Mapping Immersive-Technology Disclosure in 10-K Filings

Description

Overview This replication package contains all code and data necessary to reproduce the analysis. The study measures how metaverse-index constituent firms disclose immersive and enabling technologies across two sections of their 10-K filings: Part 1 (Business Description) and Item 7 (MD\&A). Instead of a predefined keyword dictionary, it uses a reproducible, embedding-based measure of immersive-technology disclosure, and then groups relevant paragraphs into interpretable themes using embedding-based clustering.

Steps to reproduce

### Step 1: Set up the Python environment Python 3.10 or later is required. ```bash cd replication_package pip install -r requirements.txt ``` The first run downloads the sentence-transformer model `all-mpnet-base-v2` (about 420 MB) from the Hugging Face model hub, so an internet connection is required for the first run. ### Step 2: Run the analysis ```bash python code/run_analysis.py ``` The script builds the paragraph corpus, encodes every paragraph, scores each paragraph for immersive-technology relevance, aggregates firm-year measures, selects the number of themes by silhouette analysis, labels the themes, and writes all tables and figures to `output/`. **Expected runtime:** Approximately 8 to 15 minutes on a standard desktop CPU, most of which is paragraph encoding. A CUDA-capable GPU reduces this substantially. ### Step 3: Verify output The console prints a summary matching the Key results table above, including the descriptive statistics block. Generated files appear in `output/tables/` and `output/figures/`.

Institutions

• Auckland University of Technology

  Auckland, Auckland

Categories

Licence