Quantitative Finance

Risk, portfolios, CAPM, Kelly and option pricing.

Quantitative Finance

This note covers returns and risk, mean–variance portfolios, the capital asset pricing model, position sizing, and option pricing.

Returns and risk

Return – The relative change in price from P0P_0 to P1P_1. Simple returns R=P1/P0−1R = P_1/P_0 - 1 combine across assets as a weighted average. Log returns r=ln⁡(P1/P0)r = \ln(P_1/P_0) add up across time periods.

Volatility – The standard deviation σ\sigma of returns. Over TT periods with independent returns, the mean scales with TT and the volatility with T\sqrt T. There are about 252 trading days a year, so annual volatility ≈ daily volatility × 15.9.

The expected log return is about μ−σ2/2\mu - \sigma^2/2, so volatility drags down long-run growth even when the average simple return μ\mu is unchanged.

Risk-free rate – The return rfr_f on a riskless asset such as short-term government debt. An excess return is a return minus rfr_f.

Sharpe ratio – Excess return per unit of risk:

SR=μ−rfσ\text{SR} = \frac{\mu - r_f}{\sigma}

The information ratio applies the same idea to returns relative to a benchmark. The fundamental law of active management says IR≈ICbreadth\text{IR} \approx \text{IC}\sqrt{\text{breadth}}, where IC is the correlation between forecasts and outcomes and breadth is the number of independent bets.

Value at risk (VaR) and expected shortfall – VaR at level 95% is the loss exceeded only 5% of the time. Expected shortfall is the average loss in that worst 5%. Unlike VaR, expected shortfall never penalizes diversification (it is subadditive).

Mean–variance portfolios

Portfolio – A vector of weights ww over NN assets with ∑iwi=1\sum_i w_i = 1. Weights may be negative, which means short selling. Write μ\mu for the vector of expected returns and Σ\Sigma for the covariance matrix of returns, where Σij=Cov⁡(Ri,Rj)\Sigma_{ij} = \operatorname{Cov}(R_i, R_j). Then

μp=w⊤μ,σp2=w⊤Σw\mu_p = w^\top\mu, \qquad \sigma_p^2 = w^\top\Sigma w

Diversification – With two assets, σp2=w12σ12+w22σ22+2w1w2ρσ1σ2\sigma_p^2 = w_1^2\sigma_1^2 + w_2^2\sigma_2^2 + 2w_1w_2\rho\sigma_1\sigma_2, where ρ\rho is their correlation. Whenever ρ\rho is less than 1, mixing the two lowers risk below the weighted average of their volatilities.

Efficient frontier – The portfolios with the lowest variance for each target return mm (mean–variance analysis, Markowitz). They solve

min⁡w w⊤Σwsubject tow⊤μ=m,  w⊤1=1\min_w\ w^\top\Sigma w \quad \text{subject to} \quad w^\top\mu = m,\ \ w^\top\mathbf{1} = 1

Setting the gradient of the Lagrangian to zero (with multipliers λ1,λ2\lambda_1, \lambda_2 for the two constraints) gives a linear system:

[2Σ−μ−1μ⊤001⊤00][wλ1λ2]=[0m1]\begin{bmatrix}2\Sigma & -\mu & -\mathbf{1}\\ \mu^\top & 0 & 0\\ \mathbf{1}^\top & 0 & 0\end{bmatrix}\begin{bmatrix}w\\ \lambda_1\\ \lambda_2\end{bmatrix} = \begin{bmatrix}0\\ m\\ 1\end{bmatrix}

Plotted with volatility on the horizontal axis and return on the vertical, the frontier is the upper half of a hyperbola. An equivalent penalized form traces the same curve as the risk tolerance q≥0q \ge 0 varies: q=0q = 0 gives the minimum-variance portfolio, and larger qq moves further along the frontier.

min⁡w w⊤Σw−q μ⊤w\min_w\ w^\top\Sigma w - q\,\mu^\top w

Special portfolios –

PortfolioWhat it maximizes or minimizesWeights (before normalizing to sum to 1)
Minimum variancelowest σp\sigma_pΣ−11\Sigma^{-1}\mathbf{1}
Tangencyhighest Sharpe ratioΣ−1(μ−rf1)\Sigma^{-1}(\mu - r_f\mathbf{1})
Mean–variance optimalμ⊤w−γ2w⊤Σw\mu^\top w - \tfrac{\gamma}{2}w^\top\Sigma w for risk aversion γ\gammaΣ−1μ/γ\Sigma^{-1}\mu/\gamma, no normalization

Capital allocation line – Once a risk-free asset is available, the best risky portfolio to hold is the tangency portfolio. Every efficient portfolio is then a mix of the risk-free asset and the tangency portfolio (two-fund separation), lying on the straight line from rfr_f tangent to the frontier. Its slope is the tangency portfolio's Sharpe ratio.

The estimated Σ−1\Sigma^{-1} amplifies estimation noise. In practice Σ\Sigma is shrunk toward a simpler target (Ledoit–Wolf), modelled with factors, or the weights are constrained.

