Taxes & growth

Ryan Bourne accuses lefties (via) of ignoring Karel Mertens paper (pdf) which claims that lower top marginal tax rates in the US have led to faster GDP growth.

However, this result is achieved by torturing the data. Martens estimates, Simple univariate regresssions show, says Mertens, that "tax rate increases are associated with higher income growth." It is only by imposing various controls that he finds that tax rate cuts lead to faster growth. And this opens up a potentially interminable argument about what we should control for and what not.

To see my point, consider the UK data. In 1988, Nigella's dad cut the top marginal tax rate from 60% to 40%. In the following 22 years - until the rate rose to 50% - real GDP grew by 2.4% per year. But in the previous 22 years, it had grown by 2.6% a year*. A cut in the top tax rate, then, led to slower growth. There are three possible reactions to this:

1. The tax cut actually made no difference to growth: 0.2 percentage points is well within the sort of difference we'd expect from random fluctations. Indeed, it's possible that there are almost no feasible policies which noticeably affect trend growth.

2. The tax cut actually reduced GDP growth. This might be because the income effect encouraged wealth creators to retire early, or because it incentivized rent-seeking or the growth of a banking sector which thrived only by creating risk pollution.

3. The tax cut on its own would have raised GDP growth, but other factors offset this benefit.

People like Ryan must argue for point (3). But this is tricky. Many rightists - in fact, not (pdf) just rightists -  would argue that other Thatcherite reforms of the 1980s and 90s should have raised trend growth: the reduction in trades union power; privatization; the abandonment of wage and price controls; lower corporate tax rates; and perhaps (pdf) inflation targeting.

To claim that Lawson's tax cuts boosted GDP growth thus requires one to argue either that the many pro-market reforms of the 80s and 90s had no great beneficial effect or even a negative effect, or that some other factor greatly depressed growth and offset all these benefits**. This requires some arguing.

For me, the message of all this is simply that aggregate GDP data just don't permit us to draw robust inferences about policy effects. Not only is theory ambiguous, but so is the data. For this reason, among others, Merten's paper is certainly not (pdf) the consensus view.

I suspect the argument about what the top tax rate should be is not one that can be settled by the data.

* I think it's important to take as long a period as possible here, firstly because doing so reduces the standard error around growth estimates, and secondly to allow for the (very plausible) possibility that tax rates have long-run effects by shaping social norms.

** We can leave public spending out of this. It averaged 37% of GDP from 1988 to 2010, compared to 37.5% in 1966-88.

Comments

[Luis Enrique]

are you calling adding control variables (aka thinking about omitted variables bias) "torturing the data"?

[chris]

Not in itself. The problem is that the possible omitted variables are so many there's no universally-agreed way of selecting them. Not also that Mertens is quiet about the precise mechanism through which tax cuts raise GDP - that's a pretty big omission.

[Luke]

Interesting. At the risk of being a naive numpty, I will take Ryan at his word, which is not so much about tax rates as saying that there's no point using evidence. On the same day,St Worstall at the ASI says that he doesn't believe in macroeconomics. Not really sure what he is saying, but is this really a case of rightists acknowledging that there is no evidence for their beliefs, so deciding to ignore evidence?

TW - http://www.adamsmith.org/blog/economics/i-really-dont-believe-in-macroeconomics-you-know

[Luis Enrique]

"possible omitted variables are so many there's no universally-agreed way of selecting them"

is that true? It is universally agreed that if the omitted variable is has a material impact upon the outcome and is correlated with explanatory variable you are interested in, then it must be included. And if it isn't correlated, it can be left out.

Admittedly things get tricky if you think one variable is the channel another variable works through (i.e. if you have investment in an aid-growth regression, and aid works via raising investment). This tells us that careful empirical work has to think about all these things and perform sensitivity analysis. Or you have to estimate a system of equations rather than just one. I've only scanned the paper, but isn't that what she does? An SVAR?

