Podcast: Piterbarg on Medians and Machine Learning

0

This episode of Quantcast introduces Vladimir Piterbarg, Head of Quantitative Analysis and Quantitative Development at NatWest Markets, and two-time winner of Risk.netQuant of the Year Award.

His latest work, The arc sine law for quantile derivatives, discusses the implications of using the median – or, more broadly, quantiles – to define the gain of a derivative.

The document is based on the fallback rate of Libor contracts developed by the International Swaps and Derivatives Association. Under this protocol, Libor-linked contracts will move to a risk-free rate plus a Libor Adjustment Spread (LAS), which is calculated by taking the median of the difference between the Libor and a replacement risk-free rate over a period. given time. .

“As is often the case with my papers, this was a question from a trader,” Piterbarg explains. “He was looking at what spread was implied by the market that day and he saw numbers he couldn’t quite understand.”

Piterbarg realized that calculating the LAS using the median could lead to inaccurate estimates, primarily due to the dependence of these measures on volatility. In the article, he proposes a fast and stable alternative to estimate the expected value of quantiles.

The LAS for various currencies was set on March 5, de facto put an end to the uncertainty that surrounds it, but the applications of Piterbarg’s article extend beyond these specificities; any derivative that uses quantile expectations can be affected by the same statistical bias. One example is a Napoleon option, which pays the performance of a selected security by its rank in the underlying basket.

Piterbarg also explains his skepticism about machine learning: applying it to finance, he says, often feels like a hammer looking for a nail.

Yet this is an area that he cannot ignore. Piterbarg admits that machine learning is great at interpolating data, although its ability to extrapolate from data is a bit more uncertain. Earlier this year he published Deep asymptotic, an article co-authored with Alexandre Antonov and Michael Konikov, in which they propose a mathematical technique to control the extrapolation limits of neural networks.

He is also not convinced by recent developments in volatility modeling, which he says have generated more hype than significant results. While models like raw volatility have added new weapons to the quantitative arsenal, he believes that, at least in the fixed income markets, they are unlikely to lead to a major paradigm shift.

Index

00:00 Introduction

01:28 The motivation for the research of quantile derivatives

04:42 The results of the study

08:10 The arcsinus law and its relevance for the pricing of derivatives

13:40 Application to Napoleon options and other quantile derivatives

15:15 Is machine learning in finance more hype than substance?

20:25 Current research projects

24:50 The biggest open challenges of quant finance

To listen to the full interview, listen in the player above or download. Future podcasts in our Quantcast series will be uploaded to Risk.net. You can also visit the main page here to access all tracks, or go to itunes store or Google podcasts to listen and subscribe.

Share.

About Author

Leave A Reply