Neural networks and arbitrage in the VIX: A deep learning approach for the VIX

Joerg Osterrieder*, Daniel Kucharczyk, Silas Rudolf, Daniel Wittwer

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review

2 Downloads (Pure)


The Chicago Board Options Exchange Volatility Index (VIX) is considered by many market participants as a common measure of market risk and investors’ sentiment, representing the market’s expectation of the 30-day-ahead looking implied volatility obtained from real-time prices of options on the S&P 500 index. While smaller deviations between implied and realized volatility are a well-known stylized fact of fnancial markets, large, time-varying diferences are also frequently observed throughout the day. Furthermore, substantial deviations between the VIX and its futures might lead to arbitrage opportunities on the VIX market. Arbitrage is hard to exploit as the potential strategy to exploit it requires buying several hundred, mostly illiquid, out-of-the-money (put and call) options on the S&P 500 index. This paper discusses a novel approach to predicting the VIX on an intraday scale by using just a subset of the most liquid options. To the best of the authors’ knowledge, this the frst paper, that describes a new methodology on how to predict the VIX (to potentially exploit arbitrage opportunities using VIX futures) using most recently developed machine learning models to intraday data of S&P 500 options and the VIX. The presented results are supposed to shed more light on the underlying dynamics in the options markets, help other investors to better understand the market and support
regulators to investigate market inefficiencies.
Original languageEnglish
Pages (from-to)97-115
Number of pages19
JournalDigital Finance
Early online date13 Aug 2020
Publication statusPublished - Sep 2020
Externally publishedYes


Dive into the research topics of 'Neural networks and arbitrage in the VIX: A deep learning approach for the VIX'. Together they form a unique fingerprint.

Cite this