Algo Suite Technical Architecture

Document Sample
Algo Suite Technical Architecture Powered By Docstoc
					Algo Suite Technical Architecture White Paper
APRIL 2005

SPONSORED BY:

COPYRIGHT NOTICE © ALGORITHMICS INCORPORATED, 2005 This document contains information proprietary to Algorithmics Incorporated and/or its affiliates (“Algo”) and is protected by international copyright law. The information contained herein may not be copied or duplicated in whole or in part, without the prior written consent of Algo. ALGO, ALGORITHMICS, AI & design, KNOW YOUR RISK, MARK-TO-FUTURE, RISKWATCH, ALGO CAPITAL, ALGO COLLATERAL, ALGO CREDIT, ALGO MARKET, ALGO OPVANTAGE, ALGO OPVANTAGE FIRST, ALGO RISK, and ALGO SUITE are trademarks of Algorithmics Trademarks LLC. All other trademarks referenced herein are the property of others. Algorithmics Incorporated 185 Spadina Avenue Toronto, Ontario M5T 2C6 Canada

Algorithmics Incorporated

2

Table of Contents
Algo Suite: Overview .................................................................................................................................. 4 Algo Suite: The Technology ................................................................................................................... 5 Mark-to-Future: The Methodology ........................................................................................................ 7 Business Benefits of Mark-to-Future..................................................................................................... 8 Functional Architecture Overview............................................................................................................... 9 Data .......................................................................................................................................................... 10 Data Acquisition and Processing Layer ............................................................................................. 11 Analytics .................................................................................................................................................. 14 Scenario Generation and Calibration .................................................................................................. 14 Mark-to-Future Simulation Engine ...................................................................................................... 14 Simulation and Analytics Layer .......................................................................................................... 15 Distributing the Simulation and Analytics Layer across a Grid ........................................................... 17 Presentation and Reporting ..................................................................................................................... 19 Aggregation and Reporting Layer – Post Cube Environment ............................................................. 19 Distributing the Aggregation and Reporting Layer on a Grid ............................................................. 22 User Experience .................................................................................................................................. 23 Algo Suite and Linux on HP BladeSystem .............................................................................................. 26 About Algorithmics ................................................................................................................................... 29 Algo Services ...................................................................................................................................... 29 Further Information ............................................................................................................................. 30 Contacting Algorithmics ...................................................................................................................... 30 About HP ................................................................................................................................................... 30 ACKNOWLEDGEMENTS:
Algorithmics would like to thank Hewlett-Packard for sponsoring this white paper.

Algorithmics Incorporated

3

Algo Suite: Overview
Algorithmics provides an advanced, market proven, enterprise-wide risk management solution. By integrating different risk functions with a single, unified, scalable architecture, Algorithmics helps financial institutions reduce costs, improve the quality of risk information, meet regulatory requirements and allocate capital more efficiently. Currently, Algorithmics provides enterprise-wide solutions to an extensive client base including global banks, asset management firms, insurance companies, pension funds, central banks, exchanges, regulators, corporations, utilities and brokerage firms. Senior management, traders, fund managers, risk managers and IT specialists all benefit from Algorithmics’ advanced risk analysis and decision support. Algorithmics’ suite of solutions meet the challenges of today’s financial environment with an integrated series of advanced solutions that set a new standard in enterprise risk management. Algo Suite is the first software of its kind to provide a consistent, proven platform for the integration of market, credit, asset liability and operational risk functions. By calculating the optimal risk and reward trade-off across the enterprise, Algo Suite enables institutions to realize substantial business benefits, remain competitive, meet regulatory requirements and maximize shareholder value.

