How Our ETF Screening & Comparison Methodology Works

Most “AI in investment” content stops at the adjective. It names the technology and skips the arithmetic. These pages do the opposite: each article takes one screening or comparison question — how two similar equity ETFs differ, what a factor screen really ranks, where document parsing earns its keep — and shows the workflow that produced the answer.

The steps below are the standing recipe. Individual pieces deviate when the subject demands it, and the deviation is printed inside the piece.

The six working steps

1. Frame the universe

   Start by fixing what counts as comparable: region, asset class, index family, and the replication method (physical or synthetic). A screen that mixes these produces confident nonsense.

   Public sources: exchange listings and issuer product pages.

2. Collect the raw records

   Pull the publicly reported numbers a comparison needs — net asset value history, total expense ratio, assets under management, daily traded value, bid-ask spread and the full holdings file for every fund in the universe.

   Public sources: issuer factsheets, exchange data pages, fund filings.

3. Parse the documents at machine speed

   This is where the AI enters first. Text-analysis routines read prospectuses and factsheets, extracting index-construction rules, securities-lending terms and dividend treatment — clauses that change what a number means, and that a ratio table never shows.

   Inputs: the filings themselves; flagged passages then get a human editorial read.

4. Run the screens

   Cost screens (expense ratio plus spread), tracking screens (difference and error against the stated index) and liquidity screens (traded value and assets) rank each fund within its universe — soft ranks, printed openly, never verdicts.

5. Cluster the peers

   Similarity routines group funds by holdings overlap and factor profile rather than by marketing label. This is how two “different” products turn out to be one portfolio in two wrappers — and one genuine alternative appears where none was advertised.

6. Write the comparison, show the work

   Tables are built under identical criteria for every fund in the cluster, uncertainties are labelled as such, and the checklist that produced them is appended to the finished piece.