The MSCI Barra Global Equity Model (GEM) family is the original and most widely used commercial holdings-based equity risk model, tracing back to Rosenberg (1974). The models are used for portfolio risk management, construction, and optimization. Jose Menchero was Managing Director at MSCI responsible for building GEM2 and GEM3 before moving to Bloomberg to build MAC3.
Evolution
| Model | Launch | Key advance |
|---|---|---|
| GEM | ~1990s | Original global equity factor model |
| GEM2 | ~2008 | Expanded factor set, broader coverage (Menchero, Morozov, Shepard) |
| GEM3 | 2012 | VRA, Optimization Bias Adjustment, frontier markets, factor splits |
| GEMLT | ~2016 | Systematic Equity Strategies, multi-horizon, point-in-time data |
| Global Equity Factor Model | 2022 | Crowding, Machine Learning, Sustainability, adaptive covariance |
GEM3 (2012)
Factor structure
11 style factors, 34 GICS-based industry factors, 77 country factors. Key changes from GEM2:
| GEM2 factor | GEM3 replacement |
|---|---|
| Volatility | Split into Residual Volatility + Beta |
| Value | Split into Book-to-Price + Earnings Yield + Dividend Yield |
Remaining style factors: Size, Non-Linear Size, Momentum, Growth, Leverage, Liquidity.
Compare with MAC3: both split the original Barra volatility factor into market beta and residual volatility. MAC3 (2020) has 14 style factors, adding Mid-Cap, Valuation (separate from Earnings Yield), Variability, Profit, and Long-Term Reversal.
Methodology innovations
Volatility Regime Adjustment (VRA): calibrates factor volatilities to current market levels using cross-sectional dispersion as a regime indicator. Reduces underforecasting when entering high-volatility regimes and overforecasting when exiting. Similar in concept to MAC3’s Cross-Sectional Volatility (CSV) scaling.
Optimization Bias Adjustment: scales risk up where underforecast and down where overforecast within the factor covariance matrix. Improves accuracy for optimized portfolios without degrading accuracy for non-optimized ones. Related to the eigen-adjusted covariance methodology in Menchero, Wang, and Orr (2011).
Daily model updates: factor exposures, covariance matrices, and specific risk forecasts all updated daily. History back to January 1997.
Coverage
77 countries (developed + emerging + 22 frontier markets), 63 currencies, 72,500+ assets including depositary receipts and cross-listed securities. Estimation universe based on MSCI ACWI & Frontier Markets Index.
GEMLT (~2016)
The Global Total Market Equity Model Suite expanded GEM3 with:
- Systematic Equity Strategy (SES) factors: first time in a Barra global model. Enable alpha generation, strategy evaluation, and measuring sensitivity to crowded trades.
- Multi-industry exposures: insight into different business segments of diversified firms.
- Predicted betas: Bayesian adjustment for more accurate cost-of-capital estimation.
- Point-in-time fundamental data: reduces forward bias in backtesting.
- Multi-horizon structure: Long-Term (portfolio construction), Medium-Term (attribution/monitoring), and Trading (short-term hedging) variants.
- Coverage expanded to 87 countries, 72 currencies, 75,000+ assets including equity index futures and 1,000+ global equity ETFs.
- History back to January 1995.
Global Equity Factor Model (2022) — the latest model
The next-generation model, launched June 2022. Developed in consultation with large institutional investors. Represents the current state of the art in the MSCI/Barra lineup.
Factor structure
9 factor categories with 25-26 style factors (regional models vary):
| Category | Factors (where known) |
|---|---|
| Volatility | Beta, Residual Volatility |
| Yield | Dividend Yield, Earnings Yield |
| Quality | Profit, Leverage, Variability |
| Momentum | Momentum (enhanced IPO handling), Short Interest (enhanced, non-orthogonalized) |
| Value | Book-to-Price |
| Size | Size, Non-Linear Size |
| Growth | Growth (enhanced coverage) |
| Liquidity | Liquidity |
| Sustainability | ESG, Carbon Efficiency |
Plus three novel factors not in any previous Barra model:
Crowding: quantifies “bubbliness” of stocks and portfolios. Built from four metrics:
- Valuations: whether prices are bid up relative to history
- Pairwise correlation & volatility: stocks in a factor moving together with wider swings
- Factor reversal: strong recent performance promoting performance-chasing
- Short interest spread: bottom quintile heavily shorted relative to top
These are combined into a single standardized crowding score per factor. Historically, factors with crowding scores >1 experienced significantly higher frequency of large drawdowns in subsequent periods. Available since 2018 as a standalone model; integrated into the factor model in 2022.
Machine Learning: captures non-linear relationships between factor exposures and returns using data science and NLP. Details of the ML factor construction are not publicly documented.
Sustainability: ESG factor and Carbon Efficiency factor measuring a company’s emissions relative to its size. Enables performance attribution with a dedicated Carbon column (e.g., MSCI Climate Paris Aligned index attribution).
Methodology advances
- Adaptive factor covariance: goes beyond VRA (GEM3) to improve forecast accuracy across changing market regimes. Empirically shows improvement over both the baseline and the optimization bias adjustment alone.
- Cluster risk identification: detects groups of highly similar companies that may create hidden concentration risk in a portfolio.
- SPAC coverage: pre-merger Special Purpose Acquisition Corporations included for the first time.
- Enhanced existing factors: Beta, Momentum, Residual Volatility (improved IPO handling), Growth (improved coverage), Short Interest (improved stability, non-orthogonalized).
Coverage and delivery
- 85+ countries, 80,000+ securities, 45-50+ GICS-based industry factors
- Asset classes: stocks, depositary receipts, cross-listed securities, equity index futures, ETFs, SPACs
- Available in Long-Term (Stable and Responsive variants) and Trading variants
- Distributed via Snowflake, BarraOne, Barra Portfolio Manager, Models Direct, and third-party platforms
What’s still not public
MSCI has not released detailed methodology notes for the next-gen model comparable to the GEM3 empirical notes. The publicly available documents are marketing factsheets. Factor construction details (especially for Crowding, ML, and Sustainability) and covariance estimation methodology are available only to licensed clients.
Comparison with other commercial models
| MSCI Barra GEM3 | Bloomberg MAC3 | Two Sigma Venn | |
|---|---|---|---|
| Type | Holdings-based | Holdings-based | Return-based |
| Style factors | 11 | 14 | 18 (macro + styles) |
| Industry factors | 34 (GICS) | BICS hierarchy | N/A |
| Country factors | 77 | 13 local models | Macro factors |
| Industry/country exposures | Binary (0/1) | Beta-based | N/A |
| Regression weights | sqrt(market cap) | Inverse residual variance | N/A |
| Risk horizons | 2 (short/long) | 6 (daily to long-term) | Single |
| Launched | 2012 | 2020 | 2018 |
The key methodological difference between GEM3 and MAC3 is that MAC3 uses industry/country betas instead of binary exposures and inverse residual variance regression weights instead of sqrt-market-cap. Both innovations were introduced by Menchero after he moved from MSCI to Bloomberg.