Do you understand the information contained in your ETF documents?
An ETF (Exchange-Traded Fund) is a type of investment product that allows buyers to invest in diversified securities at once, for a single price. They are traded on stock exchanges and can be bought and sold during a trading day, in a similar manner to stocks. This allows an individual to participate in the gains and losses of the underlying securities held within the ETF. ETFs typically follow one particular index or sector which means that they replicate that particular focus. The vast majority of ETFs are not managed by humans, which has, in part, led to their rise in popularity among investors.
ETF regulatory documents are typically lengthy and many investors don’t take the time to read or understand their contents
With the widespread adoption of fintech companies and robo-advisors, the ease with which an individual can invest in an ETF has increased, and so has their popularity.1 Securities regulations in the United States and Canada require that a Prospectus and other disclosure documents be prepared for each ETF, and be made available to investors. Given that these documents are lengthy, a tool that allows an investor to quickly extract the key information within those documents is valuable, since it is prudent for investors to understand the details of their investments, particularly as investors continue to hold more ETFs in their portfolios.
Using Zuva AI Trainer, we built 5 custom AI fields for use with DocAI
Zuva’s 1300+ built-in AI fields are generally created broadly, to apply to a wide variety of customers across a wide range of industries. Check out Zuva’s AI field library to learn more. For our purposes, we required highly specific ETF AI fields that don’t currently exist as built-in AI fields; we solved this gap by using Zuva AI Trainer to create them.
AI Trainer is our user interface product that allows you to quickly and easily build custom AI fields for use within Zuva DocAI. While DocAI has built-in AI fields that apply to many different use cases, there may be a need to build custom AI fields. Typically, our customers use AI Trainer to build more specific or narrow AI fields, as well as AI fields which may not currently exist as built-in AI fields in DocAI.
We used the Zuva AI Trainer tool to build 5 ETF-related AI fields and observed how they performed. Our goal was to create AI fields that would help an investor quickly find key information about an ETF, within lengthy regulatory ETF documents. Using 50 ETF documents (Prospectuses, Summary Prospectuses, and Fact Sheets), we were able to build AI fields to capture Distributions, Investment Objectives, Listing Exchanges, Management Fees, and Portfolio Managers. While there are many other important pieces of information that an investor should understand, for the purposes of our testing, we distilled it down to the 5 points above. In our view, these are among the most relevant since they give an investor a summary of varying information about their ETF. This includes when the investor can expect to receive cash, how much the ETF is paying to the manager, where the ETF is listed, what index the ETF is tracking, and who is managing the ETF.
Our 50 ETF documents were all from the United States or Canada. There were 7 ETF Fact Sheets, 15 Summary Prospectuses and 28 long-form Prospectuses, totalling approximately 2,000 pages. Most of the long-form Prospectuses contained disclosure information for multiple funds, including one which contained disclosure information for 22 different ETFs.
To train our custom AI fields, we had to upload our documents to AI Trainer, create our AI fields and begin highlighting (annotating) the relevant language for each AI field. After 10 documents were annotated and added to our training, Zuva trained the AI fields automatically. After the first training, we continued annotating our AI fields, added them to training, and monitored our results.
Zuva’s suggestions facilitated a quicker review and training process for our custom AI fields
After our first set of training on 10 documents, Zuva’s AI began to generate suggestions that we could then accept, edit, or reject. As training progressed further, Zuva’s suggestions became more accurate which sped up our review, and the training process as a whole. Suggestions for the more conceptual-based, consistently worded AI fields, such as Investment Objective and Distributions, were appearing and were more accurate quicker than the more narrow Listing Exchange, Management Fee and Portfolio Manager AI fields.
The total training on 50 documents took a single Legal Knowledge Engineer with limited prior exposure to ETFs approximately 4 hours to complete. After the custom AI fields created in AI Trainer are published to DocAI, Zuva can use those custom AI fields to quickly produce extractions. These AI fields can be applied to an unlimited number of documents, so the information can easily be extracted from however many documents you have.
Examples of some of the suggestions that Zuva’s AI provided for the custom AI fields during our training were:
Distributions: The Fund intends to pay out dividends and interest income, if any, at least quarterly, and distribute any net realized capital gains to its shareholders at least annually.
Investment Objective: To provide investment results that correspond generally to the price and yield performance, before fees and expenses, of the Underlying Index.
Listing Exchange: NASDAQ
Management Fee: 0.68%
Portfolio Manager: 1832 Asset Management L.P.
Zuva DocAI’s technology can be used to quickly and easily extract relevant information from documents
We were able to easily create our own custom AI fields in Zuva AI Trainer and, as our training progressed, could monitor the results by assessing the quality of Zuva’s suggestions and the analytics (Precision, Recall, F-Score and Highlight Overlap) of each AI field. The speed of our training also progressed as Zuva’s AI began to produce higher quality suggestions, which occurred early in our training for our more simple AI fields.
When our training was complete, we had the option to publish these AI fields to Zuva DocAI. DocAI allows its users to find the information they are looking for within documents by providing extractions of relevant information for each desired AI field contained in the document. DocAI creates significant time savings when reviewing documents through these extractions, which require far less manual effort.
Curious about the Zuva AI Trainer technology? Head over to the Zuva website to find out more!
Share this article:
