Conclusion first
A deterministic model computes one fair value from one expected cashflow path and a fixed set of assumptions. A stochastic model simulates many paths over input distributions and takes an expectation. For private credit, the deterministic approach is the right default: contractual cashflows are largely fixed, and a market-calibrated discount rate produces a defensible exit price under IFRS 13. Stochastic modelling earns its complexity only where value is genuinely path-dependent. Where it does, the stochastic output is reconciled against the deterministic baseline.
What each approach does
- Deterministic. One path, one discount curve, one answer. Inputs are explicit; the result is reproducible and decomposable into its drivers.
- Stochastic. Many simulated paths over distributions of defaults, rates, recoveries or prepayments; the fair value is an expectation across them. Captures non-linearity and path-dependence a single path cannot.
The trade-off
| Property | Deterministic | Stochastic |
|---|---|---|
| Reproducibility | Exact — same inputs, same mark | Seed- and path-count-dependent |
| Explainability | Every figure traces to an input | Result is an aggregate over paths |
| Audit defence | Re-run from recorded inputs | Requires the simulation harness + seeds |
| Captures optionality | Via explicit scenario weights | Natively, across the distribution |
| Cost / latency | Low — closed-form per path | Higher — many repricings per mark |
When stochastic is worth it
- Material embedded options whose value is convex in a stochastic factor — for example, equity conversion or warrant features.
- Prepayment behaviour that is strongly rate-path-dependent rather than reducible to a single expected assumption.
- A restructuring with several genuinely distinct outcome paths that a three-scenario blend cannot adequately represent.
How Verus handles path-dependence without giving up determinism
Verus stays deterministic and represents uncertainty explicitly rather than by simulation:
- Scenario blending. Base, downside and recovery paths are repriced and combined on stated probabilities — a transparent, reproducible stand-in for a distribution.
- Confidence range. Each facility is repriced at seven spread percentiles, giving a valuation range without a stochastic engine.
- Sensitivity grid. Fair value is recomputed across explicit input bumps, which is what IFRS 13 Level 3 disclosure actually asks for.
The result carries the upside of a probabilistic view — a range, not a false point — while remaining reproducible to the basis point and decomposable into named drivers.
Related
- Deterministic IFRS 13 valuation — how the deterministic mark is built, end to end.
- Methodology — the engine, the approach router and the scenario weights.
- Model risk — validation, including the determinism check.
- Valuation approaches — engine vs advisory vs platform.
To see the deterministic engine on your own book, book a demo.