MatLogica | Client Results

Client Results

Proven Results in Production

Real-world performance benchmarks and case studies from banks, risk management firms, and research institutions using MatLogica AADC in production

Production-Validated Performance

MatLogica AADC is deployed in production at major financial institutions, risk management firms, and research organizations worldwide. These results demonstrate real-world performance improvements across diverse applications from XVA and CVA calculations to tail risk analysis and neural network training.

October, 2023

Cross-Gamma Approximations of Portfolio CVA Value

Key Results:

  • 7x faster than tape-based AAD for second-order Greeks
  • Efficient Cross-Gamma Hessian computation for CVA portfolios
  • Code Generation AAD™ vs Tape-Based AAD comparison
  • Production-ready "bumping over AAD" methodology

A presentation by Stephan Bosch comparing the performance of "bumping over AAD" for Tape-Based AAD versus Code Generation AAD to compute second-order Greeks, focusing on Cross-Gamma Hessian entries.

Impact: Enables practical calculation of full Hessian matrices for large CVA portfolios, critical for comprehensive risk management and hedging strategies.

Download PDF Summary

Contact us at info@matlogica.com for the full presentation!

Cross-Gamma Performance: Code Generation AAD vs Tape-Based AAD for CVA

September, 2023

Estimating Expected Shortfall Sensitivities using AADC

Extraordinary Results:

  • From 8+ hours to <6 seconds (4,800x+ speedup)
  • AAD for tail risk and stress testing
  • Model risk quantification at scale
  • Production deployment at Probability&Partners

Highlights from the WBS 2023 presentation by Svetlana Borovkova from Probability&Partners on Algorithmic Adjoint Differentiation for Tail Risk, Model Risk & Stress Testing.

Impact: Transforms Expected Shortfall (ES) sensitivity calculations from overnight batch jobs to near-real-time analysis, enabling intraday risk monitoring and dynamic stress testing.

Download PDF Summary

Contact us at info@matlogica.com for the full presentation!

Expected Shortfall Sensitivities: 8+ hours to under 6 seconds with AADC

September, 2023

Comparing AAD Techniques & Performance for XVA

Comprehensive Comparison:

  • Bump & Revalue (baseline method)
  • Tape-Based AAD (traditional approach)
  • Code Generation AAD™ (MatLogica approach)
  • Production XVA calculation benchmarks

Highlights from the WBS 2023 presentation by Stephan Bosch comparing three approaches for XVA calculations: traditional bump-and-revalue, tape-based AAD, and MatLogica's Code Generation AAD™.

Impact: Provides quantitative evidence for choosing AAD implementation strategy for production XVA systems, demonstrating clear performance advantages of code generation approach.

Download PDF Summary

Contact us at info@matlogica.com for the full presentation!

XVA Performance Comparison: Bump & Revalue vs Tape-Based AAD vs Code Generation AAD

June, 2022

Case Study: Major European Bank Front-Office Risk Revolution

Business Outcomes:

  • New revenue streams unlocked through faster pricing
  • Infrastructure costs reduced significantly
  • Risk management improved with real-time Greeks
  • AAD introduced for comprehensive risk computations
  • Scenario analysis accelerated for trading desks

How a major European bank revolutionized their front-office risk management using MatLogica AADC. The bank supercharged their analytics by introducing AAD for risk computations and accelerating pricing and scenario analysis.

Impact: Transformed front-office operations from overnight risk batches to intraday real-time risk, enabling traders to respond to market moves with complete Greeks information.

Download Case Study

MatLogica Case Study: Major European Bank Front-Office Risk Management

June, 2022

Neural Networks: Automatic Synthesis for Time Series Analysis

Research Results:

  • Up to 3x more accurate than cutting-edge methods
  • Several times faster training using MatLogica AADC
  • State-of-the-art architecture synthesis for recurrent neural nets
  • Automatic neuron design for time series problems

Prof. Roland Olsson and his team used MatLogica's AADC to design state-of-the-art neural network architectures for time series analysis, demonstrating breakthrough accuracy and training speed improvements.

Impact: Opens new possibilities for financial forecasting, trading signal generation, and time series modeling where both accuracy and training speed are critical.

Download Research Paper

Neural Network Time Series Analysis: 3x More Accurate with MatLogica AADC

Proven Performance Across Use Cases

Speed Improvements

  • 4,800x+ for Expected Shortfall sensitivities
  • 7x for Cross-Gamma vs tape-based AAD
  • Several times faster neural network training
  • Real-time XVA from overnight batches

Business Impact

  • New revenue streams unlocked
  • Infrastructure costs reduced
  • Intraday risk monitoring enabled
  • 3x accuracy improvement for forecasting

See What AADC Can Do For Your Organization

Let us benchmark MatLogica AADC on your actual production workloads

Schedule a Benchmark

info@matlogica.com

Related topics: AAD performance benchmarks, code generation AAD vs tape-based, cross-gamma Hessian calculation, Expected Shortfall AAD, XVA Greeks acceleration, CVA second-order Greeks, tail risk sensitivities, stress testing AAD, front-office risk management, real-time XVA calculation, AAD case studies banking, neural networks time series AAD, production AAD deployment