A Dynamic parallel algorithm for derivatives pricing and hedging on GPU-CPU heterogeneous systems

Simple item page
dc.contributor.authorMuganda, B. W.
dc.date.accessioned2023-10-02T10:44:09Z
dc.date.available2023-10-02T10:44:09Z
dc.date.issued2023
dc.descriptionFull - text thesis
dc.description.abstractThe use of artificial intelligence in the financial services industry has the potential to transform the sector through greater efficiencies, cost reductions and better tools to draw intelligence from large datasets. The access to computing power which is scalable, accurate and reliable has consequently become a major requirement for the industry due to increased competition, increased products and complexity in models, increased volume of data, stricter regulatory environment and desire for competitive advantage. In this regard, this research provides methodological solutions that would result in accurate and fast system throughput, cost saving and speed acceleration for a financial institution’s financial engineering system by adoption of heterogeneous CPU-GPU parallel architecture with algorithms which are freshly created by drafting from dynamic copula framework for option pricing. This price estimation of options and the assessment of their risk sensitivities under stochastic dynamics namely: stochastic interest rate, stochastic volatility and jumps for varying strikes, maturity and asset classes is a computationally intensive task given the complex nature of the pricing methodologies applied. Models that are much more fully analytical and less complex for pricing derivatives under stochastic dynamics are desirable for much more accurate pricing, investment portfolio construction and risk analysis; and with it an associated system prototype that would provide real-time results. This thesis formulated dynamic parallel algorithms for derivative security pricing and hedging on GPU-CPU heterogeneous systems. This was achieved through the design and implementation of a real-time derivative pricing system prototype supported by a parallel and distributed architecture. The parallel architecture was implemented using hybrid parallel programming on CPU and GPU in OpenCL C, Python and R to provide computational acceleration. The GPU implementation resulted to a peak speed acceleration of 541 times by reducing compute time from 46.24 minutes to 5.12 seconds with the dynamic models under stochastic volatility and stochastic interest rates improving pricing accuracy by an aggregate of 46.68% over the Black-Scholes framework. This adopted approach in this thesis is of practical importance in the harnessing idle processor power, reducing the financial institution’s computational resources requirements and provision of accurate and real-time results necessary in trading, hedging, risk assessment and portfolio optimization processes. Keywords: Dynamic copula, empirical dependence, stochastic volatility, stochastic interest rates, jumps, hybrid GPU acceleration, parallelism
dc.identifier.citationMuganda, B. W. (2023). A Dynamic parallel algorithm for derivatives pricing and hedging on GPU-CPU heterogeneous systems [Strathmore University]. http://hdl.handle.net/11071/13515
dc.identifier.urihttp://hdl.handle.net/11071/13515
dc.language.isoen
dc.publisherStrathmore University
dc.titleA Dynamic parallel algorithm for derivatives pricing and hedging on GPU-CPU heterogeneous systems
dc.typeThesis
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
A Dynamic parallel algorithm for derivatives pricing and hedging on GPU-CPU heterogeneous systems.pdf
Size:
3.23 MB
Format:
Adobe Portable Document Format
Description:
Full- text thesis
Download
License bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
license.txt
Size:
1.71 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
SIMS PhD Theses