README.md

---
license: other
tags:
  - finance
  - nlp
  - sec-filings
  - item-1a
  - volatility
  - bert
  - bertopic
pretty_name: Dissertation Phase 1 Dataset — Hybrid Financial Risk Modelling
---

# Dissertation Dataset — Phase 1 Data Pipeline

**Project:** Hybrid Topic and Domain-Adaptive Modelling for Financial Risk and Forecasting
**Institution:** University of Edinburgh, School of Informatics
**Supervisor:** Prof Tiejun Ma

> This is a **private** research dataset. All files are used exclusively within the scope of this dissertation project and must not be redistributed.

---

## What This Repo Contains

This dataset repo holds all Phase 1 pipeline outputs needed to run Phases 2–5 of the dissertation. It is the single source of truth for the cluster environment.

| File | Size | Description |
|------|------|-------------|
| `sp500_1A.tar.gz` | ~440 MB | Item 1A Risk Factor text corpus — 8,247 pickle files, one per `{TICKER}_{YEAR}`, covering S&P 500 firms 2006–2025. Each pickle is a plain string of the full Item 1A section. |
| `feature_table.parquet` | ~0.2 MB | Phase 1 master table — 8,105 rows, one per filing. Columns: `ticker, cik, permno, sic, fiscal_year, filing_date, report_date, lagged_vol_30d, fwd_vol_30d`. Placeholder columns for `embedding` (Phase 2) and `topic_vector` (Phase 4). |
| `filings_index.csv` | ~0.9 MB | EDGAR submissions API output — filing dates, SIC codes, and `has_pickle` flag for all 656 CIKs. 11,447 total 10-K filings (2006–2025). |
| `volatility_labels.csv` | ~0.7 MB | 30-day annualised volatility windows per filing. `lagged_vol_30d` = AR(1) baseline; `fwd_vol_30d` = prediction target. 94.9% coverage. |
| `permno_linkage.csv` | <0.1 MB | CIK → PERMNO mapping from WRDS CCM. 651/656 firms matched. |
| `sp500_union_constituents(1).csv` | <0.1 MB | S&P 500 universe — 656 unique CIK/ticker pairs (historical constituents). |
| `README.md` | — | This file. |

---

## How to Download on the Cluster

```bash
# One-time login
huggingface-cli login

# Download all files
huggingface-cli download SarthakVishnu/dissertation-dataset \
    --repo-type dataset \
    --local-dir ~/dissertation/datasets/

# Unzip the Item 1A corpus
cd ~/dissertation/datasets/
tar -xzf sp500_1A.tar.gz && rm sp500_1A.tar.gz
```

---

## How Each File Is Used Per Phase

| Phase | Files Used |
|-------|-----------|
| **Phase 2 — DAPT** | `sp500_1A/` (MLM pretraining corpus, train split ≤2024) |
| **Phase 3 — Contrastive FT** | `sp500_1A/` + `filings_index.csv` (SIC codes for sector-view positive pairs) |
| **Phase 4 — BERTopic** | `sp500_1A/` + `feature_table.parquet` (per-filing topic vectors) |
| **Phase 5 — Ablation** | `feature_table.parquet` + `volatility_labels.csv` (downstream vol forecasting) |

---

## Temporal Split

| Split | Criterion | Approx. Filings | Role |
|-------|-----------|-----------------|------|
| Train | `filing_date < 2025-01-01` | ~7,700 | DAPT, contrastive FT, BERTopic |
| Val | `2025-01-01 ≤ filing_date < 2026-01-01` | ~280 | Perplexity checkpointing, FinMTEB eval |
| Test | `filing_date ≥ 2026-01-01` | ~125 | Held-out volatility forecasting |

---

## Licence & Ethics

- **Item 1A text** and **EDGAR-derived data** are from publicly available SEC filings.
- **WRDS/CRSP-derived data** (`permno_linkage.csv`, `volatility_labels.csv`, `feature_table.parquet`) is used under the University of Edinburgh's institutional licence and must not be redistributed externally.
- No personal data or human subjects are involved.