Jiayi Bai

Analyzed the impact of tariff increases on China's domestic oil consumption by integrating multi-source production, trade, oil prices, and macroeconomic data (1.2M+ records) using SQL & Python to support the Chief Economist team. Identified industrial production growth and export activity as the strongest leading indicators of oil-consumption fluctuations using lag features and feature-importance analysis, informing downstream forecasting and scenario analysis. Benchmarked ARIMA, SARIMA, and Prophet for 7/14/30-day oil consumption forecasting using rolling time-series cross-validation, establishing an organizational baseline for future model comparison.

Data Scientist Intern - China National Petroleum Corporation - Beijing, China

(2025-06 - 2025-09)

• Analyzed the impact of tariff increases on China's domestic oil consumption by integrating multi-source production, trade, oil prices, and macroeconomic data (1.2M+ records) using SQL & Python to support the Chief Economist team.
• Identified industrial production growth and export activity as the strongest leading indicators of oil-consumption fluctuations using lag features and feature-importance analysis, informing downstream forecasting and scenario analysis.
• Benchmarked ARIMA, SARIMA, and Prophet for 7/14/30-day oil consumption forecasting using rolling time-series cross-validation, establishing an organizational baseline for future model comparison.
• Developed Random Forest and XGBoost to detect macro-driven turning points in oil consumption by incorporating macroeconomic, trade, and commodity-market features; conducted 100+ hyperparameter searches, improving F1-score by 20%, precision from 72% to 85%, and recall from 65% to 78% compared with rule-based monitoring approaches.
• Conducted exploratory data analysis, subgroup analysis, and diagnostic checks to identify exposure-response patterns, surface potential confounders, and support interpretable reporting with data visualization using ggplot2.
• Built Tableau dashboards to visualize oil-consumption forecasts, leading demand drivers, and tariff-scenario impacts, increasing reporting frequency by 300%, enabling faster strategic decision-making for energy reserve planning.

Quantitative Research Intern - Haitong Securities - Shanghai, China

(2023-07 - 2023-09)

• Integrated 30M+ financial news articles, research reports, market data, and option datasets using SQL and PySpark; constructed unified datasets and built end-to-end preprocessing pipelines for downstream modeling.
• Fine-tuned FinBERT in Python using HuggingFace Transformers for 3-class financial sentiment classification, improving macro-F1 by 6% over the pre-trained baseline and generating daily sentiment scores for implied-volatility modeling.
• Conducted causal event-study analysis to quantify the impact of market sentiment on implied volatility, finding that high-sentiment news events were associated with a 4.2% increase in cumulative abnormal IV over a 3-day window.
• Incorporated sentiment-driven implied-volatility adjustments into Black-Scholes option-pricing models and Greeks-based hedging simulations, improving hedging accuracy by 20% for OTC structured products.
• Created KPI reporting dashboards and leveraged GPT-based workflows to automate market summaries, translating market insights into visual reports for stakeholder communication and decision support.

Data Scientist Intern - National Bureau of Statistics of China - Beijing, China

(2023-05 - 2023-07)

• Integrated 100K+ observations across consumer spending, employment, public-health, and economic indicators from 31 provinces using SQL; engineered a province-level monthly panel from multi-frequency data to support recovery analysis.
• Modeled regional recovery patterns using ANOVA, panel regression and time-series analysis in Python, revealing recovery-rate differences of up to 35% across provinces and industries in the post-pandemic period.
• Quantified the impact of consumer spending, labor-market conditions, and public-health indicators on recovery through forward & backward feature selection, finding consumer spending explained 45% of recovery variation across provinces.
• Conducted data analysis using K-means clustering to support marketing segmentation strategy and expansion planning
• Developed Tableau dashboards to track recovery trajectories and consumer-spending trends across 31 provinces, helping policymakers identify regions at risk of slower economic recovery and evaluate targeted support strategies.

M.S. - Operation Research - Columbia University (2024-09 - 2026-02)

B.A. - Economics and Mathematics - New York University (2021-01 - 2024-05)