The 1st HIPERFIT workshop takes place at DIKU, Universitetsparken 1, Copenhagen, May 25th, 9:00-14:00, in room 3-1-25.

Room changed: The workshop will be held in the small auditorium (Lille UP 1).

(Follow the signs from DIKU’s main entrance.)

The workshop is open to HIPERFIT partners, university faculty and students, and the public in general. Participation is free. Registration is required, however, due to space limitations. (We hope not to run out of space, but may have to limit participation if we do.)

To register, please fill out this form by Monday, May 23rd, 23:59 CET.

Students are encouraged to attend to get a first-hand impression of the exciting mix of mathematical finance, programming language and high performance systems problems addressed, be it because they are looking for interesting project or thesis topics or whether they first want to get a sense of which course prerequisites are advantageous to ready oneself for that.

Agenda:

  • 8:30: Arrival and Registration

  • 9:00: Welcome and introduction (Fritz Henglein, DIKU)

  • 9:10: Invited Talk: Logic-based modeling and analysis tools from Microsoft Research (Nikolaj Bjørner, MSR Redmond)

  • 9:45: Break

  • 10:00: The HIPERFIT Research Center: Mission and Outline (Jost Berthold, DIKU)

  • 10:15: Application of Functional Programming at SimCorp (Niels Hallenberg, Simcorp)

  • 10:40: Mortgage refinancing and FX trading strategies - Can domain specific languages and high performance computing help? (Kourosh Rasmussen, IMM/DTU)

  • 11:05: Break

  • 11:20: Data-Parallel Functional Programming (Andrzej Filinski, DIKU)

  • 11:45: Parallel Contract Valuation (Michael Werk & Joakim Rønne, DIKU)

  • 12:10: GPU Parallelisation of Non-Linear Pricing Problems (Jesper Andreasen, Danske Bank)

  • 12:35: Generalizing Execution of Vectorizable Computations by Generating Vector Oriented Byte Code (Troels Blum, eScience Center)

  • 13:00: Wrap up, followed by catered buffet lunch and networking (Fritz Henglein, DIKU)

  • 14:00: End of the workshop

Selected abstracts:

  • Nikolaj Bjørner (MSR Redmond): Logic-based modeling and analysis tools from Microsoft Research.

    This talk takes as starting point the symbolic theorem prover Z3, developed in the Research in Software Engineering group (RiSE) at Microsoft Research, Redmond. Z3 is used in several software engineering tools, including static analysis, test-case generation and verification tools, but also increasingly for model-based development. I provide some details of some the tools that are of interest for HIPERFIT.

    FORMULA (Formal Modeling Using Logic Programming and Analysis) is a modern formal specification language targeting model-based development. It supports concise specifications of abstractions and model transformations using strongly-typed constraint logic programming with algebraic data-types. A main advantage of FORMULA is its model finding and design space exploration facility, using logic programming and Z3. FORMULA can be used to construct complete system models from partial specifications and complex domain constraints. Model search and exploration is then used to traverse the space of complete models until it finds a globally satisfactory design. The F* language integrates refinement types into the core of the functional programming language F#. It uses Z3 to check type checking obligations arising from refinement types. I demonstrate F* on a set of functional program examples.

    Nikolaj Bjørner is a Senior Researcher and head of the Foundations of Software Engineering group at Microsoft Research Redmond.

  • Jost Berthold (DIKU): The HIPERFIT Research Center: Mission and Outline.

    This talk provides a quick overview of the HIPERFIT Research Center. We present our research vision in HIPERFIT, describe the organisation of research and projects in the center, and outline its research areas and themes. One particular focus of this talk is to position HIPERFIT in the landscape of parallel functional languages.

    Jost Berthold is assistant professor at DIKU and researcher in HIPERFIT FP. He received his doctorate degree from the Philipps University of Marburg, Germany, in 2008. His research interests are parallel programming concepts, parallel functional programming. In particular, he has worked intensively on the implementation of parallel Haskell extensions.

  • Niels Hallenberg (SimCorp): Application of Functional Programming at SimCorp.

    SimCorp has used functional programming for a number of years. One example is building short time-to-market financial instruments using a contract language developed by LexiFi (http://www.lexifi.com). Another example is developing automatic test programs using a domain specific language written in F#. In this talk we give a few examples of this and raise challenges and problem areas that might be interesting thesis projects.

    Niels Hallenberg is manager at SimCorp and External Lecturer at IT University of Copenhagen.

  • Kourosh Rasmussen (DTU): Mortgage refinancing and FX trading strategies - Can domain-specific languages and high-performance computing help?

