OREANDA-NEWS. March 13, 2017. Milliseconds matter in the financial industry.

Lightning-quick insights, executed instantly, drive profits. The key is making smart decisions faster than the competition, and it all comes down to leveraging big data.

Faster analytics offer a big advantage. With conventional computing pushed to its limits, the financial industry is moving toward GPUs.

Banks and investment companies are turning to NVIDIA GPUs and NVIDIA DGX-1, the world’s first purpose-built system for deep learning and AI accelerated analytics, and Kinetica’s GPU-accelerated, in-memory, distributed database for truly real-time analytics demands, including fraud analysis, risk management and algorithmic trading.

Interactive Portfolio Risk Management

For financial traders and portfolio managers, it comes down to making portfolio risk calculations. Five years ago, a trader had to extract data and transfer it to specialized systems to perform advanced analytics and modeling. Mathematically intensive risk calculations were typically performed overnight in batch, which made it difficult to respond to market changes in real time.

With the advancements in GPUs and deep learning, it’s now possible to perform data exploration, model development/scoring and model consumption on a single compute-heavy platform with Kinetica and NVIDIA GPUs.

Customers can perform complex queries on demand without needing to move data between systems. Quants are able to run sophisticated data science workloads on the same database housing the rich information needed to drive trading decisions. This solves the data movement challenge and enables a more simplistic architecture for AI workloads.

With Kinetica’s GPU-accelerated user-defined function capabilities, customers can deploy a model from deep learning frameworks like TensorFlow, Torch, Caffe or Spark ML via a simple API call. This allows quants and analysts to experience the performance and parallelization benefits of the GPU without needing to learn new programming languages.

Trade Execution

Trade execution involves figuring out how to get the best price for a security when you’re looking at a limit order book. Whether the future is a few hundred milliseconds out or a minute, as you’re trading larger and larger quantities of a particular security, you want to know that you’re getting the best price now versus a few seconds from now.

By having the quantitative data backed up in a deep learning framework, you can begin to understand where the millions of trades made in this security are going. You’ve trained on a ton of data, and then you can do inference on that data in real time to see whether you should trade now, in a couple hundred of milliseconds, in a second or in a minute. That intelligence really boosts the potential of algorithmic trading.

Combining Multiple Data Sources

An emerging science in deep learning is combining many different data sources together. We’ve had market data forever, so you can do technical trading, and you can look at what your securities are doing. However, if you want to get a jump on your competition, you need to look at what the other data elements are that you can integrate together.

That is a deep learning and complex analytics issue. It’s not just looking at one set of data and turning that into a prediction of where things are going. It comes down to figuring out how to combine your traditional securities data with social media data, web-based data and other proprietary data feeds, as well as figuring out how unknown events such as Brexit are going to affect your stocks. And then combining all of those things together in a deep learning model.