IRRBB, ALM & Regulatory Capital
Comprehensive Asset-Liability Management, NII sensitivity, and structural liquidity statement automation.
ALM & Capital Modules
Interest Rate Risk in the Banking Book (IRRBB) modelling
Non-Maturity Deposit (NMD) behavioural modelling for EVE & NII
Gap analysis and NII sensitivity
Net Stable Funding Ratio (NSFR)
Internal Capital Adequacy Assessment Process (ICAAP)
Structural Liquidity Statement (SLS) automation
RBI-compliant reporting outputs
Precision Compliance
Automate the generation of complex regulatory reports. Eliminate spreadsheet risk in your IRRBB and ALM calculations with institutional-grade processing pipelines.
NMD Volume Decomposition
Before any decay curve or replicating portfolio is built, every non-maturity deposit balance is split into the buckets that actually carry interest-rate risk. Getting this split right matters more than the sophistication of anything downstream — a euro of non-stable or non-core volume placed in the overnight bucket contributes exactly zero to EVE, by construction, regardless of how it is modelled afterward.
Total NMD Volume
Every non-maturity deposit account, segmented by product type (retail transactional, retail non-transactional, wholesale) as the starting point for behavioural classification.
Stable vs. Non-Stable
Volume is split based on historical volatility. Non-stable volume — the portion that has shown it can leave quickly — is placed in the overnight bucket, so its EVE contribution is zero by construction: there is nothing left to discount over time.
Core vs. Non-Core (within Stable)
Stable volume is split again using the product's pass-through rate: Core = (1 − pass-through) × Stable Volume. The pass-through portion reprices with the market and is treated like non-stable volume; only core volume is slotted into longer behavioural buckets.
Why this matters for floating-rate products: a centrally-priced product that tracks the risk-free rate with no repricing lag has a pass-through rate of 1 — so Core = (1 − 1) × Stable Volume = 0. The product correctly carries no EVE risk, because both the discount rate and the contractual rate move together. Modelling core volume directly with a decay function, without first passing it through this pass-through-based split, can fail to reproduce that result.
Behavioural Decay Modelling
Core volume is modelled to decay over time using a hazard-rate framework borrowed from survival analysis. The decay factor at time t is not built from the spot hazard rate directly, since a time-varying spot hazard offers no guarantee of monotonic decay. Instead, the model uses the cumulative hazard:
Cumulative Hazard Construction
The survival function is S(t) = exp(−Λ(t)), where Λ(t) is the integral of the spot hazard rate from 0 to t. Because Λ(t) is monotonically increasing by construction, S(t) is guaranteed to be monotonically decreasing — solving the instability that a raw, directly-applied hazard rate can introduce.
Bounding the Duration
For a constant hazard λ, average duration approximates 1/λ — which means an unconstrained estimate of λ approaching zero implies unbounded duration and unbounded risk. The model constrains λ to stay strictly positive and applies the same maximum average-life and longest-pocket caps regulators prescribe by product type, rather than relying on optimisation alone to keep the result sensible.
From Decay Curve to Replicating Portfolio
A replicating portfolio is a mix of tradable fixed-income instruments — typically G-Secs across several tenors — chosen so that their cashflows and interest-rate sensitivity track the behavioural decay profile of core NMD volume, letting EVE be computed with standard bond risk analytics instead of a bespoke model.
Matching Duration Alone Is Not Enough
A single weighted-average duration target — say, 20% in a 2-year G-Sec, 60% in a 5-year, and 20% in a 7-year, averaging to 4 years — is not a unique solution. Many different tenor-weight combinations produce the same average duration while implying very different convexity and cashflow timing. Worse, if those instruments are newly originated rather than structured as an amortising ladder, the portfolio's actual duration can differ materially from the 4-year target the weights imply on paper.
The more robust construction matches the full behavioural cashflow vector — the amount expected to run off in each future bucket, not just its average — using a continuously-rolling, laddered ("caterpillar") structure that is rebalanced each period. This matches both the duration and the shape of the behavioural runoff curve, rather than a single summary statistic that many different portfolios could satisfy.
EVE and NII Are Computed Separately
It's tempting to treat the replicating portfolio as the single source of truth for both metrics — but EVE and NII answer different questions and are driven by different inputs. Using the replicating portfolio's own coupons to infer NII conflates a valuation proxy with an income-statement projection.
Economic Value of Equity (EVE)
Computed by shocking the discount curve and re-valuing the behavioural cashflow profile of core volume — proxied by the replicating portfolio's duration and convexity. EVE answers: how much does the value of this deposit book move under a rate shock?
Net Interest Income (NII)
Computed directly from the deposit's own repricing dynamics — its pass-through rate and any repricing lag — with volumes held fixed. NII is never inferred from the coupons of the replicating bonds; a long-tenor replicating portfolio does not imply the deposit itself behaves like a fixed-rate instrument for income purposes.
Stress Testing & Macro Conditioning
The Supervisory Outlier Test shocks rates by ±200bp (and the other prescribed parallel and non-parallel scenarios) and re-runs the entire EVE and NII pipeline above under each one. Where a behavioural model also depends on macro factors beyond the rate path itself, those factors need a forecast that is internally consistent with the rate shock being tested — which is harder than it sounds, since a rate rise can reflect either an overheating economy being cooled or stagflation being fought, with opposite implications for every other macro variable.
A cointegrated VAR is a natural way to forecast a macro factor set jointly and consistently with a given rate path — the same conditional-scenario logic used in CCAR/DFAST-style stress testing. That said, more macro variables are not automatically better: where a candidate macro factor is itself largely a function of the rate level, adding it alongside the rate level risks multicollinearity without a demonstrated improvement in predictive accuracy — so each factor is only added to the behavioural model where it earns its place against that test.