Algorithmics Incorporated

4

Algo Suite: The Technology
Institutions are under pressure from regulators and shareholders alike to quantify their exposures more accurately and on an enterprise-wide basis. In response to these needs, Algorithmics has built an integrated, scalable and powerful architecture that allows institutions to respond to market developments, new business demands, and evolving regulatory requirements in a way that is both informed and timely. POWERFUL, FLEXIBLE ENTERPRISE DATA MANAGEMENT AND ANALYTICS. Algorithmics’ risk technology is built upon a set of components that cover data management, scenario generation, market, credit and operational risk analytics, limits management, collateral management and reporting. All components share a common data model, which maintains consistency among applications and streamlines the way data is passed among the components. This approach not only allows institutions to select those components that cover their specific areas of business, but also enables institutions to scale up their systems to meet the advanced analytical demands of enterprise risk management. OPEN, SCALABLE FRAMEWORK. The component-based approach and Algorithmics’ adherence to industry standards make it easier for firms to integrate Algo Suite into their IT infrastructure. Algo Suite can be easily extended to accommodate new asset classes, financial models, simulation methods, post-processing applications and emerging lines of business. Deploying Algo Suite on Linux on the HP BladeSystem platform makes it easy to tailor the computing infrastructure to meet the specific needs of each task in a dynamic and adaptive way. This provides comprehensive flexibility and scalability across both software and hardware dimensions. Thus, firms can seamlessly extend their risk management infrastructure to incorporate new risk management tools, business divisions, geographical areas, instruments and methods, and lines of business as they evolve. ROBUST DATA HANDLING. Algo Suite supports real-time, batch, and intraday data acquisition. Algorithmics data handling components are built using platform- and database-independent technologies such as C++, Java and Enterprise Java Beans, and the XML standard. To communicate and exchange messages and data between components, Algo Suite has the potential to support any third-party middleware. Algorithmics has also introduced its own version of an XML-based language, called Algo Input Markup Language (AIML) as a new standard for trades, to simplify the integration of middle-office risk management/measurement systems and front-office trading systems.

Algorithmics Incorporated

5

REAL-TIME LIMITS AND RISK ANALYSIS. Algo Suite’s Mark-to-Future technology is built upon proven, high-performance, real-time risk analytics delivered via user-friendly, web-enabled interfaces. By empowering users to obtain various types of risk analysis on demand, such as limits checking, ‘what if’ trade analysis, portfolio ‘slicing and dicing’, and marginal risk analysis, a firm can measure and monitor its risk without the need to draw on unrealistic and unreliable analytical approximations. WEB-ENABLED RISK MANAGEMENT SERVICES. Algo Suite’s component-based architecture, use of industry standards, and scalable processing capabilities make it an adaptable technology for Application Service Provision (ASP) by firms that wish to offer their customers advanced risk management services online.

Algorithmics Incorporated

6

Mark-to-Future: The Methodology
At the heart of Algo Suite is Mark-to-Future, a robust and forward-looking framework that links disparate sources of risk and provides a means for calculating the risk-reward trade-off within a single, unified framework. By explicitly incorporating the passage of time, the evolution of scenarios, and the dynamics of multiple portfolio holdings over time, Mark-to-Future provides a flexible and unifying platform for managing future risk, which supports a wide range of advanced analytical tools and risk measures. New sources of risk, as well as innovations in risk management best practice, can be readily accommodated, so that neither financial institutions nor regulators are locked into a particular formulaic approach. Furthermore, Mark-to-Future’s extensible risk architecture can be leveraged within one institution or across several disparate business units. Through Mark-to-Future, Algo Suite’s innovative approach to risk analysis offers many distinct advantages: FULL INTEGRATION OF ENTERPRISE-WIDE RISKS. Mark-to-Future defines risk factor scenarios to compute future distributions of value. Because individual risk factors can evolve jointly (and arbitrarily) over time, Mark-to-Future allows users to capture the relationships between disparate sources of risk, over multiple time steps. For example, many wellknown financial disasters, including recent corporate failures, occurred precisely because of the high correlation between market and credit risk in stress periods. Such occurrences are naturally modeled through scenarios where adverse changes in market conditions trigger adverse changes in credit quality. With Mark-to-Future, market, credit, asset liability and operational risk may naturally be integrated within a common enterprise-wide framework. A TRANSPARENT, EASILY UNDERSTOOD APPROACH. Within Algo Suite, Mark-to-Future scenarios are the drivers of all future uncertainty and, as such, become the language of risk. Since the risk measures that result from scenarios can be explained in a straightforward manner, individuals with different levels of sophistication can contribute to risk-reward discussions through the focus on plausible scenarios. A FORWARD-LOOKING FRAMEWORK. Mark-to-Future’s scenario-based framework produces risk and reward measures that explicitly capture the passage of time. Consequently, challenging issues such as credit mitigation techniques, portfolio path dependency, settlement and reinvestment, dynamic re-balancing, liquidity and thin market effects can be effectively addressed.

