Is the oil price being driven by a weak dollar?

Crude oil prices, against all rationality, are skyrocketing and the phenomenon is being widely attributed to the decline in the oil currency, namely the US dollar. Can such a correlation be explained so easily when it might not even exist in the first place? Futures examines the case of "strong oil versus weak dollar".

Experts have been falling over themselves for weeks trying to explain the dramatic rise in crude oil prices. But one argument crops up repeatedly in various guises: namely that the falling dollar rate is driving up oil prices - this hypothesis is based on the fact that the oil prices are quoted in US dollars. The following release by a news agency on October 18 received cross-country recognition and was quoted in numerous media:

„When the dollar rate falls the purchasing power of oil producing countries in other currencies also falls. They try to close the gap by increasing the prices. ‘When the dollar falls, commodities prices rise, that‘s nothing new,’ explains oil market expert Klaus Matthies of the Hamburg Institute of International Economics. ‘Producers want to avoid any further losses,’ he says.

Another effect of the weak dollar is, say experts, the rising oil price: for speculators outside of the dollar area it is becoming more favorable to invest in oil securities. But cheaper oil futures make for a more interesting investment. This growing demand also drives prices up.

Given the decline in the dollar, many investors are also shifting their assets from the dollar to commodities such as gold, as well as oil, in order to stop the fall in value. This is also driving up the price of oil.“

These seem like logical arguments. But if you were to look more closely, you‘d be entitled to raise a few objections. The idea that oil producers could fix the price of the oil they are selling is far from reality. In the twenty-first century, the prices are determined by the market. One fact of which the quoted commodities expert, as one of the senior economists of the Hamburg Institute of International Economics, is of course also aware. In response to Futures, certified economist Klaus Matthies says: „My statement referred to commodities in general, and I used coal and non-ferrous metals as examples. Crude oil is different in that the selling prices of the oil producers (oil producing countries) are now aligned with stock exchange prices. Which means that producers no longer have a direct influence over the price of oil.“

It is also hard to corroborate the suspicion that „most investors“ would prefer to „shift“ their assets towards gold and crude oil because of the falling dollar rate. Using gold to hedge against the falling dollar may generally work (even if there are better ways). But oil is a different story altogether: an oil buyer would increase rather than decrease his dollar risk because the value of his investment also decreases as the dollar declines. A hedge would at best be possible with a short oil position.

But perhaps there is a grain of truth in there somewhere. It is feasible that large-scale consumers or speculators - e.g. in the euro area - might stock up on relatively cheap oil during weak dollar periods, ultimately driving prices through the roof. Alternatively, producers could still influence the price using the much quoted „oil tap“ (rising prices as a result of curtailed production).

Trends on both futures markets between December 2001 and October 2007: the chart shows crude oil to be on a long-term upward trend and the dollar to be on an almost continual downward spiral since the spring of 2002. However, this isn‘t enough to confirm any correlation between the two market trends.

Oil versus dollar: data check

We wanted to know precisely and to attempt to check, based on statistical methods, whether the hypothesis surrounding the influence of the dollar on the oil price has any truth in it. More precisely, whether the dollar rate could be used to forecast future oil prices. To do this, we are using those prices which accurately reflect trends in oil prices and the dollar rate:

1.) As a benchmark for oil price trends we will take the world‘s most liquid crude oil futures contract (the crude light future on the New York NYMEX).
2.) As a benchmark for trends in the dollar rate we have opted to use the FINEX US dollar index future, which reflects the value of the dollar using an average of the six strongest reference currencies (including the euro, which replaced the D-Mark from 2001, the yen, the British pound and the Swiss franc). If the price of this contract increases, then the value of the dollar on the currency market increases and vice versa.

Using this data, the following first has to be measured in case there is a linear and stable correlation (oil up, dollar down and vice versa): the correlation (in the form of a correlation coefficient) between the change in value of the crude light and the dollar index should be highly negative and stand up to a regression analysis.

Test 1: Monthly comparison

Let‘s begin with the longest available period - the change in value of the crude oil future and the dollar index between January 1986 and October 2007. First we perform a regression analysis based on 262 monthly changes. The correlation coefficient of -0.01 already tells us that there is no direct correlation between oil and the dollar - at least in a direct comparison of the individual months. There may be a slight overhang of months with opposing price trends and those periods during which oil and the dollar headed in the same direction. Specifically, the ratio was 113:158. But a glance at the scatter plot on page 3 shows that despite the slightly negative correlation, we are dealing with a random distribution.

Scatter diagram based on linear regression: the monthly changes in the crude light future (X-axis) and the dollar index future (Y-axis) in the 161 months examined. The data points in the red quadrants indicate months in which trends for the two values are opposed, points in the green quadrants indicate months with identically oriented value changes. Overall, we get a picture of random distribution.

