What Causes A Global Crisis? - Oil? - Yeah, You Got It!

| About: iPath S&P (OIL)

Summary

In this piece I reexamine and extend previous findings to see whether oil price volatility causes a global crisis.

One-year lagged coefficients of variation of oil prices seem to do an outstanding job at determining the occurrence, magnitude and extent of global crises in the last 29-30 years.

All possible correlation coefficients calculated with increasing sample sizes were found to stabilize to a value of about -0.50, confirming a strong negative relationship between the two variables.

It then comes as no surprise that WTI oil price volatility appears to Granger cause global recessions.

Although the current global crisis may not be as severe and extensive as previous ones, we can’t say the same about the upcoming crash due to increased EV sales.

Much discussion is going on right now on the possibility of a new global crisis. Most analysts have placed their eyes on the U.S. economy. A common belief is that when the U.S. sneezes the rest of the world catches a cold. But to what extent are these fears justified? While it might be hard to draw any definitive conclusions at first glance, there appear to be some indications that something is in the works. First, the U.S. economy is experiencing difficulty to grow at rates comparable to pre-crisis averages. Second, U.S. manufacturing is shrinking, Americans are spending less, and goods are piling up on warehouse shelves. Third, U.S. corporate profits went down between September 2014 and September 2015. Lastly, U.S. stock prices have shown no growth in the last 12 months or so (See Figure 1). Behind all of these warning signs seems to be oil.

Figure 1

Click to enlarge

In a previous contribution I advocated using the coefficient of variation of oil prices (i.e. the yearly standard deviation divided by the yearly average oil prices multiplied by 100) to predict a global crisis. Unlike standard deviations, I argued, coefficients of variation give us an indication as to the magnitude or relative importance of each crisis. Next I assumed that low and/or negative U.S. GDP growth rates can be thought of as a proxy for global economic and financial crises. I then found that the highest coefficients of variation - which also appear to be above the total average for the whole period - always precede the lowest U.S. GDP growth rates - which also seem to be below the total average for the whole period - confirming the occurrence of the world's three most significant crises (Gulf War, Early 2000s Recession, and Great Recession) between 1986 or 1987 and 2014. In this article, I reexamine and extend my previous findings to see whether there is a causal negative relationship between coefficients of variation of oil prices and U.S. real GDP growth rates.

In closing the main argument in my previous piece I suggested two reasons why I thought a new global economic/financial crisis might be around the corner. First, a coefficient of variation not only above the total average for both periods of analysis under consideration but also higher than all such indicators besides those specifically related to the other global crisis already identified. And second, growth rates consistently below the total average for the whole period under scrutiny for a considerable number of years. As shown in Figures 2 and 3, the trend identified last year continues to hold for 2015, further giving support to my intuition that we are already in the middle of a global crisis.

Figure 2

Click to enlarge

Sources: EIA and U.S. Bureau of Economic Analysis.

Figure 3

Click to enlarge

Sources: EIA and U.S. Bureau of Economic Analysis.

In what follows, I elaborate on the idea that the highest coefficients of variation always precede the lowest U.S. GDP growth rates confirming the occurrence of the world's three most significant crises during the period under examination. This is reflected in Figures 4 and 5 where we can see a clear negative relationship between one-year lagged coefficients of variation and U.S. GDP growth rates, both for WTI and Brent prices. Note also that this allows us to ascertain more neatly both the occurrence and relative magnitude of two additional crises (the 1987 Stock Market Crash and the current crisis) with WTI oil prices and one other crisis (the current one) using Brent oil prices. In addition, using this methodology, we are also able to determine the relative extent of the different global crises as depicted by the area formed by the longest distances (i.e. largest differences) between one-year lagged coefficients of variation of (NYSE:WTI) oil prices and U.S. growth rates (considering their most pronounced peaks and troughs) from the beginning to the end of each global crisis. Interestingly enough, there appears to be a perfect correspondence between the highest peaks of the first variable and the lowest troughs of the second one, and much coincidence between beginning and end of each crisis (usually described by a deceleration process followed by very low or negative U.S. growth rates and their recovery) and rise and fall of oil price volatility, both in the Gulf War and the Great Recession. However, in the Early 2000s Recession, these relationships don't seem to be as clear-cut. In fact, in this case we see that the U.S. growth rate starts decelerating when oil volatility peaks, bottoms up when the coefficient of variation collapses and completes its recovery as oil uncertainty rises and falls again. This boils down to a rather extensive global crisis. Based on these results, we can safely argue that: i) Except for the Early 2000s Recession, 1-year lagged coefficients of variation do an outstanding job at establishing the occurrence of (i.e. predicting) global crises; ii) the most severe crisis was the Great Recession, followed by the Gulf War, the Early 2000s Recession, the Stock Market Crash and the current crisis; and iii) the most extensive crisis appears to have been the Great Recession followed by the Early 2000s Recession and the Gulf War.

