Augmented Performance Officer

Inspiration

Finastra's solution Fusion Invest supports Investment Managers across the globe Front to Back to Risk.
To help them manage their portfolios and report on their activities, Fusion Invest produces a large amount of numerical data. Checking manually the analytics and catching all potential errors in the input data is a challenge - especially for smaller institutions which are often under-staffed and/or under-equipped. Business intelligence can help automate those checks, leading to productivity gains and increased trust in Fusion Invest.

What it does

Augmented Performance Officer App first runs automatic consistency checks: for example, it verifies that the sum of the performances of all financial positions is equal to the overall performance of the fund they relate to. Second, it runs error detection checks based on machine learning on the historical times series of prices, trade and other static & market data. The power of computers in both those checks far exceeds the number of cases that human employees in financial institutions could possibly achieve. This frees time for humans to focus on more complicated business cases, value added tasks, and reduces costs.

How we built it

The first type of consistency checks was implemented in classical C++ programming. The second check uses Microsoft's machine learning library (https://dotnet.microsoft.com/apps/machinelearning-ai/ml-dotnet), to automatically detect anomalous daily returns in the historical series of prices. The results of those checks are presented in user-friendly dashboards leveraging Microsoft's Power BI (https://powerbi.microsoft.com/en-us) in the cloud.

Challenges we ran into

The cleaning of data was a work we did not anticipate. For example dividends produce jumps in the daily returns, that should not be classified as errors. Also different closed business days in different regions of the world make the comparison of time series harder. Finally, Microsoft's adaptive kernel density estimations require independent statistical data, which blocked our first attempt to use prices instead of daily returns.

Accomplishments that we're proud of

We had a great teamwork on this short-term project, with talented people. We found a robust solution to address an old pain point for our customers, finding errors in large amounts of data. We successfully executed our solution on an internal database that help us to find some data issues.

What we learned

We taught ourselves Power BI and the basics of error detection with machine learning.

What's next for Augmented Performance Officer

At the moment, this solution focuses on performance attribution for funds. Next we want to apply those error detection techniques to other data-intensive parts, like value at risk, data reconciliation with external systems, and the series of net asset values (NAV).