Skip to main content

Model Risk Management Benchmark and Review

Client

US Investment Bank

Function

E-Trading, Front-office, Compliance

Asset Class

Equities, FX

Geography

Europe

Duration

2 months

The Ask

Our client, a Tier-1 US bank, commissioned an investigation of the current practices employed across the banking industry regarding the governance and risk management processes for e-Trading algorithms (algos) that incorporate a model component (or feeder).

SR 11-7 forms the basis for the governing policies and processes for bank’s 2LoD MRM function, but also contributes to many of the issues regarding the management of models via the current algo Trading business, and the model risk management lifecycle. This is due to the fact that SR 11-7 does not provide any guidelines on how to deal with model components of algos and how these two worlds should communicate in such cases.

GreySpark was engaged to carry out an independent review across the banking sector by interviewing 10 banks of different Tiers and of various levels of maturity. In addition, GreySpark also investigated the regulatory trends and expectations through multiple discussions with FED and PRA representatives and research from publicly available resources.

Methodology

A senior GreySpark team of SMEs and analysts partnered with the client supported the business with a benchmarking exercise of model risk management practices against the general practice and approach of the bank’s peer group, industry best practice, and external regulator (FED and PRA) requirements and expectations.

10 institutions with active and established algorithm trading businesses were interviewed with focus on direct comparative analysis with the client’s approach to the area in order to gain insight into peer activity and an aggregate view on what best-in-breed looks like in practice.

Our approach consisted of :

  • Leveraging proprietary data accumulated by GreySpark’s Capital Markets Intelligence research practice
  • Collecting all tacit knowledge gathered through specific client interviews with the peer group
  • Analyzing data gathered from the client’s relevant to model components of algos to form current state framework
  • Establishing an objective view of each competitor’s algo/models business and governance environments

Outcome

GreySpark provided the client with continuous updates throughout the project allowing the client view of the process and a sense of collaboration where possible. The ultimate delivery was a 60+ page document covering the following areas in forensic detail:

  • Current State: GreySpark assessed the client’s current state by gaining knowledge on the current lifecycle for model components of algos within the bank. To do so, GreySpark analysed the existing policies, procedures, frameworks and documentary evidence. The team also conducted meetings with key stakeholders within the bank.
  • Regulatory Landscape: GreySpark provided a breakdown of prescribed statements and conducted interviews with the representatives from the FED and PRA to obtain the non-prescribed ‘regulatory expectations’ and unwritten interpretation of current papers such as SR 11-7 and SS 3-18.
  • Industry Practice: Leveraging our relationships across the street we conducted 25 interviews with both 1LoD and 2LoD representatives from 10 different banks with the objective to understand the industry’s approach to model components of algos, and the impact of the current processes on the lines of defence.

GreySpark was able to provide an independent view on the industry’s interpretation of regulatory guidelines such as SR 11-7 and SS 3-18 as well as how banks approached algos with model components, from both a 1LoD and a 2LoD perspective.

By interviewing banks of different tiering, GreySpark was able to introduce new approaches to the MRM Lifecycle Process and factual information on industry’s approach to the model governance Framework.

GreySpark was also able to provide a clear view on the tools and behaviours which shape a mature model risk management lifecycle.

Our analysis of the data gathered through this benchmark exercise allowed us to provide our client with a real snapshot of the current state of the industry regarding  the treatment of algos with model components across the entire model lifecycle. 

Do you have a similar project that you need help with?