Machine learning’s advance is changing how securities lending data is used, particularly by quantitative traders, said panellists on a webinar organized by consultancy firm Finadium.
Securities lending markets are a collection of market participants across the world that make short-term loans of stocks or bonds in exchange for cash or noncash collateral.
Borrowers include hedge funds for activities like short selling, while lenders, called beneficial owners, tend to be the big buyside like pension and insurance funds that want to make fees on inventory.
In the middle are prime brokers and agent lenders, generally from the big sellside institutions, which provide a price for such transactions. Securities lending data is traditionally used by desks to benchmark rates.
But technological progress is not only attracting quantitative entrants but also evolving how this type of alternative data is being used.
“With the onslaught of new technologies like machine learning and the like, you are starting to see people trying to pair data in different ways,” said Nick Delikaris, global head of trading and algorithmic strategies at State Street Securities Finance.
“People are able to find new correlations and alpha signals that you weren’t anticipating before.”
One of the changes Delikaris mentioned has been observed across the industry: from a modelling perspective, there’s been a shift from machines helping to prove a hypothesis to machines creating a hypothesis.
“It’s a lot more about how much data can I give it, and then have it tell me what some of these correlations are and look at what is going on amidst the data,” he said.
In that respect pairing securities lending data is necessary, and a strategy where he believes alpha can be found.
It’s no cakewalk however, since there’s plenty of work yet to be done to get timing right.
“When you pair against other market data, and figuring out that sweet spot for predictability, that’s where I think your additive value is, especially if you are thinking about it from a hedge fund side,” he said.
“Those are the two dynamics that are really going on these days: how you are mixing it with other data, and then how are you helping find the sweet spot of correlation there?”
Exchange-derived seclending data
Bob Levy, head of business development at Hanweck, shared Delikaris’ sentiments about adding to human power with machines.
Hanweck produces real-time option analytics, and Levy said he is currently focused on developing alternative data sets to expand this scope. The company publishes the “Borrow Intensity Indicator”, which is built on exchange-derived securities lending data from billions of quotes daily across 15 options exchanges in the US.
The underlying data is quite massive and very real time, he added.
“Exchange data is inherently neutral, it’s very transparent and it’s very timely, and we, through our option analytics feed, are monitoring quotes and generating analytics really at the millisecond level that’s slowed down and aggregated and transformed quite a bit before it goes into what we generate as the Borrow Intensity Indicator,” Levy explained.
The Indicator’s dataset includes two series, one that is “dirtier” than the other, which is the one favoured by quant traders, Levy noted. While users looking to draw information to relate to securities lending could prefer more “smooth” data.
Industry contributed seclending data
Alongside exchange-derived data is industry contributed data.
DataLend receives and processes entire books of open loans from prime brokers and agent lenders globally. That translates to about 3.25 million open loans a day of which between 120k and 180k are new.
The firm covers about 47,500 securities on loan on any given day across the globe worth about $2.3 trillion, and its dataset has grown to about 25 terabytes, which translates to 2.2 million phone books worth of data, explained Chris Benedict, director at DataLend.
Transactions are the basic building blocks of the dataset. So, for example, if a trader wants to know what’s happening in Hong Kong, they may find volume average fees to borrow Hong Kong equities are down over the month. Or volume average fees to borrow IT stocks in Taiwan are hitting multi-month highs at around 300 basis points, for example.
“It’s very interesting to be able to start with small building blocks with the data and then expand on that to allow users to go in and see macro-level data based on those underlying securities and those underlying transactions,” said Benedict.
Both contributed and derived securities lending data are boosted by new tools making automated analysis more accessible, said Levy, pointing to Google’s TensorFlow and other open source approaches.
“You are going to see another wave of tools being thrown at the problem, which will be more demanding of data,” he said.
Panellists also discussed the size and shape of the securities lending market, geographic and other user trends, as well as other types of datasets and arbitrage opportunities, and lessons learned from backtesting.