Factor model methodology

Factor Model Methodology

Audience: Allocators, portfolio managers, risk analysts, and compliance teams evaluating OMNI’s factor model for portfolio construction, attribution, and risk management.

Overview

OMNI Datastream provides an allocator-grade factor model spanning 230+ factors across 7 categories: market, style, macro, sector, industry, country, and thematic. The model follows a hierarchical purification architecture consistent with institutional standards pioneered by MSCI Barra and adopted by Axioma (Qontigo), Bloomberg PORT, and Northfield.

Every factor return is recomputed daily with intraday snapshots at 1-minute and 5-minute intervals during US market hours (9:30–16:00 ET).

Factor categories

Category	Count	Description
market	1	Broad US equity risk premium (MKT_US)
style	24	Cross-sectional return drivers (momentum, value, quality, size, low-volatility, etc.)
macro	16	Macroeconomic regime factors (rates, credit, commodities, currencies)
sector	21	GICS sector and sub-sector return attribution
industry	27	Industry-level return attribution
country	61	Single-country and regional equity factors
thematic	81	Named investment themes (nuclear energy, genomics, REITs, etc.)

Factor construction pipeline

Step 1: Universe construction

Factors are computed over a point-in-time US equity universe refreshed monthly:

• Minimum $100M market capitalization
• Minimum $5M average daily dollar volume
• Minimum $1.00 share price
• Excludes warrants, units, rights, ETFs, ETNs, ADR certificates
• Exchanges: NYSE, NASDAQ, AMEX, ARCA, BATS

Universe membership is determined as of each rebalance date using data available at that time — no look-ahead bias.

Step 2: Constituent sourcing

Each factor type uses a different data source for membership and characteristic computation:

Source	Factor types	Example
Point-in-time fundamentals (SEC XBRL)	Style factors	Book/Market ratio from latest 10-K filing
SIC classification	Sector and industry factors	SIC 7372 → Technology / Software
ETF holdings (Polygon API)	Thematic baskets with ETF proxy	URA holdings → Nuclear Energy basket
Curated constituent lists	Thematic baskets without ETF proxy	Hotel REITs: APLE, HLT, MAR, HST…
Single-country ETF	Country factors	EWJ → Japan equity premium

Step 3: Return computation

Style factors: Long-short decile portfolio sorts. Stocks are ranked by a characteristic (e.g., book-to-market for Value), divided into decile portfolios, and the factor return is the spread between the top decile (long) and bottom decile (short). Portfolios are cap-weighted within each leg and rebalanced monthly.

Thematic baskets (ETF-backed, 66 factors): Factor return is the ETF return minus a relevant benchmark. The ETF provider (Global X, ARK, VanEck, iShares, etc.) handles constituent selection and rebalancing. Example: THEMATIC_NUCLEAR_ENERGY = R(URA) - R(SPY).

Thematic baskets (curated, 15 factors): Factor return is the equal-weight constituent basket return minus a relevant benchmark. Constituents are sourced from public factor composition data and reviewed quarterly. Example: THEMATIC_CASINO_LEISURE = (1/9) · Σ R(casino stocks) - R(SPY).

Macro and country factors: ETF proxy spread versus a reference instrument. Example: RATES = R(TLT) - R(SHV).

Step 4: Hierarchical purification (orthogonalization)

Each factor declares 1–3 parent factors and is purified against them using rolling 156-trading-day OLS regression to extract the residual return not explained by parent factors.

Market (MKT_US)
├── Style factors (SIZE, VALUE, MOMENTUM, ...)
│   └── purified against: MKT_US
├── Sector factors (SECTOR_ENERGY, SECTOR_TECH, ...)
│   └── purified against: MKT_US
├── Thematic baskets
│   ├── Energy themes → purified against: MKT_US, SECTOR_ENERGY
│   ├── Tech themes → purified against: MKT_US, SECTOR_TECH
│   ├── REIT themes → purified against: MKT_US, THEMATIC_REITS
│   └── Country themes → purified against: MKT_US, regional parent
└── Country factors
    └── purified against: MKT_US, regional parent

Why hierarchical purification?

