Yet more on CAD/USD and WTI

This post continues on the discussion of the degree to which CAD/USD is influenced by WTI, and is primarily meant to add some colour to the discussion of a post by Mike Moffat over at WCI.

In earlier posts (here and here) I had argued that the price of CAD/USD was not directly influenced by the price of oil, but rather by US dollar strength (or weakness). My claim was supported by the fact that CAD/EUR, which should gauge Canadian dollar strength independent of the US dollar, was not particularly correlated to the price of oil.

Having thought about this a bit more, I want to expand on this discussion. Specifically, while it is true that CAD/USD is primarily influenced by the strength of the US dollar (as is the price of oil), one can discern periods where the price of oil is higher than one would expect solely from US dollar strength. During these periods, the Canadian dollar seems to outperform the Euro (that is CAD/EUR rises). This suggests that when the price of oil is higher than expected based on USD, it correlates with Canadian dollar strength. In other words, CAD/USD and the price of oil are correlated independent of the US dollar.

Below is a chart with WTI in the top panel on a log scale (OHLC bars). Superimposed in the panel is 1/US Dollar Index on a linear scale (which plots US dollar weakness). EUR/USD and CAD/USD are plotted in the second panel in blue and red respectively. CAD/EUR appears in the third panel.

My first comment is that log(WTI) seems to correlate much better with 1/USD than WTI, at least over the 2003-2010 period.

Second, I've highlighted three periods in tan where US dollar weakness (i.e., 1/USD) outperforms log(WTI) strength. During these periods, EUR/USD outperforms CAD/USD and CAD/EUR falls. I've also highlighted three periods in yellow where log(WTI) outperforms US dollar weakness. During these periods, CAD/USD outperforms EUR/USD and CAD/EUR rises.

Thus when WTI is stronger than what one would predict from USD (that is when log(WTI) outpeforms 1/USD), the Canadian dollar outperforms the Euro. Conversely when WTI is weaker than what one would predict from USD (1/USD outperforms log(WTI)), the Canadian dollar underperforms the Euro.

From this data I can't help but conclude that the price of oil has an independent correlation to the Canadian dollar (indpendent of the US dollar that is). But evidence of the correlation only appears when the price of WTI diverges from what one would predict solely on the basis of the US dollar.

I owe Stephen Gordon an apology for nitpicking to such a degree about his earlier post.

More on CAD/USD & WTI

Mike Moffat on WCI posted a "toy" model relating CAD/USD to the price of oil and interest rate differential between the U.S. and Canada. I've reproduced his model below:

Using the daily data from the years 2001-2006 a linear regression was run to calculate values for the toy:

MERT value = 48.62 + (0.6211*OIL) + (1.8306*INTGAP)

where:

MERT = The predicted value of the Canadian dollar (CAD/USD)

OIL = WTI Cushing Spot price in dollars

INTGAP = Target for the Overnight Rate (Canada) - Target for the Federal Funds Rate (U.S.). A 1.25% target for the overnight rate enters the equation as 1.25.

Really quite a lovely "toy" which does a pretty good job in estimating the CAD/USD exchange rate (see the above link for some graphs of its predictive capabilities). I especially like the use of WTI as a predictive term. Now WTI is strongly correlated with the inverse of the US dollar. I believe the power of the "toy" in predicting CAD/USD stems from this strong correlation.

So I was wondering about my earlier use of CAD/EUR to gauge Canadian dollar strength independent of the US dollar and thought to examine that technique in light of this new "toy". Specifically, how does CAD/EUR compare to both WTI and the inverse of the US dollar (or the US Dollar Index).

WTI appears in the top panel as OHLC bars along with the inverse of the US Dollar Index as a blue line. CAD/EUR appears in the second panel as a green line along with CAD/USD as a red line. My apologies for the busy chart.

Note: WTI is plotted on a log scale, while the inverse of USD is on a linear scale!

via Stockcharts.com
It's immediately apparent that log WTI and the inverse of the US Dollar Index are strongly correlated (I looked at straight WTI versus inverse US Dollar Index, and the log appears to be a much better fit).

I've highlighted several periods where the two signal diverge. The yellow periods indicates instances where the US dollar weakens faster than WTI strengthens (that is inverse US dollar index rises above WTI). In these periods, CAD/EUR drops as the Euro outperforms the Canadian dollar. The tan periods are instances where Inverse US Dollar Index underperforms WTI (i.e., the US dollar strengthens while WTI remains strong). In these periods, CAD/EUR rises.

What I find very interesting is that when Inverse US Dollar Index diverges from WTI, CAD/EUR follows the US dollar (the dollar index itself, not the inverse). When the US dollar weakens faster than WTI falls, CAD/EUR falls (yellow periods), and when US dollar rises faster than WTI falls, CAD/EUR rises (tan periods).

If you examine CAD/USD (red line), whatever influence CAD/EUR has on the Canadian dollar appears to be dwarfed by whatever the US dollar is doing. Post 2003, CAD/USD and CAD/EUR appear correlated, but previous to this date, changes in CAD/EUR appear to have only a small correlation to CAD/USD.

Finally, it's important to note that this approach ignores other macroeconomic variables which could influence exchange rates. In particular, Mike Moffat's "toy" model includes the interest rate gap between Canada and the US.

Update: After further thought, I'm not sure that the above figure speaks one way or another on WTI's independent influence on CAD/USD. Now it is true that CAD/EUR tends to follow USD more than WTI (yellow and tan regions). But, this is probably happening because USD is moving more than WTI in each of those bands.

I think there are better approaches, but I'll post more when I get a chance.