F for Fail

The Joint Economic Committee (JEC) Republican analysis of the impact of monetary policy on gasoline prices

I’ve been grading papers for the past half week, so when this popped into my mailbox yesterday morning, I was in a “grading” mood. And when I finished reading it, I determined I would give it an F. From “The Price of Oil and the Value of the Dollar: Declining Value of the U.S. Dollar Adds to the Price of Oil and Gasoline,” Republican Staff Commentary (May 16, 2011):

Arguably, there are other factors affecting the price of gasoline than just the price of oil. However, the retail price of gasoline in the United States moves in tandem with the price of oil. In fact, the correlation between the two is greater than 98%. Given that oil is the primary input to gasoline and the close correlation we can perform a similar analysis to determine how much of the current price of gasoline is attributable to the declining value of the dollar.
The final chart shows what the price of gasoline would be if the value of the dollar had not declined. In other words, the dollar’s decline accounts for 56.5 cents of the \$3.963 current price of gasoline. [emphasis added in bold – mdc]

Clearly, the objective of the study is to argue tha t QEI and QEII raised oil prices, I think by arguing that they caused inflation. Well, this may very well be the case, but here I’m just going to critique the analytics of the memo.

The JEC-Republican Staff Commentary Analysis

The core of the analysis is summarized by this graph from the Commentary:

Figure from “The Price of Oil and the Value of the Dollar: Declining Value of the U.S. Dollar Adds to the Price of Oil and Gasoline,” Republican Staff Commentary (May 16, 2011)

What the authors have (apparently) done is to plot the price of Brent Crude, and then plot Brent assuming the Fed’s trade weighted dollar index (broad) had stayed constant at 2008M11, levels. The gap is \$17.04, which they then convert to \$0.565 gap per gallon of gasoline (by the way, the implicit conversion factor is slightly different from the 0.25 cents per gallon for a \$10/barrel.

Math and Regression Estimates

This is an interesting procedure. It is so interesting, I don’t know why one would ever do it. I can certainly replicate it; since the dollar broad index is roughly 14.8% weaker in 2011M04 than 2008M11, then one can divide the actual Brent price by 1.148 to obtain a \$15.85 dollar figure for 2011M04, close enough to the \$17.04 cited by JEC-Republicans.

The analysis presupposes the relationship between oil and dollar were at equilibrium in 2008M11. More importantly, it presupposes a unit relationship between the two. The graph seems to suggest the existence of a unit coefficient, but regression analysis does not uncover such a relationship. Over the sample period shown in the JEC-Republican graph (2008M01-2011M04), ∂Poil/∂DOLLAR = 1.32 if the constant is suppressed (R2 = 0.27), and negative 7.92 if the constant is allowed (adj-R2 = 0.87).

Interestingly, the coefficients I obtain for the 2008M11-2011M04 period are remarkably dissimilar to those for the period conforming to monetary policy easing, starting in 2007M07 (when the Fed started dropping the Fed funds rate). Then, the regression coefficient (no constant) is 0.16. If one thought monetary easing was the culprit, this would be the right time to start the analysis, not with QE I. Re-doing their calculation, based on 2007M07, one finds the implied difference is only \$8.85.

Figure 1: Current dollar price of Brent Crude (blue), and Brent indexed to 2007M07 value of inverted Fed trade weighted value of dollar index (red). Source: IMF, International Financial Statistics, St. Louis Fed FREDII, and author’s calculations.

Of course, one has to wonder if one can trust the correlation coefficients as representative of the underlying population parameters (technically, do the point estimates converge to the population moments?). I find that the series are integrated of order one, but the two series are only possibly cointegrated using asymptotic critical values, and a 5% significance level, over the sample period investigated by the JEC-Republicans. That relationship disappears if the sample is extended to 2007M07-11M04. (In log terms, Brent appears to be stationary, but the nominal dollar appears stationary, over the 2008M11-2011M04 period; and both appear to be integrated of order one for the longer period).

