By Qasim Nasar-Ullah
Arabesque recently created an AI and data engineering centre based in Singapore. The centre is supported under the Financial Sector Technology & Innovation – Artificial Intelligence & Data Analytics (FSTI – AIDA) scheme, which aims to strengthen the AIDA ecosystem in the Singapore financial sector. The FSTI – AIDA scheme is funded by the Financial Sector Development Fund, administered by the Monetary Authority of Singapore. Arabesque team members work in close collaboration with the wider Arabesque team, primarily in London, but also in Tokyo, New Delhi, Frankfurt and Boston.
A key differentiator of the centre is its engineering-first approach. This has a number of benefits; for example: we can quickly experiment and iterate our ideas; our work can be tested at-scale; and we can move our code to production quickly. We continue to maintain our strong engineering partnership with GCP, enabling us to remain at the cutting-edge. See a recent example of our outreach with Google on March 11th here.
Team members based in Singapore support the production of our key deliverable, AutoCIO. AutoCIO is an easy-to-use platform that builds customisable investment strategies. For example, by using AutoCIO, investment strategies can be customised to target certain CO2 emissions or certain gender diversity levels. The platform is utilised by Arabesque’s clients- such as DWS, one of the world’s leading asset managers, and BIMB Investment, a Shariah-ESG investment management company in Malaysia.
Our engineering centre works on specific KPIs as indicated in our press release, and broadly these cover:
- Implementing NLP data to our financial Knowledge Graph;
- An AI Financial Analyst, creating new approaches for financial modelling and analysis; and
- Understanding data bias with application-to-transfer learning
1. Implementing NLP data to our financial Knowledge Graph
Problem: The growth of structured data (like quarterly revenues or sales) and unstructured data (like news or social-media) is problematic if we do not have the appropriate tools to meaningfully organise the data. Such diverse sources of data (and particularly unstructured data) are often rich in interconnectedness despite their apparent heterogeneity. Such interconnectedness is often obscured in traditional database structures, thereby limiting the value and AI-based insights that these datasets may otherwise provide.
In order to address the problem statement above, the project has three key areas:
- The Knowledge Graph as a means to organise interconnected data.
- The tools to traverse the graph, extracting relevant data and study the correlations of entities in the Knowledge Graph.
- The Knowledge Graph being used to train new AI predictive models.
2. An AI Financial Analyst, creating new approaches for financial modelling and analysis
Problem: Financial analysts typically devote a substantial amount of their time pouring over financial statements, industry reports etc. in order to build complex models for the purpose of corporate valuation, or the prediction of company financials. Aside from the substantial cost of human resource, the absence of modern data-driven AI approaches can ultimately impact the accuracy and integrity of the financial analyst’s output.
Our approach implements a general machine learning approach that can enable a more
accurate and holistic approach to the work typically carried out by a financial analyst targeting the following areas:
- DCF (discounted cash flow) predictions, including company/project valuation.
- The inclusion of ESG data within DCF approaches.
- Company financials predictions, e.g. quarterly sales, including fraud detection.
- Event analyses, e.g. Merger likelihoods; Post-IPO stock price analyses; Supply chain impact analyses.
3. Understanding data bias with application to transfer learning
Problem: High-quality, labelled datasets are difficult to obtain or produce because of the large amount of time and effort required to label such datasets. Furthermore, these datasets are often subject to various forms of data bias, thereby hindering trust with a machine learning model using such data.
In order to address the problem statement above, this project has three key areas:
1. Data sourcing: To deliver datasets of high quality and quantity suitable for financial machine learning.
2. Data biases: To develop automated tools and systems for capturing and identifying bias in datasets.
3. Transfer learning: To make advances in the area of transfer learning. In particular, applying the knowledge obtained from one market to other markets for applications such as making investment predictions.
The current focus of the centre is a reimaging of our data processing pipelines to maximise our ability to onboard and process our big data sets. Our current job roles reflect this, and we aim to add additional roles soon. We look forward to continuing to expand our team and leading this initiative to provide innovative financial research in the APAC region.