Rolling IC (60D) - Single Factor Analysis
Rolling statistics begin once 60 trading days of data are available.
Rolling ICIR (60D)
ICIR = mean(IC_60d) / std(IC_60d)
Rolling t-stat (60D)
t-stat = mean(IC_60d) / std(IC_60d) × √N
Long-Short Portfolio Performance (by Factor)
Definition:
• Long Portfolio = Top quintile (20%) stocks with highest factor scores
• Short Portfolio = Bottom quintile (20%) stocks with lowest factor scores
• Long-Short Return = long_return - short_return (the difference, not cumulative subtraction)
• Note: If Short portfolio outperforms Long portfolio (Long-Short is negative), this indicates the factor is negatively predictive. In such cases, reverse the factor sign for trading.
• Long Portfolio = Top quintile (20%) stocks with highest factor scores
• Short Portfolio = Bottom quintile (20%) stocks with lowest factor scores
• Long-Short Return = long_return - short_return (the difference, not cumulative subtraction)
• Note: If Short portfolio outperforms Long portfolio (Long-Short is negative), this indicates the factor is negatively predictive. In such cases, reverse the factor sign for trading.
Factor Correlation Matrix
Using Pearson correlation method
Multi-Factor Risk Exposure (Barra-style)
Factor Distribution by Barra Style (IC vs ICIR)
Note: This is NOT machine learning clustering. Factors are grouped by Barra-style economic definitions used in the Multi-Factor Risk Exposure model. The purpose is to evaluate factor quality and stability within each risk style bucket.