From Berkan Sesen’s experience

Using natural language processing to build thematic equity funds

At J.P. Morgan Asset Management, Berkan co-managed machine-learning thematic funds that used natural language processing to screen global stocks, with analysts validating the selections.

Industry
Financial services
Organization
J.P. Morgan Asset Management
Berkan Sesen’s role
Senior quantitative portfolio manager and fund co-manager
When
2017-2021

The business need

A thematic fund depends on finding the listed companies most exposed to a theme. J.P. Morgan Asset Management announced its Genetic Therapies fund in Europe in November 2019 to give investors diversified exposure to that theme.

The evidence for exposure sits in a very large body of text: news articles, company profiles, research notes and regulatory filings, across more than 10,000 stocks worldwide.

What Berkan Sesen did

  1. Berkan worked as a research analyst and portfolio manager in the firm's Quantitative Beta Strategies group, a team that builds machine-learning based investment technology.
  2. The team's thematic engine, ThemeBot, used natural language processing to screen more than 10,000 stocks and flag those most exposed to a theme, accounting for liquidity, market capitalization and profitability.
  3. The funds' methodology, which ranks companies by combining a textual relevance score with a revenue exposure score, was filed as a US patent application in 2020, with Berkan as a named inventor.
  4. Before any stock entered the portfolio, the team and the firm's equity analysts vetted and validated the engine's selections.
  5. Berkan co-managed the Genetic Therapies fund with two other portfolio managers; the fund drew on research by the UBS Global Wealth Management Chief Investment Office and was distributed through UBS initially.
  6. Berkan also co-developed the News Filter, a machine learning and natural language processing system built at J.P. Morgan Asset Management.

Technology and approach

  • Natural language processing over news articles, company profiles, research notes and regulatory filings
  • Composite scoring of textual relevance and revenue exposure (US patent application US20200320633, filed 2020)
  • A convolutional neural network that classifies publications by relevance, trained on manually labeled examples (a separate JPMorgan patent application, US20200142962, filed 2018, with Berkan as a named inventor)

Outcome

  • The Genetic Therapies fund was announced as J.P. Morgan Asset Management's first actively managed fund combining machine learning and active equity insights.
  • The machine-learning thematic funds reached more than $2bn in combined assets under management.
  • The News Filter won at the Banking Technology Awards UK and the American Financial Technology Awards in 2017, according to Berkan's site.

Decisions this experience speaks to

  • Where does AI actually earn its place in this product, and where are we just chasing the hype?
  • We're being told the model performs well. How do we know that number is real and not an artefact of how it was tested?

Related services and use cases

Sources

  1. Berkan Sesen’s profile on CTO Grand Rounds
  2. Fund launch announcement - Funds Society (opens in a new tab)
  3. Berkan Sesen bio - J.P. Morgan Asset Management (opens in a new tab)
  4. Thematic portfolio patent application - Google Patents (opens in a new tab)
  5. Content filtering patent application - Google Patents (opens in a new tab)
  6. Berkan's experience and awards - sesen.ai (opens in a new tab)