Hans Buehler, global head of equities analytics, automation and optimisation at JP Morgan, visited our London offices to record a podcast on a recently published paper he co-authored on a new technique called deep hedging.
The quant argued this new machine learning technique can hedge derivatives without the need to use classical models such as Black-Scholes. Typically, banks use risk sensitivities known as Greeks derived from classical models to hedge their options books, but these methods are limited in their ability to factor in transaction costs and additional market information. With deep hedging, machines can learn from large amounts of historical data to make more precise hedging decisions, said Buehler.
“Every trader who uses all these classical models will tell you there are some overrides, [such as] delta skew or barrier shifts… none of these are systematic usually in the strict mathematical sense, so it keeps requiring human input to maintain those [overrides],” said Buehler. “Then the machine learning techniques came up which made it possible to do much heavier, much more precise calculations.”
Buehler argued the approach also aligns with the way traders actually think about hedging, as the objective is mainly to reduce hedging error, or the difference between the hedged item and the hedge.
“It fundamentally does what people actually do when they trade. In reality, they [ask], ‘What do I need to do in order to minimise my hedging error in the sense of P&L uncertainty?’ rather than saying, ‘How much vega do I have?’. It is very data driven,” Buehler added.
Another advantage is that the technique allows for more automation of hedging, as machines can run in parallel to identify appropriate hedges. This can make the process faster.
“If I wanted to run a lot of books of options in parallel, it is very difficult for humans to observe the vega exposures on a lot of single stocks in parallel because each book is very specific,” said Buehler.
The technique is currently being applied to index options books, although this can be expanded to more liquid vanilla products, Buehler said. One caveat is that less liquid over-the-counter products may be hard to apply this technique to, as data is sparse for these instruments.
1:20 What is deep hedging?
4:32 Applications of deep hedging
6:12 Advantages of not using Greeks
8:08 Sparse datasets
9:10 Asset classes applicable
10:23 Operational changes from adoption of the technique
11:42 Benefits to exotics and illiquid products
12:47 Caveats to deep hedging
13:35 The black box problem
14:27 Will banks adopt this on a larger scale?
To hear 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 the iTunes store or Google Podcasts to listen and subscribe.