FRM Exam Part II · Future Value and Exposure

Monte Carlo Simulation for Counterparty Exposure Modelling

Updated 11 October 2026 · Fact-checked

Monte Carlo exposure simulation generates many random paths for market risk factors, revalues every trade in the netting set at each future date, applies netting and collateral, and records the exposure max(V, 0). Averaging across paths gives EE; the chosen percentile gives PFE.

Understand Monte Carlo Simulation for Exposure Modelling

Exposure is the amount you would lose if a counterparty defaulted today or on a future date. Future exposure is uncertain because market prices move. You cannot solve it with one formula for a whole portfolio of swaps, options and FX forwards, so banks simulate it.

The process has four parts. First, choose the risk factors (interest rates, FX rates, equity prices, credit spreads) and a model for how each moves, such as geometric Brownian motion for FX or Hull-White for rates. Second, simulate thousands of paths for those factors over a grid of future dates. Third, on each path and date, revalue every trade, net the values within each netting set, and apply collateral terms. Fourth, take exposure = max(net value, 0) and summarise across paths: the mean is expected exposure (EE) and a high percentile (for example 95% or 97.5%) is potential future exposure (PFE).

The key modelling choice is the measure. Under the risk-neutral measure, drifts are set so that discounted assets are martingales, and parameters are calibrated to market prices (implied volatilities, current curves). This is the right choice for pricing, so CVA uses it. Under the real-world (physical) measure, drifts and volatilities are estimated from historical data and reflect expected returns, including risk premia. This is the right choice for risk management: PFE, limits and capital-style measures.

Calibration links the two. Historical calibration uses past data, is stable and fits the real world, but looks backward. Implied calibration uses current option prices, is forward-looking and consistent with pricing, but embeds risk premia and may be thin for long maturities. Using the wrong measure changes the numbers. Real-world drift and volatility typically differ from risk-neutral ones, so the exposure profile differs.

Two practical points matter. Trades with early exercise or path dependence need extra care, because you must value them on every future date and path, which is costly. Banks often use regression (such as least-squares Monte Carlo) or approximations to do this. Also, simulation must capture correlations between risk factors, or netting benefits will be wrong.

Key formulas to remember

Exposure on a path and date: E(t) = max(V(t), 0); V is the net value of the netting set after collateral terms are applied. Negative value is not a loss to you.
Expected exposure (EE): EE(t) = (1 ÷ N) × Σ max(V_i(t), 0), summed over paths i = 1 to N; Average of positive exposures across simulated paths at date t.
Potential future exposure (PFE): PFE_α(t) = the α-percentile of the distribution of E(t); Typically 95% or 97.5%. Pick the percentile from the sorted simulated exposures.
Expected positive exposure (EPE): EPE = time-weighted average of EE(t) over the horizon; A single-number average of the EE profile, used in capital and CVA contexts.
GBM step (risk-neutral FX or equity): S(t+Δt) = S(t) × exp[(r − q − σ²÷2)Δt + σ√Δt × Z]; Z is a standard normal draw. Under the real-world measure, replace (r − q) with the expected return μ.
Monte Carlo standard error: SE ≈ s ÷ √N; s is the sample standard deviation. Quadrupling paths halves the error.

How to solve Monte Carlo Simulation for Exposure Modelling questions

Use this order for any question on simulating exposure, whether it is conceptual or numerical.

1Identify the purpose. Pricing or CVA means risk-neutral. Risk limits, PFE or stress means real-world.
2List the risk factors and the model for each, and check that correlations are included.
3Check calibration: historical data for real-world parameters, market-implied prices for risk-neutral parameters.
4Simulate paths on a time grid that includes trade cash-flow dates and margin dates.
5Revalue each trade on each path and date, then net within the netting set and apply collateral.
6Take max(V, 0) per path and date, then compute EE as the average and PFE as the percentile.
7Read the profile: peak date, shape by product, and effect of netting and collateral.
8Check accuracy: number of paths, standard error and model risk.

Quickest way: Measure, then mean or percentile

When to use it: Use when an MCQ gives a short list of simulated values or asks which measure or calibration fits a stated purpose.

