Exotic components
Exotic products are not quoted on the market: their price is only an estimate given by the model. An exotic component is useful to represent a specific risk driver of the product.
Termsheet (core components)
2Y USD 14% p.a. Autocallable Multi Reverse Convertible (55% European) on Microsoft, Oracle
- Barrier = 55%, European
- Strike = 65%
- Autocall level = 100%
Payoff
Below the barrier, the investor still receives the coupons (28). The put he shorted is exercised: he must buy $100/65$ shares of the worst performer at 65 (% of initial) while these are only worth $W$, i.e. a loss of $\frac{100}{65}(65 - W)$. Hence P&L $= 28 - \frac{100}{65}(65 - W) = 28 + 100 \cdot W/65 - 100$.
Investor exposure
The exposure can be described in two ways:
- Exposure to components where components are what the investor holds: they change the cash flows, hence they appear in the contract. It is calculated by pricing the product with and without the component, everything else equal.
- Sensitivities are how the price reacts when a market input changes: they do not change the contract. It is calculated like a derivative: the price is computed with a small change in the market input, everything else equal.
The sign gives the direction: positive means long, negative means short.
Exposures to components of the BRC:
- Long zero-coupon bond -> vanilla
- Short worst-of down-and-in put -> exotic
- Long coupons -> vanilla (replicable with Treasury STRIPS, i.e. a strip of zero-coupon Treasury bonds)
- Short autocall -> exotic
Sensitivities in the BRC:
- Long correlation -> in general, if the assets are uncorrelated, there are more chances that one of them falls and hits the barrier.
Calibration step
The calibration is done in 3 steps: market implied vols $\rightarrow$ SVI surface $\rightarrow$ local vol surface.
1/ Implied vol surface
For each maturity $T$, the forward $F(T)$ is deduced from the call-put parity (using near-ATM strikes):
\[C - P = e^{-rT}\big(F(T) - K\big) \quad \Rightarrow \quad F(T) = K + e^{rT}(C - P)\]The implied vol is then obtained by inverting the Black formula (European option on the forward):
\[C = e^{-rT}\big(F(T)\,\mathcal{N}(d_1) - K\,\mathcal{N}(d_2)\big)\]2/ SVI surface
The market IVs are only available at a discrete set of strikes and maturities. To get a continuous surface, we fit an SVI (Stochastic Volatility Inspired) parametrization on each maturity slice. SVI is just a “sophisticated” interpolation method linking all the market IVs: it has no dynamics, it only gives a smooth smile shape in log-moneyness $k = \log(K/F(T))$:
\[w(k) = a + b\Big(\rho\,(k - m) + \sqrt{(k - m)^2 + \sigma^2}\Big)\]with $w(k) = \sigma_{imp}^2(k)\,T$ the total implied variance and $(a, b, \rho, m, \sigma)$ the 5 parameters fitted to the market IVs of the slice.
Fitting means finding the parameters that make the curve as close as possible to the $n$ market points $(k_i, w_i^{mkt})$ of the slice, in the least-squares sense:
\[\min_{a,b,\rho,m,\sigma}\ \sum_{i=1}^{n}\Big(w(k_i;\,a,b,\rho,m,\sigma) - w_i^{mkt}\Big)^2\]Since $w$ is nonlinear in the parameters, there is no closed-form solution: the minimum is found with a numerical optimizer. With only 5 parameters, the curve does not go through every quote; it smooths out the noise.
3/ Local vol surface
The local vol $\sigma_{loc}(K,T)$ is found from the implied vol surface with the Dupire formula. Since SVI gives $w(k,T)$ as a smooth function, its derivatives in $k$ and $T$ can be computed and plugged into the Dupire formula written in terms of total implied variance:
\[\sigma_{loc}^2(k,T) = \frac{\dfrac{\partial w}{\partial T}}{1 - \dfrac{k}{w}\dfrac{\partial w}{\partial k} + \dfrac{1}{4}\Big(-\dfrac{1}{4} - \dfrac{1}{w} + \dfrac{k^2}{w^2}\Big)\Big(\dfrac{\partial w}{\partial k}\Big)^2 + \dfrac{1}{2}\dfrac{\partial^2 w}{\partial k^2}}\]This local vol surface is the one used in the SDE of the Monte Carlo simulation: $dS_t = (r - q)S_t\,dt + \sigma_{loc}(S_t, t)\,S_t\,dW_t$.
Note: at low spot, the local vol surface has holes. These come from our SVI fit/interpolation that is too simplistic (missing a no-arbitrage constraint).
Pricing steps
1/ Load the inputs
2/ Simulate correlated paths under local vol
- Use Cholesky to generate correlated returns
- Build the grid -> list of uniform (all trading dates + insertion of important dates like observation/calling dates)
- Simulate the price at each step using the local vol SDE
3/ Check the simulations
- Check martingale: $\mathbb{E}[S(t)]$ (simulation) should be equal to the $F(t)$ (deduced from Call-put parity)
4/ Compute the worst-of
- Observed at valuation dates
5/ Build the cashflows
- Cash flows are mostly principal repayment (if called) + coupon
- Cash flows are paid at payment dates
6/ Discount and average
- Use of the risk-free rate
7/ Describe the redemption profile
- Autocall probabilities: proportion of paths that triggers an autocall
- Barrier probabilities
-> computed in order to understand better the price
8/ Solve for the fair coupon at inception
- Based on the redemption price (price decomposition)
9/ Run sensitivities
- Stress test some parameters like correlation or funding spread