Alpha Hedge ServicesSearch / Rechercher / 搜索

Strategy, processes and change management →

From strategic question to reviewable decision

Original framework proposed by Alpha Hedge Services, adapted from targeted reading. These five steps are not attributed as a sequence to any author. Examples are hypothetical, with no client engagement or measured effect claimed.

  1. Diagnose

    Define the decision, break down the problem and confront competing hypotheses with data, interviews and process observations.

    Proposed example and deliverable

    Proposed banking example: reconcile P&L differences and data breaks. Deliverable: diagnosis, evidence register and hypotheses to challenge.

  2. Frame Strategic Options

    Compare distinct options, including the status quo. Make resources, constraints, dependencies and adverse scenarios explicit.

    Proposed example and deliverable

    Example: repair an interface, change accountability or redesign the chain. Deliverable: options memo and comparison criteria.

  3. Decide

    Present the proposed conclusion first, then arguments, uncertainties and trade-offs. The client owns the decision and risk acceptance.

    Proposed example and deliverable

    Example: approve a conditional pilot rather than full deployment. Deliverable: decision memo, disagreements and reassessment conditions.

  4. Execute

    Assign responsibilities, plan dependencies and verify controls, training and adoption. Match resources to the work required.

    Proposed example and deliverable

    Example: product-scope pilot with acceptance criteria. Deliverable: roadmap, milestones, risks and governance arrangements.

  5. Measure Impact

    Compare with a documented baseline; distinguish delivery, adoption and economic effect. Examine other factors before attributing an outcome to the engagement.

    Proposed example and deliverable

    Example: track time, rework and incidents after the pilot. Deliverable: monitoring table with sources, owners and a continue-or-revise decision.

Sources and passages consulted →
TECHNICAL RESEARCH NOTE · AHS-TRN-01 Version 1.0 · septembre 2026

Architecture quantitative · Gouvernance des modèles

De l’équation au contrôle : une architecture de recherche quantitative traçable.

Cette note décrit comment une sélection de démonstrateurs quantitatifs est spécifiée, calculée et examinée. Elle ne prétend ni à une validation indépendante de chaque méthode ni à un usage en production.

Explorer les méthodes documentées ↗Consulter les méthodes documentées

Note technique web publiée — non évaluée par les pairs

01 / THESIS

Principe directeur

Un résultat quantitatif n’est défendable que si son contrat, ses données, sa mesure de probabilité, sa discrétisation et ses contrôles peuvent être reconstruits indépendamment. La précision numérique sans traçabilité n’est pas de la rigueur scientifique.

Le Quant Lab est volontairement un environnement de recherche pédagogique : calcul local, paramètres explicites, graines déterministes lorsque la simulation intervient, et limites d’usage visibles.

02 / ARCHITECTURE

Cinq couches, une chaîne de preuve

05

Contrôles & gouvernance

Invariants, benchmarks, limites, versionnage et revue indépendante.

04

Moteur numérique

Formule fermée, arbre, quadrature, optimisation ou Monte Carlo.

03

Modèle & mesure

Dynamique des facteurs, numéraire, calibration et paramètres.

02

Données & courbes

Horodatage, conventions, unités, interpolation et qualité.

01

Contrat & payoff

Term sheet, calendriers, cash-flows, collateral et événements.

03 / DOCUMENTED METHODS

Méthodes documentées sélectionnées

Le tableau ci-dessous illustre le niveau de documentation retenu dans le Quant Lab. Chaque méthode visible précise sa finalité, sa mesure, son approche numérique et son invariant de validation. Ce catalogue pédagogique ne constitue pas une affirmation de maturité pour la production institutionnelle.

