The risk model

Risk is estimated at four levels.

Sector, sub-sector, styles and trading factors, and idiosyncratic risk at each contract. Two models are jointly estimated so that the explanatory power of the style factors stays consistent across levels of aggregation.
The ARC nested commodity factor modelThree sectors resolve into eleven sub-sectors and are surrounded by seven cross-sectional style and trading factors.BASISMOMENTUMSHORT MOM.OPEN INTERESTACTIVITYVOLATILITYSHORT VOL.PRECIOUSBASECRUDENAT GASREFINEDBIOFUELSCOALGRAINSSOFTSPROTEINSLUMBERMETALSENERGYAGSARC

Three sectors / eleven sub-sectors / seven styles and trading factors

Structure

How the model is built.

Returns are expressed through a small set of common drivers plus a return specific to each contract.

Loadings are observed. Factor returns are estimated cross-sectionally, across all contracts, for each period.

The loadings are sector and sub-sector membership together with styles and trading factors, with sub-sectors nested inside sectors so that the two views agree.

The output is a set of exposures, a factor covariance matrix and factor returns, from which risk, value at risk, attribution and stress tests follow.

Why a factor model

1,200+Futures
11Sub-sectors
4Levels
26yHistory

Style factors

Seven, cutting across the hierarchy.

Styles and trading factors cut across the hierarchy rather than sitting inside any one sector, and are estimated jointly with it.

FactorDefinition
BasisLog difference between the price of a contract at a maturity beyond the front month and the front-month contract.
MomentumReturn over the full year excluding the last 30 days.
Short-term momentumReturn over the last 30 days.
Open interestOpen interest for each contract.
Trading activityOne-day change in open interest for each contract.
VolatilityOne-year historical daily standard deviation of contract returns.
Short-term volatilityDaily historical volatility over the last 30 days.

Every descriptor is standardized and winsorized. Exposures are reported as z-scores, where zero is the average across the model universe.

Output

Exposure fingerprint.

An illustrative, non-client-specific book of 26 equally weighted futures, 13 long and 13 short, scored against the model universe.

Exposure

Solid bars are net long exposure; hollow bars are net short.

Decomposition

Where the risk sits.

Sector exposure nets to almost nothing, so the book reads as hedged at the sector level. The risk is in the styles: basis and momentum together account for most of the systematic contribution.

Annual ex-ante volatilityContribution
Total risk7.4%
Agriculture0.2%
Energy0.0%
Metals0.2%
Basis2.5%
Open interest−0.2%
Momentum1.5%
Short-term momentum0.0%
Trading activity0.0%
Volatility0.3%
Short-term volatility0.7%
Specific risk5.4%

Figures illustrate model outputs and are neither realized performance nor a forecast.