Research Library

RP-001: Why Hardcoded Cells Cause Financial Model Failure

This paper examines the mechanism by which a hardcoded value embedded in a formula cell causes financial model failure, treating it as a distinct structural failure mode rather than a stylistic convention. It grounds the argument in FMAE's own R001 rule rationale — the specific row-pattern detection logic used to identify a hardcode overriding an otherwise-consistent formula — and situates it within the broader, independently published empirical literature on spreadsheet error rates, which consistently finds that manual, undetected errors are a persistent feature of spreadsheet development regardless of developer experience.

RP-002: Structural Model Risk — A Working Definition

This paper formalizes structural model risk as a distinct category of risk, separate from forecast risk, market risk, credit risk, and operational risk, defined specifically as the risk that a financial model produces an incorrect result due to errors in its own construction, logic, or structure rather than due to any flaw in the underlying business case it evaluates. The definition extends and formalizes content already established on the Knowledge Centre's Model Risk pillar page into a standalone academic-register treatment intended to be citable independently of any single product's marketing framing.

RP-003: Aggregation Logic Errors in Financial Models

This paper examines aggregation range gaps — the specific failure mode where a formula-bearing cell sits immediately adjacent to, but is not included in, an aggregation formula's range — as a distinct class of silent structural error. Unlike most formula errors, an aggregation range gap produces no visible error value and no obviously wrong-looking number; the total simply understates or overstates what it should represent, with the excluded line item continuing to display normally elsewhere on the sheet. This paper documents the structural logic FMAE's R023 rule uses to detect this pattern precisely, including the specific suppression conditions that distinguish a genuine gap from an expected cascading-subtotal structure.

RP-004: Financial Model Audit Taxonomy

This paper presents a formal taxonomy of structural financial-model audit findings, organized into six categories — Structural, Assumptions Governance, Integrity Controls, Structural Hygiene, Aggregation Logic, and Model Governance. Unlike a taxonomy constructed for editorial or classification purposes, this one is derived directly from, and verified against, the category attribute each of 26 rules declares in a deployed, production rule engine, making it a description of an existing operational classification rather than a proposed one.

RP-005: Evidence-Driven Rule Calibration — The R015 Case Study

This paper documents a single, real calibration event in the FMAE rule engine's history as a case study in evidence-driven rule design. R015, then named "Addition Chain Risk," originally fired on any formula joining five or more cell references by addition, at high severity. A review of that rule's findings against a real institutional model found the overwhelming majority — both non-contiguous addition chains (representing deliberate, named component selection) and contiguous ones (representing fixed, named business subtotals) — to be legitimate modelling practice rather than structural defects. The rule was redesigned in response — the minimum chain length was raised, non-contiguous references were excluded entirely, severity was lowered, and its weight was reduced proportionally. This paper treats that redesign as a case study in what evidence-driven calibration of a deterministic rule looks like in practice.

RP-006: Model Validation vs. Model Audit — A Methodological Comparison

This paper compares model validation and model audit as distinct methodological disciplines within financial model risk management, extending the Knowledge Centre's existing buyer-facing comparison into an academic register. The two disciplines differ in the kind of question each is designed to answer — audit asks whether a model, as built, calculates what it claims to calculate; validation asks whether the modelling approach and assumptions chosen are appropriate for the model's intended purpose. The two are complementary rather than substitutable, since a model can pass one and fail the other independently.

RP-007: Deterministic vs. Generative Approaches to Structural Model Review

This paper compares deterministic, rule-based structural review and generative large-language-model review of financial models, focused on the single property that most clearly distinguishes them — reproducibility. A deterministic rule engine applies a fixed, disclosed methodology to every formula in a model, such that the same workbook produces the same findings on every run. A generative language model reading the same workbook and offering commentary is not guaranteed to produce the same output twice, nor to apply the same standard of scrutiny consistently across every formula in a large model. This difference has direct consequences for whether either approach's output can serve as defensible evidence behind a material financial decision.

RP-008: Materiality and Severity — Distinguishing Two Axes of Model Risk

This paper distinguishes two axes along which a structural finding in a financial model can be assessed — severity, how serious the specific defect is on its own terms, and materiality, how much weight that finding should carry within an overall assessment of the model's reliability. The distinction is formalized from a concrete mechanism already implemented in FMAE's own scoring methodology, the critical-override rule, which caps a model's overall grade regardless of its aggregate numeric score whenever specific severe findings occur — a design decision that only makes sense once severity and aggregate scoring weight are treated as related but non-identical concepts.