Leveraging data science for next-generation risk and P&L

Alexei Tchernitser, director of analytic solutions at Quantifi, discusses the impact data science has had on next-generation risk, and profit and loss (P&L)

How has data science transformed finance?

Alexei Tchernitser, Quantifi

Over the past few years, there has been a step change in the role of data and technology in trading, risk management and investment decision-making. Data analytics techniques previously considered emerging or experimental are becoming mainstream. Firms are deploying data science tools to improve risk assessment and business response strategies, and bring more rigor to their operations. On-demand access to significant computational resources via the cloud, with high-performance data stores and in-memory architectures, enables firms to conduct more ad hoc analysis, testing and validation. This is performed using the most granular levels of data and without the need to pre-aggregate or preformat the data. Firms, however, face the challenge of how to enable their quants and data scientists to produce high-value work without compromising the security and restrictions on who can access and adjust ‘official’ risk and P&L numbers.

Historically, risk managers have had to ‘lock down’ the official platforms. This created two main problems that hampered advanced analysis.

The first is duplication of work: to carry out sophisticated analysis using the firm’s actual trade and/or market data, quants sometimes needed to redevelop the pricing models implemented as part of the firm’s official P&L platforms. This is time-consuming and complex, especially for derivatives products or when an accurate representation of all trade details is required.

The second is inconsistency: inevitably, the models implemented independently by quants/strategists/data scientists differ from the official ones, and divergence typically grows as the products become more complex. This is particularly true when a technology provider incorporates pricing models into the P&L platform.

Modern data science tools integrated into the latest generation of mark-to-market (MtM) platforms solve both problems.

Integrating data science tools into existing processes

Open-source data science tools provide many possibilities for building various machine learning models and analysing vast amounts of data. However, to be helpful in real-world applications, the data underlying any analysis must come from a source system. For modelling even simple products, such as bonds, one usually needs relatively complex building blocks, such as interest rate curves, reference data, exact product definitions and market quotes for interest rates and bonds. Collecting, representing and normalising this data is a complicated and tedious task, and inaccuracies in modelling these components can affect or even invalidate further analysis.

A new set of application programming interface (API) tools is now emerging. These tools are designed to seamlessly integrate open-source data science packages and programming environments with more traditional MtM risk platforms, such as Quantifi. These new APIs enable innovative integration between a standalone risk platform and programming environments that quants and traders can easily use on their desktops.

This framework provides the next level of interoperability by allowing the transfer of fully calibrated complex objects (such as curves, volatility surfaces, product or trade representations) to various parts of the risk ecosystem. Using these APIs, quants and quant traders can take an existing portfolio of trades from the risk platform and perform backtesting, custom value-at-risk calculations, ad hoc scenarios or sensitivity analysis independent of the primary risk platform. Alternatively, users can simply extract the required data objects such as curves, quotes and reference data, and construct new trading strategies. They can also apply these objects to price bespoke derivatives not handled by the P&L platform.

Another advantage of this new technology is that it allows users to work in their preferred programming environment, for example, Python in Jupyter Notebooks or other popular programming languages and integrated development environments. Users can also perform their analyses and/or build advanced models using the APIs on their local machines. At the same time, the primary P&L platform can be operated elsewhere, even hosted in cloud-computing platforms.

Achieving consistency in a ‘low-code’ setting

Using the framework described above, users can ensure consistency between pricing trades in an official MtM platform and their local development environment. This is because all the business objects required for calculations are passed directly from the platform, where they were created using the predefined official set of pricing rules and parameters.

In addition, users benefit from a truly low-code environment, where the risk platform handles the setup of complex trade pricing logic. Users can therefore focus on adding value by implementing high-level tasks, such as portfolio backtesting, custom scenario or risk measure calculations or portfolio optimisation. All of this can be achieved without spending time on setting up the underlying risk factors, security or trade details.

Furthermore, this framework can facilitate the transfer of objects representing any trade type supported by the MtM platform, no matter how complex. As a result, users can run analyses on portfolios consisting of mixed trades and hedges, vanilla and derivatives. By using high-level code, users avoid the extra layers of complexity inherent in low-level code. Moreover, because the process runs standalone from the central MtM platform, users can be confident they will not negatively affect the performance of the primary platform or the integrity of the data. The framework also allows integration with popular machine learning libraries, which often require extensive use of computing power, again without affecting the performance of the primary MtM platform. Consistency between the two is also maintained.

Additionally, the MtM platform takes care of the data dependencies (such as market and reference data) required to price trades. The process involves automating the trade and market data feeds and preserving the relationships and hierarchies of data from multiple sources. Users can therefore focus on implementing new functionalities instead of ‘cleaning’ the data.

Integrated risk platform, powered by data science

Advanced machine learning models can be set up using accurate trade and product representation, and consistent market data and pricing rules. This adds new levels of flexibility and robustness while ensuring consistency throughout the modelling process.

Quantifi’s data science-enabled platform enables quants and traders in multiple financial institutions to automate and outsource the task of manually collecting and processing data. Users can focus on implementing the required custom business logic, complementary to the functionality of the platform, and significantly reduce delivery times. This new data science platform provides clients with the ability to perform complex data analysis and flexible reporting using Python, Jupyter Notebook and other popular data science tools. Integrated with Quantifi’s advanced model library, clients benefit from complex data-driven analysis, strategy backtesting and ad hoc portfolio what-if scenarios, all using mixed datasets from diverse sources.

Quantifi’s integrated pre- and post-trade solutions allow market participants to better value, trade and risk-manage their exposures, and respond more effectively to changing market conditions. Quantifi’s investment in the latest technology – including data science, machine learning and APIs – provide clients with new levels of usability, flexibility and integration.

Learn more about Quantifi’s data science-enabled platform.

Sponsored content