Directive 369: where a validation team finds each item

Last updated 16 November 2026

Section 93 of the Bank of Israel's Proper Conduct of Banking Business Directive 369, Management of Model Risk, sets out what a banking corporation must obtain from the supplier of a model it uses. This page takes those items in turn for GIAnalyzer and the Gamma Impact API, and says where each one is published.

What it points to is developer documentation and test evidence, for your own validation. It is not a validation of the models and does not replace one: under §93(d) the bank validates its use of a supplier's product, and under §93(j) it does not rely solely on reports the supplier provides. The page is our reading of the directive, for information; the Bank of Israel has not reviewed it. The directive itself is on the Bank of Israel's site. For the US, UK and Canadian rules, see Model risk regulation.

Components, design and intended use

§38, §93(b): documentation from the supplier that allows appropriate validation, and evidence from development of the product's components, design and intended use.

Coverage & models lists the instruments covered in each asset class, the outputs for each, and the models by name. One methodology page per area describes how each instrument is valued: the model, the market data it takes, and the numerical method.

The function reference documents every function: its syntax, its arguments and a worked example. What each product is for, and how positions and market data reach it, is on the GIAnalyzer and Gamma Impact API pages.

Mathematical specification and sources

§47: the methodology, including the mathematical specification and the numerical techniques and approximations, based on professional publications, with comparison against alternative approaches.

Each methodology page names the model behind each instrument and ends with its references. Where more than one method applies, the methods run side by side on the same inputs, so their results can be compared and the method behind any number named. The options models, for example, include Black-Scholes, binomial trees and Monte Carlo, with Bjerksund-Stensland for American exercise.

Test results

§93(b): results of appropriate tests showing that the product operates as expected.

Every function in the function reference has a worked example, and several show the manual calculation behind the result. In GIAnalyzer, the Manual Calculations workbooks work a valuation out step by step and compare the result with the function's. They cover basis swaps, caps and floors, cross-currency swaps in all four leg combinations, interest rate swaps, overnight index swaps, swaptions, real rate swaps and zero-coupon inflation swaps. See Workbooks.

Clients also receive manual calculations of further results, among them swap values, sensitivities and Value at Risk.

Limitations and assumptions

§93(b): clear and detailed information on the model's limitations and assumptions, and on cases where its use may be problematic.

Each methodology page has a section on what the models do not do:

Credit and commodity derivatives are not covered.

Sensitivity analysis and benchmarking without the code

§93(e): where a supplier model does not give full access to its code, the bank relies more on sensitivity analysis and benchmarking, and on compensating controls. §81 names derivative pricing models as an example of benchmarking.

The code is not open, so the products are built to be tested from outside:

  • Every input and parameter comes back with the result, so each value can be reproduced and checked.
  • In GIAnalyzer, the functions run in your own workbook. Change any input and the result recalculates, so sensitivities can be run on any argument, not only those the product reports. In the Gamma Impact API, scenarios you define do the same across a portfolio.
  • Methodological choices are parameters, such as interpolation, volatility weighting and the VaR method. They are not assumptions inside the code.
  • Monte Carlo runs take the number of paths and the random seed from you, so a run can be repeated exactly.
  • For benchmarking, the models are named with their references, so an independent implementation can be built against them. Where more than one method applies, the methods can also be compared with each other on the same inputs.

Reproducing a result

§33: documentation that lets someone unfamiliar with the model understand how it works, its limitations and main assumptions, and reproduce it independently.

The methodology pages and the function reference describe the calculation. The inputs and parameters returned with each result give what it was run on. In the Gamma Impact API, each run also writes a log, with counts at every step and every skipped position with its reason. A missing input is reported, never filled in.

Data and assumptions from the supplier

§93(g), §93(h): input data or assumptions supplied by the supplier, and the data used to develop the model.

No market data is bundled with the calculation. It comes from you: Bloomberg and LSEG files are read as exported, or you can use curves of your own, or mix the two curve by curve. The Value at Risk and the risk measures are computed on histories you supply. No holiday calendars are supplied either: Saturdays and Sundays are the only non-business days. Where a method has a default, such as linear interpolation of zero curves, the methodology page states it.

Settings chosen by the bank

§93(f): adjustments made to a supplier model for the bank's circumstances, documented, justified and reviewed in validation.

The choices a bank makes are parameters it sets, and they are returned with every result, so the settings behind any number are on record.

Ongoing monitoring and changes

§93(c): ongoing performance monitoring by the supplier, with continuous disclosure to the bank, and updates to the model over time.

Clients receive a message with each new release, and each client chooses when to upgrade. Each methodology page states the version of GIAnalyzer it describes.

If the product is no longer supported

§93(i): contingency plans for the case that the supplier model is no longer available or the supplier can no longer support it.

The Gamma Impact API can run on your own server, and the mappers on your own machine, with no connection to us; licensing works offline. GIAnalyzer has an offline version. A licence keeps working until its expiry date, and support continues until then. Source-code escrow can be arranged on request.

Building internal knowledge

§93(d): through validation, the bank gains enough internal knowledge of the product to continue if the relationship ends.

The user guide, the FAQ and the glossary are public, as is everything listed above. We also provide training to clients.

Outside this page

If we host the Gamma Impact API for you, Directive 359A on outsourcing applies as well; section 93 is without prejudice to it. This page does not cover it.

Position Mapper and Confirmation Mapper read position files and confirmations rather than value positions. Their rules are described in the Position Mapper guide and the Confirmation Mapper guide.

Who designed the models

The models were designed by Leon Sandler. He has validated the pricing and risk models of vendor platforms at most of Israel's banks, and helped the Bank of Israel establish the model-risk function in its banking supervision division. That work was advisory, and is not an approval of these products by the Bank. Because he designed them, this page is documentation for your validation, not an independent validation.

Going through the directive with your validation team

Send us the clauses your team is working through, and we'll go through each one with you, on your own data if you want. Request a demo