Score oil-complex tightness before divergence or regression

Start with the oil-linked currency complex

An oil-linked currency complex is a four-market set used to place one currency trade in context: the Canadian dollar, a large domestic producer, an oil-service basket, and crude.

A correlation matrix is a table of pairwise sample correlations that ranks which legs of the complex actually move together before a model is chosen.

What the sample matrix ranked first

In a 2005 to 2015 sample, the Canadian dollar's correlation with crude oil was 0.81, higher than its correlation with a large Canadian producer (0.54) or an oil-service basket (0.53).

In that same matrix the producer correlated more strongly with crude oil (0.62) than with the Canadian dollar (0.54).

Tight coupling leaves few divergence setups

The producer and the oil-service basket printed a 0.93 pairwise correlation and, on a weekly overlay with crude, tended to travel together, leaving fewer distinct divergence setups.

Tight coupling is a very high pairwise correlation that leaves few lasting separations for a divergence rule.

High-price decoupling can break the currency-oil map

The currency-oil mapping was described as able to break down at very high crude prices. In 2008, with oil near 140, the Canadian dollar peaked almost eight months earlier. In 2011, with oil near 110, the currency formed a double top before following oil lower.

High-price decoupling is a regime in which the usual Canadian-dollar and crude mapping weakens after oil has already printed an extreme.

A price comparison of the producer, crude, and the Canadian dollar showed the producer turning down before crude in 2008 and later diverging. In 2011 both crude and the producer diverged versus the currency.

Tight coupling and name-specific decoupling

When a producer and its sector basket are that tightly coupled, a later split cannot be confidently read as a temporary market inefficiency rather than a name-specific problem.

Name-specific decoupling is a split between a producer and its sector basket that may reflect a company event rather than a temporary market inefficiency.

When regression gives way to band-based divergence

A single-variable linear regression was the first design tried for a Canadian-dollar system and was then replaced with a band-based divergence method after the regression was judged inadequate.

Single-variable regression is a linear forecast that maps one intermarket driver onto a target over a defined lookback. Band-based divergence is a price-indicator condition that flags even small separations between related markets and is used when those markets are tightly correlated.

Band-based divergence was argued to suit tightly correlated markets because it can register small separations and therefore produce more signals, whereas multiple regression was said to add value only when the independent variables themselves are weakly correlated.

A regression of the producer on the oil-service basket did not raise r-squared when crude was added as a second independent variable, and substituting the Canadian dollar for crude improved the fit only slightly. Multiple regression is a linear forecast that adds a second driver only when that driver is not already highly collinear with the first.

Editorial reading of the workflow

Editorial reading: the sample matrix made crude the dominant Canadian-dollar driver, so a single-variable map from crude to the currency was the first quantitative design. The producer-basket pair was too tightly coupled for a second crude regressor to lift the fit, and substituting the Canadian dollar helped only slightly.

Editorial reading: band-based divergence is assigned to tightly correlated legs that can still print small separations, not to a later producer-basket split that cannot be separated from name-specific decoupling. High-price decoupling is a reminder that even the tightest currency-oil rank can weaken after crude has already printed an extreme.