Machine learning for stock trading: 97% faster data processing

Innowise developed a machine learning stock trading solution that capitalizes on exchange price discrepancies.

Customer

Industry

FinTech

Region

Client since

2023

Our client is an Irish proprietary trading firm. The company’s primary focus is trading highly correlated products while capturing minor price discrepancies.

Detailed information about the client cannot be disclosed under the provisions of the NDA.

Challenge

Delays in the client's trading system made it too slow to keep up with fast-moving market data, causing missed trading opportunities.

The client’s previous trading system just couldn’t keep up with fast-moving data. It had major lag issues, taking 2-3 seconds to process information, which was way too slow for making quick trading decisions.To make their new trading strategies work, the client needed a fast system that could handle big volumes of financial data in real-time. It was key for them to spot and analyze short-term differences between related assets since those opportunities can pop up and disappear in seconds. The new system had to process that data in milliseconds to deliver accurate calculations and successful trades.To tackle these challenges, we set out to build a new platform for machine learning stock trading, designed to deliver a fast, reliable, and customized solution.

Solution

A low-latency, machine learning-driven trading platform that quickly identifies optimal trading opportunities

Innowise revamped the client software with a low-latency infrastructure for quantitative crypto trading. This new platform allows the client to respond to market changes fast and execute trades with almost no delay, giving them an edge in capturing arbitrage opportunities.

We applied machine learning techniques to identify the best times to buy assets and catch market anomalies that hinted at solid buying opportunities. The system also integrated with Grafana, a tool for querying, visualizing, and analyzing various trading metrics, along with customizable alerts.

The machine learning stock trading platform incorporates five main modules:

Market data module
Order management system
Positions manager
Risk manager
Strategy manager

Market data module

To manage exchanges in different regions, the trading system uses a geo-distributed setup. The main system runs on a central server, working as the hub for collecting and processing market data. Smaller gateways are set up near each exchange server to pull data directly from them. This setup allows the central system to gather real-time data from multiple exchanges — like quotes, order book status, funding rates, and more — giving our client a full market overview.

Order management system

The order management module lets our client keep an eye on multiple orders in real time, giving them a clear view of both full and partial executions. Traders get instant updates on order statuses, so they can jump on good price opportunities fast. It also comes with order-level approvals, allowing traders to approve orders based on specific criteria for extra control and accuracy.

Positions manager

The positions manager gives traders real-time insights into their active trades, balance control, and a full view of their available funds. This tool lets traders monitor their portfolios and evaluate their exposure to various assets. It also provides key details like average purchase price, current market value, and unrealized gains or losses for each position. Additionally, this module works closely with the risk manager to oversee trading operations and enforce limits to keep trades within set risk parameters.

Risk manager

The machine learning stock trading platform gives traders full control over orders, purchases, and risk management. A set of algorithms helps keep purchase prices within set limits, and by comparing executed prices to the current market price, the platform helps traders avoid big deviations that could impact profitability.The module tracks Profit and Loss (PnL) in real time, giving traders a clear view of their current profits and letting them set custom loss limits based on their risk tolerance and strategies. It also comes with advanced tools to help assess the risks of individual trades or the entire portfolio. By looking at things like asset volatility, past price trends, and correlations, traders get a better sense of their risk exposure and can fine-tune their risk management strategies.

Strategy manager

At the core of the module is the strategy, set up as a distinct class that captures the trading logic and defines actions for different market situations. By working with relevant datasets and using machine learning for stock trading, the module identifies key data points to train models that automatically execute strategies based on real-time market conditions.The process kicks off by training machine learning models with selected datasets. These models then analyze market data, such as trading volumes, to spot anomalies and pinpoint the best entry or exit points for specific assets. The models use boosting algorithms to generate asset price predictions within extremely short time frames, sometimes in just milliseconds.The machine learning models work with the trading system’s backend, where their predictions are stored in a database for further analysis and decision-making. As fresh market data comes in from exchanges, the models evaluate the conditions against set criteria. By combining trading volume data with machine learning-powered anomaly detection, the tool boosts the chances of executing profitable trades.

Technologies

Back-end

C#, ML.NET, Python

Cloud

AWS

CatBoost, XGBoost, NumPy, pandas, SciPy, scikit-learn

Integrations

Grafana, Prometheus

Process

During the development process, Innowise took a clear and efficient approach to keep things running smoothly with the client. We broke the project down into three key stages:

Requirements gathering: We started with in-depth discussions and consultations with the client to really understand their trading strategies and what kind of system would fit their needs best. This meant having several meetings over Google Meet, where we worked together to set clear goals and outline the benefits of using machine learning for stock trading platform.

Planning and architecture design: We used Jira to manage the project, setting up a clear roadmap, defining key milestones, and assigning resources. This kept everything organized and made sure the development process ran smoothly from start to finish.

Development, training, and testing: We kicked off the development phase by building and deploying the core machine learning system on the main server, setting up gateways to link with cryptocurrency exchanges. This phase also involved data mapping and training the machine learning models to make sure everything worked well for real-time trading integration.

Integration, deployment, and enhancement: Once each module was developed and tested, the team worked on bringing all the components of the trading platform together. We ran thorough integration tests to make sure everything communicated properly and functioned as a unified system.

Our team is expanding the project by adding more data collection exchanges to make it stand out in the market. To level up, we’re rewriting the codebase in C++ to boost speed and performance. We’re also considering rebuilding frequently used connectivity libraries from scratch to further elevate the system’s performance and improve machine learning stock trading techniques.

Team

Lead Developer

DevOps Engineer

C# Developers

Python Developers

Quantitative Researchers

Results

Machine learning for stock trading delivers 97% faster information processing and a 34 ms market response time

Building the custom quantitative trading platform made a huge difference for the client. We slashed processing delays from 2-3 seconds to just 34 milliseconds, speeding things up by about 97%. By using machine learning for stock trading, the platform sharpened the client’s strategies and boosted their profitability. Plus, its quick response to market moves and ability to spot arbitrage opportunities gave the client a solid edge over the competition.Innowise developed a user-friendly API that simplifies strategy development and testing. Now, the client doesn’t need to rely on third-party resources, as everything is handled within our unified system. On top of that, the API delivers clear, detailed metrics for each strategy, helping the client quickly assess whether it fits their risk profile.

Project duration

April 2023 - Ongoing

97%

faster trading information processing

milliseconds market response time

Related cases

Need a technological solution? Contact us!

What happens next?

Having received and processed your request, we will get back to you shortly to detail your project needs and sign an NDA to ensure the confidentiality of information.

After examining requirements, our analysts and developers devise a project proposal with the scope of works, team size, time, and cost estimates.

We arrange a meeting with you to discuss the offer and come to an agreement.

We sign a contract and start working on your project as quickly as possible.