IDMéthode / finalitéMesure & méthodeInvariant de validation
QL-01Multi-curve IRSValorisation OIS / projectionCash-flows déterministes ; courbes séparéesAu taux par, NPV = 0 à l’arrondi
QL-02HJM bond optionOption sur zéro-couponMesure T₁-forward ; Black sur forward bondLimite intrinsèque à volatilité nulle
QL-03BGM / LMM capletBrique de volatilité forwardMesure forward de paiement ; Black-76Limite intrinsèque ; vega positif
QL-04Credit / CDSJambes prime et protectionIntensité constante ; intégration discrèteAu spread par, les deux jambes coïncident
QL-05xVACVA · DVA · FVAExposition gaussienne ; hazard ; somme temporelleHazard et funding nuls impliquent xVA nul
QL-06Portfolio VaR / ESAgrégation du risqueDelta-normal ; matrice de covarianceMatrice PSD ; ES ≥ VaR
QL-07Asian Monte CarloOption asiatique arithmétiqueGBM risque-neutre ; variables antithétiquesGraine déterministe ; intervalle de confiance
QL-08BSM & implied volVanille européenneFormule fermée ; inversion par dichotomieParité put-call ; bornes d’arbitrage
QL-09American CRRExercice anticipéArbre recombinant ; induction rétrogradeValeur américaine ≥ valeur européenne
QL-10Bond analyticsPrix, duration, convexitéCash-flows actualisés ; moments temporelsCohérence prix–rendement–DV01
QL-11Mean–VarianceGMV et portefeuille tangentMarkowitz ; inverse de covarianceSomme des poids = 1 ; covariance définie positive
QL-12Credit portfolioPertes corréléesFacteur latent gaussien ; Monte CarloEL simulée ≈ EL analytique ; ES ≥ VaR
QL-13SABR smileVolatilité implicite de tauxDiffusion CEV stochastique ; asymptotique de HaganContinuité ATM ; volatilité positive
QL-14Heston SVVolatilité stochastique actionsFonction caractéristique ; Simpson ; Little Heston TrapBornes d’arbitrage ; parité put-call ; convergence de quadrature
QL-15Capital-protected noteProtection + participationZéro-coupon + call Black–ScholesPayoff à maturité toujours supérieur ou égal au nominal
QL-16Reverse convertibleCoupon contre risque de baisseGBM risque-neutre ; Monte CarloGraine déterministe ; erreur standard publiée
QL-17Autocall PhoenixAutocall et coupon mémoireGBM path-dependent ; Monte Carlo annuelGraine déterministe ; conservation des flux par trajectoire
QL-18Digital optionCash-or-nothing européenBSM fermé ; probabilité N(d₂)Parité digitale exacte à l’arrondi
QL-19Variance swapVariance réalisée contre strikePayoff linéaire ; variance forward exogèneValeur nulle au strike forward
QL-20Basket optionCall arithmétique à deux actifsGBM corrélés ; Monte Carlo antithétiqueGraine fixe ; erreur standard publiée
QL-21Hull–White ZCB optionOption sur zéro-couponTaux court gaussien ; formule fermée forwardC−P=P(0,T)−KP(0,S)
QL-22Merton structural creditLevier et défaut structurelValeur d’entreprise lognormale ; equity-callE + D = V
QL-23Rating migration bondRevalorisation à un an par étatsProbabilités, spreads et recouvrement exogènesΣpᵢ = 100%
QL-24European swaptionDroit d’entrer dans un swapMesure annuité ; formule Black‑76Payer−Receiver=N·A·(F−K)
QL-25Callable bond treeOption de remboursement émetteurArbre de taux ; induction rétrogradeVcallable ≤ Vstraight
QL-26Barrier Brownian bridgeCall up-and-out continuGBM ; Monte Carlo antithétique ; correction bridge0 ≤ Vbarrier ≤ Vvanilla
QL-27Shifted-Black capletCaplet shifted-lognormalBlack‑76 décaléF+s>0 ; K+s>0 ; V≥intrinsic
QL-28CDO tranche LHPPerte de tranche synthétiqueLHP ; quadrature0 ≤ ETL ≤ 100% ; A < D
QL-29FX ForwardChange à termeF=S·exp[(rᵈ−rᶠ)T]F/S = DFᶠ/DFᵈ
QL-30FX Vanilla GKOption FX européenneGarman–Kohlhagen · closed formC−P=S·DFᶠ−K·DFᵈ
QL-31FX DigitalCash-or-nothingΦ(d₂) · closed formDigital call + put = payout·DFᵈ

04 / SCIENTIFIC STANDARD

Protocole de validation

01

Spécification

Payoff, conventions, numéraire, mesure et hypothèses explicités avant le calcul.

02

Analyse dimensionnelle

Unités, échelles de taux et sensibilité aux conventions contrôlées.

