Learning Outcome Statement
Estimate and interpret VaR under the parametric, historical simulation, and Monte Carlo simulation methods.
On the exam: The stem gives a portfolio value, horizon, confidence level, and either parametric inputs or empirical or simulated tail information, then asks you to estimate VaR, interpret it, or compare method-consistent results in an MCQ or item set.
You are not forecasting the exact loss. You are locating a loss threshold for a stated horizon and confidence level. The portfolio value today, $V_0$, converts a percentage loss into a positive dollar VaR, while the parametric method uses daily mean return $\mu_d$, daily volatility $\sigma_d$, and adverse standard-normal cutoff $z_c$ at confidence level $c$ for the one-day case. Historical simulation supplies $L_{c,H}$ from observed returns, while Monte Carlo supplies $L_{c,MC}$ after simulated risk-factor scenarios reprice the portfolio.
| Method | Supplied input | Tail source |
|---|---|---|
| Parametric | $\mu_d$, $\sigma_d$, $z_c$ | Normal return model |
| Historical | $L_{c,H}$ | Observed portfolio returns |
| Monte Carlo | $L_{c,MC}$ | Simulated scenarios and repricing |
The parametric method creates the adverse return from a distributional assumption. Each simulation method takes its tail loss from the stated data-generating process, so the source in the stem tells you which formula belongs.
$$ \begin{aligned} \text{Parametric}\quad \mathrm{VaR}_c &= V_0(z_c\sigma_d-\mu_d) \ \text{Historical}\quad \mathrm{VaR}c &= V_0L{c,H} \ \text{Monte Carlo}\quad \mathrm{VaR}c &= V_0L{c,MC} \end{aligned} $$
$V_0$ scales a loss rate into dollars. The worked horizon is one day, so daily inputs enter directly. In the parametric line, $z_c\sigma_d$ pushes VaR up as the adverse cutoff or volatility rises, while a positive $\mu_d$ pushes it down because expected return offsets loss. A higher confidence level usually moves farther into the adverse tail. In the simulation lines, $L_{c,H}$ and $L_{c,MC}$ already contain the tail loss, so a z-score does not belong there. The amount is a threshold at the stated horizon, not the maximum possible loss.
Use this path when the stem mixes model inputs and tail losses. First identify whether the question gives normal-return parameters, observed portfolio returns, or simulated scenarios. That choice determines where the tail estimate comes from before you do any arithmetic. For Monte Carlo, remember that scenarios must reprice the portfolio; for the other methods, follow the input source stated in the question. Then check the horizon and express the selected loss threshold in dollars.
Loading diagram…
Worked example
You are handed a portfolio and the relevant model inputs for a one-day risk report. The question asks you to estimate VaR at the stated confidence levels under each method and interpret each dollar threshold from the investor's perspective. The returns, volatility, and loss rates are stated as decimal fractions.
| Given | Value |
|---|---|
| Portfolio value | USD 10,000,000 |
| Horizon | 1 day |
| Confidence levels | 95% and 99% |
| Daily mean return | 0.0002 |
| Daily volatility | 0.012 |
| Parametric cutoffs | 1.645 and 2.326 |
| Historical tail losses | 0.018 and 0.027 |
| Monte Carlo tail losses | 0.019 and 0.029 |
The line shows parametric VaR as confidence moves from 95% to 99%. The table below keeps the historical and Monte Carlo estimates beside it. Read the chart as a comparison of two thresholds for the same portfolio, horizon, mean return, and volatility. The higher confidence level uses the larger adverse cutoff, so its parametric estimate is higher. This pattern follows from the formula; it does not imply that either amount is a forecast of the exact next-day loss. The next table lets you compare that model result with both supplied simulation tails.