The choice to purify against a small number of theoretically motivated parent factors — rather than all factors simultaneously — is a deliberate methodological decision supported by decades of factor research:

• Parsimony: Fama & French (2018, Journal of Financial Economics) argue that factor models should include only factors that earn their place. Redundant factors should be excluded, not controlled for via regression.
• Overfitting resistance: Harvey, Liu & Zhu (2016, Review of Financial Studies) show that including many correlated regressors inflates false discovery. Kozak, Nagel & Santosh (2020, Journal of Financial Economics) demonstrate that shrinkage estimators dominate unrestricted high-dimensional OLS.
• Institutional alignment: MSCI Barra’s US Equity Model (USE4) uses hierarchical nesting: market → country → industry → style. Axioma and Bloomberg follow similar architectures. No major institutional risk model provider uses unrestricted regression against 100+ factors.
• Signal preservation: Purifying a housing theme against market + real estate sector preserves the thematic signal. Over-purification against all known factors strips out the very characteristics that define the theme.

Step 5: Dynamic volatility targeting

After purification, each factor’s residual return series is dynamically leveraged to target 10% annualized volatility:

• Rolling realized volatility computed over the same 156-day window
• Leverage = min(target_vol / realized_vol, 3.0)
• Prevents any single factor from dominating portfolio-level attribution
• Consistent with standard practice at AQR, MSCI, and major factor ETF providers

Step 6: Z-score computation

Scaled returns are converted to rolling z-scores for cross-factor comparability. Z-scores are the primary output for dashboards, screening, and signal generation.

Intraday factor snapshots

During US market hours (9:30–16:00 ET), OMNI computes 1-minute and 5-minute factor snapshots using real-time ETF and constituent prices. Each snapshot includes raw return, purified return, scaled return, and z-score.

Data provenance and traceability

Every factor return observation includes:

• requestId and traceparent for end-to-end request tracing
• modelName identifying the computation pipeline version
• methodology.inputs listing the data sources consumed
• sourceRightsNotes documenting licensing posture for each factor family

Methodology comparison

Feature	OMNI	MSCI Barra (USE4)
Purification	Hierarchical, 1–3 declared parents	Hierarchical nesting
Estimation window	Rolling 156-day	Rolling 252-day with exponential decay
Factor hierarchy	Market → Sector → Theme	Market → Country → Industry → Style
Weighting	Cap-weight (style), equal-weight (thematic)	Cap-weight
Volatility targeting	10% annual, dynamic leverage	Factor-specific
Recomputation	Daily + intraday (1m, 5m)	Daily
Factor count	230+	Varies by model

Thematic basket methodology

OMNI’s 81 thematic baskets cover named investment themes across energy, technology, healthcare, financials, real estate, consumer, transport, ESG, and regional equity.

Two construction modes:

1. ETF-backed (66 baskets): An institutional ETF provider handles constituent selection and rebalancing. OMNI computes the spread versus a relevant benchmark and applies hierarchical purification. Examples: genomics (ARKG vs XLV), nuclear energy (URA vs SPY), cybersecurity (CIBR vs XLK).

2. Curated constituent (15 baskets): Equal-weight portfolio of named constituents versus a benchmark. Constituents reviewed quarterly. Examples: hotel REITs (18 stocks vs VNQ), trucking (16 stocks vs XLI), alternative asset managers (10 stocks vs XLF).

Both modes pass through the same hierarchical purification and volatility-targeting pipeline.

Academic references

• Ang, A., & Kristensen, D. (2012). Testing conditional factor models. Journal of Financial Economics, 106(1), 132–156.
• Asness, C. S., Moskowitz, T. J., & Pedersen, L. H. (2013). Value and momentum everywhere. Journal of Finance, 68(3), 929–985.
• DeMiguel, V., Garlappi, L., & Uppal, R. (2009). Optimal versus naive diversification. Review of Financial Studies, 22(5), 1915–1953.
• Fama, E. F., & French, K. R. (2018). Choosing factors. Journal of Financial Economics, 128(2), 234–252.
• Harvey, C. R., Liu, Y., & Zhu, H. (2016). …and the cross-section of expected returns. Review of Financial Studies, 29(1), 5–68.
• Kozak, S., Nagel, S., & Santosh, S. (2020). Shrinking the cross-section. Journal of Financial Economics, 135(2), 271–292.
• Menchero, J., Orr, D. J., & Wang, J. (2011). The Barra US equity model (USE4). MSCI Barra Research.
• Patton, A. J., & Verardo, M. (2012). Does beta move with news? Review of Financial Studies, 25(9), 2789–2839.