[Blissex]

I am always astonished that such debates happen on small details like 0.2% differences in growth rates or subtle disquisitions on which factors to discount and which not to discount, as if "ceteris paribus" were attainable.

it is of course not surprising that the well paid propagandists of the property interests grasp at whichever straws they can and studiously disregard the big loud braying purple elephants in the room, which are as per my usual remarks the government policies to use North Sea oil to fund tax cuts for the wealthy, and to blow up a massive debt boom to create an asset price bubble, as Tony Blair wrote in 1987 quite presciently, as to his own actions tend years later too:

«Without oil and asset sales, which themselves have totalled over £30 billion, Britain under the Tories could not have enjoyed tax cuts, nor could the Government have funded its commitments on public spending. More critical has been the balance-of-payments effect of oil. The economy has been growing under the impetus of a consumer boom that would have made Lord Barber blush. Bank lending has been growing at an annual rate of around 20 per cent (excluding borrowing to fund house purchases); credit-card debt has been increasing at a phenomenal rate; and these have combined to bring a retail-sales boom – which shows up dramatically in an increase in imported consumer goods. Previously such a boom and growth in imports would have produced a balance-of-payments deficit, a plunging currency and an immediate reining-back on spending, with lower rates of growth.

Instead, oil has earned foreign exchange and also produces remittance payments from overseas investments bought with oil money. The situation is neither stable nor healthy in the long term: but in the short term it allows the living standards of the majority to rise rapidly, even though the industrial base, the ultimate foundation of a successful economy, is still only achieving the levels of output of 1979. The fact that we have failed to use oil to build a productive and modern industry for the future is something historians will deplore. Nevertheless, oil has been utterly essential to Mrs Thatcher’s electoral success. Academics and commentators may ruminate on the Thatcher ethos and its effect on social attitudes, but the voters are looking in their pockets.»

Indeed as a more recent commentator confirmed:

«Another of Thatcher’s magic potions was ‘home equity withdrawal’ or remortgaging – drawing down the equity in the borrowers home for (mainly) consumption purposes – new cars, holidays, and so forth.

Under the two Prime Ministers that preceded her, James Callaghan and Ted Heath, home equity withdrawal as a percentage of GDP growth was around 36% for both.

Under Thatcher, this exploded to over £250bn across her premiership – a staggering 104% of GDP growth. [ ... ]

[ ... ] But Blair did his homework and let loose – as did Thatcher – a wave of cheap credit, financial deregulation, house price inflation and an equity withdrawal-led consumption boom.

Withdrawals under Blair’s leadership totalled around £365bn,
that’s a full 103% of GDP growth over the same period,»

So those are two of the big loud braying elephants (another one is the yen carry trade arising in the 1990s from the Japanese ZIRP) in the room that are being bravely disregarded in this argument, and I find it amazing that it is not just propagandists of the right that do so, but also ChrisD, who usually seems rather more informative, but with few exceptions seems to like more to discuss second-order effects or fig leaves like managerialism (which I detest too BTW) and tax cuts.

Is there a "conventio at excludendum" that in "polite" argument one should not mention massive but vulgar things like the near-doubling of median wages of property owners by debt-fueled capital gains redistributing income upwards?

[Blissex]

«It is universally agreed that if the omitted variable is has a material impact upon the outcome and is correlated with explanatory variable you are interested in, then it must be included. And if it isn't correlated, it can be left out.»

Ohhh please enough with the usual comical oversimplification of matters based on the usual "ceteris paribus" delusion.

The argument quoted above is vaguely reliable only "ceteris paribus", so that with a judicious choice of variables their separate effects can be discerned and discounted.

But the economy is not simple like that, in particular being subject to *strong* exogenous factors, and one needs a strong theory then to separate what is separable.

However given the outcome of the Cambridge Capital Controversy, the concomitant lack of a theory of capital, money and interest in mainstream "Economics", the theorem of second best and other well known results, using an artful, targeted choice of what to discount and what not to discount is just propaganda.

Put more coarsely, "explanatory" usually is just clever handwaving...

And never mind that the periods under consideration include one in which the colossal impact of North Sea oil policy and of maximum-leverage policy make any attempt at meaningful comparison moot.

[Luis Enrique]

Blissex, I don't follow you at all. I am talking about empirical data analysis, I did not mention economic theory at all.

Explanatory variables are whatever you choose in an attempt to explain variation in whatever you are studying.

You have suggested oil revenues, credit expansion, home equity withdrawal etc.

These are examples of variables that are correlated with tax cuts (you claim because they funded them) which also explain the outcome (GDP growth) and which therefore ought to be included in the set of explanatory variables otherwise the estimated impact of tax cuts could be wholly misleading.

