Three snapshots of where computational finance actually is now. Parallel computing: the post-2008 explosion of valuation workload (XVA needs billions of valuations per night) outgrew CPUs, and the GPU programming model — with its strengths, bottlenecks and Amdahl's-law limits — is the response. XVA itself: the 2008 crisis broke the single-risk-free-rate, default-free world of Notes 1–3, and the family of valuation adjustments (CVA, DVA, FVA, MVA, KVA) is the repair — conceptually the deepest material in this note, because it makes pricing non-linear and dealer-dependent. ML/AI: the quant equity stack of Note 4 re-implemented with gradient-boosted trees, LLM text embeddings and alternative data, where the binding constraint is not model capacity but the brutal signal-to-noise ratio of financial returns. Each topic ends in the same place: more computation, more model risk, same fundamentals.
Learning objectives. By the end of this note you should be able to:
- explain why nested-MC workloads (XVA) forced the move to GPUs, and contrast CPU vs GPU architectures;
- state Amdahl's law and the data-transfer caveat; describe the MC and FDM case studies and the software stack;
- tell the pre/post-2008 story; define exposure, EE, ENE; write and interpret the CVA formula; describe DVA/FVA/MVA/KVA, wrong-way risk, netting and margin;
- describe the ML stack for cross-sectional stock selection, the signal-to-noise problem, LLM/NLP signal extraction and alternative data.
In this lesson
- 7.1 Parallel computing and GPUs
- 7.2 XVA: the paradigm shift
- 7.3 ML & AI in quant investing
The full lesson, with worked examples and exercise solutions, is for members.
Read the full lessonNew to SelfMadeQuant.com? See membership options or browse the Portfolio Theory and Quantitative Investment syllabus.