Algorithmics Incorporated

7

INCREASED ACCURACY. Whether for market, credit or operational risk, Mark-to-Future’s use of scenarios overcomes the limitations of simplistic analytical methodologies that make onerous assumptions about risk factors or instruments for mathematical tractability. A wealth of techniques can be used to generate scenarios from history, advanced Monte Carlo models or subjective views of the world. Furthermore, Mark-to-Future’s extensive valuation coverage provides user flexibility in modeling even the most complex portfolios. COMPUTE ONCE, USE MANY TIMES. The realization of Mark-to-Future values across positions, scenarios and time steps determines any risk or reward measure. Thus, the addition of a new position, scenario or time step requires only the simulation of the new values, which are then appended to the previously computed results. Previously simulated results need not be recalculated; only the calculation of risk or reward measures need be repeated. Mark-to-Future provides computationally efficient, scenario-based risk management tools that allow a firm to identify a portfolio’s most significant sources of risk, indicate how potential trades impact on its risk-return trade-off, and recommend optimal portfolios. These tools can be used for both end-of-day risk analysis and real-time risk decision support.

Business Benefits of Mark-to-Future
The advanced technology used in the Mark-to-Future framework allows effective risk management with fewer resources. Algo Suite’s server architecture can persist Mark-to-Future data and deliver following business benefits: • Facilitates ad-hoc ‘slicing and dicing’ of results by performing multiple re-aggregations within a single simulation. • Extends the overnight processing window by computing exposures of some positions while information is gathered for portions of the portfolio. • Enables multiple simulators, sharing the same scenario set definition, to individually process different portions of the portfolio. • Allows error correction without having to re-compute the whole overnight run. • Provides on-demand, ‘what-if’ risk analysis without re-computing the full portfolio exposure profiles. Shares data across the organization, eliminating the need to re-compute risk figures at the branch as well as head-office levels.

Algorithmics Incorporated

8

Functional Architecture Overview
Algo Suite is comprised of a series of individual components. Conceptually, the system is divided into three layers: • • • Data Acquisition and Processing Simulation and Analytics Reporting and Presentation

Algo Suite is a flexible risk management solution because it allows end users to customize the reporting and presentation layer. In this framework, Algo Suite essentially creates a fourth “virtual” layer whereby different users share the same underlying risk architecture.

Figure 1: Algo Suite Functional Overview

Algorithmics Incorporated

9

Data
To meet the complex task of enterprise data management, a risk management solution must deliver a data structure that is robust, easily customizable and fully extensible. Algo Suite’s data layer design is based on a common meta-data structure that is shared by all of the analytical and data components. This flexible structure allows all applications to adapt easily to changing requirements, by modifying existing objects or adding new risk analyses such as, credit or asset and liability. Algo Suite’s input database is the first staging area as it collects all the data necessary for the analytical process, including; instrument terms and conditions, positions/holdings, portfolio information, risk factor data, portfolio structure, and reporting dimensions. Drawing upon many years of client experience, the data model and the intuitive GUI-based user utilities have been designed to optimize risk management. The input data staging area consists of the following components: • • • A flexible data model defined by meta-data within the database. A GUI-based administration utility to manage and extend the meta-data definition. A GUI-based utility to facilitate the input data management and data filtering capabilities of the database. • A data communication layer that serves multiple data requests for several client applications.

In conjunction with Algo Suite’s mapping utility, the input database supports daily batch and intra-day risk management functions. Numerous features, such as complex filter query building, data mapping documentation, data reconciliation and data manipulation allow for easy maintenance and distribution of portfolio and risk factor information to various clients and user sessions. The input database is critical to managing an effective distributed processing environment.

Algorithmics Incorporated

10

Data Acquisition and Processing Layer
The data requirements to run an enterprise-wide risk system are large. The risk system must load the “terms and conditions” (T&C) or descriptive data for each instrument (i.e. coupon rate, maturity date, strike price, etc.) as well as all the market data required to theoretically value each instrument (e.g. U.S. LIBOR Curve, option volatility surface, etc.). The system must load the amount owned of each instrument (i.e. the portfolio positions) as well as any portfolio hierarchy and aggregation attributed uses to group the instruments and positions for reporting. Finally, market data must be captured for historical and Monte Carlo scenario generation. Typical Algorithmics clients load data from a variety of source systems such as front-office trading systems; middle-office position keeping systems, internal data warehouses, and external data feeds. Data from each of these potential sources must be extracted and loaded in Algo Suite. Algorithmics offers a pre-packaged data loading system that works with Bloomberg Data License. Figure 2 describes how data is acquired through the Bloomberg Data Loading System.