Which seems to be your point.

[Anonymous]

All this suggests to me that economics is closer to religion that it is to science.

[rogerh]

My money is on 1).

I ended up wondering if all we are measuring year on year are noisy shifts in a very slow moving data series. So far as I can see any fool can stimulate or dampen an economy for a short while but the long-term trends happen nontheless and with inexorable force. The question to my mind is 'what is needed on a 20 year timeframe to make a real shift (preferably +ve)'? With the likely causes and cures and confounding factors so slow acting what hope is there for a 5 year policy timeframe.

A naive view is that the long-term is merely a concatenation of short terms - works fairly well in the corporate world. But for mature nations it does seem that some success factor for the long term is missing - too obvious, too inconvenient, too expensive, not possible? I don't know.

[WNY-WJ]

Using a mere 0.2% GDP difference in one country between two time periods is indeed problematic, given the well-known controversies. A much more robust analysis is required to persuade anyone in this debate - more countries, more time periods, various statistical techniques, etc. The analysis in question will not cause anyone to reexamine their prior beliefs. It is too simplistic to be good economic analysis, but it could be good politics, though.

[Blissex]

«But for mature nations it does seem that some success factor for the long term is missing»

Look all relevant statistics are heavily biased for the obvious reasons, but some people think they discern trends, and a very plausible hypothesis is that the key to running a "complex society" with high value added is *cheap energy*. More precisely, cheap and *dense* and *divisible* energy.

Because that's the only way of having onboard motive power (cars are also called "auto-mobiles" which shows how wondrous that fact is). For things like lawnmowers or combine harvesters or ships or motorbikes.

Because having autonomously powered engines that don't rely on human or animal or wind or water as external power sources is a fantastic thing.

Once upon a time dozens or hundreds of people were needed to harvest what a self-powered combine harvester can do today.

No fuel is know that is both as cheap and as power dense and easily subdivided as oil.

The price of oil drives the entire value added chain; the more expensive oil is, the less "productive" people and capital are.

"Economics" as taught in most contemporary university is of course (deliberately) blind to such issues; but the "classics" could have well understood by making an analogy with the sudden discovery of extremely fertile marginal land, but with sharply decaying fertility after a time.

"Developing" nations are currently plateauing into the so-called "middle income trap". Perhaps, or even probably, because cheap labour and more capital matter, but cheap energy to power the self-motive capital items matters even more.

[Blissex]

«I did not mention economic theory at all.»

That's why your points amount to fatuous hand-waving. Pure statistical analysis can only discover ex-post correlations, which not only are not causation, but they are not even hints of future correlations by themselves, and no meaning can be imputed to them.

There is a core theorem of econometrics that says that a model where every variable correlates with every other can't make sense; one must make a choice, strongly driven by theory, on ascribing specific roles to the variables, a-priori, and tht choice matters a very great deal.

Therefore what you call "explanatory variables" aren't:

«Explanatory variables are whatever you choose in an attempt to explain variation in whatever you are studying.»

In the study of the political economy there are a series of really big problems with a purely statistical approach:

* The economy is what Soros calls "reflexive": it is the result of the actions of wilful agents who react to events. It is not a once-built, always-the-same, machine.

* The economy therefore is an aggregate whose shape, and its "correlations" change with time; it is a self modifying aggregate, and even has phase transitions.

* Even worse, in particular each national economy can be subject to massive exogenous impulses, that can generate a lot of noise that swamps whatever specific signal one is seeking like a needle in the haystack.

Therefore this seems to me a deep misunderstanding of my position:

«You have suggested oil revenues, credit expansion, home equity withdrawal etc.
These are examples of variables that are correlated with tax cuts (you claim because they funded them) which also explain the outcome (GDP growth) and which therefore ought to be included in the set of explanatory variables otherwise the estimated impact of tax cuts could be wholly misleading.
Which seems to be your point.»

What I have meant is that I don't regard them as explanatory variables, but as exogenous noise, as some other commenters seem to have thought as well. Noise of such magnitude that trying to figure out the effects of tax cuts on their own is futile, especially without a strong a-priori of well understood theory.

Sure control variables can help separate different influences on the aggregate outcome, but they require strong enough signal-to-noise ratios in a slowly changing situation.

