Monte Carlo you can reproduce, seed and all
Updated
In short
How does Kelvyn quantify uncertainty on a portfolio?
Kelvyn runs seeded Monte Carlo simulation over the portfolio — NPV distributions, portfolio value-at-risk, recovery-only and timing-only runs — and renders the result as P5-to-P95 fan charts rather than a single point estimate. Seeded means reproducible: the same inputs and the same seed regenerate the same distribution exactly, so a figure shown to a committee can be reproduced when someone asks about it later. Sensitivity analysis ranks what actually moves NPV and IRR — tornado charts, heatmaps, break-even finding — and probability-weighted expected NPV across scenarios is computed as exactly that: NPV, never dressed up as expected credit loss.
Distributions, not point estimates
A single NPV is an opinion with false precision. The simulation produces the distribution — P5, P25, P50, P75, P95 — as a fan chart, alongside portfolio value-at-risk computed from the same runs.
Recovery-only and timing-only simulations separate how much arrives from when it arrives, which are different risks with different mitigations.
Reproducible by construction
The random number generator is seeded, so a simulation is a deterministic function of its inputs. That is a governance property, not a technical footnote: a distribution shown in a committee paper can be regenerated months later and produce the same numbers, which is the difference between a figure and an anecdote.
What moves the number
Tornado analysis ranks variables by their impact on NPV and IRR; heatmaps show how two assumptions interact; a break-even finder answers the inverted question of how bad an assumption may get before the case fails.
The simulation library runs to 26 registered tools, including curve fitting, roll rates, stage migration and concentration analysis.
Scope
What this does not do
Probability-weighted output here is expected NPV across scenarios. It is not probability-weighted IFRS 9 expected credit loss, and the product keeps the two apart deliberately.
A simulation samples from the assumptions you author. It quantifies the uncertainty in your model; it does not validate the model, and no simulation can.
There are no hedging analytics. Rate and FX paths exist for pricing debt facilities; hedge instruments and hedge optimisation do not exist in the product.
Questions
- Are simulation runs reproducible?
- Yes. The generator is seeded, so the same inputs and seed produce the same distribution exactly.
- What does the VaR calculation cover?
- Portfolio value-at-risk derived from the simulated NPV distribution, alongside recovery-only and timing-only runs.
- Is this expected credit loss?
- No. Probability-weighted expected NPV across scenarios is a different calculation from probability-weighted ECL, and Kelvyn never presents one as the other.
Related