This post was written by Mike Veall of the Department of Economics at McMaster University.
Sorry, I am still on about the research that was used to formulate the Ontario Progressive Conservative jobs plan. Again I emphasize this is about the underlying econometric research by Benjamin Zycher, Senior Fellow of the American Enterprise Institute. It is not about what I think the plan’s proposals actually will or won’t do.
In a Financial Post piece on this topic, I was the point and Mr. Zycher was the counterpoint. I didn’t know there was going to be a counterpoint, but this is not a complaint. Terence Corcoran should be credited with running a piece critical of his own writing. And as I argue that Mr. Zycher’s time series econometrics (see pages 13 to 18) is incompetent, Mr. Zycher deserved a chance to respond. (BTW, I have sent him this post and renew my offer to debate him about this.)
My piece was edited by the Financial Post. While I understand space constraints and that it may be impractical to consult authors about such edits, it was unfortunate that among other things it cut the reference to Jim Stanford’s comment on Mr. Zycher’s research (which is much more thorough than mine and anticipates much of what I write below). It also cut the reference to my more detailed comment in Worthwhile Canadian Initiative.
Regardless Mr. Zycher’s counterpoint is just plain weird. To see this, let’s go back to his original work and focus on his employment equation:
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
Mr. Zycher takes the -3992.22 Corporation tax coefficient, multiplies it by negative 3.5 because that is the percentage point corporate income tax that the Ontario Progressive Conservatives propose and concludes the cut will increase “full-time equivalent employment by about 14,000”.
But even though there is no indication of it in his original paper, Mr. Zycher now says that the equation can be interpreted as saying that the increase would be 14,000 the first year, 28,000 the next year and so on to get 112,000 after 8 years.
But his equation has no dynamics and is inconsistent with the cumulative interpretation. The one-year effect equals the two-year effect equals the eight-year effect by construction.
I tried to get this across in the Post by pointing out that if Mr. Zycher could pretend there were dynamics in the econometric equation to cumulate the corporation income tax cut effects, one could equally pretend the same for other variables that enter the equation in exactly the same way, such as the population variables. That approach would imply that a 1000 one-time increase in the stock of working-age immigrants would increase employment by a clearly ridiculous 13,000 after eight years.
But I should have stayed out of the pretending game: one should never use the reductio ad absurdum argument with someone whose work (in this case) is already absurd.
Hence Mr. Zycher responds with…”he [Veall] fails to distinguish between economic shifts that play out over several years and those where adjustments would be quick”. But this is surreal. This is precisely what his own econometric model fails to distinguish.
Hence when he writes it that a “cut in the corporation income tax…would increase investment and induce other effects that would not be complete after a year”, that may be so, but again the problem is that his own econometric model excludes any multi-year effect. Yes, he has no dynamics, he has no dynamics today.
There are other problems including a regressing-an-identity point regarding his GDP equation to which he doesn’t respond. (For detail on that and other problems, such as endogeneity, again see my original post.)
But the issue is more fundamental. For Mr. Zycher, step one is to estimate an econometric equation and then step two is to pretend that it says what he wants it to say. My suggestion is that he skip the econometrics step.
Note: this is a slightly revised version than the one originally published.
Mike: Great advice in your last sentence!
Also, "one should never use the reductio ad absurdum argument with someone whose work (in this case) is already absurd." True, but sometimes it is just so, so, tempting... :-)
Posted by: Dave Giles | June 08, 2014 at 02:01 PM
"one should never use the reductio ad absurdum argument with someone whose work (in this case) is already absurd."
;) ;) ;)
Posted by: Min | June 08, 2014 at 02:16 PM
Wow, this is strong stuff, Stephen. Hats off for trying to rid the world of bad arguments, one step at a time.
[Thanks, but I'm just posting what Mike Veall wrote - SG]
Posted by: Determinant | June 08, 2014 at 04:31 PM
Another problem: Most of his variables are not stationary (Employment, GDP, Population, Energy Production, possibly Electric Cost Index). Massive issue of spurious regressions.
Posted by: Joseph S | June 08, 2014 at 05:42 PM
"Dear colleague: you have regressed an accounting identity. And that identity is false."
No matter how partisan he is, no physicist would state that a proton electric charge is -3. What is it in economics that some of our colleagues are willing to make the equivalent statement?
Posted by: Jacques René Giguère | June 09, 2014 at 01:19 AM
Jacques: “No matter how partisan he is, no physicist would state that a proton electric charge is -3. What is it in economics that some of our colleagues are willing to make the equivalent statement?”
Your observation about the physics holds because there is no special interest group who stands to benefit if they could convince people that the proton charge was three electron charge units.
While the observations by Mike Veall on this subject are fun (yes, if you are going to do this work correctly, perhaps you should start with a full DSGE specification), it misses the more central issue: The point of an election is to win – it is not about the search for truth. All of this “million jobs” nonsense is simply a reflection of public choice theory. If you really want to create a million jobs overnight, make mechanized farm equipment illegal.
Society still has a largely mercantilist view of the economy and politicians parrot those positions. Politicians, sell much more damaging fairy tales than bad regressions: a low Canadian dollar increases prosperity (not subsidizing exports at the expense of imports), minimum wage helps poor people (not protecting unions from wage competition), rent control, zoning and planning permissions creates affordable housing (not protecting landlords), the Competition Bureau protects consumers (not producers), supply management ensures food sovereignty (not making food more expensive for the public, but making the farmer richer), deposit insurance and CMHC protects our financial system (not a subsidy on runnable debt that encourages banks to take reckless positions), immigration needs to be controlled (not because immigrants would compete in the labour market), corporations need to pay their fair share of taxes (not a tax on consumers, workers, and shareholders), CPP premiums and other government pension schemes are shared between the employer and the employee (not that the employee pays it all in form of lower take-home pay), and on, and on, and on,...
