return for a specified risk under the model’s assumptions and constraints, or the lowest risk for a specified expected return. It is not universally optimal independently of those inputs.

Expected returns are particularly uncertain. Small changes can create large shifts in unconstrained portfolio weights. Covariance estimation can also be unstable when many assets are estimated from limited observations. Sensitivity analysis should therefore vary inputs and constraints rather than presenting one frontier as a precise map of future opportunity.

Transaction costs, turnover, liquidity and concentration restrictions matter for implementation. A theoretically superior portfolio that cannot be traded at the assumed prices is not operationally superior. Validation should distinguish mathematical optimisation from feasible portfolio construction.

Beta is a covariance ratio, not a universal multiplier

For asset or portfolio return $R_i$ and market return $R_m$, beta is

$$\beta_i=\frac{\operatorname{Cov}(R_i,R_m)}{\operatorname{Var}(R_m)}.$$

In words: Divide variation shared with the market by the market’s own variation to measure sensitivity to that market.

It measures sensitivity to the chosen market factor under the estimation convention. It does not describe all risk and need not remain stable across regimes. Residual risk can matter for a concentrated portfolio even when a broad equilibrium model treats diversifiable risk differently from systematic risk.

The Capital Asset Pricing Model writes an expected-return relation as

$$E[R_i]=r_f+\beta_i\big(E[R_m]-r_f\big),$$

where $r_f$ is the risk-free return and the bracketed term is the expected market risk premium. This is a model relation under strong assumptions, not a promise that a realised return will equal the right-hand side.

The source teaching lists frictionless trading, homogeneous expectations, a common horizon and unrestricted borrowing and lending among its simplifying assumptions. A real validation should ask which assumptions are acceptable for the intended use and how deviations affect interpretation.

Factor models distinguish exposures from surprises

A simple realised-return factor model can be written

$$R_i=E[R_i]+\sum_k\beta_{ik}F_k+\varepsilon_i,$$

where $F_k$ are factor surprises relative to their reference expectations and $\varepsilon_i$ is the residual. The factor definition is essential. Using a level where the model expects a surprise changes the meaning of the coefficient.

The source’s hypothetical example starts with expected return of 10%, a growth-surprise beta of 2 and expected growth of 3%. Revising growth to 4% supplies a one-percentage-point surprise. Under that stated linear model, expected return updates by two percentage points to 12%, assuming no other information changes.

This is a conditional arithmetic example. It does not establish a forecast for a named company, a stable macro sensitivity or a causal estimate from observed data. The distinction keeps useful teaching numbers from becoming unsupported security recommendations.

A factor hedge is only as complete as its factor set

The source considers a portfolio with growth-factor beta 0.50 and sentiment-factor beta 0.30. In its idealised construction, a unit-growth factor portfolio has exposure one to growth and zero to sentiment. A 50% short position in that factor portfolio offsets the original growth exposure while preserving the sentiment exposure.

Similarly, a 30% short position in a unit-sentiment factor portfolio offsets the second beta. These are exact within the stipulated factor model. In practice the hedge instruments may contain residual exposures, transaction costs and unstable coefficients.

An expected-return difference between two factor-matched portfolios is therefore not automatically a risk-free arbitrage. Matching selected betas does not prove identical payoffs in every state. The teaching material’s 12% versus 10% example gives a two-percentage-point expected spread under its assumptions; a guaranteed arbitrage requires stronger conditions, including no residual risk and feasible financing and execution.

This is a useful validation question: which risks are cancelled algebraically, and which remain outside the model? A report should answer it before using the word hedged without qualification.

Performance ratios require the right denominator

Performance denominators answer different questions

The choice should match the decision. A benchmark-constrained manager may be assessed on active risk, while a standalone portfolio’s total variability remains relevant to its owner. A strategy with nonlinear tail exposure can look attractive on a volatility ratio while carrying substantial downside beyond the sample.

Validation should examine return frequency, compounding, fees, benchmark alignment and annualisation. A ratio calculated from monthly returns cannot be compared casually with one based on daily returns if the scaling assumptions differ. Stale valuations can also suppress measured volatility and inflate apparent performance.

Option models provide exact controlled benchmarks

For a non-dividend-paying European call under the basic Black–Scholes assumptions,

$$C=S_0N(d_1)-Ke^{-rT}N(d_2),$$

$$d_1=\frac{\log(S_0/K)+(r+\tfrac12\sigma^2)T}{\sigma\sqrt{T}}, \qquad d_2=d_1-\sigma\sqrt{T}.$$

In words: The two formula inputs compare spot with strike after allowing for interest, uncertainty and time until expiry.

Here $S_0$ is spot price, $K$ strike, $r$ the continuously compounded risk-free rate, $T$ time to expiry in years, $\sigma$ annualised volatility and $N$ the standard-normal cumulative distribution function. The formula’s assumptions include the stated payoff and exercise style as well as its dynamics and financing conventions.

A unit error in $T$ or $\sigma$ can produce a plausible but incorrect option value. A good implementation test therefore includes dimension checks, expiry limits, monotonicity, put–call parity where applicable and comparison with an independently implemented benchmark.

The handwritten source shows formula implementation and Greek interpretation. The validation lesson is broader than one formula: reproduce known limiting cases before trusting a general pricing engine.

Exercise style changes the problem

A European option can be exercised only at expiry; an American-style option permits earlier exercise under its contract. A binomial tree can compare continuation value with immediate exercise at each permitted node. Applying the European closed-form expression to every American option would ignore that choice.

A tree should converge as its step count increases under consistent assumptions. Oscillating or slow convergence may require a better numerical scheme or careful treatment of strike alignment. A Monte Carlo estimate of a European payoff can provide another benchmark, but basic terminal-payoff simulation does not automatically solve an early-exercise problem.

The source notes also distinguish moneyness, time decay, gamma and volatility term structure. Those concepts define useful test cases: near expiry, near strike, deep in or out of the money and across changes in the volatility surface. A model validated only at one ordinary spot and tenor has not been challenged where its numerical behaviour is most difficult.

Connect model tests with broader risk assessment

The June 2024 notes organise stress testing around objective definition, scenario selection, modelling, impact analysis, communication and mitigation. That sequence is useful for deciding what a valuation benchmark must support. An accurate price under ordinary assumptions does not establish suitability for stress, liquidity planning or a portfolio with different exercise features.

A practical validation pack should therefore separate benchmark identity, numerical correctness, parameter estimation, economic applicability and production use. It should include controlled examples, adverse cases and limits of interpretation. Private examination questions and answer keys are not needed to communicate those methods.

Use models as disciplined reference points

Portfolio and option models are strongest when their assumptions remain visible. Covariance explains how exposures combine, factor models explain the selected sources of return, and pricing formulas provide reproducible values under specified conditions.

The validation task is to determine where those conditions approximate the intended use well enough and where additional treatment is required. A controlled benchmark is a foundation for that judgement, not a substitute for it.