The last 25 years in the UK have been far from a smooth ride with just a bump in taxes and/perhaps a few other bumps whose effects can be easily subtracted.

Because there is nothing "endogenous" about oil discoveries, the fiscal policy chosen to spend that oil, and the credit policy, or the decisions of the Japanese government to adopt ZIRP or of the Chinese and Saudi ones to build immense stocks of foreign IOUs. They are all deeply exogenous or political acts.

So one can use statistical analysis to discern consequences in a relatively calm, steady state, and social scientists and not just economists are fabulously happy when they discover "natural experiments" like that, otherwise it is ridiculous.

ChrisD has pointed out that the magnitude of the consequences teased out has been too small to be significant, but without mentioning the big, purple, braying elephants in the room that make a polite discussion of the finer points of statistical analysis a joke of the "apart from that, Mrs. Lincoln, did you enjoy the play?" sort.

[Blissex]

«Noise of such magnitude that trying to figure out the effects of tax cuts on their own is futile, especially without a strong a-priori of well understood theory.»

BTW this is exactly what particle physicists do nowadays: since some critical aspects of their experiments *cannot be measured*, they collect data, *simulate* (by computer) the parts that cannot me be measured based on their favourite theory, and the small data that remain must be new physics, if the theory that underlies the simulations is right and the simulations have been coded correctly. A truly colossal amount of top level human intelligence and computing power goes in getting that right. Even more amazingly since the data collected is truly immense, the theory tells them to discard almost all of it before subtracting the estimated "invisible" bits, because most of the data is "obvious".

But that's that best we can do so far and some top physicists have their doubts sometimes that's good enough.

The particle physicists need to run their accelerators for a long time with lots of particles (luminosity measured in inverse "barns" amusingly) because they are really noisy machines, especially proton-proton colliders, and non-random patterns in the residual noise emerge only very slowly. But even those need a strong theoretical underpinning to be judged credible, and not just enormous accumulations of data and high luminosity to improve the story with the noise.

And particle physicists believe their particles to be non-reflexive, the laws of physics to be non self-modifying (self interacting fields are bad enough), and that their experiments involve negligible exogenous impulses.

[Blissex]

«noisy shifts in a very slow moving data series.»

I like that another commenter sees much the same thing I see even if I see it a bit differently and with a slightly enriched perspective.

I agree with the short term noise, but it is not just noise: there are also short term (relatively) effects of specific policy choices. In other words there is noise, but not just noise, and in some periods the noise is not even that significant (natural experiments do happen, rarely...).

Again as to "very slow moving data series" I tend to agree, again with some differences. On reflection I would rather talk of "evolving" rather than "moving", because of the self-modifying nature of the engine.

A large part of the slowness is due to the fact that, in Keynes' magically enlightening and always forgotten observation "almost all capital is long term". In that observation nearly the whole of supply-side economics, done properly, gets summarized. And Keynes was very interested in long term capital formation, strongly deploring that in speculative cultures casinos drove it. It is particularly despicable that despite his very clear thinking on the importance of long term and capital formation some (well compensated) propagandists of the right have accused him of short termism because of some of his Oxbridge-style hobbies :-).

Relatedly as to «economics is closer to religion that it is to science» the core topics of the original discipline, political economy, were capital formation and distribution of income, and these topics have been handwaved out of the Economics "religion" taught in most "successful" Departments of Economics, a situation of such (well rewarded) intellectual treason.

[Kaleberg]

The real debate should be about whether a 2.4% growth rate that applies only to the upper tail is anywhere close to a 2.6% growth rate that applies across the entire economy. Arguing that sprinters have gotten faster, because automobiles can travel at 150kph while the humans have gotten slower is not a realistic way of assessing the sport. It makes a lot of sense to do some historical research and start digging up some of Tukey's work on robust statistics.

[Kaleberg]

Actually, a better analogy would be from Mary Renault's The Last of the Wine. Bringing in a 9 foot Persian monster to clobber the best on the wrestling circuit was not a sign of the glory days of the sport.

