Explain factor leakage checks and IC/ICIR filtering
Company: Citadel
Role: Data Scientist
Category: Machine Learning
Difficulty: medium
Interview Round: Technical Screen
You’re interviewing for a quantitative/alpha role and have built predictive factors (features) for returns.
Answer the following (conceptual) questions clearly:
1. **How do you determine whether a factor leaks future information (uses “future data”)**?
- Give concrete examples of common leakage sources in financial datasets.
- Describe how you would test for leakage.
2. **How would you use IC and ICIR to select factors**?
- Define **IC** and **ICIR** (state any assumptions, e.g., cross-sectional vs time-series).
- Describe a practical factor screening and monitoring workflow.
3. **If you train a model using factors, how do you train it and what is the objective**?
- Describe the training setup (labels, horizon, universe, sampling).
- Provide typical objectives used in alpha modeling.
4. **How do you prevent overfitting in factor/model research**?
- Cover data-splitting methodology specific to time series and finance.
- Mention regularization / model complexity controls and research-process controls.
Quick Answer: This question evaluates competency in factor-based predictive modeling, including detection of information leakage, use of information coefficient (IC) and ICIR for factor screening, construction and training of alpha models, and controls for overfitting in finance-specific time series.