When I teach econometrics, one of the things my students get sick of hearing is “correlation is not causation”. Nonetheless, I think it’s a warning that should be heeded in all sorts of instances — including this one. From the “commentary”:

Analysts and pundits often cite, correctly or incorrectly, the turmoil in the Middle East, a strengthening global economy, or speculation as the causes for the run up in crude oil prices. What is rarely discussed as an important factor in the rise of the dollar price of oil is the role played by the dollar itself. Oil is an international commodity that trades in dollars. The value of the unit of exchange, in this case the dollar, plays an important role in determining the “headline” price for the underlying commodity.

The authors of the commentary then rush headlong into modeling the oil price as a function of the nominal exchange rate, holding all else constant. As I noted in this post, such an approach is untenable.

while there is a negative relationship between the dollar’s value and the price of oil (in logs), that relationship is not statistically significant after accounting for serial correlation; nor is it significant in first differences.
Second, the idea that it’s just a numeraire issue — weak dollar implies more dollars per barrel of oil — does not seem to be consistent with a negative correlation between the real price of oil and the real value of the dollar, plotted in Figure 3.
In point of fact, one should expect two-way causality. A higher relative price of oil should weaken a country’s real exchange rate if it worsens the country’s terms of trade (i.e., the country is a net importer of oil). In addition, if the change in the relative price induces obsolescence of some of the capital stock, this would induce an economic contraction that might depreciate or appreciate the currency, depending on variety of assumptions (home bias in consumption, capital/labor ratios in the nontradable versus tradable sector, complementarity of capital and labor with energy, etc.). In Chinn and Johnston (1996) [pdf], a 10 percentage point rise in the real price of oil induces a 2 percentage real depreciation in a typical OECD country real exchange rate. That estimate relies upon exogeneity of real oil prices (an assumption not invalidated by the data).
Close readers will see that my discussion of how to apportion how much of the dollar decline is causing — versus being caused by — oil price increases is related to the issue of how to identify oil price shocks. There are numerous ways of accomplishing this goal. For one instance, see the IMF’s April 2006 World Economic Outlook Chapter 2, which uses a particular VAR to identify the shocks (thanks to Alessandro Rebucci for reminding me about this study). In that case, there is essentially zero effect of the oil shock on the real value of the U.S. dollar.

Exploiting Other Correlations

Consider another correlation – between rest-of-world GDP and Brent. Figure 2 illustrates the strength of the relationship over the 2001Q1-2010Q4 period.

Figure 2: Log current dollar price of Brent Crude (blue), and log rest-of-world real GDP (purple). Source: IMF, International Financial Statistics, Federal Reserve Board, and author’s calculations.

Running a regression of log Brent on log RoW GDP (and current and lagged first differences) over the 2001Q1-08Q3 period yields a specification that explains a large proportion of variation in Brent. The Adj.-R2 = 0.96 (!!). (It’s 0.91 over the entire 2001Q1-2010Q4 period). In fact, this relationship rejects the no-cointegration null hypothesis at all conventional levels, unlike the exchange rate-oil price relation exploited by JEC-Republicans.

One can then ask how much lower oil prices would’ve been if RoW GDP had held constant at 2008Q4 levels: \$10.78. A conclusion consistent with this view is that we could have had lower oil prices if only the RoW had stopped growing.

I don’t literally believe this result (nor the implied conclusion). What this exercise shows is that there are many plausible drivers of oil prices. Trying to tease out the impact of monetary policy on oil prices is a laudable goal, but trying to do it by assuming exchange rates have zero covariance with the fundamental determinants of real oil prices is a mistake.

What Would Be Better

Despite the previous assessment, I wouldn’t say the JEC-Republicans necessarily wrong in their estimate. Merely that if they are right, it would be by accident. In fact, one could come up with numbers that are bigger.

