Skip to content

FRM Exam Part II · Non-parametric Approaches

Strengths and Weaknesses of Non-parametric VaR Methods

Updated 11 October 2026 · Fact-checked

Non-parametric VaR methods, such as historical simulation, estimate risk straight from past data without assuming a distribution. Strengths: simple, captures fat tails and skew in the sample. Weaknesses: depend on the window, create ghost effects, react slowly, and cannot say much beyond the worst observed loss.

Understand Strengths and Weaknesses of Non-parametric Methods

A parametric method assumes returns follow a known shape, usually normal. You estimate a few parameters, such as mean and standard deviation, and read VaR from the formula. A non-parametric method makes no such assumption. It lets the historical data speak and takes the VaR as a quantile of the observed losses.

This is the main strength. If the data show fat tails, skew or odd correlations, the method picks them up automatically. There is no need to estimate a covariance matrix or fit a distribution. It is easy to explain to a board and works with non-linear positions because you revalue the portfolio under each past scenario.

The main weakness is that everything depends on the sample. The past must be representative of the future. If the window holds only calm days, VaR will be too low. If it holds a crisis, VaR may stay high long after markets settle. So window length is a trade-off. A long window gives more data and more precise quantiles, but it includes stale and irrelevant observations and reacts slowly. A short window is more responsive, but it gives fewer observations in the tail and noisier estimates.

A ghost effect (or shadow effect) arises because every observation in the window has equal weight and then drops out suddenly. When a large loss enters the window, VaR jumps. Exactly one window length later, that loss leaves and VaR jumps down, though nothing happened in the market that day. The VaR is distorted by an event that is no longer news.

Finally, non-parametric methods are weak in the extremes. You cannot estimate a loss beyond the worst one in your sample. At high confidence levels, such as 99.9%, there are very few or no observations in the tail, so the estimate is unreliable. Quantile estimates also have large standard errors there. Extreme value theory or weighted and kernel-based refinements are used to address these limits.

Key formulas to remember

Historical simulation VaR
VaR at confidence c = the loss at the (1 − c) quantile of the sorted P&L sample
With n observations at 99%, the VaR is near the (0.01 × n)th worst loss. Interpolation conventions vary, so follow the question.
Expected tail observations
Observations beyond VaR ≈ n × (1 − c)
For n = 500 at 99%, only about 5 observations lie in the tail. This shows why precision is poor.
Ghost effect timing
An observation affects VaR for exactly n days, then drops out
n is the window length. The VaR jump occurs on entry and again on exit.
Equal weighting
Each observation has weight 1 ÷ n
This is the standard historical simulation. Weighted variants replace it with declining weights.

How to solve Strengths and Weaknesses of Non-parametric Methods questions

For any question comparing non-parametric and parametric methods, or asking about a weakness, use this routine.

  1. 1Identify the method: plain historical simulation, bootstrap, weighted, kernel, or a parametric benchmark.
  2. 2Note what the question stresses: distribution assumption, data, window length, ghost effect, or extremes.
  3. 3State the relevant strength or weakness in one sentence, with the cause. For example: equal weights cause the ghost effect.
  4. 4If numbers are given, compute n × (1 − c) to see how many tail observations exist, or locate the quantile in the sorted data.
  5. 5Link the cause to the effect on VaR: too low in calm windows, too high after crises, or a sudden jump when an observation exits.
  6. 6Check each answer option for overstatement. Words like 'always' or 'eliminates' are usually wrong.
  7. 7Pick the option that matches both cause and effect.

Quickest way: Cause-and-effect matching

When to use it: Use this for conceptual MCQs where you have under a minute per question.

  1. No distribution assumption: strength. Captures fat tails and skew in the sample.
  2. Window length: long means stale and slow; short means noisy and few tail points.
  3. Equal weights: ghost effect, with VaR jumps on entry and exit.
  4. Beyond the sample: no estimate worse than the worst observed loss, and weak at very high confidence.
  5. Pick the option with the right cause. Reject any that claim the method is free of data dependence.