03

Non-arbitrage

Parités, bornes, monotonicités et signes des Greeks servent de tests structurels.

04

Convergence

Pas temporel, nombre de scénarios, erreur standard et stabilité sont observés.

05

Reproductibilité

Paramètres visibles, calcul local et graines fixes rendent les résultats rejouables.

06

Risque de modèle

Limites, domaine de validité, alternatives et besoin de validation indépendante sont déclarés.

05 / USE BOUNDARIES

Quand un démonstrateur n’est pas un modèle de production

R0Note de rechercheÉquations et hypothèses
R1Prototype contrôléTests, benchmark, reproductibilité
R2Validation indépendanteExamen contradictoire, limites, approbation
R3Production gouvernéeDonnées, suivi, maîtrise des changements

06 / BIBLIOGRAPHY

Références structurantes

  1. Cairns, A. J. G.Interest Rate Models: An Introduction — HJM ch. 5–6; BGM ch. 9; numerical methods ch. 10; credit ch. 11.
  2. Schönbucher, P. J.Credit Derivatives Pricing Models — survival/CDS ch. 3; rating-transition, firm-value, portfolio-credit and intensity implementation.
  3. Savine, A. & Andreasen, J.Modern Computational Finance — xVA, Part V, ch. 23.
  4. Fries, C.Mathematical Finance — BSM ch. 4; Greeks ch. 7; Monte Carlo ch. 13; Bermudan ch. 15; sensitivities ch. 17.
  5. Cornuéjols, G. & Tütüncü, R.Optimization Methods in Finance — mean–variance and quadratic programming, ch. 8.
  6. London, J.Modeling Derivatives in C++ — BSM/American ch. 1; Monte Carlo ch. 2; CRR ch. 3; implied volatility ch. 7.
  7. Rebonato, R., McKay, K. & White, R.The SABR/LIBOR Market Model — SABR ch. 3; calibration ch. 5–6.
  8. Rouah, F. D.The Heston Model and Its Extensions — European pricing ch. 1; integration issues ch. 2; numerical schemes ch. 5.
  9. Wu, L.Interest Rate Modeling: Theory and Practice — HJM ch. 6; Hull–White / extended Vasicek §6.5.2; bond options and correlation adjustment §6.7.
  10. Andersen, L. B. G. & Piterbarg, V. V.Interest Rate Modeling, Vol. II — one-factor Gaussian short-rate models, European options, Jamshidian decomposition and swaption calibration, ch. 10.
  11. Löffler, G. & Posch, P. N.Credit Risk Modeling using Excel and VBA — structural approach and Merton model, ch. 2; transition matrices and hazard rates, ch. 3.
  12. Corb, H.Interest Rate Swaps and Other Derivatives — swap pricing and risk, ch. 1–3; swaptions, ch. 5.
  13. Smith, D. J.Valuation in a World of CVA, DVA, and FVA — callable debt, ch. IV; bilateral-credit swaps, ch. V; FVA, ch. VI.
  14. Pricing and Hedging Financial Derivatives — replication and risk-neutral valuation, ch. 5; Greeks, ch. 6; volatility and exotic barriers, ch. 7–8.
  15. Jewitt, G.FX Derivatives Trader School — vanilla pricing and Greeks, ch. 5–10; volatility surfaces, ch. 11–13; exotic FX derivatives, ch. 18–31.
  16. Duffy, D. J.Numerical Methods in Computational Finance — finite differences, ADI and splitting, ch. 14–23; Asian options, ch. 24; CIR/Heston PDE, ch. 25.
  17. Duffy, D. J.Financial Instrument Pricing Using C++ — two-asset pricing, ch. 16; PDE/FDM frameworks, ch. 20–23; Monte Carlo architecture, ch. 31–32.
Alpha Hedge ServicesQuantitative Research & Model Engineering

Document de recherche à finalité informative. Ni conseil en investissement, ni valorisation indépendante.

Governance and strategy: organising judgment

Decision frameworks connecting assumptions, accountability, investment and operations. These proposals describe neither certified procedures nor client engagements.

Published research note · Methodological synthesis

Who owns the decision when the model remains uncertain?