    A number of Stochastic Programming (SP) models on mortgage choice and refinancing for Danish households have been introduced during recent years (Nielsen & Poulsen 2004, Rasmussen and Clausen 2007, Rasmussen and Zenios 2007). Whereas most mortgage banks in Denmark today advise private home owners to finance their property with one loan only, our models suggest that most households are better off with two loans. With regards to mortgage refinancing, our models suggest a higher level of refinancing activity than what is observed today. The empirical study, which is the main subject of this talk, is designed to perform extensive historical ex-ante tests of the advice generated by our model framework for the period 1995-2010. We will compare SP-based advice with current practice which is based on rules of thumb and short-sighted forecasts and show there is significant potential for improvements in refinancing strategies for private home owners. The ex-ante testing of several strategies requires huge amount of computational work and here massive parallelization, for example on a cloud environment, can make similar analysis possible on the desktops of analysts and advisors. The talk is concluded with a few remarks on the possible applications of domain-specific languages within testing of refinancing and trading strategies. We suggest that tools combining domain-specific languages and high-performance computing will give financial institutions a competitive edge on product and strategy development and assessment.

    Kourosh Rasmussen holds a Ph.D in Operations Research. He is a partner in FinE Analytics Aps and an associate professor at DTU.

  • Andrzej Filinski (DIKU): Data-Parallel Functional Programming.

    Data-parallel functional programming aims to combine the expressivity and transparency of the applicative (functional) programming model with the efficiency of bulk-oriented approaches to data processing on modern, highly parallel computing platforms, such as GPGPUs. This talk is intended to serve as an introduction to the theory and practice of data parallelism in a functional setting, and to help dispel some common misconceptions about the subject area.

    Andrzej Filinski is associate professor in the Algorithms and Programming Languages (APL) researcher group at DIKU, working primarily on applied programming-language semantics. He heads the Functional Programming (FP) research area of HIPERFIT.

    Andrzej Filinski is associate professor at DIKU, heading the Functional Programming research area in HIPERFIT. He holds a PhD in computer science from Carnegie Mellon University.

  • Michael Werk and Joakim Rønne (DIKU): Parallel Contract Valuation (work in progress).

    As previously shown by Peyton Jones and Eber (2003), the building blocks of a large class of financial contracts can be expressed as a combinator library. This talk summarizes this approach and then focuses on how to efficiently price financial contracts expressed in this generic way. The strategy chosen is to build an embedded domain specific language for Haskell which is simultaneously expressive enough to implement observables, models and pricers and which supports Monte Carlo simulation on a GPGPU.

    Michael Werk and Joakim Ahnfelt-Rønne are final-year master students in computer science at DIKU, with particular interests in programming languages, functional programming, parallelism and concurrency.

  • Jesper Andreasen (Danske Bank): GPU Parallelisation of Non-Linear Pricing Problems.

    A range of non-linear pricing problems such as Value-at-Risk and CVA (credit value adjustment) can be solved using recursive regression techniques similar to those previously applied to American option pricing problems, as for example in Longstaff and Schwartz (2001). This dramatically reduces the complexity of the calculations and enables accurate calculations even for exotic derivatives.  Technically, it shifts the burden of the calculation from path generation to payoff evaluation. If, in turn, the payoff evaluation can be GPU parallelised it opens for real time calculation of VaR and CVA and tactical risk management based on such quantities. We discuss different strategies for such parallelisation in the context of the Danske Markets’ Jive scripting language.

    Jesper Andreasen heads the Quantitative Research Department at Danske Bank in Copenhagen. Prior to this, Jesper has held positions in the quantitative research departments of Bank of America, Nordea, and General Re Financial Products. Jesper’s research interest include: term structure modeling, volatility smiles, and numerical methods. In 2001 Jesper received Risk Magazine’s Quant of the Year award. Jesper holds a PhD in Mathematical Finance from Aarhus University, Denmark.

  • Troels Blum (eScience Center, KU): Generalizing Execution of Vectorizable Computations by Generating Vector Oriented Byte Code.

    In my work I have used a new approach to the problem of closing the gap between productivity tools, and highly parallel systems, which allows a high degree of modularity and there by reuse. My approach involves creating a model, cphVB, in which the computing devices (hardware) are viewed as engines processing vectorized instructions, called Vector Engines. It defines a clear and easy to understand byte code language, which is processed by these Vector Engines. cphVB also contains a protocol to govern the safe, and efficient exchange, creation, and destruction of model data.

    I have made a proof of concept implementation of the cphVB model for NumPy, that uses CUDA devices for executing the byte code. This implementation has proven very efficient, and cemented that the cphVB model, with its byte code, is a viable solution to the problem of exposing exotic hardware to languages, and other front-ends, that scientists, and other programmers use every day.

    Troels Blum is working in the research group of Brian Vinter, who is head of the High Performance Systems research area in HIPERFIT. He works on the HPHP (High Performance - High Productivity) project at the eScience center. His primary focus since obtaining his M.S. degree is GPGPU programming.