It’s incumbent upon the critic to propose alternatives. Before doing that, it might be useful to think of what could drive up oil prices denominated in US dollars:

• US GDP
• Rest-of-world GDP
• Energy intensity of economic activity
• Oil production capacity
• Cost of production of marginal producer
• Nominal interest rates
• Inflation rates
• Expected price of oil (itself a function of trends in the above factors)
• Speculative activities of non-fundamentalist traders
• Numeraire issues

The JEC-Republican study essentially focuses on the last point, assumes changes in the dollar’s value against other currencies has no other impacts on the underlying price of oil. For instance, this approach is consistent with dollar depreciation that induces no re-allocation of aggregate demand across borders, or alternatively, the oil intensity of the US and the rest-of-the-world is the same.

The above list of potential determinants suggests that one needs more than a bivariate approach, and a multivariate (multiple equation) approach is required: either a structural multi-equation model, a VAR or SVAR. From Alquist, Kilian and Vigfusson, “Forecasting the price of oil,” forthcoming Handbook of Economic Forecasting, edited by Graham Elliott and Allan Timmermann:

There are several reasons to expect the dollar-denominated nominal price of oil to
respond to changes in nominal U.S. macroeconomic aggregates. One channel of transmission is
purely monetary and operates through U.S. inflation. For example, Gillman and Nakov (2009)
stress that changes in the nominal price of oil must occur in equilibrium just to offset persistent
shifts in U.S. inflation, given that the price of oil is denominated in dollars. Indeed, the Granger
causality tests in Table 1a indicate highly significant lagged feedback from U.S. headline CPI
inflation to the percent change in the nominal WTI price of oil for the full sample, consistent
with the findings in Gillman and Nakov (2009). The evidence for the other oil price series is
somewhat weaker with the exception of the refiners’ acquisition cost for imported crude oil, but
that result may simply reflect a loss of power when the sample size is shortened.
Gillman and Nakov view changes in inflation in the post-1973 period as rooted in
persistent changes in the growth rate of money. Thus, an alternative approach of testing the
hypothesis of Gillman and Nakov (2009) is to focus on Granger causality from monetary
aggregates to the nominal price of oil. Given the general instability in the link from changes in
monetary aggregates to inflation, one would not necessarily expect changes in monetary
aggregates to have much predictive power for the price of oil, except perhaps in the 1970s (see
Barsky and Kilian 2002).

…there is considerable lagged feedback from narrow measures of money such as M1 for the refiners’ acquisition cost and the WTI price
of oil based on the 1975.2-2009.12 evaluation period. The much weaker evidence for the full
WTI series may reflect the stronger effect of regulatory policies on the WTI price during the
early 1970s. The evidence for broader monetary aggregates such as M2 having predictive power
for the nominal price of oil is much weaker, with only one test statistically significant.
A third approach to testing for a role for U.S. monetary conditions relies on the fact that
rising dollar-denominated non-oil commodity prices are thought to presage rising U.S. inflation.
To the extent that oil price adjustments are more sluggish than adjustments in other industrial
commodity prices, one would expect changes in nominal Commodity Research Bureau (CRB)
spot prices to Granger cause changes in the nominal price of oil. Indeed, Table 1a indicates
highly statistically significant lagged feedback from CRB sub-indices for industrial raw materials
and for metals.
In contrast, neither short-term interest rates nor trade-weighted exchange rates have
significant predictive power for the nominal price of oil. According to the Hotelling model, one
would expect the nominal price of oil to grow at the nominal rate of interest, providing yet
another link from U.S. macroeconomic aggregates to the nominal price of oil. Table 1a,
however, shows no evidence of statistically significant feedback from the 3-month T-Bill rate to
the price of oil. This finding is not surprising as the price of oil clearly was not growing at the
rate of interest even approximately (see Figure 1). Nor is there evidence of significant feedback
from lagged changes in the trade-weighted nominal U.S. exchange rate. This does not mean that
all bilateral exchange rates lack predictive power. In related work, Chen, Rossi and Rogoff
(2010) show that the floating exchange rates of small commodity exporters (including Australia,
Canada, New Zealand, South Africa and Chile) with respect to the dollar have remarkably robust
forecasting power for global prices of their commodity exports. The explanation is that these
exchange rates are forward looking and embody information about future movements in
commodity export markets that cannot easily be captured by other means.