Alpha Hedge Services ·

Receiving a dashboard does not, by itself, make a committee accountable for a decision. Separate the person proposing an assumption, the reviewer examining the evidence and the owner accepting the consequences. For a valuation change, the developer explains the mechanism, a distinct reviewer challenges its limits and the business owner determines the permitted use.

We propose a short decision record: the question, rejected alternatives, conflicting evidence, residual exposure, named owner and review date. Greater uncertainty should change the decision, its scale or its reversibility rather than disappear into an average.

This is a proposed method. It establishes neither effective review independence nor organisational compliance. Role separation must be verified in practice, particularly where the same people produce and approve analyses.

Sources and attribution

Cynthia A. Montgomery, “Putting Leadership Back into Strategy”, Harvard Business Review, January 2008; David J. Snowden and Mary E. Boone, “A Leader’s Framework for Decision Making”, Harvard Business Review, November 2007. Supplied Essential Guide to Leadership, pp. 31–40 and 59–70.

Published research note · Methodological synthesis

Fund transformation through decision stages

Alpha Hedge Services ·

A transformation budget becomes intelligible when it answers an explicit economic question. Which avoidable cost, business capability or unmanaged risk justifies investment? Feature counts cannot answer that question. Establish a baseline, a benefit hypothesis and an owner responsible for checking whether the benefit occurs.

Staged funding can attach each tranche to evidence: minimum data quality, pilot results, operating capability and observed benefits. Sunk expenditure alone does not justify the next tranche. Compare remaining costs, still-plausible benefits and alternatives, including stopping.

Flexibility has value only where decisions are genuinely available. A contractual or political inability to stop must not be valued as a certain right. NPV scenarios illustrate trade-offs; they do not predict programme success.

Sources and attribution

Johnathan Mun, Advanced Analytical Models, Wiley, 2008, Real Options Models; AXELOS, ITIL 4: Digital and IT Strategy, TSO, 2020, §2.5–2.6, pp. 16–18 (supplied English manual).

Published research note · Methodological synthesis

Connect strategy to operating services

Alpha Hedge Services ·

Choosing to invest in technology does not settle its value. A service also depends on skills, processes, data and partners. In a valuation chain, an exact formula loses usefulness if data arrive late or no one owns exception handling. Governance therefore needs to cover the delivered service and its dependencies.

We propose linking each priority to an intended outcome and then to the resources required. The service owner follows delays, incidents, manual rework and critical dependencies; the decision-maker examines cost, quality and resilience trade-offs. Each indicator needs a source, denominator and measurement horizon.

A map supports discussion; it is not evidence of operational control. Compare it with observations, recovery tests and responsibilities actually exercised. No quantified service level is claimed here.

Sources and attribution

AXELOS, ITIL 4: Digital and IT Strategy, TSO, 2020, §2.6 “Products and services”, pp. 17–18; supplied English manual, PDF pp. 31–32.

Published research note · Methodological synthesis

Choose the AI use case before the technology

Alpha Hedge Services ·

An AI proposal should start with an observable task or decision: classify an exception, reconcile documents or assist a review. Specify its frequency, the cost of error and the current solution. This allows comparison of a complex model with a simple rule and may show that AI is unnecessary.

A pilot should measure net benefit: time saved after verification, errors corrected and additional control effort. High average accuracy can conceal expensive failures in rare categories. We therefore propose error-type analysis and an explicit human escalation path.

Scaling depends on data stability, operating cost and the ability to withdraw the model. Pilot results do not establish future performance. This framework does not imply operational AI capabilities in the firm’s public tools.

Sources and attribution

Thomas H. Davenport and Rajeev Ronanki, “Artificial Intelligence for the Real World”, Harvard Business Review, January–February 2018; supplied HBR’s 10 Must Reads on Technology and Strategy Collection, volume on AI, Analytics, and the New Machine Age.

Published research note · Methodological synthesis

Approve a model for a defined use

Alpha Hedge Services ·

Model approval needs a use, version and limits. A model useful for explaining sensitivity is not automatically suitable for financial reporting or hedging. The record should connect contractual definition, data sources, assumptions, numerical method and conditions under which the result becomes unreliable.

Review compares calculations with benchmark identities, boundary cases and alternative methods. Explain discrepancies instead of merely widening tolerances. The change log separates code correction, parameter change and model change because their consequences differ.