Figure 2: Data Acquisition and Processing Layer – Bloomberg Example

Algorithmics Incorporated

11

Data License Back Office: This is the process whereby Bloomberg automatically creates several generic data files that are posted to the Bloomberg Data License FTP site throughout the day. This process is used for most exchanged traded products. This back-office push mechanism is used to extract the terms and conditions of vanilla fixed income securities, equities, futures, and other derivative products. The main advantage of this mechanism is that it is easier to use when dealing with large volumes of data. This data, however, must also be filtered, transformed, and reformatted at a later stage to become “RiskMapper compatible.” As a result, some certain files must be merged and rudimentary data formatting routines applied. The Data Filter is a specialized software tool from Algorithmics used to create input files from the Bloomberg Data License files based on each region and instrument type for RiskMapper. The Data Filter removes securities from the overall pool. For example, in order to select only fixed income instruments that are issued in G7 countries, one would identify instruments for which the Country of Issue is not a G7 country and instruct Data Filter to remove it. Algorithmics provides specific filtering rules for many instrument types. The Bloomberg Data License Data Transformer software application can be used to replace field values, make field name changes and multiple instrument comparisons. Its main use, however, is to select the required data fields for input into RiskMapper. This selection process usually reduces the number of fields from several hundred to one dozen required for pricing. The application contains specific transforming rules for each instrument type. Files are normally “RiskMapper compatible” after this last process and therefore, ready to be mapped by RiskMapper. Data License per Security: This process extracts only the information needed from a Bloomberg data feed. This process extracts only pre-specified data fields on a per security basis. The pull mechanism is similar to the Bloomberg functions widely used in by terminal users. The main advantage of this process is that the data is presented in the right format from the onset (since Algorithmics created the templates). The ASP production environment at Bloomberg uses the pull mechanism (per Security) to extract current risk factor information such as curves, FX rates, and volatility surfaces. It can also extract historical time series of risk factor information on a daily basis to populate Algo Scenario Engine. Finally, this mechanism is used to extract the terms and conditions of treasury bonds, callable bonds, callable convertibles bonds, etc. Algo Scenario Engine (ASE): ASE is an advanced scenario generation and management application which is a primary component of Algo Suite’s Mark-to-Future framework. It is a component-based application that allows the user to customize and define scenarios, and then link them together to build a scenario set. The scenarios created in ASE are based on historical time series of data for observable market rates and prices – referred to by Algorithmics as risk factors. ASE uses risk factors to generate scenarios, which are later used for simulation and stress testing of instruments.

Algorithmics Incorporated

12

RiskMapper (RM): RiskMapper is the Algo Suite component that receives, translates and transforms financial data from its original input form so that it can be used by other Algo Suite components. RiskMapper is used to load instance data such as trades, positions and terms and conditions data into AIDB. Algo Input Database (AIDB): AIDB stores and maintains all the data required to define financial instruments and to perform market and credit risk calculations in Algo Suite, including financial instruments terms and conditions, positions, portfolio structures, market data and pricing models. AIDB consists of the following components: • • A flexible data model defined by meta-data within the database A supporting set of administration applications that includes the Meta-Data Manager (MDM) and the Input Data Manager (IDM) • A data communication layer supported by the Algo Data Server (ADS)

AIDB supports daily batch and intra-day risk management calculations. It also stores and maintains a consistent source of validated data used by Algo Suite risk analytics tools.

Algorithmics Incorporated

13

Analytics
At the core of Algo Suite is a powerful analytics layer comprised of a series of analytical engines, including a scenario generation engine and the Mark-to-Future (MtF) simulation engine. SCENARIO GENERATION AND CALIBRATION Algo Suite’s core scenario generation server allows multi-user access to scenarios directly with client applications. It produces and stores all types of historical scenarios and single- or multi-step Monte Carlo scenarios. The scenario generation engine can also simulate curve tilts, perform Principal Components Analysis, generate conditional scenario simulations, and view simulated data. It has a powerful and extensible scenario generation library that ensures a robust, accurate and transparent scenario generation process. Scenarios can be conveniently requested by the applications through a direct server connection. MARK-TO-FUTURE SIMULATION ENGINE The MtF simulation engine inside Algo Suite is the most advanced analytic risk engine available in the market today. It provides a complete set of methodologies to simulate, measure, restructure and manage both market and credit risk. With extensive instrument coverage and a flexible and extensible design, it is the world’s leading engine for enterprise risk management. Furthermore, MtF has a distributive nature and allows multiple simulation engines to be run simultaneously. This separates the computationally intensive task of pricing securities across scenario and time from the comparatively simple function of extracting the significant summary statistics. The ability to distribute intensive computations over an unlimited number of simulation engines across the entire network enables Algo Suite to tackle even the largest portfolios. The Linux on HP BladeSystem platform provides the ideal distributive system environment to complement the MtF simulation engine. IT planners can determine a tradeoff point for performance versus cost and deploy the MtF software on the optimum number of blade servers to meet that target. An increase in computer power is easily accomplished by simply plugging more blade servers into the BladeSystem infrastructure or by upgrading to new, more powerful server blades as CPU technology evolves. Finally, the workloads that are distributed among all of the blade servers in the infrastructure can be easily shifted to adapt to both acute and long-term processing requirements.

