Winnie (Xiaowen) YU

Alpha Research: Construct alpha signals by applying Machine Learning and Natural Language Processing techniques on alternative textual datasets, including annual reports, earnings call transcripts and Google patent dataset.

• Executive’s communication style in earnings calls, examples: Concreteness (sematic consistency score of managers’ answers from a trained topic model); Evasiveness (the topic similarity between analyst’s question and manager’s answers); Linguistic complexity (separation of informative disclosure and intended obfuscation); Contextualized sentiment (word window-based double search of contextual and sentimental dictionaries); CEO-CFO language style matching (the level of automatic coordination and consistency).
• Investment opportunities prediction: identify keywords whose frequency best predict firm’s future actual investments using LASSO regression and predict firm’s investment opportunities.
• Corporate culture: created an expanded culture dictionary from seed words of five values (innovation, integrity, respect, quality, and teamwork) and measure firm level culture scores.
• Textual PEAD (post earnings announcement drift): train a Logistic Regression model to generate word level beta score that predicts market surprise around earnings announcement and predict future PEAD.
• Technology quality, measured by patent centrality and forward citation frequency (importance, stability, most influential time).
• Technology-linked momentum, technology similarity weighted momentum, similarity measured using patent classification and content.
• Innovation strategy, measured by the change of technology focus.
• Technology leader, fast follower and beneficiaries detection.
• Other alpha signals include complexity mispricing, violation penalty, management compensation, geographical exposure linked momentum, cross-shareholdings in Japan, etc.
• Factors have good out-of-sample performance in live production.

Model and Portfolio Management

• Responsible for managing, maintaining, and recalibrating Europe large cap and small cap models, from universe generation and customization, new factor additivity tests to optimal factor weights assignments.
• Manage the European components of global portfolios as a co-PM, conduct portfolio attribution analysis, work with other PMs on the optimization and rebalance.

Marketing and Production

• Responsible for research market slides used to communicate with current and prospect clients.
• Work with Data Science to tackle production issues and generate stable model alpha scores for trading.

• Built ESG dictionary for NLP research by extracting words from 500 sustainability reports, generated ESG Manager Sentiment Score which has high IC in R2000 (t-stats 3.21) and International Universe (t-stats 2.34)
• Conducted web scraping on Google News platform, trained n-gram language models to capture unusualness in news using cross-entropy, and combined unusualness with sentiment scores to predict stock price movement.

• Tested how traditional factors work in Chinese A-shares, analyzed the background and direct cause of the Chinese stock market crash in 2015, and built portfolios capturing technical and fundamental alpha signals
• Investigated the hedging performance of onshore and offshore index futures and A-H share dual-listed stocks.

• Conducted research with topics covering macro economy, capital markets and government policies.
• Worked on economic development index for low-carbon economy.