This site’s demonstrators are educational. Published numerical checks are neither independent validation nor production approval. Professional use requires review specific to the relevant contracts and data.

Sources and attribution

Johnathan Mun, Advanced Analytical Models, Wiley, 2008, applications 37, 41–42, 55 and 66; AHS methodological proposal on review and release boundaries. Numerical tests: tools/test-analytical.mjs, version dated 2026-09-27.

Management: from strategic priority to sustained control

A synthesis of management methods applied to markets operating chains. Examples are illustrative and deliverables proposed: they describe neither completed engagements nor demonstrated gains.

Deploy strategy into daily decisions

Hoshin Kanri connects objectives, means, ownership and reviews. In a markets business, the practical question is how to translate an executive priority into reviewable decisions: which chain to improve, who arbitrates and which capabilities to mobilise?

Framework and implementation

Start with internal and external diagnosis; select priorities; discuss feasibility with teams; connect each initiative to an owner, a measure and a review. A deployment matrix should expose dependencies and capacity trade-offs.

Markets application

Illustrative application: clarify Front Office–Finance reconciliations. Proposed deliverable: objective → process → initiative → owner → outcome evidence matrix. A delivered schedule alone does not establish process improvement.

Assumptions and limits

Assumptions: explicit priorities, available owners, comparable data and realistic capacity. Limits: formal alignment can conceal conflicting incentives; measures must remain challengeable when context changes.

Sources and synthesis status

T. M. Kubiak, The Certified Six Sigma Master Black Belt, ch. 1, Hoshin Kanri, pp. 7–13; ch. 2, Strategic Plan Alignment.

Qualify a project before prioritising it

Identification, qualification, selection, prioritisation and resource assignment are distinct decisions. An attractive project should not start before its problem, sponsor and data are sufficiently defined.

Framework and implementation

The charter records the observable problem, scope, exclusions, process owner, quality requirements and expected evidence. Then assess strategic alignment, control constraints, dependencies and available skills. Retain the rationale for starting, deferring or stopping.

Markets application

Illustrative application: choose between stabilising market data and rebuilding a pricing library. Proposed deliverable: qualification sheet and decision register; prerequisites and obligations take precedence over an aggregate score.

Assumptions and limits

Assumptions: agreed criteria, comparable scopes and horizons, documented costs and dependencies. Limits: ordinal scores are not financial valuations; avoid false precision and double-counted benefits.

Sources and synthesis status

T. M. Kubiak, ch. 5, Opportunities for Improvement, pp. 71–78; ch. 6, Risk Analysis of Projects and the Pipeline. CSSC, Certification Training Manual, ch. 12, Define, p. 133.

Improve a process or redesign it?

DMAIC structures improvement of an existing process: define, measure, analyse, improve and control. DMADV addresses design: define, measure, analyse, design and verify. Choose from the problem, not from a preferred tool.

Framework and implementation

Define the gap and population; verify collection; segment possible causes; test causes before acting; pilot a change and transfer control to the process owner. Reviews may require returning to an earlier stage.

Markets application

Illustrative application: recurring valuation breaks. Examine conventions, data, models, lifecycle events and interfaces separately. Proposed deliverables: process map, measurement plan, cause analysis and control plan.

Assumptions and limits

Assumptions: defined defect, traceable data and before/after comparison on a consistent scope. Limits: correlation differs from cause; portfolio or regime changes can distort comparisons. The method does not constitute independent model validation.

Sources and synthesis status

CSSC, Certification Training Manual, ch. 11, Introduction to DMAIC and DMADV, pp. 123–124; ch. 12–16. T. M. Kubiak, ch. 4, Six Sigma Methodologies, pp. 54–60.

Measure quality without creating false assurance

A useful measure specifies population, period, defect rule and source. Distinguish failure costs, prevention costs and appraisal costs. A quality measure needs an owner and an action when deterioration occurs.

Framework and implementation

Create a measurement dictionary and verify classifications. The control plan records frequency, alert rules, actions, escalation and retained evidence. Statistical limits describe a process; they do not replace business tolerance thresholds.