Algorithmics Incorporated

14

Simulation and Analytics Layer
The Mark-to-Future methodology is based on full revaluation where all instruments are revalued across every scenario and at every time step and the simulation results are written to disk in the form of Markto-Future cubes. Specifically, the simulator, RiskWatch, must load all the instruments and market data in order to perform a mark-to-market on each instrument. RiskWatch then reads the scenarios, which are defined as changes to the market data and other risk factors or inputs to the valuation models, and revalues each instrument across every scenario.

Figure 3: Simulation Layer

Algorithmics Incorporated

15

Algo Scenario Engine (ASE): ASE is an advanced scenario generation and management application, which is a primary component of Algo Suite’s Mark-to-Future framework. It is a component-based application that allows the user to customize and define scenarios and then link them together to build a scenario set. The scenarios created in ASE are based on historical time series of data for observable market rates and prices – what Algo refer to as risk factors. ASE uses risk factors to generate scenarios, which are then used by RiskWatch for simulation and stress testing of instruments. Algo Input Database (AIDB): AIDB stores and maintains all the data required to define financial instruments and to perform market and credit risk calculations in Algo Suite, including financial instruments terms and conditions, positions, portfolio structures, market data and pricing models. AIDB consists of the following components: • • A flexible data model defined by meta-data within the database A supporting set of administration applications that includes the Meta-Data Manager (MDM) and the Input Data Manager (IDM) • A data communication layer supported by the Algo Data Server (ADS)

AIDB supports daily batch and intra-day risk management calculations. It also stores and maintains a consistent source of validated data used by Algo Suite risk analytics tools, such as RiskWatch. RiskWatch: The RiskWatch application is the main analytical tool underlying all of Algorithmics’ solutions. RiskWatch is a robust enterprise-wide risk application designed to simulate financial instruments through time and across scenarios using a full-valuation methodology. RiskWatch comes with a full set of valuation models for more than 200 instrument types. In addition, Algorithmics provides bridges to many common third-party valuation libraries and can be extended to incorporate proprietary pricing functions.

Algorithmics Incorporated

16

Distributing the Simulation and Analytics Layer across a grid
Distributing the simulation and analytics layer across a grid requires partitioning the instrument data within AIDB into small groups and then starting up multiple instances of RiskWatch to create MtF cubes for each partition or unit of work. In the figure below, each simulation or RiskWatch process is run on a separate CPU. These CPUs can be all on a single server or multiple servers. Each RiskWatch process loads a set of instruments and simulates those instruments across all scenarios and for all time steps. At the end of the simulation, RiskWatch writes out the results to a file called a MtF cubelet. The “Logical MtF Cube” used for aggregation and reporting consists of all physical files (i.e. all the MtF cubelets) created parallel during the risk run.

Figure 4: Distributing the simulation layer across a grid

AIDB Filters: The AIDB filters are used by RiskWatch to load data from AIDB. For example, a filter named “All Callable Bonds” might be defined to contain all the instruments in AIDB with type equal to “callable bond.” When RiskWatch reads this filter, all of the callable bonds are extracted from AIDB. Filters can be designed to load dependencies so that a filter loading a derivative, such as an equity option, will also load any underlying instruments required for valuation. Furthermore, filters can be defined to load data stored in AIDB including; instruments, positions, portfolio hierarchies, and market data. AIDB contains a tool named the Filter Builder Utility that defines and stores filters for use by RiskWatch.

