Two Sigma·Data Scientist·Technical Phone Screen
May 2026
Two Sigma DS interview that was basically a single extended case study: you get six years of web search logs and have to design a full pipeline to predict a company's stock price. The interviewer pushed hard on every decision, nothing was left unchallenged.
- You have six years of raw search log data. How do you turn that into model features for a specific company, and which events are even relevant to that company in the first place?
- What is your prediction target and why? Once you've settled on a target, what model class do you start with?
- Financial time series data is noisy. How do you avoid overfitting, and how do you validate the model properly given the temporal structure?
- Search behavior and market regimes both change over six years. How do you handle data drift, variance drift, and seasonality?
- You've trained several models based on different hypotheses about how search activity relates to stock price. How do you decide which one to use, or whether to combine them?
- Search interest is partly reactive: a price move causes news, which causes searches. How would you test whether your features are actually predictive rather than just reflecting moves that already happened?
- How would you show that the search-derived features actually add value over a baseline that only uses past price and volume?
- If the deployed model's live performance decays after six months, how do you diagnose whether the root cause is input drift, a regime change, or backtest overfitting?
- How would your feature pipeline and model change if predictions were needed intraday, within minutes, rather than once per day?
“The entity linking piece tripped me up more than I expected.”