Can Memes Predict Markets? Using Social Media Sentiment for Stock Forecasting
summary
Claims that social media sentiment predicts prices get tested on mid‑liquidity Asian stocks (14 assets) and cryptocurrencies (≈13,000 hourly Reddit points). Sentiment uses a log ratio of bullish to bearish posts with weighted rolling windows; classification moved from fine‑tuned BERT to OpenAI API for better labels. Vectorized backtests show typical gains of 10–20% over buy‑and‑hold on many assets and a portfolio peak of 600% normalized return for crypto; Twitter proves unreliable, XGBoost beats fancy deep models, and per‑asset rolling re‑optimization is required.
AI-written summary · about
chapters (9)
- 0:00Meme-based market prediction
- 2:11Efficient market hypothesis limits
- 6:28Social media sentiment matters
- 10:40AI challenges in finance
- 13:13Data sources setup
- 17:28Sentiment analysis results
- 21:47Modeling techniques
- 25:45Q&A data methodology
- 32:54Q&A commodity limits
related talks
30mIncreasing the trading prediction by mining aggregated human texting messages
33mPredicting Cryptocurrency Exchange Rates
39mThe art and science of making computers understand sentiment and emotion
30mPredictive trading
53mA Machine Learning mood-o-meter
01h 05mSocial Text, Sentiment, and Tone Analysis
37mAI/ML Strategies & Techniques for a Post Pandemic World
30mHow to build a stock market money printing machine with AI