Interview: Javed Ahmed, Senior Data Scientist, Metis Corporate Training
Metis as a Solution Provider
How are you supporting clients during this challenging period?
Many of our clients found themselves having to adjust to their teams working from home for an extended period of time. A change like that can be stressful, especially when there are few opportunities to add variety to the workday and to connect with your team. First, we helped a number of our clients transition from in-person to live online training. We wanted to do more so we decided to launch a new program, the Metis Corporate Training Series. It’s a weekly one-hour live online webinar covering a popular data science topic. We held our first session, Intro to Python, throughout the month of April. It was a great success. Over 1,000 people registered. Managers encouraged whole teams to join. At the end of the four weeks it really felt like a community had formed. We recognized attendees, had a few laughs and, most importantly, we helped people on their journey to mastering Python.
Next, we’re hosting a live webinar, “Deep Learning Approaches to Forecasting and Planning,” on June 18th. During that workshop I’ll focus on the intuition behind various deep learning approaches to explore how managers can tackle highly complex models by asking the right questions and evaluating the models with familiar tools.
Metis POV on Industry Topics
What do you think are the biggest challenges facing data scientists/AI experts/quantitative investors in 2020/2021? Why are they important?
Data professionals, including quantitative investors, have to contend with many challenges as they apply AI and ML techniques throughout their business. For example, investors implementing predictive approaches must determine how to incorporate the unusual historical patterns arising from new events, like the Covid Pandemic. Quantitative and ML models mostly look backwards, and assume historical patterns are representative of the future. Because of how unusual Covid-related disruptions are, automated and backward-looking models will have a hard time incorporating them. Forecasts require judgement regarding how forward-looking conditions differ from historical conditions. This is very hard and not generally the area where ML models outperform traditional analytical approaches
There seems to be a lot of cynicism surrounding the use of alt data and its role in alpha generation and if you can truly find value from these datasets. How is your organisation working around this/what are your views on the future of alternative data?
At Metis we advise clients on the use of alternative data in various contexts, including alpha generation (we don't invest in financial markets ourselves). Alternative data can mean many different things. Investment firms have traditionally been relatively open to using new data sources for alpha generation. The offerings of many independent providers of alternative data are adopted to varying degrees by investment management firms. As with any technology there are different levels of adoption, and attendant risks (data availability, privacy, reliance on material nonpublic information etc.). Organizations are generating more and more new data, and by some estimates over 70% of this data goes unused for analytics. This presents a major opportunity, and effective use of all data sources (including alternative data) will influence future success in investment management.
A portion of the industry are adamant that advanced ML techniques such as Reinforcement Learning and Deep Learning cannot be applied to financial data – do you agree? What are the main challenges in preventing this from happening?
Deep and reinforcement learning are already being applied to financial data by several large firms, in varying contexts. Deep learning techniques are prevalent across several areas including customer relationship management and risk management. Many investment firms have some automated or high-frequency trading operations that often feature these approaches. Primary challenges of implementing these approaches include data sufficiency and implementation detail. These approaches are complex, and require both a lot of data and experienced people who understand their implementation. This knowledge needs to be communicated more broadly to management, who also need the ability to understand these approaches and associated risks. Finally, given the disruption (in the data) represented by Covid 19, implementation of these approaches will be more difficult as conditions change and previously-measured relationships in data may not hold after the pandemic has passed.
Javed is an economist and data scientist with experience in banking, finance, forecasting, risk management, consulting, policy, and behavioral economics. He has led development of analytic applications for large organizations including Amazon and the Federal Reserve Board of Governors, and served as a researcher with the Office of Financial Research (U.S. Treasury). He holds a PhD in financial economics and MA in statistics from U.C. Berkeley, as well as undergraduate degrees in operations management and systems engineering from the University of Pennsylvania. In his spare time, Javed enjoys tennis, squash, and reading.