Update, 1pm Pacific: If you thought I shouldn’t waste my time dissecting this “Commentary”, see how many places this study has been cited as authoritative: [SF Examiner] [Houston Chronicle] [Weekly Standard] [a] .

22 thoughts on “F for Fail”

1. Steven Kopits

If oil were priced in, say, Euros, then the dollar price of oil would vary directly with the dollar/euro exchange rate (less any short term marginal impact on US demand). The exchange rate, in the short term (days to weeks), would be a material, and likely decisive, factor in oil prices in dollar terms, and an important factor in the long-term, I would think.
Now, if oil is priced in dollars, then the exchange rate doesn’t matter if the price of oil is sticky in dollars, eg, the Chinese or Europeans won’t buy more even if it becomes cheaper in home country terms. Do we believe that?
I am inclined to believe that exchange rates do indeed affect oil prices, certainly in the short term. But that’s a different question from whether QEI or QEII was good policy. QEII cannot be judged solely on its impact on oil prices.

2. westslope

MC: Great stuff! 10^3 thanks. Luckily, 99.8% of Republican Party members cannot read or understand your post otherwise I am sure that you would be receiving those 3:00 a.m. telephone calls that endear so many to American democracy.

I have not read Chen, Rossi and Rogoff (2010) but wouldn’t exchange rates forecast commodity prices because net capital flows are driving up the exchange rates of small net resource exporters based on forecast commodity prices?

3. dave

If the last few years don’t show a rough correlation between the dollar and oil prices, I don’t know what does. The causal case for the correlation is common sense.
But hey, I’m just a guy getting priced out of the overseas vacation I want because a banker with his QE2 bonus money is outbidding me.

4. Menzie Chinn

dave: QE I and QE II could’ve induced higher oil prices — I just don’t think the JEC-Republican approach can identify the impact. QE 1 and QE 2 could’ve induced faster RoW growth, which in turn could’ve increased demand for oil, for instance. Many, many other possible linkages exist. But appeal to common sense — well, that justified the use of leeches in the past.

westslope: Chen et al. (2010) don’t isolate the channels by which this effect operates (although they conjecture); merely that it is exploitable for statistical purposes.

Steven Kopits: Yes, two-way effects are plausible, and some one-way effects might dominate at high frequencies.

5. The Rage

Dave, your mistake is looking at nominal dollar settings. Go beyond that. Trades are trades. Right now, the “pro-growth” Euro-China trade is on. That trade dies, so does the “lower” dollar and oil prices crash while the global economy contracts.

6. ppcm

As seen trough J Hamilton Econbrowser post (Brent-WTI spread),WTI and Brent Ns follow the same pattern. A Short term range pattern graph is provided through the same post .
When using transitivity since both data (usd,Wti) are available on a longer time frame with Fred, one may notice that on a long term pattern the correlation usd,WTI price does not hold.
U.S. / Euro Foreign Exchange Rate (EXUSEU)
http://research.stlouisfed.org/fred2/series/EXUSEU?cid=95
Crude Oil: West Texas Intermediate (WTI)
http://research.stlouisfed.org/fred2/series/DCOILWTICO
J Hamilton comment “Traditionally WTI had less sulfur than Brent and sold for a slightly higher price.”
I tend to think the conclusion may be something more sulfurous.
My main assumption: On a longer term the correlation WTI,NS Brent holds.

7. Ricardo

Menzie wrote:
what could drive up oil prices denominated in US dollars:

• US GDP
• Rest-of-world GDP
• Energy intensity of economic activity
• Oil production capacity
• Cost of production of marginal producer
• Nominal interest rates
• Inflation rates
• Expected price of oil (itself a function of trends in the above factors)
• Speculative activities of non-fundamentalist traders
• Numeraire issues

What is interesting is that this list can essentially be summarized into three categories: 1. Keynesian measurements of government spending (money creations?)2. weakened dollar, with only one in category 3. Energy usage.
As Menzie demonstrates by the number of potential cuases on his list, energy usage is the weakest. So, since the biggest factor in a weak currency is excess money creation we should see that of Menzie’s 10 items that could drive up oil prices 90% are related to a weak currency.