As far as a misspecified regression goes, that's pretty tame, if not boring, fiction.
Posted by: Avon Barksdale | June 09, 2014 at 09:31 AM
In addition to the well-founded points Mike has made, something else about these specifications stuck out for me. Zycher uses the level of employment (or annual GDP) as the dependent variable and the corporate tax rate is an independent variable in a regression where the unit of observation is the province-year.
So, PEI is there and so is Ontario. In this specification, the corporate tax rate is predicted to have the same effect in PEI and Ontario.
I wouldn't let a student get away with this--you would always want to use either a rate or a log specification.
I don't think this is as bad as the errors Mike has picked out, but just something to add to the weight of the evidence that this 'economic analysis' was not quite up to undergrad-level econometric standards.
Posted by: Kevin Milligan | June 09, 2014 at 12:55 PM
Re: Avon Barksdale.
Imagine two worlds:
World 1: politicians just say *whatever they want* and experts like Mike Veall just shut up and don't speak out because hey it's just politics and what you say doesn't matter. Campaigns descend even further into fact free BS-fests with the only important decision being which flavour of scary music to put in the attack ads.
World 2: when politicians say things that are different than the facts, experts point this out. While such 'fact checks' aren't ever going to be decisive in a political campaign, these fact checks do inflict at least a little reputation cost on the politician. Therefore, when the politician and advisors are considering a communications policy based entirely on fact-free BS, they may pause--even a little--to consider the damage--even if it is slight--that the fact checking will inflict on their credibility.
Me, I prefer to live in World 2.
Posted by: Kevin Milligan | June 09, 2014 at 01:00 PM
We know that, for any object, mass = volume x density. That's true by definition.
It's like estimating mass = a + b.volume + c.density + d.luminosity
Then saying that d > 0 shows that more luminous objects have greater mass.
The only thing that d =/= 0 is telling us is that we have misspecified the functional form. We have taken a linear approximation to a multiplicative relationship.
Posted by: Nick Rowe | June 09, 2014 at 02:15 PM
Kevin,
I'm with you. But the problem is that World 2 isn't a possibility, because no politician consciously adopts a communications policy based on fact free BS. They BELIEVE their fact-free BS. The absence of factual basis for those believes is neither here nor there. Some of them may continue to believe it long after it been refuted (witness the NDP socialist caucus). For them its not a choice between spouting fact-free BS and somewhat factual BS, the facts are whatever their BS is.
That being said, PJ O'Rourke and the Cato institute have an interesting (and very funny) commentary on the history and role of BS in American politicans in a recent submission to the US Supreme Court: http://www.bubblews.com/news/2536935-a-satirical-amici-brief-to-the-us-supreme-court-by-pj-orourke. One of the introductory sentences gives you a flavour of the piece:
"The campaign promise (and its subsequent violation), as well as disparaging statements about one’s opponent (whether true,
mostly true, mostly not true, or entirely fantastic), are cornerstones of American democracy."
Apart from being a compelling legal argument in favour of the right to commit acts of political BS (and, indeed, the merit of doing so), it's wickedly funny (see footnote 15 for a particularly pointed comment about the "truthiness" of the US Supreme Court's Obamacare decision. It takes a lot of nerve to make a submission to the US supreme court were you openly mock one of their highest profile recent decisions, but I suspect that Justice Roberts is a fan of PJ O'Rourke).
Posted by: Bob Smith | June 09, 2014 at 03:55 PM
In fact, PJ O'Rourke has a comment that applies equally to your hope for a World 2 as to to the impguned Ohio state elections law:
"Not only does this Pollyannaish hope stand in the face of all political history, it disregards the fact that, in politics, truths are felt as much as they are known."
Posted by: Bob Smith | June 09, 2014 at 04:05 PM
Kevin Milligan,
I not arguing that we shouldn't criticize poor economic analysis, which Zycher's work clearly is. The point is that an election should not be about who can do better regressions. This is a time for normative debates along with issues of efficiency. A million more jobs is not even right question. All economic models are metaphors - simplifications - stories or parables. They help sharpen our understanding and create insight. But they should not be used by experts to spin ridiculous fairy tales. Yes, the PC's calculations are nonsense, but so is the NDP's claim that corporations can pay taxes (they can't, only people can pay taxes), or that the Liberals's new pension plan will be a shared cost between the employee and the employer (in equilibrium, the price of labour is its marginal product, regardless of who actually receives the pay cheque). All of this is silly, and economists would do well to point it all out. The focus on a egregiously poor regression analysis seems rather moot considering all the other elephants in the room.
Who cares about some silly regression? Even if done "right" they are almost always wrong. How long has the BoC's models failed in their GDP growth predictions? Four years now? I suspect they have rather state-of-the-art DSGE models and the like. Perhaps economists should point out to the public that unconditional forecasting is essentially impossible, even by the experts, but with conditional forecasting we can at least usually get the signs right.
But, of course, as with any election, we are watching the capture process of public choice theory at play. The welfare state has no difficulty in creating special interest groups or economists who support them. Perhaps that is where we should start the debate.
Posted by: Avon Barksdale | June 09, 2014 at 06:35 PM
You think we were any better before the welfare state, Avon? The Welfare State has nothing to do with it; George Brown's Globe, now the Globe and Mail is a particularly Canadian example of that.
Posted by: Determinant | June 09, 2014 at 09:02 PM
"You think we were any better before the welfare state, Avon?"
Yes.
Posted by: Avon Barksdale | June 09, 2014 at 09:25 PM
Hilarious.
Posted by: Determinant | June 09, 2014 at 10:03 PM