Markets application

Illustrative application: reconciliation monitoring. Present the case-level exception rate together with age, materiality and recurrence. Proposed deliverable: dashboard with a response plan; released capacity is not equivalent to realised cash savings.

Exception rate = n_exception / n_reviewed × 100%

n_exception: cases with at least one exception; n_reviewed: cases reviewed in the same period; rate: %. This definition counts cases, not multiple defects.

CoPQ = internal failure costs + external failure costs

CoPQ: cost of poor quality; C: failure costs in euros over the same scope and period. Prevention and appraisal are tracked separately within total quality cost.

Assumptions and limits

Assumptions: same population and period, non-zero denominator and controlled classification. Limits: a low rate can conceal a severe incident; measured improvement does not establish causality. Reconcile costs and remove double counts.

Sources and synthesis status

CSSC, Certification Training Manual, ch. 8, The CoQ and the CoPQ, p. 95; ch. 16, Control / Create a Control Plan, p. 183.

Lead change and test the diagnosis

A diagnostic framework organises questions without predetermining the answer. State a working hypothesis, specify what would disprove it, examine facts and revise the recommendation. The sponsor retains decision accountability; the improvement team supplies method and evidence.

Framework and implementation

Separate symptoms from causes, stakeholder needs from execution capabilities. Compare explanations using consistent criteria. For adoption, document role changes, training needs, resistance and user feedback after the pilot.

Markets application

Illustrative application: slow reporting may originate in data, interfaces, manual rework or approvals. Proposed deliverable: question tree, hypothesis → test → observation → decision register and adoption plan. Summarise options, uncertainties and the next review point.

Assumptions and limits

Assumptions: access to facts, room for challenge and an engaged sponsor. Limits: WetFeet addresses case interviews; its structure is adapted here for diagnosis, without empirical evidence of effectiveness. Communication alone does not guarantee adoption.

Sources and synthesis status

WetFeet, Ace Your Case VI: Mastering the Case Interview, PDF pp. 27–28 and 106. T. M. Kubiak, ch. 7, pp. 110–111; ch. 8, Change Management, p. 123.

Decision governance and control of biases

A funded project may be continued to defend a past decision. Selective perception and self-attribution can make successes feel personal and failures external. The book also connects incentives with governance and presents nudging as choice architecture that should preserve freedom and transparency.

Application to the workflow

Proposed application: independent reviews, prewritten continuation and stopping criteria, comparable option presentation and recorded exceptions. For UI: explain defaults, allow changes and disclose costs alongside risks. Synthesis: make challenge practicable and measure the effects of presentation changes before claiming effectiveness.

Assumptions and limits

Assumptions: comparable options, estimated future flows and explicit risks; test alternative costs, delays and benefits. Positive NPV does not replace capital, liquidity or compliance constraints. Biases are review hypotheses, not evidence of wrongdoing or individual motives.

Explore behavioural finance notes →

Read the result according to its economic meaning

The Labs distinguish model prices, risk scenarios and decision tools. The research companions make their assumptions and limits explicit; the same word “value” does not give these different outputs the same status.

ObjectRequired reading contextHow to interpret itExplore
Pricing modelContract, pricing measure, numeraire, inputs and numerical method.A conditional model value; calibration and market quotation are separate evidence.BGM/LMM interpretation
Risk scenarioExposure, horizon, shock and dependence assumptions.A scenario loss or sensitivity, not automatically an observed probability or regulatory capital.Risk and control framework
Economic valueCommon scope, valuation date, currency and costs counted once.A financial value bridge, distinct from a contractual price or an accounting journal entry.Acquisition value bridge
Preference modelReference point, utility scale, subjective beliefs and period length.An illustrative preference comparison; it does not supply risk-neutral pricing probabilities.Intertemporal decision discipline
Control frameworkOwner, execution evidence, exception, remediation and retest.A documented decision process, not an automatically effective control or a monetary risk score.Financial contract monitoring
Accounting and legal interpretationApplicable source text, entity, jurisdiction, date and recorded judgment.A reading and review framework; entity-specific treatment requires its own assessment.Accounting and obligation evidence

Keep the source, convention, observation date and interpretation together. A book-based mechanism, a historical association and a currently applicable rule require different evidence.

MBS example: cash-flow scenario, price and static spread →

Economic interpretation: identification before causal inference →