8. Anonymous

The Rage,
That proves my point completely. Oil crashes when the dollar goes up and soars when the dollar goes down. To the fact that the dollar and oil also move in concert with growth rates and other metrics I say, DUH! Of course they do. Is that some kind of revelation that disproves the thesis?
QE2 = more dollars = more demand for oil = higher oil prices
(actually, I don’t think QE1 or QE2 are that expansionary, but credit market conditions have certainly changed due to other Fed actions)
If you didn’t need extra money (say you had a productive job in a non bubble sector that still had tight labor and no debt) then all QE really got you was more expensive stuff, including oil.

9. Jeffrey J. Brown

Given declining global net oil exports, shouldn’t we expect to see a declining dollar, since the US is currently the world’s largest net oil importer?
In other words, I suspect that the declining dollar is primarily a function of declining global net oil exports, especially what I define as define as declining Available Net Exports.
Following are what we show for global net oil exports for 2002 to 2009 (oil exporters with net oil exports of 100,000 bpd or more in 2005, which account for 99% plus of global net oil exports). I have added an estimate for 2010 (actual data will be out in a month or so).
Note that global net oil exports increased at about 5%/year from 2002 to 2005, and then we had flat to declining global net oil exports. I suspect that this inflection point was quite a shock to oil importing countries, especially developed oil importing countries.
Also shown are Chindia’s combined net oil imports. The difference between the two is what I define as Available Net Oil Exports (ANE), i.e., global net oil exports not consumed by Chindia.
As you can see, ANE fell from 40.8 mbpd in 2005 to 35.7 mbpd in 2009. A plausible estimate is that ANE could be down to about 27 – 30 mbpd by 2015.
Global Net Oil Exports Less Chindia’s Combined Net Oil Imports (BP + Minor EIA data, mbpd):
2002: 39 – 3.5* = 35.5 (ANE)
2003: 42 – 4.0 = 37.4
2004: 45 – 5.1 = 39.9
2005: 46 – 5.2 = 40.8
2006: 46 – 5.5 = 40.5
2007: 45 – 6.1 = 38.9
2008: 45 – 6.6 = 38.4
2009: 43 – 7.3 = 35.7
2010: 44 – 8 = 36**
*Chindia’s combined net oil imports
**Estimated
This table shows the detailed data for 2005 to 2009:
http://i1095.photobucket.com/albums/i475/westexas/Slide3-1.jpg

10. Steve

Let us not forget that China has been securing large buys of oil for the next 5 years by depositing hard currency and opening credit facilities in exisitng and future oil producers. This has speculators looking at oil for long term gains. If the speculators were not allowed to by futures, but instead it was limited to the actual users of the oil and oil products, the price what begin to circle down to a level around \$75 per barrel. The dollar falling is a maybe 11%-15% part of the overall price increase!

11. Frank

Ricardo,
You must have very little understanding of even basic statistics to make such a spurious claim.
Just because there are 10 different possible variables within an equation does not mean that 9 of the variables in aggregate have necessarily greater impact than just 1 variable alone…
Not to mention your buckets hardly make sense, but that’s a discussion for some other time.

12. Jonathan

Thanks for the post, Terrific. Really appreciate it. I wish you were better at coming up with the 10 to 25 word summary, something that could be repeated to the illiterate. I know in this case it’s difficult to explain in any simple manner that oil is traded in dollars and thus the actual price of oil changes to some degree as the dollar’s value changes, so it is not at all as simple as the dollar goes down and that means we pay more. The idea is pretty darned complicated.

Great work – you can’t repeat this correlation is not causation too many times.

14. don

Figure 3?
The intuitive appeal of the Republican position is certainly stronger, in my view, than the evidence you present. First, consider the effect of a weakening dollar on the price of oil. In the very short run (where inflation in the U.S. is zero or can be ignored) a fall in the value of the dollar is just the other side of an increase in the average of the prices of all traded goods. No causality is needed for this. True, the change in the price of oil could be smaller or greater than this average, but I think the Republican position is the most reasonable alternative given the available evidence. The better question is whether QE caused the dollar fall/oil price increase. Although the counterfactual can be argued forever and would probably be impossible to pin down empirically, I think that the primary effect of QE was to cause dollar depreciation, exactly as happened when Japan adopted the policy (causing a strident yen carry trade) and exactly as trade partners have been claiming.

