21 May Apply Natural Language Understanding in Measuring Diversity
At TenPoint7, we partnered with industry experts to measure and monitor culture metrics of the 13000 registered investment management firms in the US. The investment management industry is lagging in achieving diversity goals according to a recent self-reporting survey by the CFA Institution, https://www.cfainstitute.org/en/about/press-releases/2018/cfa-institute-releases-diversity-and-inclusion-guide
When asked how their firm is doing on diversity, most participants scored their firms as “average” (3.1 on a 5-point scale), with San Francisco firms reporting the highest (3.5) and Boston firms reporting the lowest (2.6).• When asked to score how the industry overall is doing on diversity, the overall score was low (1.8 on a 5-point scale). [1*]
Experts believe that it has become more and more important for investors and investment management firms to be able to measure and monitor diversity against industry guidelines.. However, doing so requires analysis of a large amount of data. This is where the TenPoint7 product, Addy, helped by applying Natural Language Understanding.
CFA Institute is a global, not-for-profit professional organization that sets new industry guidelines and provides investment professionals with finance education.
Amidst demand for greater diversity and inclusion (D&I) in the investment management industry, CFA Institute has released a guide titled, “Driving Change: Diversity & Inclusion In Investment Management.” The guide examines the current state of the industry and outlines 20 actions to encourage more diverse and inclusive workplaces, to ultimately achieve better outcomes for investors. [1*]
To illustrate what Addy can do in this blog, I will focus on measuring firm gender diversity against CFA Institute guide. The basic idea is as follows:
- Industry experts identify best reference sentences about gender diversity from CFA Institute Guide.
- For each firm, Addy collects text data related to “diversity” from the firm’s official websites.
- Using natural language understanding or NLU, Addy measures the firm’s alignment with gender diversity guidelines.
Industry experts identify best reference sentences about gender diversity from CFA Institute Guide.
Women are a minority in investment management across global markets, making them the “universal diversifier”. Our experts identified the following goals of women in investment management initiatives proposed by CFA Institute.
Increase the number of women who join the profession and earn industry recognition such as the CFA charter
Retain women in the profession and influence culture from within
Create demand for diversity as an industry imperative
CFA Institute found that of more than 800 institutional investors, 83% said gender diversity is important to them. The majority (55%) believed that gender diversity in investment teams leads to better performance through diverse viewpoints, and 28% simply preferred to hire an investment firm that is supportive of gender diversity.
For each firm, Addy collects text data related to “diversity” from the firm’s official websites.
Addy can efficiently combine the power of a web search engine with Addy text data ingestion pipelines, and intelligent text extraction technology..
For example, Addy can execute a search query such as “site:firmwebsite diversity” in order to ingest any web data related to diversity from a firm’s official website.
Using natural language understanding or NLU, Addy measures the firm’s alignment with gender diversity guidelines.
Let’s use two anonymous but very large investment firms (Firm A and Firm B) in the US to explain how it works. Addy scans text data sentence by sentence and quickly detects most similar sentences to the previously mentioned reference sentences from the CFA Institute. Addy combines the BERT language model  and FAISS (Facebook AI Similarity Search)  to accomplish this.
In Firm A, the most similar sentences are
The makeup of our intern classes has become more diverse over the past five years. We have also seen progress at the more senior levels, with the percent of women managing directors reaching 19 percent in 2017, a steady increase from 14 percent in 2012.
“Proactively increasing the representation of women in senior management roles and nurturing the future pipeline is a key strategic priority for our global leadership team as well as for me personally.”
Our language model shows that those statements have high similarity to the women in investment management initiative. On a scale of 0-100, the similarity measure is 78. The initiative advocates for “Increase the number of women who join the profession” and Firm A states a concrete 5% increase in “percent of women managing directors”. The alignment of Firm A with the initiative would seem very strong to a human reader as well.
In Firm B, the most similar sentences:
Socially responsible investing has been around for years, but a desire for asset managers to screen companies with an ESG (“Environmental, Social and Governance”) lens has risen in 2020…Under social, examples include workforce diversity, health and safety, training and development, human rights, community relations, supply chain transparency and political contributions.
While “workforce diversity” is mentioned, it is just a general statement about investment trends and there is nothing specific about what Firm B did. The alignment seems weak to a human reader and the language model shows a similarity measure of just 22.
While we should not rate an investment management firm solely based on the official website, what a firm discloses to the public on the website reflects the firm’s emphasis and understanding of the diversity initiatives. Therefore our experts would use website alignment as one of the metrics in firm diversity. They convert the previously mentioned similarity measure to a normalized rating system used for benchmarking firms compared to each other or even compared to specific peer groups.
In summary, the new NLU feature in Addy enables industry experts to create customized metrics based on text data in order to monitor and compare a large number of companies.
Our next steps are:
Expand data sources to include third party data such as Glassdoor employee reviews.
In addition to similarity measures, apply document classification models based on expert labeling.