Asset-management companies are starting to incorporate big-data analytics and the artificial-intelligence tools to drive efficiencies and improve portfolio performance.
Operations and data officers at global fund houses shared some of their recent achievements at a conference last week organized by the Investment Management Association of Singapore (IMAS).
Today we look at Franklin Templeton’s experience.
Franklin Templeton
The firm is leveraging data science and analytics to enhance its investment-management lifecycle, says Chris Pham, senior vice president and investment-management chief data officer, based in the U.S.
She says the starting point was to amass a centralized data lake (an aggregation of all types of data, both traditional and alternative) in order to deliver new use cases to the firm’s various investment teams.
“We want to be in the insights business, not the data management business,” she said, noting the firm outsourced a lot of the management operations to cloud services vendor Snowflake.
Franklin Templeton now has amassed more than 100 data sets covering indices, benchmarks, prices, economic factor, ESG, and more, with some data going back more than 25 years. The firm has built its own machine-learning and other tools to work with the underlying data.
Improving performance
Initial use cases include applying data science tools to fixed-income fundamental data to identify market mispricings. “We leverage our data on bond fundamentals, macroeconomic and business cycles, and build a model using machine learning to combine these factors and test them,” she said.
In equities, Franklin Templeton is using more alternative data (data other than traditional sources such as financial statements or market prices). Pham says alt data can supplement traditional, and even substitute for it in specialized instances. Alt data such as trending searchwords on Google, property valuations, and job postings are ways to help measure an economy.
“The best alternative data is when we apply data science to our own collective data set,” Pham said. For example, the firm has records of its historical positions, transactions, performance, risk-management decisions, and analyst recommendations. By using AI to comb through these, the firm can begin to understand its own behavioral biases.
Pham says there are two big themes going forward. One is using unstructured data (data that is non-organized, and therefore not machine-readable) to drive outperformance in portfolios. For example, company prospectuses for bond issuance may be unstructured. If the firm can find ways to automate the understanding of these texts, it can begin to drive similar insights into new areas such as private equity.
A second opportunity, still being explored, is the possibility of tapping information on blockchain networks as an additional type of data, and determining if it exhibits useful correlations.
Pham did say that some AI tools are not practical. For example, natural-language processing, a methodology of layering neural networks, is specific to language. Building an NLP to make sense of texts written in English is useless for texts written in Japanese.
We will continue this week with a look at two other buy sides’ use of data and AI.