15. JTapp

Dr. Chinn,
I was just reading Paul Krugman’s Peddling Prosperity (1994). He has a page discussing how Art Laffer produced a similar spurious analysis as the JEC study above in the early 1970s. His (and his co-horts’) goal was to show that de-linking the exchange value of the dollar to gold was the cause of the double digit inflation in the U.S. It’s apparently a cornerstone of “supply-side” economists that any devaluation of the dollar means high rates of inflation.

16. JTapp

You disagree with that claim?
The inability to reconcile 70s stagflation has always been a big problem for Keynesians. It shows you can’t print to prosperity.

17. Menzie Chinn

[deleted JTAPP]/dave: It’s hard for old-style Keynesians. But almost nobody is a completely demand-determined Keynesian anymore, so that’s a straw man you’ve thrown up. For thirty years the standard textbook has incorporated a supply side. You would be well advised to read one.

18. JTapp

It appears Dave mistakenly posted using my name in the above comment. My point was that the JEC study’s disingenuous nature seems similar to the Mundell-Laffer ideas of 1973. The goal of Laffer then, as Paul Ryan and co. now, was to show that having a fiat currency was causing the dollar to depreciate and oil prices to rise as a one-to-one result. According to Krugman, the list of other variables that Dr. Chinn listed above was also routinely ignored in 1973.
If I’m way off-base, please let me know, but the similar politics seem more than coincidental to me.

19. 2slugbaits

Menzie I find that the series are integrated of order one, but the two series are only possibly cointegrated using asymptotic critical values, and a 5% significance level, over the sample period investigated by the JEC-Republicans. That relationship disappears if the sample is extended to 2007M07-11M04. (In log terms, Brent appears to be stationary, but the nominal dollar appears stationary, over the 2008M11-2011M04 period; and both appear to be integrated of order one for the longer period).
Sorry for being thickheaded, but I’ve read and reread this paragraph several times and it’s just not sinking in. So let me see if I’ve got this right. Which series did you find integrated of order one? It’s the smaller JEC series, right? And under the JEC interpretation, the possible cointegrating relationship that holds the two series together would be QE1 and QE2? But you say that extending the series back to beginning of a relaxed monetary policy in July 2007 dissolves the I(1) relationship? But then this parenthetical note keeps throwing me for a loop:
(In log terms, Brent appears to be stationary, but the nominal dollar appears stationary, over the 2008M11-2011M04 period; and both appear to be integrated of order one for the longer period).
Can you disentangle that for me? I’m confused. Just an eyeball “analysis” of the data suggests that the shorter period is either trend stationary or perhaps a random walk with a strong drift component. If you look out over the longer series the natural interpretations would be either trend stationary with a level shift in July 2007 or a random walk with strong drift and an impulse shock in July 2007. Like I said, just looking at the data visually, that’s what I see in each series individually. If they are both trend stationary, then there is no cointegrating relationship so you’d have to use something like an SVAR to uncover any structural relationship (if any). If the two series are individually I(1), then you’d want to look for a cointegrating relationship recognizing some (orthogonal?) shock in July 2007. Like I said, that parenthetical note left me in the dust.

20. Menzie Chinn

2slugbaits: I packed in a lot. First, looking at things in levels (which I rarely do, but that’s how JEC-Republicans implicitly looked at the data) shows both series possibly cointegrated during the sample they investigated. That cointegration result disappears for the longer period encompassing all monetary easing.

In log terms, which is a more conventional way to model the series, one series appears trend stationary and another difference stationary, so no cointegrating relationship should hold. Over the longer period, both appear to be I(1) series.

In other words, it’s hard to come to conclusions because the inferences regarding stationarity/nonstationarity are sensitive to sample period. The JEC-Republicans can do what they do in part because they picked a very specific time period.