How do I fix a Monte Carlo VaR workflow that mixes log returns and simple returns?

The fix is to be consistent about what object you are aggregating.

For multi-asset portfolios, the weighted sum relationship applies cleanly to simple returns, not to log returns. A safer workflow is:

1. Simulate correlated log returns for each asset.
2. Convert each simulated asset log return into a simple return.
3. Aggregate asset simple returns using portfolio weights.
4. Translate the resulting portfolio return into portfolio profit and loss.
5. Sort losses and read the VaR percentile.

Why this matters:

• log returns are convenient for statistical modeling
• portfolio aggregation is naturally done on value changes or simple returns
• skipping the conversion can bias the simulated loss distribution

If the portfolio includes options or other nonlinear products, go one step further: simulate the risk factors, reprice the instruments, and compute portfolio P&L directly instead of aggregating asset-return formulas that no longer describe the true payoff.