HIPERFIT News April 2015

News from the Research Center on Functional High-Performance Computing for Financial IT

This month HIPERFIT is welcoming Oleksandr Shturmov (Oleks), a new HIPERFIT/DIKU PhD student who will be working with developing Domain Specific Language technology for expressing financial models. Please see the HIPERFIT People List, for information about the complete set of people associated with HIPERFIT.

In April 2015, HIPERFIT is hosting two seminars:

April 14, 15:00-16:00, AUD 7, HCØ (Tuesday)
The Fundamental Theorem of Derivative Trading - Exposition, Extensions, and Experiments
Rolf Poulsen (Math/University of Copenhagen)

April 28, 15:00-16:00, AUD 7, HCØ (Tuesday)
Streaming in Data Parallelism
Andrzej Filinski and Frederik Meisner Madsen (DIKU/University of Copenhagen)

See below for details about the two April seminars. As always, more information about HIPERFIT activities is available from the HIPERFIT web site, including access to recent publications.

The Fundamental Theorem of Derivative Trading - Exposition, Extensions, and Experiments
Rolf Poulsen (Math/University of Copenhagen)
Time: Tuesday, April 14, 15:00-16:00
Place: AUD 7, HCØ

When estimated volatilities are not in perfect agreement with reality, delta hedged option portfolios will incur a non-zero profit-and-loss over time. There is, however, a surprisingly simple formula for the resulting hedge error, which has been known since the late 90s. We call this The Fundamental Theorem of Derivative Trading. This paper is a survey with twists of that result. We prove a more general version of it and discuss various extensions (including jumps) and applications (including deriving the Dupire-Gyöngy-Derman/Kani formula). We also consider its practical consequences both in simulation experiments and on empirical data thus demonstrating the benefits of hedging with implied volatility. The paper is available online.

Streaming in Data Parallelism
Andrzej Filinski and Frederik Meisner Madsen (DIKU/University of Copenhagen)
Time: Tuesday, April 28, 15:00-16:00
Place: AUD 7, HCØ

The presentation will consist of two shorter talks:

Streaming NESL (speaker: Andrzej Filinski, associate professor, DIKU)

Streaming NESL, under development in the HIPERFIT Center, is a refinement of the nested data-parallel functional language NESL. It extends NESL's flattening-based implementation strategy and language-integrated cost model for time (in the form of work and step measures) to also achieve predictable (and optimistic) bounds for space usage. This is done by avoiding complete materialization of intermediate vector-typed values that do not actually require random access. Instead, such values are transparently materialized and processed in "chunks", of size proportional to the available parallel computing resources - from SIMD instructions on single-core CPUs to large GPGPUs.

We give an overview of the status of the project, including a summary of the source language and cost model, a sketch of the implementation strategy in terms of a chunked-dataflow abstract machine (which may also be relevant to other language-technology projects within HIPERFIT), and an outline of some current challenges, both theoretical and practical.

Streaming Accelerate (speaker: Frederik Meisner Madsen, PhD student, DIKU)

The relatively mature embedded language Accelerate brings the power of hardware-accelerated, shape-polymorphic programming with regular arrays to Haskell. Extending Accelerate with streaming functionality, along the lines of our current research, will hypothetically improve its usability and performance for many important use-cases. This talk presents an extension to Accelerate that enables streaming by lifting a non-trivial subset of the language from working on arrays to working on sequences of arrays. The programmer obtains a sequence by slicing an array into a sequence of sub-arrays, and may return to ordinary Accelerate through reduction. Sequence expressions are evaluated efficiently in fixed-size chunks by reusing memory and compiled kernels, allowing sequences to scale in length without requiring additional computing resources except computation time.

Your text caption goes here. You can change the position and width of the caption from the block settings tab.