Decide the measure from the purpose: pricing or CVA means risk-neutral, risk limits or PFE means real-world.
Set any negative net value to zero before doing anything else.
For EE, add the positive values and divide by the number of paths.
For PFE, sort the values and pick the percentile position (for 95% of 100 paths, around the 95th ranked value).
Eliminate options that mix measures, such as historical drift used for CVA pricing.

Common mistakes in Monte Carlo Simulation for Exposure Modelling

Using the risk-neutral measure for PFE.
Candidates assume one model serves all purposes.
Fix: Remember that PFE is a real-world risk measure. Risk-neutral is for pricing and CVA.
Averaging negative values into EE.
They average the raw trade values instead of the exposure.
Fix: Apply max(V, 0) to the netted value on each path first, then average.
Taking the percentile of the trade values before netting.
They forget that netting is applied within the netting set on each path.
Fix: Net on each path, floor at zero, and only then compute the percentile.
Treating implied calibration as free of risk premia.
Market prices look objective.
Fix: Implied parameters are risk-neutral and embed risk premia, so they can differ from real-world estimates.
Ignoring correlation between risk factors.
Each factor is simulated on its own for simplicity.
Fix: Simulate with a joint correlation structure, or netting and cross-asset exposure will be misstated.
Thinking more paths removes model risk.
Candidates confuse sampling error with model error.
Fix: More paths reduce standard error (s ÷ √N) only. A wrong model or calibration stays wrong.

Worked examples

Example 1

A bank simulates the net value of a netting set at one future date on five paths, in USD million: +8, −3, +12, 0, −5. What are EE and the exposure on each path?

Show the solution

Apply max(V, 0): 8, 0, 12, 0, 0.
Sum the exposures: 8 + 12 = 20.
Divide by 5 paths: 20 ÷ 5 = 4.

Answer: Exposures are 8, 0, 12, 0 and 0, so EE = USD 4 million. The raw average of values would be 2.4, which is wrong because it ignores flooring at zero.

Example 2

A risk team must produce (a) a CVA number for a swap book and (b) a 97.5% PFE for limit monitoring. Which measure and calibration should each use, and why?

Show the solution

CVA is a price. It must be consistent with market prices, so use the risk-neutral measure.
Calibrate the risk-neutral model to current market quotes, such as the yield curve and implied volatilities.
PFE is a risk estimate of what could happen. Use the real-world measure.
Calibrate the real-world model to historical data for drift and volatility, with attention to stress periods.
Take the 97.5th percentile of the simulated exposure distribution at each date.

Answer: (a) CVA: risk-neutral, implied calibration. (b) PFE: real-world, historical calibration, 97.5th percentile of exposure at each date.

Exam tips

Link the measure to the purpose in one line: pricing is risk-neutral, risk management is real-world.
Watch for a distractor that uses historical volatility in a pricing context or implied drift in a PFE context.
If numbers are given, floor at zero first, then average for EE or rank for PFE.
Know the strengths and weaknesses of historical versus implied calibration in a sentence each.
Remember that more simulations cut sampling error, not model error.

Practice questions from Future Value and Exposure

Monte Carlo Simulation for Exposure Modelling in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Monte Carlo Simulation for Exposure Modelling: frequently asked questions

What is the difference between risk-neutral and real-world measures in exposure simulation?

The risk-neutral measure sets drifts so that discounted assets have no excess return, and calibrates to market prices. The real-world measure uses historical estimates of drift and volatility. Use risk-neutral for pricing and CVA, and real-world for PFE and risk limits.

How do you calculate PFE using simulation?

Simulate risk factor paths, revalue and net the trades on each path and date, and floor at zero. For each date, sort the exposures and take the chosen percentile, such as 95% or 97.5%. Plot these across dates to get the PFE profile.

Is historical or implied calibration better?

Neither is better in every case. Historical is stable and suits real-world risk measures but is backward-looking. Implied is forward-looking and suits pricing but includes risk premia and may lack data for long maturities.

Why is exposure floored at zero?

If the netted value is negative, you owe the counterparty, so a default does not cause you a loss on that path. Your exposure is therefore max(V, 0).