First-Order Portfolio Loss Approximation and the Role of the Time Horizon
Summary
This explanation derives a first-order approximation of portfolio loss when value is modeled as a function of time and observable risk factors. Applying a Taylor expansion separates the change in portfolio value into a time derivative and exposures to changes in each risk factor. The apparent difference between two written formulas comes from notation: the subscript on the value function denotes its derivative with respect to its first, time, argument.
The answer also distinguishes the discrete time index from the actual horizon length. If each step represents a duration Δ, value can be written using elapsed time tΔ, so the time contribution is scaled by Δ. For a sufficiently short horizon, the answer says this term is often small and omitted in practice. This is a local linear approximation; the discussion does not quantify approximation error or address nonlinear exposure effects captured by higher-order terms.
Key ideas
- A first-order Taylor expansion approximates portfolio value changes using time and risk-factor sensitivities.
- The time derivative notation refers to differentiation with respect to the value function’s time argument.
- When each step has duration Δ, the time contribution is scaled by that horizon length.
- The time term may be omitted for short horizons when it is small, while nonlinear effects remain outside the first-order approximation.
Tags
Full text
# Intuition behind the first-order approximation of loss of a portfolio given finite observable risk factors
# Intuition behind the first-order approximation of loss of a portfolio given finite observable risk factors
I am trying to get an intuition behind the first-order approximation, $L_{t+1}^\Delta$ of the loss of a portfolio, $L_{t+1}$ from time $t$ to $t+1$ defined as $$L_{t+1}=-[V_{t+1}-V_t]$$ $$L_{t+1}^\Delta=-\left[f_t(t,Z_t)+\sum_{j=1}^df_{Z_j}(t,Z_t)\cdot X_{t+1,j}\right]$$ where, $V_t$ is the value of the portfolio at time $t$ (notations in $L^\Delta_t$ are defined below). We model $V_t$ as a function of time and $d$ risk factors, $Z_t=(Z_{t,1},Z_{t,2},..Z_{t,d})$ (assumed observable at time $t$) i.e., $$V_t=f(t,Z_t).$$ Hence,$$L_{t+1}=-[f(t+1,Z_{t+1})-f(t,Z_t)].$$
Now, using Taylor's first order approximation formula on $f$, we can write $$f(t+1,Z_{t+1})\approx f(t,Z_t)+\frac{\partial f}{\partial t}(t,Z_t)\cdot(t+1-t)+\sum_{j=1}^d \frac{\partial f}{\partial Z_{t,j}}(t,Z_t)\cdot (Z_{t+1,j}-Z_{t,j})$$ and setting risk-factor changes $X_{t+1,j}:=Z_{t+1,j}-Z_{t,j}$ and partial derivatives of $f$ wrt $(j+1)$th component, $f_{Z_j}:=\frac{\partial f}{\partial Z_{t,j}}$, $j=1,...d$, we have $$f(t+1,Z_{t+1})\approx f(t,Z_t)+\frac{\partial f}{\partial t}(t,Z_t)+\sum_{j=1}^d f_{Z_j}(t,Z_t)\cdot X_{t+1,j}.$$
Using this we get $$L_{t+1}\approx -\left[\frac{\partial f}{\partial t}(t,Z_t)+\sum_{j=1}^d f_{Z_j}(t,Z_t)\cdot X_{t+1,j}\right]$$ whereas, $$L_{t+1}^\Delta=-\left[f_t(t,Z_t)+\sum_{j=1}^df_{Z_j}(t,Z_t)\cdot X_{t+1,j}\right]$$
Why is there a difference? Are there any assumptions that will lead us to the same equation? Is my calculation incorrect?
## Answer by zaira (score 1, accepted)
https://quant.stackexchange.com/a/81602
$f_t$ is the derivative of $f$ wrt the first component. The approximation is hence, correct.
In practice, some specific unit of time (horizon length) $\Delta$ is taken. Portfolio value is defined using the mapping $g(s,Z)$ as $$V_t=f(t,Z_t)=g(t\Delta,Z_t)$$ and using first-order Taylor approximation, as above we get $$L_{t+1}\approx-\left[\Delta\cdot g_s(t\Delta,Z_t)+\sum_{j=1}^d g_{z_j}(t\Delta,Z_t)\cdot X_{t+1,j}\right]$$ If $\Delta$ is a short time horizon then $g_s(t\Delta,Z_t)$ is small and dropped in practice.Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.