Bucketed Vega and Factor-Based P&L Attribution for Volatility Surfaces

# Defining and Calculating Vega PnL for Basket Options


# Defining and Calculating Vega PnL for Basket Options












### Defining and Calculating Vega PnL for Options Dependent on the Volatility Surface

I am working with exotic options, such as accumulators, whose value V depends on the entire volatility surface σ(K, T), encompassing both the term structure and the smile/skew across different strikes K and expiries T.

Conceptually, such an option might sometimes be viewed as a portfolio of vanilla options with various expiries T1, T2, ..., TN.

The challenge lies in accurately defining and calculating the Vega PnL when the volatility surface changes from σ1(K, T) to σ2(K, T). A naive calculation using a single "Total Vega" figure (e.g., `∂V / ∂σ` derived from a flat bump to the entire surface) multiplied by some average volatility change is generally insufficient. This is because the change Δσ(K, T) = σ2(K, T) - σ1(K, T) is typically not flat; it involves shifts, twists, and changes in the shape of the smile/skew across different expiries.

Furthermore, decomposing the exotic option into an equivalent static portfolio of vanilla options is often not feasible or accurate due to path-dependency, early exercise features, or other complexities. If such a decomposition were possible, one could calculate the Vega 𝒱i for each vanilla component i (at expiry Ti, strike Ki) and approximate the total Vega PnL by summing the individual PnLs: ∑i 𝒱i ⋅ Δσ(Ki, Ti). But, as mentioned, this approach is often impractical.

I have considered a couple of intuitive ideas for how to handle this when we might only have access to the total option value V and perhaps a single aggregate Vega sensitivity, or sensitivities to a few specific points on the surface:

- Equivalent Flat Volatility Shift: If the option's value V is monotonic with respect to overall volatility levels, could we find an "equivalent flat volatility" &barσ such that pricing with a flat surface σflat = &barσ yields the same value as pricing with the actual surface σ(K, T)? If so, could we then approximate the Vega PnL based on the change in this &barσ? This feels heuristic and relies on potentially strong monotonicity assumptions.

- Surface Discretization / Bucketed Vegas: Could we calculate local Vega sensitivities `∂V / ∂σ(Kj, Tk)` at specific grid points (Kj, Tk) on the surface (perhaps using finite differences on the input surface points)? The Vega PnL could then be approximated by summing the contributions from these points: ∑j,k `∂V / ∂σ(Kj, Tk)` ⋅ Δσ(Kj, Tk). This seems more plausible but might suffer from basis risk depending on the granularity of the grid and how Δσ is measured or interpolated at these points.

These approaches feel intuitive but lack formal rigor.

My questions are:

- How is the Vega PnL (or more accurately, Vega PnL attribution/decomposition) formally defined for options whose value is sensitive to the entire volatility surface σ(K, T)?

- What are the rigorous and practically accepted methodologies to calculate or approximate this Vega PnL, especially in scenarios where a clean decomposition into simpler instruments is not possible? How do practitioners typically handle this?

- Are there standard approaches for parameterizing the surface movement Δσ(K, T) (e.g., using Principal Component Analysis (PCA) on surface changes, or decomposing into factors like parallel shift, term-structure twist, smile steepness change) and attributing the Vega PnL to these factors?

- Could you please point me towards relevant literature (textbooks, papers) that discuss Vega PnL decomposition, volatility surface risk management, and PnL attribution for exotic options?

Thank you!

Suggested Tags: `options`, `greeks`, `volatility-surface`, `pnl-attribution`, `exotic-options`, `risk-management`

## Answer by Dimitri Vulis (score 2, accepted)

https://quant.stackexchange.com/a/82369

The usual approach to Taylor expansion for vega P&L is to do it two ways: with individual volatility points and with summary parameters, such as level, slope, and curvature.

You have a 2-dimensional implied volatility surface or even a 3-dimensional volatility cube. You calculate the first-order sensitivities of $V$ to each point on the surface or cube, as you wrote (bucketed vega).

To explain the P&L due to the change in volatilities, you multiply the first-order sensitivities to the changes in each volatility point.

For more complicated products, in order to reduce unexplained P&L, you may want to include second-order sensitivities to the volatilities; or some cross-gammas between different volatility points, or between volatilities and other model inputs such as underlying prices and interest rates, and passage of time.

If you define your volatilities using only a small number of parameters (such as level, slope, curvature) - you should calculate $V$'s sensitivities to these parameters for one Taylor expansion. However it is useful to also back out the volatilities at each point, and to calculate the sensitivities to each of these volatilities, and to display these sensitivities on a risk report; and to use these for the other Taylor expansion as well.

It is a good practice to run principal component analysis on the history of the volatility surface. The first 3 principal components are likely to look like level, slope, and curvature, and may or may not be similar to the parameters defining a parametric surface. You can calculate sensitivities to these historical PCs, and show every day how the change in the volatility surface is explained by the historical PC movements, and how much is unexplained, and how much P&L is attributable to each historical PC.

Related: PnL Explained Using Scenario(Full Reval Model) , Good references on PNL explain? , Vega of option with market derived parameters , Attributing change in option prices to greek components

A couple of recent papers describing other approaches to option P&L attribution: Olivier Daviaud. Rethinking P&L attribution for options. https://ssrn.com/abstract=4495530

Peter Carr, Liuren Wu. Option Profit and Loss Attribution and Pricing - a New Framework. The Journal of Finance 2020 DOI: 10.1111/jofi.12894 https://ssrn.com/abstract=3148796