Figure 4

Click to enlarge

Sources: EIA and U.S. Bureau of Economic Analysis.

Figure 5

Click to enlarge

Sources: EIA and U.S. Bureau of Economic Analysis.

In addition, to give this analysis a more statistical flavor, I calculated all possible correlation coefficients between the two variables (one-year lagged coefficients of variation and U.S. real GDP growth rates) with increasing sample sizes during 1987-2015 and put all of them on a graph. This is shown in Table 1 and Figure 6. As it can be observed, there appears to be a robust pattern in the data so that as the sample size increases, the correlations coefficients go down, stabilizing to a value of about - 0.50, reflecting a noticeable downward trend which confirms a relatively stable/strong negative relationship between the two variables.

Table 1

Click to enlarge

Note: Rows 1 and 2 are empty because at least 3 values are necessary to compute a correlation coefficient.

Figure 6

Click to enlarge

We are now in a position to check for unidirectional negative causality running from coefficients of variation of oil prices to U.S. GDP growth rates (as a measure of a global crisis). This requires: testing for stationarity of the two variables to find their order of integration; (ii) check for cointegration if the variables appear to be integrated of order one; and test for unidirectional negative Granger causality.

As shown in Table 2, the two variables were found to be stationary (or integrated of order 0), which implies that no differencing is necessary and that they don't share a common trend and therefore can't be cointegrated. In this context, to test for Granger causality, we have three possibilities: One, use the two variables as they are in the Granger causality regressions. Two, include a time trend variable in the Granger causality regressions. And three, de-trend the variables, by regressing each one of them (if appropriate) on a time trend and using their corresponding residuals in the Granger causality regressions. Lastly, on each Granger causality regression we need to perform corresponding Wald tests to reject or accept the null hypothesis of non-causality.

Table 2

Results of Unit Root Tests

Click to enlarge

For simplicity, in what follows we describe in detail the first possibility only with comments on the other two.

Causality Equations:

Click to enlarge

Where USRGDP t is U.S. real GDP growth rate, WTI_COV t is the Coefficient of Variation of oil prices; α, β, γ, δ are the coefficient parameters; p is the optimal lag length selected based on the Schwarz Information Criterion (SIC); μ t and ω t are the white noise error terms.

In Table 3 the (one-year lag) best-fit regression results for Equation 1 are presented. They were obtained based on the sample size of our variables, using the SIC after trying a maximum of 7 lags. As we can see, except for the coefficient parameter of USRGDP, all coefficients are highly significant. Note, however, that the t-statistic for this variable is > 1, which fully justifies its inclusion in the estimating equation.

Table 3

Click to enlarge

Next, we perform a Wald test to see if we can reject the null hypothesis of non-causality that βi is equal to zero against the alternative hypothesis that the opposite is the case. In Table 4, we show the results of the Wald test where it's clear that the null hypothesis of non-causality is rejected at relatively low p-values (0.0368 and 0.0277 for our F and Chi-square statistics, respectively). The relative strength of our one-year lagged coefficient of variation of oil prices in Granger causing a global crisis is therefore confirmed.

Table 4

In Tables 5 and 6 the best-fit regression and Wald test results for Equation 2 are presented. We attained the (one-year lag) best-fit, based on the sample size of our variables and using the SIC after trying a maximum of 7 lags. As we can see, except for the intercept, all coefficients appear to be insignificant, and the Wald test can't reject the null hypothesis of non-causality from U.S. GDP growth rates to coefficients of variation of oil prices at any reasonable level of significance. This confirms the unidirectional character of the causal relationship previously identified.

Table 5

Click to enlarge

Table 6

In terms of the other two possibilities, first, it was found that when a trend time variable is incorporated, the goodness of fit of the causal relationship (as measured by the SIC) diminished somewhat, though leaving our previous results about causality and non-causality intact, and, second, de-trending the only variable on which a clear association between it and time was encountered (USRGDP), actually contributes to improving (albeit only marginally) the overall goodness of fit. Needless to say, in both cases the other results about reverse non-causality are also confirmed.

Thus we are ready to answer the question of our title in the affirmative. Oil prices volatility (as measured by their coefficients of variation) do seem to Granger cause global recessions (as proxied by low and/or negative U.S. real GDP growth rates). This conclusion must be taken with care though as Granger causality is not the only way to approach the essential notion of causality. But that's subject of another research well beyond the present one.

In closing, it all seems to indicate that the current crisis may not be as severe and extensive as the previous ones. But we are incapable of saying the same about the upcoming crash to be prompted in about a decade by increased EV sales in the advent of a new techno-economic paradigm with lithium as its key factor in the world.

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.