Flogging a Dead Horse: Research the Progressive Conservatives misinterpreted was flawed

This post was written by Mike Veall of the Department of Economics at McMaster University.

David Reevely and Jim Stanford, followed up by Paul Boothe, Mike Moffat, the Globe and Mail and others have shown that the Ontario Progressive Conservative millions job plan does not follow from the research it is based on. For example, one of their commissioned studies says that a reduction in regulation (Ontario level to Alberta level, however measured) will yield a total of 10,600 jobs. But the PC plan says that will create that many jobs per year, so that over 8 years, that is 8 × 10,600 = 84,800 jobs. They make the same mistake looking at other proposals.

I had been looking at this too but was slow to the draw. One point that I don’t think that has yet been covered is that one of the pieces the PCs misinterpreted, the study by Washington D.C. economist Benjamin Zycher, is not exactly technically watertight itself.  So let’s flog the dead horse. I am grateful to WCI for giving me the space to do so.

Here is the Zycher two-equation model that led to the projections the PC used for the employment effects of reducing electricity prices (“ending subsidies on wind and solar”) and participation in the New West Partnership Trade Agreement (NWPTA): 

GDP = 0.97×lagged GDP + 2.23GDP growth rate + 7.37×Total Population +19.47×Total working-age population  - 2.41×Corporate tax rate – 0.88×Personal tax rate + 10.87×Energy Production  - 0.71×Electricity cost index + 0.53×NWPTA participation + Provincial Fixed effects

Employment = 394.12×GDP + 1738.38×GDP growth rate + 667.55×Total Population + 934.66×Total working-age population  - 3992.22× Corporate tax rate – 38.77×Personal tax rate + 1381.07×Energy Production  - 179.26×Electricity cost index + 198.88×NWPTA participation + Provincial Fixed effects

Probably anyone who is still with me at this stage can do their own critique of this model but issues to flag include:

• For the GDP equation, normally we would say lagged GDP and the GDP growth rate are complete determinants of GDP (i.e. current GDP = lagged GDP × (1 + the GDP growth rate). Mr. Zycher is not estimating an identity only because he linearizes this relationship. Mr. Zycher writes about this issue, writing, “Since lagged GDP and the GDP growth rate should explain current-year GDP fully, this identity allows us to examine the effects of the proposed reforms in isolation”.  I do not understand this statement: my view is that given the identity, the other variables have no place in the equation.
• Referring to the GDP equation, Mr. Zycher writes “Note that each of these variables can be viewed as exogenous or pre-determined”. I don’t see how this statement can apply to energy production. GDP affects energy production. Energy production affects GDP.
• The employment equation has a recursive-ish structure with the GDP equation. Here Mr. Zycher rules out simultaneity by fiat but there are clearly some identification issues which are probably not worth detailing. Mr. Zycher does not include the indirect effects through GDP in assessing the employment effects of policy changes but reads them straight off the second equation alone.

Do the results make sense?  Mr. Zycher estimates that the effect of ending the wind and solar subsidies will be to reduce the energy cost index from 160.9 (the Ontario index level) to 132.7 (the Canadian average). Hence he multiplies the Electricity cost index coefficient -0.71 times (132.7-160.9) and obtains 20.022. Since he is measuring GDP in billions, he estimates an increase in GDP of about $20 billion.

For comparison, suppose the working-age population went up by 1000 because of immigration. The units for both the population variables are 1000 so, using the coefficients from the equation, GDP would increase by (7.37 + 19.47) = $26.84 billion.  That is almost $27 million per immigrant. That seems unreasonable.

None of this is a statement about the PC program, just a statement about the research that underlies it. But why didn’t they find/commission research based on growth equations for GDP and employment rather than level equations?

Note: See also Mike's follow-up post.

Comments

[Nick Rowe]

"For the GDP equation, normally we would say lagged GDP and the GDP growth rate are complete determinants of GDP (i.e. current GDP = lagged GDP × (1 + the GDP growth rate). Mr. Zycher is not estimating an identity only because he linearizes this relationship."

OMG!

[Stephen Gordon]

No kidding. I wonder what the R2 is on that equation - 0.9999?

[BSF]

With a coefficient of 0.97 and a t-stat of 38 on lagged GDP I'd start thinking about nonstationarity issues.

[Avon Barksdale]

I find this kind of forecasting a mug's game in the first place - there are too many other unobserved variables that are almost certainly correlated with the error term here. Politicians should be arguing from principle - this is the time for us to decide whether or not subsidizing anything makes sense. Let economists research the positive issues - a political campaign is the time for largely normative debates. If anything stating, "My plan will create a million jobs.", is a distraction that opens you up to endless debates about how good your regressions are. Fine for a seminar, not for a political campaign.

[Dave Giles]

Clearly a spurious regression. What a great example of the latter for us all to use in our classes.

[Chris J]

Avon... great handle.

[Jacques René Giguère]

"Dear colleague: You regressed an accounting identity. Anyway, that identity is false."
As my father used to say:"Get your degree. Then you'll be allowed to say anything." When did thet meet?

[T Orwell]

So if a thousand working age immigrants cause GDP to increase by $26.84 billion and (according to the employment equation) 394.12 jobs are created per billion GDP increase, bringing in 1,000 working age workers will create 10,578 jobs.
That's more than 10 jobs created per new worker! Sign me up!

[tono-bungay]

His stream-of-consciousness argument about electricity prices is hilarious. He starts with the EIA estimates of costs of different generation sources, realizes that with those numbers the GEA mix is cheaper than what it replaces according, declares that "it is likely to be the case that the new estimate has been shaped by political pressures within the Obama administration" and therefore dismisses them, and then shops around for a prediction more in keeping with his desired conclusion. He picks a 2009 study that predicted that feed-in tariffs would significantly increase in the next few years, when in fact we know 5 years later that they decreased quite sharply. Quite apart from the fact that to get the savings of that 2009 doomsday report we would have to unbuild all the renewable power at no cost and re-commission the coal plants also without cost. He takes the dollar increase prediction from the right-wing think-tank predicting a high increase in prices from green energy, divides it by the predicted total bill from Environmental Defence, which predicts a small increase, to calculate a percentage increase.

As this post noted, it seems quite clear that Zycher has taken a list of single-variable regressions and strung them together, including the correlation between energy production and employment. So it seems that a large generation surplus would be just what the doctor ordered to create jobs! "An increase in primary energy production of 1000 cubic meters-equivalent is predicted to increase employment by about 1381 jobs."