Algorithmics Incorporated

17

FilterSplit: The FilterSplit tool dynamically splits a filter into partitions based on a set of policies. This allows a large filter to be divided into small units of work for optimal load balancing across a grid. For example, suppose there is a master filter defined as “All Instruments” that contains all of the instruments that need to be simulated. One splitting policy might be “Instrument type -> fixed groups of 1000.” In this case, all of the instruments are sorted by type and then partitioned into groups of 1000. Using this policy, each RiskWatch loads a filter partition that contains at most 1000 instruments, effectively limiting the simulation time for each RiskWatch job. By using FilterSplit to create many filter partitions, it is possible to running multiple RiskWatch jobs in parallel across a grid. Filter split policies can be defined using any aggregation key available within AIDB. Common keys include; portfolio, counterparty and instrument type. GridServer: GridServer from DataSynapse manages multiple RiskWatch jobs across the grid. GridServer provides several important features for optimal performance. GridServer schedules the filter partitions for simulation across the available servers in the grid. Scheduling jobs is very flexible with rules such as “Run a job on any Linux server” to “Run a job on the server with hostname ‘Algo33’”, GridServer provides resiliency and fault tolerance by restarting and simulation jobs that do not complete successfully. GridServer also allows the Algo Suite implementation to virtualize the machines used as compute engines at any point in time. In other words, a typical Algo Suite implementation does not need to know the number of compute engines in the grid; a job can start with 24 engines and if 12 more become available at any time afterwards, GridServer will immediately start to schedule jobs across all 36 engines. In practice, the most robust and flexible computing infrastructure can be achieved by using GridServer to distribute RiskWatch processes among a set of server blades in a Linux on HP BladeSystem environment within a single physical datacenter. One possible configuration might dedicate a number of blade servers to RiskWatch during normal business hours, then expand the grid after hours by adding interactive blade servers that have gone idle into the RiskWatch pool. Storing the MtF cubelets on disk: When distributing the simulation across a grid, each RiskWatch computes a small part of the entire simulation and saves the simulation results to disk. The simulation results from single RiskWatch are called a “cubelet” and all the cubelets from a single simulation make up a single “Logical MtF Cube.” By definition, all cubelets belonging to the same “Logical MtF Cube” have the same scenario and time step dimensions. Each cubelet contains a different subset of instruments. MtF cubelets are written to disk as individual files in a proprietary binary format optimized for performance. For aggregation and reporting purposes, it is not possible to mix and match cubelets from different MtF simulations. As a result, all cubelets from a MtF simulation must be written to unique set of directories. The aggregation engine assumes that all cubelets found in a specified set of directories belong to a single “Logical MtF Cube.”

Algorithmics Incorporated

18

In general, Algorithmics recommends writing all the simulation results to network-attached disk storage (SAN, NFS mounted directories, etc) rather than local disks. Aggregation and reporting will fail if a cubelet file is “missing.” When writing cubelets to local disk, there are many conditions/failures that might lead to a missing cubelet file required for reporting (server crash, disk full error, network connectivity problem, etc.) Writing cubelets to network storage is a requirement for virtual grids where machines are dynamically allocated and can be reassigned during a risk reporting cycle.

Presentation and Reporting
Algo Suite ensures effective dissemination of risk information across the entire firm by supporting multiple reporting technologies to meet diverse user requirements. The tools and reporting applications are tailored to the end-user needs, and provide the appropriate level of functionality and detail for the user. At the core of the presentation layer are two fully integrated components that provide a single reporting platform. The first is the Mark-to-Future aggregation engine and its associated report server for computing and querying risk information, and the second is a flexible and robust database for storing results. Algo Suite provides both the diagnostic tools and the end-user applications to analyze and present risk information in an effective and comprehensive manner. For example, risk managers can use a flexible, Java-based diagnostic tool to interrogate, dissect, and drill down on all Mark-to-Future data. On the other hand, the bank’s front office can use the intuitive and easy-to-use web interfaces to access advanced risk analytics, perform what-if analysis, and check and administer position limits.

Aggregation and Reporting Layer – Post Cube Environment
Aggregation and reporting is part of the post-cube architecture. The MtF aggregation engine is a distributed server that calculates portfolio and other aggregation level results from MtF tables. Acting upon requests from client applications, it retrieves data from the MtF cubes generated by the simulation engine based on an aggregation hierarchy and performs a range of statistical analyses. Therefore, the batch overnight run can be used several times to perform risk analysis for different portfolios

Algorithmics Incorporated

19

Figure 5: Aggregation and Reporting

Algorithmics Incorporated

20

Risk Portal: Each Risk Portal is a java GUI running within a browser that an end user can use to configure a report. In other words, a user can select the type of report, the portfolio or aggregation group, the scenario set, the risk analytic(s), and the method for displaying the results (e.g. table or graph). The servlet/report generator translates the GUI selections into a series of XML requests that are used to query the MtF cubes, extract the necessary simulation results, and compute any risk analytics. Report Manager (RPM): RPM receives XML report requests from one or more Risk Portals. RPM sends the report request to the Algo Risk Engine for processing. Algo Hierarchy Server (AHS): AHS stores the portfolio hierarchies, position units and aggregation keys required to aggregate Mark-to-Future simulation results into groups for reporting. Algo Risk Engine (ARE): The Algo Risk Engine extracts the Mark-to-Future results for an aggregation group, scales the results by position units and computes the requested risk analytic. In a distributed environment, each local ARE knows about a subset of the physical MtF cubes and each local ARE can only work with the simulation results of the instruments contained in its subset of MtF cubes. The global ARE is responsible for coordinating the local AREs and passing the results back to the report manager. Data is passed back to the global ARE from all the local AREs and the global ARE computes the risk statistics that are passed back to the report manager. ARA Risk Portal: To compute most risk reports, a user will want to aggregate the instruments’ results into groups (e.g. by portfolios, by currency, etc), scale the simulation results by position units and then compute a statistical analysis (e.g. VaR.).

