Monte Carlo simulation is the foundation of institutional portfolio risk analysis. Instead of predicting a single outcome, it models thousands of possible futures — producing a probability distribution of where your portfolio could end up under varying market conditions.
Monte Carlo vs Stress Test. Monte Carlo is probabilistic path simulation: it draws many random return paths and reports a distribution. A stress test is deterministic: it applies one fixed shock vector and reports a single before/after impact. KlirInvest runs them as two separate engines; this article is about Monte Carlo only.
KlirInvest runs this kind of simulation on your recorded holdings, with its inputs, assumptions and limitations shown next to the result.
How Monte Carlo Simulation Works
- Define the portfolio: Current holdings, weights, and asset characteristics
- Calibrate the model: Historical returns, volatilities, and correlations for each holding
- Generate scenarios: Using statistical models, simulate thousands of possible return paths
- Analyse the distribution: Examine the range of outcomes — tail risk, median, and probabilities
Each simulation run generates a different sequence of returns using random sampling calibrated to the statistical properties of your holdings. Run 10,000 of these, and you get a rich picture of possible outcomes.
Two Common Monte Carlo Assumptions, and Where KlirInvest Stands
Most Monte Carlo implementations make two simplifying assumptions. KlirInvest's simulation makes the first by default and the second always; both are disclosed so you can judge the results:
1. Gaussian Returns
Standard models assume returns follow a normal distribution. Market data shows they do not. Returns exhibit fat tails (extreme events are far more common than Gaussian models predict) and negative skewness (large losses are more frequent and severe than large gains).
Explore Futures uses a Normal baseline by default, so it shares this assumption unless you switch on Tail Risk Mode, a normal location-scale mixture (df=5). Expected conditional diffusion variance is matched to Normal, but terminal log-return variance need not match because volatility and variance drag vary across paths. The comparison does not establish that either model predicts reality.
2. Static Correlations
Standard models use a single correlation matrix for all scenarios. In reality, correlations change — and they have tended to rise most during the market conditions where accurate modelling matters most, such as the 2008 financial crisis.
KlirInvest shares this assumption: it uses one Cholesky-decomposed covariance matrix estimated from your holdings' own return history (Ledoit-Wolf shrinkage on a 400-calendar-day window) for every path. It is used as measured: no market-regime adjustment is applied, so a crisis-level correlation appears in the simulation only if it is in that window. The separate stress test does not apply crisis correlations either: its scenarios state fixed shocks per asset class, and the headline is the sum of holding-level impacts with no correlation adjustment.
Key Metrics from Monte Carlo
- Median outcome: The central tendency of your portfolio's trajectory
- 5th percentile: What happens in the worst 5% of scenarios — your realistic downside
- 95th percentile: Your upside potential under favourable conditions
- Probability of meeting goals: What percentage of simulations achieve your target return or value
- Maximum drawdown distribution: Not just one worst-case number, but a distribution of how bad drawdowns could get
What Monte Carlo Reveals That Backtesting Cannot
Backtesting examines how your portfolio would have performed during specific historical periods. Its fundamental limitation: the future will not replicate the past.
Monte Carlo generates scenarios that have not occurred yet but are statistically plausible given your holdings' characteristics. It can reveal:
- How likely a drawdown deeper than a chosen level is, given your current holdings and their measured behaviour
- How likely the portfolio is to reach a target value over a chosen horizon, as a range rather than a single number
- How adding or removing a position changes the weak tail of outcomes compared with the middle of the distribution
Market Regime Is Context, Not an Input
Markets go through calmer and more volatile periods. KlirInvest shows the current market regime as context next to a simulation, but it does not change the simulation's volatility, correlation or tail settings: regime labels describe conditions and are not calibrated volatility forecasts. To see a portfolio under crisis-level assumptions, use the deterministic stress scenarios.
Limitations to Acknowledge
Monte Carlo is powerful but not prophetic:
- Model risk: Results depend on input assumptions — if volatility estimates are wrong, outputs will be wrong
- Structural breaks: Monte Carlo cannot anticipate events that change the fundamental structure of markets (e.g., a new regulatory framework)
- Computational confidence: 10,000 simulations provide statistical robustness, but they sample from the model — not from reality
The responsible approach: treat Monte Carlo outputs as probability ranges, not predictions. Use them to identify vulnerabilities and calibrate risk, not to forecast returns.
Try the free quick check at KlirInvest. It gives a rough Risk Score from asset-class averages, treating every symbol as an equal share. The Health Score, the Decision Readiness Score and a holdings-weighted analysis need a saved portfolio in an account.