Loading diagram…
Compare the estimates by method at each confidence level, rather than treating the chart as a ranking of all three methods. Parametric VaR comes from the stated mean, volatility, and cutoff. Historical VaR scales the observed tail loss, while Monte Carlo VaR scales the simulated tail loss after repricing. The results can differ because their tail sources differ; use the method named in the question and interpret each figure at the one-day horizon.
| Method | 95% VaR | 99% VaR |
|---|---|---|
| Parametric | USD 195,400 | USD 277,120 |
| Historical | USD 180,000 | USD 270,000 |
| Monte Carlo | USD 190,000 | USD 290,000 |
-
Use the normal-return inputs because the question asks for parametric VaR. The 95% calculation is $10{,}000{,}000(1.645 \times 0.012 - 0.0002)=195{,}400$, and the 99% calculation is $10{,}000{,}000(2.326 \times 0.012 - 0.0002)=277{,}120$.
-
Use each observed tail loss directly because historical simulation already provides the empirical quantile. The calculations are $10{,}000{,}000 \times 0.018=180{,}000$ and $10{,}000{,}000 \times 0.027=270{,}000$.
-
Use each simulated tail loss directly because Monte Carlo has already repriced the portfolio under simulated risk-factor scenarios. The calculations are $10{,}000{,}000 \times 0.019=190{,}000$ and $10{,}000{,}000 \times 0.029=290{,}000$.
The investor should use the threshold tied to the selected method, confidence level, and horizon, and should not describe it as the maximum possible loss.
Changed condition
| Method at 95% | Before | After |
|---|---|---|
| Parametric | USD 195,400 | USD 234,480 |
| Historical | USD 180,000 | USD 216,000 |
| Monte Carlo | USD 190,000 | USD 228,000 |
Only the portfolio value changes from USD 10,000,000 to USD 12,000,000. Because each tail rate stays fixed, you should predict before reading the after column that every dollar VaR rises with portfolio value. Before reading on, calculate the 99% historical VaR for the larger portfolio yourself.
Common error
You may apply the parametric normal-distribution formula to every method because it is the quickest route. It feels right because the formula is familiar, and every answer eventually becomes a dollar loss. The stem supplies empirical or simulated tail losses, so importing a z-score is unjustified.
Check yourself
Practice
An investor reviews a one-day VaR report for a USD 10,000,000 portfolio. Daily mean return is 0.0002, daily volatility is 0.012, and the parametric cutoffs are 1.645 at 95% and 2.326 at 99%. Observed tail losses are 0.018 and 0.027. Simulated Monte Carlo tail losses are 0.019 and 0.029.
Question 1
Estimate the one-day parametric VaR at 95% confidence using the parametric inputs.
Question 2
A one-day 95% VaR estimate is USD 195,400. Which interpretation is correct?
Question 3
The stem gives an observed portfolio-return sample and an empirical tail-loss rate of 0.018 at 95% confidence. Which method matches that information?
Question 4
Calculate Monte Carlo VaR at 95% confidence using the simulated tail-loss rate of 0.019.
A portfolio's value rises from USD 10,000,000 to USD 12,000,000 while every return, volatility, confidence, and tail-loss input remains unchanged. What is the 95% parametric VaR?
For the base portfolio at 99% confidence, which ordering of VaR estimates is correct from highest to lowest?
Revision snapshot
For the one-day LOS parametric case, use
$$ \mathrm{VaR}_c=V_0(z_c\sigma_d-\mu_d) $$
Parametric VaR uses portfolio value multiplied by the adverse standardized return threshold. Historical simulation uses the empirical upper-tail loss from observed portfolio returns. Monte Carlo simulation uses the upper-tail loss from simulated risk-factor scenarios and portfolio repricing. VaR is a loss threshold at a stated confidence level and horizon, not the maximum possible loss. On the exam, use a z-score only for supplied parametric inputs. Use empirical or simulated tail losses directly for the matching simulation method.
Unlock the full lesson and continue reading
Signup to continue reading this lesson and unlimited access questions, flashcards, AI notes, and more