Could Deep Learning Dethrone HFT?
Deep learning just may be poised to shake investment strategies based on alpha capture to their knees, according to some experts.
The machine learning sub-discipline is setting itself up to displace high-frequency trading which has dominated many of the electronically traded markets since approximately 2007.
Similar to high-frequency trading, deep learning-based strategies do not approach a trade with a preconceived hypothesis but examines the data for useful information it might contain, explained Gaurav Chakravorty, chief investment officer of online asset manager qplum and who spoke during a webinar hosted by Quantitative Brokers.
“It is the first machine learning algorithm that tries to remove noise from the data before it starts predicting,” he added.
Deep learning does this by using “layers” to summarize the content of the previous layers until a few summaries can describe all of the original data. Chakravorty cited an example where an initial layer was composed of 5,000 distinct pieces, and each subsequent layer reduced the amount of data by a third until the model had five summaries in the end.
“The deeper that you go, the literally more intelligence that you get,” he said.
However, Chakravorty has never worked on a network that processed more than ten layers because of the cost it would take to build something larger. In comparison, experts estimated that the deepest network in the human mind is 4 million layers deep, he added.
Theoretically, deep learning could lead to 100% accuracy in modeling like identifying a photograph of President Barak Obama, but the required work and resources would be prohibitive, according to Chakravorty. “Getting to 56% accuracy in finance is going to be very tough.”
Deep learning’s potential accuracy has reached a point where Wall Street professionals may see it as the ultimate way to develop investment strategies since it can adapt itself to changes in the markets.
“No one asks what will be the next search engine,” said Chakravorty, “Google has become a ‘good enough’ solution, unlike the previous solutions.”
If deep learning becomes the de facto modeling tool for investors, it could make the markets more efficient and reduce alpha capture even further.
“The main function of investing is not to make money at the expense of others,” said Chakravorty. “It is to figure out that there is all of this data coming into markets and how does it relate to pricing and how to derive information from that.”
However, it is very early days regarding deep learning adoption by asset managers. Chakravorty estimates that there are fewer of five firms that use deep learning with high quality.
Increased electronification has created useable and accessible real-time and historic trade data.
The fintech uses data so institutions can assess the environment impact of their portfolios.
Working with Riskfuel will reduce reliance on slow and expensive financial models.
The platform uses AI to match experts and users and digitize the research process.
Databricks is the first data lakehouse platform in the cloud.