Algorithmics Incorporated

21

Distributing the Aggregation and Reporting Layer on a Grid
Distributing the presentation and reporting layer across a grid requires partitioning the individual MtF cubelet files that make up a single Logical MtF cube into groups and mapping these groups to local ARE processes. The mapping between local ARE processes and MtF cubelet files can be one-to-one, one-tomany, many-to-one, or many-to-many. Although many factors determine the optimal mapping, in general a one-to-many mapping (where one local ARE process is mapped to several MtF cubelet files) is the best choice.

Figure 6: Algo Risk Engine (ARE) running in a distributed environment.

Algorithmics Incorporated

22

User Experience
The Algo Risk Application (ARA) is Algorithmics’ premier risk portal for aggregating and reporting risk. The highly customizable user interface runs within a web browser allowing for easy distribution of risk reporting capabilities throughout the organization. Security for portfolios, benchmarks and other data can be set at individual user level. Users can create report templates by choosing a screen layout, the risk data elements to display and the data formats (tables, pie charts, bar graphs, histograms, etc.) Individual reports are created by selecting portfolios, benchmarks, scenario sets, time horizons, aggregations, and other data elements for analysis. The ARA provides a rich set of analysis techniques allowing the user to monitor and investigate risk. Among these include the ability to set and monitor limits, view heat maps, apply risk hedges, and perform “what-if” trade analysis.

Figure 7: A concentration report where the portfolio is aggregated by two dimensions, first by sector, and then by country. The report shows the information in both a pie chart and a tabular grid.

Algorithmics Incorporated

23

Figure 8: A VaR report showing both the PnL distribution and the one-day VaR at several confidence intervals.

Figure 9: A stress test report showing a portfolio's interest rate
sensitivity to eighteen individual curve shock scenarios.

Algorithmics Incorporated

24

Figure 10: A risk report showing the active return of the portfolio relative to a benchmark, broken out by country and duration bucket.

Algorithmics Incorporated

25

Algo Suite and Linux on HP BladeSystem
Why Linux on HP BladeSystem?
The accelerated adoption of the Linux operating system in the financial services sector represents one of the largest growth areas in the IT industry. The adoption of blade servers by this sector is on a parallel ramp. Algorithmics solutions are optimized to run on Linux and, with DataSynapse GridServer, in the multi-server environments where Linux excels. Recent benchmark testing by Algorithmics confirms that the Linux version of Algo Suite running on a cluster of 28 dual-processor HP Linux servers provided superior performance and scalability compared to non-Linux versions running on comparable proprietary platforms. A key advantage of the HP BladeSystem platform is its deployment flexibility. Customers can use a blade server as a Linux system, then easily redeploy it in another environment—even as a Microsoft® Windows® server, for example. These personality changes can be easily implemented over the course of a year or on a daily basis, depending on business needs. The ability to dynamically reallocate and redeploy computing resources in the fast-paced financial industry is a marked improvement over the inflexible and costly proprietary environments of the recent past. AN EXAMPLE CONFIGURATION The Linux on HP BladeSystem configuration described in this section represents one possible scenario for deploying the Algo Suite models depicted in Figures 4-6 in a scalable, adaptive manner. The number of blade servers assigned to each component is loosely based on an HP rack server configuration used by Algorithmics to generate the benchmark results referenced above. Although the BladeSystem configuration depicted in this example is intended in concept to achieve a similar level of performance, actual performance of any specific configuration cannot be guaranteed until benchmark tests have been run.