Test 2: Delay

We could let things lie and say that further examination from a practitioner‘s perspective isn‘t really worth it because there is no rule with a sufficiently high strike rate which can be derived from the value pair dollar/oil. But it‘s not as simple as all that: after all, the virtually zero correlation of the monthly changes in no way means that there is no reciprocation at all between the two markets. There could, for example, be a delayed reaction. The oil price would then react later to changes in the dollar rate. Or there would only be a reaction after the dollar change reached a certain point.

There is a second approach which allows us to follow the hypothesis through from this angle: we assume that a change in the dollar rate affects the oil price not directly but rather in the medium to long term. The simplest way of doing this is to take a longer period - for example the annual changes of both markets. Our data thus provides us with 21 comparison values (31.12.1985 to 31.12.2006). In 11 out of 21 cases, the annual changes in the dollar rate and oil prices are opposed. Thorough analyses do not need to be carried out here. A quick glance at a traditional bar chart indicating the annual changes of both markets reveals the obvious: you might as well flip a coin judging by the number of times the hypothesis can be confirmed. The correlation must be virtually zero. In fact the coefficient has a value of 0.05 and this method has obviously brought us to a dead end.

So let‘s try a different approach. We need to create a monthly time series where we continually calculate the value changes of the previous six months (a „rolling window“). This gives us much more data to go on and allows us to rule out any inaccurate conclusions based on chance (which could have been the case with the year-end data).
An evaluation of this data reveals, however small, a negative value for the correlation: -0.17 is far from „significant“, but it does suggest that there could be the correlation we suspected in one form or another.

Regression analysis

In order to get somewhere using statistical means, we need to dig a little deeper into our box of tricks and pull out something called a cointegration analysis. In practice, this procedure is applied to error correction models and the economic statisticians Robert Engle and Clive Granger who developed it towards the end of the 1980s were awarded the Nobel prize for it in 2003. Cointegration analysis attempts to describe long-term correlations between several - often trend-driven - variables (e.g. the time series of various stock exchange prices). The advantage of this method over regression is that there does not have to be a precise correlation at all times in order to explain it.

Applied to our dollar and oil problem this means that via regression analysis we have proved that there is at best a very weak correlation between the value changes of the dollar rate and the oil price in the corresponding periods (monthly, six-monthly window or annually). But because we are still not sure whether there might yet be a correlation which may occur later or occur in the wake of some other trend, we use a second diagnostic procedure, namely cointegration analysis. Just as a cautious doctor may also employ expensive Computer Tomography after a conventional X-ray if he is still unsure of his diagnosis.
The more complex process enables us to examine whether we are perhaps dealing with the complex form of the following, simple example in our case of dollar and oil: imagine two phase-shifted sine curves (see figure). Interestingly, the high, low and inflection points which describe these curves - as the scatter diagram also shows - deliver a correlation of absolutely zero. It is nevertheless obvious that both curves are in absolute harmony and the identification of a subsequent point on one curve allows the relevant data point on the other curve to be precisely calculated. Statistical series like these are called „cointegrated“ and they are identified via cointegration analysis.

It‘s a method which popular spreadsheet programs don‘t generally offer and the mathematic model is somewhat more complex than that of conventional regression analysis. Fortunately the test as to whether cointegration exists also delivers an optical output which can be easily understood by non-statisticians. In the ideal case of extreme cointegration between the time series under examination (as in our example of the two sine curves, which although lacking in practical relevance is at least graphic), it would again resemble a stable periodic variation in value about a constant mean - in this case the neutral axis (see figure).

Figure: Output of test for cointegration of the two sine curves.

Statistical series such as these are referred to as „stationary“. Of course, the reality of stock exchanges is that „ideal results“ simply don‘t exist, but the closer the result of a test corresponds to a stationary statistical series such as this, then the clearer the indication that cointegration exists between the time series under examination.
We have applied this method to the dollar and oil statistical series (monthly changes) and the cointegration test graph reveals what we suspected based on previous diagnoses: the chart shows no fluctuation around a mean value for the entire period, but - by complete contrast - a typical trend as you would expect to see in stock charts. Although there are periods where the results are more or less stationary (e.g. between 1985 and 1996), overall there is no cointegration between the oil price and dollar rate.

Output of test for cointegration between oil and the dollar rate: the graph shows non-stationary data. The crude oil price and dollar rate are apparently not cointegrated.

Two independent trends

Even under torture, it‘s sometimes impossible to force data into submission. Ultimately, the fact remains that the price of oil has clearly been increasing since the winter of 1998 and that the dollar has, in the meantime, fallen. But the dollar rate also increased up to February 2002, and only after that did it move in the opposite direction. So a direct correlation between the two trends is highly unlikely - and there is certainly no correlation which could be used to make any kind of reliable forecast.