Capital asset pricing model and factors

Beta – How strongly an asset moves with the market mm. It is the slope from regressing the asset's excess return on the market's excess return. The intercept of that regression, α\alpha, is the return not explained by the market.

βi=Cov⁡(Ri,Rm)Var⁡Rm\beta_i = \frac{\operatorname{Cov}(R_i, R_m)}{\operatorname{Var}R_m}

Capital asset pricing model (CAPM) – In equilibrium, only market risk is rewarded:

ERi−rf=βi(ERm−rf)\mathbb{E}R_i - r_f = \beta_i(\mathbb{E}R_m - r_f)

Factor model – Several common drivers FkF_k (such as market, size and value in Fama–French) with loadings βik\beta_{ik}, plus independent idiosyncratic noise εi\varepsilon_i. Writing BB for the matrix of loadings, ΣF\Sigma_F for the factor covariance and DD for the diagonal matrix of idiosyncratic variances:

Ri=αi+∑kβikFk+εi⇒Σ=BΣFB⊤+DR_i = \alpha_i + \sum_k\beta_{ik}F_k + \varepsilon_i \quad \Rightarrow \quad \Sigma = B\Sigma_FB^\top + D

Kelly criterion

Kelly criterion – Choose the fraction ff of wealth to bet that maximizes the expected log of wealth, which maximizes long-run growth. For a bet paying bb to 1 that wins with probability pp:

f∗=p−1−pbf^* = p - \frac{1-p}{b}

In continuous time, for an asset with drift μ\mu and volatility σ\sigma, f∗=(μ−rf)/σ2f^* = (\mu - r_f)/\sigma^2, which gives growth rate rf+(μ−rf)22σ2r_f + \frac{(\mu - r_f)^2}{2\sigma^2}.

Betting twice Kelly brings growth back down to the risk-free rate. Many practitioners bet half Kelly to cut variance.

Option pricing

Option – A European call (put) gives the right, but not the obligation, to buy (sell) an asset at the strike price KK at the maturity date TT. Its payoff is max⁡(ST−K,0)\max(S_T - K, 0) (respectively max⁡(K−ST,0)\max(K - S_T, 0)), where StS_t is the asset price at time tt.

No arbitrage and the risk-neutral measure – If no riskless profit is available, every price equals its discounted expected payoff under an adjusted probability measure Q\mathbb{Q} (the risk-neutral measure). Under Q\mathbb{Q}, every asset grows on average at the risk-free rate rr.

V0=e−rT EQ[payoff]V_0 = e^{-rT}\,\mathbb{E}^{\mathbb{Q}}[\text{payoff}]

Put–call parity – Owning a call and selling a put has the same payoff as owning the stock and borrowing KK:

C−P=S0−Ke−rTC - P = S_0 - Ke^{-rT}

Black–Scholes formula – This assumes the price follows a geometric Brownian motion (normally distributed log returns) with constant volatility σ\sigma and rate rr, and that the position can be hedged continuously. With N(⋅)N(\cdot) the standard normal cumulative distribution function:

C=S0N(d1)−Ke−rTN(d2),d1,2=ln⁡(S0/K)+(r±12σ2)TσTC = S_0N(d_1) - Ke^{-rT}N(d_2), \qquad d_{1,2} = \frac{\ln(S_0/K) + (r \pm \tfrac12\sigma^2)T}{\sigma\sqrt T}

The price of any such option V(t,S)V(t, S) satisfies Vt+12σ2S2VSS+rSVS−rV=0V_t + \tfrac12\sigma^2S^2V_{SS} + rSV_S - rV = 0.

An at-the-money call (K=S0K = S_0) is worth about 0.4 S0σT0.4\,S_0\sigma\sqrt T.

Binomial tree – A discrete approximation in which each step of length Δt\Delta t moves the price up by a factor u=eσΔtu = e^{\sigma\sqrt{\Delta t}} or down by d=1/ud = 1/u. Prices are computed backwards through the tree using the risk-neutral up probability

q=erΔt−du−dq = \frac{e^{r\Delta t} - d}{u - d}

Greeks – The sensitivities of an option's price VV, with ϕ\phi the standard normal density:

GreekSensitivity toBlack–Scholes call
Delta Δ\Deltaprice, ∂V/∂S\partial V/\partial SN(d1)N(d_1)
Gamma Γ\Gammaprice, second order, ∂2V/∂S2\partial^2V/\partial S^2ϕ(d1)/(SσT)\phi(d_1)/(S\sigma\sqrt T)
Vegavolatility, ∂V/∂σ\partial V/\partial\sigmaSϕ(d1)TS\phi(d_1)\sqrt T
Theta Θ\Thetapassage of time, ∂V/∂t\partial V/\partial tusually negative

Delta hedging – Holding −Δ-\Delta shares against a long option cancels small price moves. What remains earns 12ΓS2(σrealized2−σimplied2) dt\tfrac12\Gamma S^2(\sigma_{\text{realized}}^2 - \sigma_{\text{implied}}^2)\,dt, so a hedged option is a bet on volatility. Here σimplied\sigma_{\text{implied}} is the volatility that makes the Black–Scholes price match the market price.