[Luis Enrique]

Blissex,

I do not think a simple statement of the problem of omitted variable bias is fatuous hand-waving, neither is using the term "explanatory variables" in this sense:

http://www.et.bs.ehu.es/~etpfemaj/teaching/iecntx/other/glossary-intro.htm

(scroll down for definition)

** Pure statistical analysis can only discover ex-post correlations ..."

shut the front door! there's a difference between correlation and causation? who knew? Hey, maybe that has something to do with what Mertens does trying to identify exogenous variation.

"no meaning can be imputed to them."

pure hogwash. observed correlations restrict the range of plausible hypotheses, for a start.

The economy is what Soros calls "reflexive" ...

oh for God's sake, what next you're going to rediscover the Lucas critique.

"... "correlations" change with time ..."

the stability of relationships between variables is an empirical question, just because you've pulled the idea out of your backside that the past is no guide to the future don't make it so.

" a lot of noise that swamps whatever specific signal one is seeking like a needle in the haystack "

again, that's an empirical question. what does the data say?

" ... I don't regard them as explanatory variables .."

so much the worse for you. credit expansion, for example, is a perfectly plausible determinant of GDP growth, hence a candidate explanatory variable in any empirical analysis.

I do not get the impression your grip on this topic is a firm as you imagine it is. But OK, you are deeply sceptical about being able to learn anything form empirical analysis, so what, we fall back on consulting the infallible oracle that lives up your backside do we?

[Blissex]

«"Pure statistical analysis can only discover ex-post correlations ... no meaning can be imputed to them."

pure hogwash. observed correlations restrict the range of plausible hypotheses, for a start.»

If you are talking about hypothesises that extend beyond the collected data, instead of correlations within the collected data, that seems (euphemism alert) violently simplistic, and perhaps (euphemism alert) desperate optimism.

To make a statement like that you need *also* some kind of theory that says that if a correlation exists between A and B then it must emerge above noise within the sample size you have.

Otherwise, there is a case of absence of proof not being proof of absence.

The difficulty that some people have is to recognize that switching from interpolation to extrapolation is a very dangerous act, and an act of pure will and faith, and can only be supported by a well understood theory.

The more so in the case of economics where "reflexivity" and self-modifiability etc. prevent (euphemism alert) brutal jumps to conclusions.

In the next post ChrisD mentions «the growth of interest in markets as emergent evolutionary (pdf) ecosystems (pdf) that this issue has returned to the forefront of economics» which I think is rather optimistic thinking, because of the social underpinnings of research in the political economy.

«you are deeply sceptical about being able to learn anything form empirical analysis,»

That seems to be a deep misunderstanding, as I pointed out in the previous posts a lot, empirical analysis is very useful indeed, and I quoted some pretty big numbers accordingly.

But I am deeply skeptical of (euphemism alert) fatuous handwaving as to what past data can tell us about the future in the absence of a robust theoretical framework which the data support and *viceversa*.

Also, I am deeply skeptical of (euphemism alert) artful glibness about discussing past data in the presence of very large noise sources which are not even mentioned in polite discourse.

More generally, poor signal-noise ratio is a big problem of econometrics, with the small data sets and the difficulty to define what is being measured (for a classic case, unemployment, for a more challenging one, service sector productivity).

Economists get only a few chances at finding good "natural experiments".

[Blissex]

«For me the message of all this is simply that aggregate GDP data just don't permit us to draw robust inferences about policy effects. Not only is theory ambiguous but so is the data. ... I suspect the argument about what the top tax rate should be is not one that can be settled by the data.»

I hope that my previous comments show that I am in strong agreement with ChrisD's conclusion.

What I find absurd is the very euphemistic way it has been reached with the usual polite disregard for the biggest stories in UK policy for the past couple dozen years.

Which include enormous shifts in the composition of GDP and in the distribution of income and wealth also geographically between sexes and in demographics.

[Luis Enrique]

"empirical analysis is very useful indeed"

Right. Is it useful if you haven't thought about omitted variable bias? Because I still can't see why my comments set you off on one.

[SteveH]

We should argue re-distribution politics, if we argue them at all, from the idea that these policies dictate the kind of society we live in, not the impact on growth. We should argue that the Scandinavian model is more desirable than the US one, for example.

Of course, it may be (or may not be) that introducing policies that affect the wealthiest may have a short term negative (or positive) economic impact on society, this is because the wealthy can and very often do attempt to hold nations to ransom and demand that entire populations bow to their will. This will ultimately lead to disaster and should at all times be resisted.