Student-built ETF forecasting pipeline with LLM-scored news sentiment dataset

Design · All designs
Evidence · 2 claims (2 for) · 1 study (1 design), q2 · 0 of 1 report an effect size · 2 claims rest on one study

Description

The project's technical artifact is a two-stage predictive system: regression models estimate the magnitude of ETF price deltas and classification models predict direction, with predictions merged by multiplying magnitude and sign. The dataset covers "a historical price dataset for 29 exchange-traded funds (ETFs), covering January 2, 2024 through September 23, 2025" harvested via the Yahoo Finance API, plus 260 scraped NASDAQ news articles scored by gpt-5-mini on a -10 to +10 scale with justifications, aggregated daily by sector. Sentiment coverage was sparse and uneven across ETFs, from 0% to 23.16% as shown in Table 1.

Design Implications

Context

Requirements

• Mentor intervention was needed for rate limiting, error recovery, and anti-bot scraping measures such as dynamic content loading.

Constraints

• Sector-level sentiment aggregation may dilute ETF-specific signals; zero-imputation for days without news may introduce bias; LLM scores were not validated against human-coded ground truth.

Target Learners

• high-school students
• early-undergraduate students

Learning Goals

• working with APIs and structured financial data
• machine learning model development and validation
• web scraping and data engineering

Claims

• Sentiment Features Did Not Improve Etf Forecasting [+M]
• Svm Regressors Outperformed Other Regressors [+M]

Related Designs

Examples

-

Key Sources

• Freyaa Chawla, Ahan Chawla, Rishi Singh, Joe Germino, Grigorii Khvatskii. (2026). Human-AI Co-Mentorship in Project-Based Learning: A Case Study in Financial Forecasting. arXiv preprint. https://arxiv.org/abs/2605.05144

Page metadata

Field	Value
type	design
id	etf-forecasting-pipeline-sentiment-dataset
title	Student-built ETF forecasting pipeline with LLM-scored news sentiment dataset
description	The project's technical artifact is a two-stage predictive system: regression models estimate the magnitude of ETF price deltas and classification models predict direction, with predictions merged by multiplying magni...
status	review
generated	by process:settle-candidates on 2026-10-11
trust	unverified — no one has confirmed this page's content
sources	1 — freyaa-chawla-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.