Replicating Stern (1993)

Last year, Energy Economics announced a call for papers for a special issue on replication in energy economics. Together with Stephan Bruns and Johannes König we decided to do a replication of my 1993 paper in Energy Economics on Granger causality between energy use and GDP. That paper was the first chapter in my PhD dissertation. It is my fourth most cited paper and given the number of citations could be considered "classic" enough to do an updated robustness analysis on it. In fact, another replication of my paper has already been published as part of the special issue. The main results of my 1993 paper were that in order to find Granger causality from energy use to GDP we need to  use both a quality adjusted measure of energy and control for capital and labor inputs.

It is a bit unusual to include the original author as an author on a replication study, and my role was a bit unusual. Before the research commenced, I discussed with Stephan the issues in doing a replication of this paper, giving feedback on the proposed design of the replication and robustness analysis. The research plan was published on a website dedicated to pre-analysis plans. Publishing a research plan is similar to registering a clinical trial and is supposed to help reduce the prevalence of p-hacking. Then, after Stephan and Johannes carried out the analysis, I gave feedback and helped edit the final paper.

Unfortunately, I had lost the original dataset and the various time series I used have been updated by the US government agencies that produce them. The only way to reconstruct the original data would have been to find hard copies of all the original data sources. Instead we used the data from my 2000 paper in Energy Economics, which is quite similar to the original data. Using this close to original data, Stephan and Johannes could reproduce all my original results in terms of the direction of Granger Causality and the same qualitative significance levels. In this sense, the replication was a success.

But the test I did in 1993 on the log levels of the variables is inappropriate if the variables have stochastic trends (unit roots). The more appropriate test is the Toda-Yamamoto test. So, the next step was to redo the 1993 analysis using the Toda-Yamamoto test. Surprisingly, these results are also very similar to those in Stern (1993). But, when Stephan and Johannes used the data for 1949-1990 that are currently available on US government websites, the Granger causality test of the effect of energy on GDP was no longer statistically significant at the 10% level. Revisions to past GDP have been very extensive, as we show in the paper:

Results were similar when they extended the data to 2015. However, when they allowed for structural breaks in the intercept to account for oil price shocks and the 2008-9 financial crisis, the results were again quite similar to Stern (1993) both for 1949-1990 and for 1949-2015.

They then carried out an extensive robustness check using different control variables and variable specifications and a meta-analysis of those tests to see which factors had the greatest influence on the results.

They conclude that p-values tend to be substantially smaller (test statistics are more significant) if energy use is quality adjusted rather than measured by total joules and if capital is included. Including labor has mixed results. These findings largely support Stern’s (1993) two main conclusions and emphasize the importance of accounting for changes in the energy mix in time series modeling of the energy-GDP relationship and controlling for other factors of production.

I am pretty happy with the outcome of this analysis! Usually it is hard to publish replication studies that confirm the results of previous research. We have just resubmitted the paper to Energy Economics and I am hoping that this mostly confirmatory replication will be published. In this case, the referees added a lot of value to the paper, as they suggested to do the analysis with structural breaks.

Buying Emissions Reductions

This semester I am teaching environmental economics, a course I haven't taught since 2006 at RPI. Last week we covered environmental valuation. I gave my class an in-class contingent valuation survey. I tried to construct the survey according to the recommendations of the NOAA panel. Here is the text of the survey:

Emissions Reduction Fund Survey

In order to meet Australia’s international commitments under the Paris Treaty, the government is seeking to significantly expand the Emissions Reduction Fund, which pays bidders such as farmers to reduce carbon emissions. To fully meet Australia’s commitment to reduce emissions by 26-28% below 2005 levels by 2030 the government estimates that the fund needs to be expanded to $2 billion per year. The government proposes to fund this by increasing the Medicare Levy.

1. Considering other things you need to spend money, and other things the government can do with taxes do you agree to a 0.125% increase in the Medicare levy, which is equivalent to $100 per year in extra tax for someone on average wages. This is expected to only meet half of Australia’s commitment, reducing emissions to 13-14% below 2005 levels or by a cumulative 370 million tonnes by 2030.

Yes No

2. Considering other things you need to spend money, and other things the government can do with taxes do you agree to a 0.25% increase in the Medicare levy, which is equivalent to $200 per year in extra tax for someone on average wages. This is expected to meet Australia’s commitment, reducing emissions to 26-28% below 2005 levels or by a cumulative 740 million tonnes by 2030.

Yes No

3. If you said yes to either 1 or 2, why? And how did you decide on whether to agree to the 0.125% or 0.25% tax?

4. If you said no to both 1. and 2. why?

***********************************************************************************


85% voted in favour of the 0.125% Medicare tax option and 54% voted in favour of 0.25% - So both would have passed. A few people voted against 0.125 and for 0.25, so I changed their votes to for 0.125 as well as 0.25.


Reasons for voting for both:


• $200 not much, willing to do more than just pay that tax 
• We should meet the target
 
• Tax is low compared to other taxes - can reduce government spending on health in future
 
• Can improve my health
 
• Benefit is much greater than cost to me
 
• I pay low tax as I'm retired, so can pay more
 
• I'm willing to pay so Australia can meet commitment
 
• Only $17 a month
 
• Tax is small
 
• Because reducing emissions is the most important environmental issue
 
 

Reasons for voting for 0.125 but against 0.25:


• Can afford 0.125 but not 0.25
• Government can cover the rest with other measures like incentives
 
 

Reasons for voting against both:


• There are other ways to reduce emissions - give incentives to firms rather than tax the middle class... 
• Government should tax firms
• Don't believe in emissions reduction fund because it is inefficient

• I prefer to spend my money rather than pay tax and reduction in emissions is not very big for tax paid


Mostly the reasons for voting for both are ones we would want to see if we are really measuring WTP - can afford to pay and it is a big issue. Those thinking it will increase their personal health or reduce health spending were made to think about health by the payment vehicle. I chose the Medicare Levy as the payment vehicle as the Australian government has a track record of increasing the Medicare Levy for all kinds of things, like repairing flood damage in Brisbane!
 I chose the emissions reduction fund because it actually exists and actually buys emissions reductions.

Most people who voted for 0.125% but against 0.25% have valid reasons - they can't afford the higher tax. However, one person said the government should cover the rest by other means. So that person may really be willing to pay 0.25% if the government won't do that.


When we get to the people who voted against both tax rates, most are against the policy vehicle rather than not being willing to pay for climate change mitigation. So, from the point of view of measuring WTP these votes would result in an under estimate. These "protest votes" are a big problem for CVM. Only one person said they weren't willing to pay anything given the bang for the buck.