Technology

One calculation for valuation and risk

Broad sensitivities without one portfolio rerun per factor

MASTIX uses Adjoint Algorithmic Differentiation (AAD) to calculate valuation and configured sensitivities together, so treasury and risk teams can run scenarios faster, explain movements consistently, and trace results to their drivers.

Why sensitivity analysis becomes a batch job

Large portfolios may need sensitivities to hundreds or thousands of risk factors.

With bump-and-revalue, each factor requires another portfolio run: change one input, revalue the portfolio, and measure the difference.

How the engine changes the calculation

Repeated revaluation versus one reverse pass

The portfolio and risk factors are the same. What changes is how the sensitivities are produced.

Repeated calculation path

Bump-and-revalue

1
Run the full portfolio valuation.
2
Change one configured risk factor.
3
Revalue the full portfolio and measure the difference.
4
Repeat for every configured factor.

One full portfolio recalculation for every configured risk factor.

One calculation path

MASTIX with AAD

1
Run the valuation while recording dependencies.
2
Propagate derivatives back through the calculation.
3
Produce the configured sensitivity set alongside the valuation.

A small multiple of one valuation, rather than one run per factor.

Architecture

Why AAD has to live inside the engine

AAD depends on the calculation's dependency structure. It must be implemented within valuation, not added later at the reporting layer.

That same foundation lets valuation, sensitivities, scenarios, and attribution share inputs, models, and calculation lineage.

What this changes for treasury and risk teams

Interactive scenario analysis

Teams can test assumptions and run follow-up scenarios in the same working session, while governed full balance-sheet runs remain controlled portfolio jobs.

Attribution tied to valuation

Teams can separate movements from rates, volumes, model changes, and new deals without rebuilding the explanation in a separate process.

Traceable results

Outputs remain linked to their drivers and assumptions, so ALCO, model governance, and audit can trace a result back to its inputs.