The Challenges of Real-Time Trade Execution

The increasing amount oftradingvolumes and numbers of markets create bottlenecks for onlinetradingsystems. Today’s market demands low latencytradingfor buy-side and sell-side. As a result,tradingfirms need ultra-fast pricing engines to ensure they find the best venues fortradesat any giventime, as well as fast-processing order management systems toexecutethosetrades.

High frequencytradingapplications demand higher throughput and lower latency than ever to meet the challenge of processing terabytes ofdatafrom various sources, which must be available to multiple parties, making system management increasingly complex. The variety and volume ofdatais overwhelming traditional classic database solutions and legacy batch-oriented systems that were not originally designed to meet todaysrealtimeanalytic requirements. Faster hardware alone will not meet the challenging requirements ofreal-timedata.

Here are four key questions regarding trading system performance.

  1. Does yourtradingapplication offer competitive latency?
    Applications are responsible for 65% of thetradingprocess latency. Tier-based architectures moving massivedataamounts cant cope with such demand levels. The financial services industry has the highest computing demands of all industries because executing trades faster, and executing more trades within the same time frame translates directly into higher revenues. Any degree of latency – even milliseconds – can lead to loss of tens of thousands of dollars.

As a result for a need for low latency, the financial industry was among the first to adopt in-memory computing (IMC), which increases the speed at which data is processed by orders of magnitude. In-memory computingis the storage of information in the main random accessmemory(RAM) of dedicated servers. RAM is up to 100,000 times faster than disk, so its an obvious choice to minimize I/O latency for trading apps.

As a result, IMC trading platforms bring the processing to the data, rather than bringing the data to the processing, thereby achieving a billion transactions per second (TPS) compared with a few million TPS with disk processing. Apps run entirely on a single platform with all the tiers collapsed into one container. The platform provides fast data access by storing data in random-access memory (RAM), fast storage device (SSD) and persistent memory (PMEM), ensuring low latency, high performance and high availability. In addition, IMC trading platforms can provide near real time access to information archived in data lakes and data warehouses to perform fast analytics.

  1. How does your system cope with volatile markets?Can your system handle changing, unpredictable volumes?
    For each trading opportunity there is a need to calculate possible price movements which, based on market volatility, can require crunching through massive amounts of data on demand requiring elastic scalability. Using IMC banks are able to recalculate the market change mathematical models and back test against the previous 60-90 days to iterate strategies to predict the most likely outcome within seconds, even during periods of intense trading. IMC can provide the computing power necessary to analyze the impact of the event in seconds enabling systems to continue trading with minimal interruption.

During periods of peak load due to high market volatility, IMC can dynamically expand applications onto additional physical resources – automatically and on-demand, to meet any service level requirement. Resiliency is guaranteed by in-memory backup within each container, and by mirroring data to a traditional database outside the runtime.

  1. How effectively do you meet regulatory requirements?
    Asset management firms are really starting to use compliance as a differentiator when looking to win new business. At the same time, compliance is becoming more difficult to manage as regulators want to tighten the controls and protect investors so the rules are becoming broader in scope and are being updated more often.

For example, on the buy side, there are at least six different places where compliance can be checked, so compliance is no longer checked just at pre-trade and post-trade but is now generally integrated into the process. The sell-side must prove that investors receive the best prices based on real-timedata feeds from multiple sources with split-second execution adding its own computing requirements.

The number of orders, times the total number of factors that need to be checked, can result in a very large number of tests. In fast moving markets, you cant have a compliance check that takes more than a matter of seconds, so fast compliance checking is important for customers. Performing compliance checks using in-memory computing can offer a significant edge and can even be used as a selling point for customers.

  1. Can your pricing engine deliver the best bid & offer?
    Processing marketdatatick by tick with an ever-growing number of quotes and markets, and the complexity of delivering different spreads to different customers, makes pricing a challenge to IT.

Among the technologies that contribute to data analytics for determining pricing are artificial intelligence (AI), exploratory data analysis (EDA), quantitative data analysis (QDA), confirmatory data analysis (CDA),predictive analytics, and data mining, just to name a few.

There are also new data sources, such as social media and news feeds that by using natural language processing (NLP) can evaluate popular sentiment which can be an important factor when predicting prices. When more data can be analyzed quicker, the final results are typically more accurate resulting in more profitable trades.

Speed is an essential element for every aspect from trading, from price predictions, to executing trades and confirming compliance. In memory computing is one of the technologies that can reduce latency to keep pace with fast trades and speed up analytics for quicker execution and more profitable trading.

Tal Doron is Director of Technology Innovation, GigaSpaces