Common mistakes in Strengths and Weaknesses of Non-parametric Methods

  • Saying non-parametric methods make no assumptions at all.

    The name suggests it.

    Fix: They make no distribution assumption, but they assume the past is representative and that returns are drawn from a stable process.

  • Believing a longer window is always better.

    More data feels more accurate.

    Fix: Longer windows give more tail points but include stale data and respond slowly. It is a trade-off.

  • Explaining the ghost effect as a market event.

    The VaR jump looks like news.

    Fix: It is a statistical artifact of equal weights. An old observation drops out of the window and VaR changes though nothing new happened.

  • Thinking historical simulation can estimate losses worse than any in the sample.

    Confusing it with a fitted distribution that has an unbounded tail.

    Fix: The method only reproduces observed losses. For extremes, use EVT or a parametric tail model.

  • Claiming non-parametric VaR adapts quickly to volatility changes.

    It uses real data, so it seems responsive.

    Fix: With equal weights it reacts slowly to new volatility. Weighted schemes fix some of this.

Worked examples

Example 1

A bank uses 250-day historical simulation at 99% confidence. A one-day loss of USD 12 million enters the window on day 1 and is the largest in the window. Explain what happens to VaR on day 1 and on day 251, assuming no other large losses.

Show the solution
  1. On day 1, the loss enters the sample and becomes the worst observation. VaR rises if it falls into the tail used for the 99% quantile.
  2. For 250 days it remains in the window with weight 1 ÷ 250.
  3. On day 251 the observation leaves the window, because only the latest 250 days are used.
  4. VaR falls abruptly on day 251 though no new market information arrived.
  5. This is the ghost effect, caused by equal weights and a hard cutoff.

Answer: VaR jumps up on day 1 and drops sharply on day 251. The drop is a ghost effect, not a change in market risk.

Example 2

A risk manager has 500 daily P&L observations and wants 99.9% historical simulation VaR. Comment on reliability.

Show the solution
  1. Expected tail observations = 500 × (1 − 0.999) = 500 × 0.001 = 0.5.
  2. Fewer than one observation is expected beyond the 99.9% quantile.
  3. So the quantile is determined by the single worst loss or by interpolation, with high sampling error.
  4. The method cannot estimate any loss worse than the worst in the sample.
  5. A better approach for this confidence level is extreme value theory or a larger sample.

Answer: There are only about 0.5 expected tail observations, so the 99.9% VaR is unreliable and limited by the worst observed loss. Use EVT or more data.

Exam tips

  • Match each weakness to its cause: ghost effect to equal weights, slowness to long windows, extreme-loss limits to sparse tails.
  • Be careful with words like 'always', 'never' and 'eliminates'. Non-parametric methods have trade-offs, not absolutes.
  • If a question gives n and the confidence level, compute n × (1 − c) first. It tells you how reliable the tail is.
  • Know the standard fixes: weighted historical simulation for ghost effects and slow adaptation, EVT for extremes, bootstrap for precision.

Practice questions from Non-parametric Approaches

Strengths and Weaknesses of Non-parametric Methods in other exams

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

Strengths and Weaknesses of Non-parametric Methods: frequently asked questions

What is the difference between parametric and non-parametric VaR?

Parametric VaR assumes a distribution, such as normal, and computes VaR from its parameters. Non-parametric VaR takes the quantile of observed historical losses. Parametric is smooth but can understate fat tails; non-parametric reflects the sample but depends on it.

What is the ghost effect in historical simulation?

It is the sudden change in VaR when a large past observation drops out of the window. Equal weights make an old event count fully until it leaves. VaR then falls though nothing new happened.

How does window length affect non-parametric VaR?

A long window gives more tail data but includes stale observations and reacts slowly. A short window is more responsive but has fewer tail points, so estimates are noisier. Choose a compromise based on data and purpose.

Why are non-parametric methods weak for extreme events?

There are few or no observations far in the tail. The method cannot produce a loss larger than the worst in the sample. At very high confidence levels, extreme value theory is a better tool.