Case Studies

Risk Analysis in the Financial Sector

Financial Firms Use Frontier to Analyze Risk and Meet DRCOP Requirements

Goals

Accelerate Monte Carlo simulations for two established global financial firms, and provide supplemental on-demand computational capacity.

Business

Investment Risk Analysis

Industry

Global Financial Services

Challenge

Nowhere is computational speed more important than in the financial markets, where seconds can mean millions of dollars - in one direction or the other. And, where seconds have such value, contingencies for continuing operations are critical. This case study describes how two global financial firms have used the Frontier Grid Platform to both accelerate the modeling on which their business depends and provide failover capacity in the event of disaster.

Scenarios

Because analysts must account for high-impact events, even if they are extremely improbable, simulations must delve deeply into the probability distributions that describe the expected behavior of candidate investments, as well as the correlation of their relative movements. Monte Carlo algorithms are well suited to capture the full subtlety of these interactions, however, they tend to be highly compute intensive and can often require days to weeks to execute on a single computer.

-----

Scenario 1: A Wall Street bank needed to more rapidly execute its proprietary models for determining the conditional value at risk (CVaR) of its derivatives portfolios. Each instrument has its own risk model based on a variety of historical factors and correlated price fluctuations. The risk profiles of myriad potential portfolios are evaluated by sampling the underlying probability distributions of the risk models in Monte Carlo fashion. The ultra-high volatility of these instruments requires extremely deep simulation because seven-sigma events, though improbable, can have enormous influence on the value at risk of these portfolios. In such situations, Frontier's grid-scale capacity is not only required; it's an essential competitive advantage.

-----

Scenario 2: A UK financial services firm wished to accelerate calibration and evaluation of its options trading system over volumes of historical market data, which simulates thousands of price paths of an asset. Each price path is used to find the lifetime return of portfolios managed under different hedging rules. On each simulated day of the varying price paths, the portfolio is hedged using cash and the underlying stock according to pre-determined hedging rules. At the end of the simulation, the results from hundreds of portfolios, each hedged using different rules, are compared to find which rules provide the greatest portfolio return. Here again, Frontier provides the capacity to perform these calculations in a timely fashion.

-----

Solutions

In both scenarios, Parabon's Frontier Grid Platform was used to cost-effectively speed these proprietary Monte Carlo simulations. In the first, Parabon's professional support team used Frontier Rapids to enable the bank's proprietary application to run on Frontier in just over a week. "They took our existing application and without changing a line of its code, adapted it for extreme-scale execution on Frontier," says a Senior Systems Engineer for the firm. "It runs in-house, but whenever we need an additional 500 C of power to knock out a job quickly, we just flip the switch and launch against Parabon's Frontier Grid Service." Furthermore, the ability to run the same application both internally and externally provides the firm a readymade failover solution that fulfills the industry's Disaster Recovery and Continuance of Operations (DRCOP) requirements.

In the second scenario, Frontier Enterprise was deployed on a combination of dedicated server-class machines and desktops. The firm's development team adapted their trading system to call the Frontier API directly. Whereas Frontier Rapids provides a fast path to employing Frontier, the Frontier SDK (and its API) give developers ultimate control over their Frontier-powered applications. In this case, so tight is the integration, new users are unaware this "desktop application," is powered by a Frontier Enterprise Grid. "Old-timers," however know the difference — back-tests that formerly required more than a week to run are completed in a few hours on Frontier.