Algorithmics Incorporated

26

CONFIGURING THE SIMULATION LAYER The logical configuration depicted in Figure 4 requires a cluster or grid of RiskWatch servers that share access to a common Algo Input Database and a common file store for the MtF cubelets. To approximate the RiskWatch cluster benchmark configuration within a Linux on HP BladeSystem environment, the example configuration uses 16 two-processor BL30p servers. These double density “compute-blades” offer the highest computing power per cubic foot of any cluster configuration. However, if these servers are to perform double duty for functions other than RiskWatch, similarly configured full-height BL20p blade servers could be used as an alternative. To provide access to the shared storage pool for MtF cubelets, a pair of full-height BL20p blade servers is used to provide NFS access to a common directory within the system SAN. The NFS server pair is configured with the PolyServe Scalable File Services software to provide both high throughput and high availability access to the MtF cubelet storage pool. The BL20p blade server was chosen for this layer on the basis of its robust support for FibreChannel connectivity via the optional SAN interconnect module. The PolyServe Scalable File Services layer can be expanded to up to 16 nodes where additional throughput and availability are required. CONFIGURING THE AGGREGATION AND REPORTING LAYER The logical configuration depicted in Figure 6 requires a cluster of Algo Risk Engine servers that share access to a common file store containing the MtF cubelets generated by the RiskWatch cluster. In the benchmark tests run by Algorithmics, the 16 two-processor rack servers were re-provisioned after the simulation phase as the ARE cluster. The total Linux on HP BladeSystem configuration for the RiskWatch and ARE cluster is shown in Figure 11. Note that although this configuration depicts 16 servers as assigned to each function, the topology of the Linux on HP BladeSystem virtual computing environment enables any number of servers to be assigned to either function at any time. This provides unparalleled flexibility to match computing resources to business needs in an adaptive manner.

Algorithmics Incorporated

27

Figure 11: HP Bladesystem configuration for Algo Suite

Algorithmics Incorporated

28

About Algorithmics
Founded in 1989, Algorithmics is a recognized leader in enterprise risk management. Following its acquisition by the Fitch Group in January 2005, Algorithmics has been combined with Fitch Risk. Together, the Algorithmics and Fitch Risk team is the world’s leading provider of enterprise risk management solutions and services that enable financial institutions to effectively understand and manage their financial risk. Algorithmics’ and Fitch Risk’s combined client base includes more than 200 of the world’s leading financial institutions, representing more than 60 of the 100 largest financial institutions in the world. The combined team represents one of the largest dedicated enterprise risk management team in the world, with more than 550 dedicated risk professionals, located in 18 offices in key international markets. Algorithmics was recently recognized as the dominant enterprise risk solution provider in market, credit and operational risk in Risk Magazine’s 2004 Technology Rankings.

Algo Services
From its earliest days, Algorithmics has been providing clients with not only advanced risk software solutions but also experienced teams to scope, design, implement and support these systems. Drawing on the resources of the largest research and development team in the risk management industry, the Algorithmics services team has helped financial institutions around the world to achieve their goals through a combination of leading-edge solutions and a broad range of consulting and support services. The services team employs a number of critical tools to facilitate their efforts including Algo Lab, an inhouse solution center designed to facilitate product testing, services support and knowledge transfer. Alchemy, a proven client engagement methodology designed to help define client requirements and execute sophisticated implementations; and Algo Academy, a program designed to deliver relevant training to clients on Algo Suite components.

Algorithmics Incorporated

29

Further Information
Algorithmics provides a variety of information related to risk management, and also to its specific solutions. These publications may be of interest: Mark to Future: A Framework for Measuring Risk and Reward Enterprise Credit Risk Using Mark-to-Future Seeing Tomorrow: Weighing Financial Risk in Everyday Life

Contacting Algorithmics
TO ACCESS ALGORITHMICS ONLINE: The corporate web site: www.algorithmics.com TO ARRANGE DEMONSTRATIONS, DISCUSS THE SOLUTIONS, OR ANY OTHER MATTERS: E-mail: sales@algorithmics.com

About HP
HP delivers adaptive enterprise solutions for mission critical environments including brokerage, trading, and asset management operations across the global financial markets. These solutions meet market demand for order routing and management, trading support, analytics, risk management, and transaction processing for the buy-side, the sell-side, and treasury operations. Optimized for performance on proven platforms: HP-UX, Linux, OpenVMS™, NonStopTM and Microsoft Windows, HP offerings include leading ISV applications and are augmented by HP Services deployments. HP is a technology solutions provider to consumers, businesses and institutions globally. The company's offerings span IT infrastructure, personal computing and access devices, global services and imaging and printing. For more information about HP, visit www.hp.com.

Algorithmics Incorporated

30


				
DOCUMENT INFO
Shared By:
Categories:
Stats:
views:601
posted:9/6/2008
language:Spanish
pages:30
Laura Trunk Laura Trunk
About