The Lexicon-Based Sentiment Analysis for Stock Price Prediction in Islamic Fashion Industry

Abstrak

This study develops a lexicon-based sentiment analysis method to examine the relationship between social media sentiment and stock price movements in Indonesia's Islamic fashion industry. We collected 6,087 social media comments from TikTok, Instagram, and Twitter over 396 days, alongside daily stock data from Elzatta (ZATA). Using an Indonesian fashion-specific lexicon with 42 positive and 38 negative indicators validated through inter-annotator agreement (Cohen's Kappa = 0.82), we employed ordinary least squares regression with heteroskedasticity-robust standard errors to test sentiment-return relationships. Results show no economically significant relationship between social media sentiment and stock returns (correlation = -0.06, p = 0.240), with models explaining only 2.09% of return variance. We identify three boundary conditions limiting sentiment-based prediction: (1) feature mismatch 82% of comments discuss product attributes rather than business fundamentals, (2) demographic disconnect social media users (78% female, aged 18-29) differ markedly from investors (68% male, aged 35-55), and (3) market microstructure constraints 43% zero-volume days and 2.3% bid-ask spreads impede price discovery. This study provides the first Indonesian fashion-specific sentiment lexicon and establishes actionable validation guidelines for practitioners: sentiment features must be verified for demographic alignment and market liquidity before deployment. For machine learning applications in emerging markets