Why Chartblogging Is Superior To Mainstream Macro

Orthodox-heterodox economical squabbling has i time once again erupted on the internet. As always, the mainstream declaration is that their methodologies are superior because they are based on mathematical models. My master copy expanse of involvement is the quantitative cease of economics, hence I attain non pay also much attending to some of the purely literary approaches to economics. But fifty-fifty so, I believe that mathematical too statistical methods are existence applied incorrectly past times mainstream economists, too hence whatever modelling wages they have got is largely illusionary. I illustrate this amongst a few examples, including an explanation why I believe the mainstream debate nearly the "natural rate" of involvement is largely meaningless.

This latest eruption has revolved around the non-prediction of the Financial Crisis. Noah Smith argued:

Personally I recollect DSGE techniques haven't  reaped dramatic benefits (yet). But what other option is better? When I inquire angry "heterodox" people "what ameliorate option models are there?", they ordinarily either cry some models but neglect to provide links too and hence rapidly alter the subject, or they link me to reports that are basically exactly chartblogging. [Link to Wynne Godley paper.]

This received responses, such every bit this i past times Michalis Nikiforos. Although it is clear that Noah Smith aims to travel provocative, I desire to address the spirit of his views. I volition at i time verbalize over examples of poorly applied modelling techniques inside mainstream macro.

The Debt-To-GDP Ratio And Bond Yields

During my career, I ran into models that "proved" that high debt-to-GDP ratios would enhance bond yields. I have got non withal created a formal literature survey, but I volition only assert that at that topographic point a lot of these models floating around.

I discussed the human relationship betwixt bond yields too the debt-to-GDP ratio inwards this article, too the primal nautical chart is shown above. It shows a real strong prediction that high debt-to-GDP ratios outcome inwards lower bond yields.

I have got stacked the deck inwards a way to brand those models await every bit stupid every bit possible. I discussed some nuances inwards that before article. But if at that topographic point was such a render too demand outcome that mattered, it should demonstrate upward inwards a nautical chart similar this. (It is alone possible that "high" debt-to-GDP ratios enhance the 10-year Treasury yield past times 0.20%; but when you lot catch that economist year-ahead yield forecasts are typically hundreds of footing points also high, such a minor outcome is effectively non-measurable.) The U.S. of A. of America is non a exceptional case; pretty good all developed countries amongst free-floating currencies have got similar charts.

The mainstream response to this obvious downward-sloping occupation is to engage inwards widespread statistical too modelling skulduggery to create an upward-sloping occupation for some wildly transformed variables. This nonsensical activity is viewed every bit "valid" only because statistical seek results are dumped into the papers.

Unit Roots And GDP

Professor Roger Farmer of UCLA provides a recent illustration of the over-selling of statistical techniques inwards "There is No Evidence that the Economy is Self-Correcting (Very Wonkish)" He describes the context every bit follows:

David Andolfatto asks inwards a twitter telephone commutation for evidence that deviations of gross domestic product from vogue are non-stationary. Here is the raw data. Figure 1 is the residuum from a regression of the log of existent gross domestic product on a constant too a fourth dimension vogue for quarterly US information from 1955q1 through 2014q4.

He provides a dump of diverse statistical tests which are looking for the presence of a "unit root"; he has a farther explanation inwards a linked article.

I am non interested inwards the details of the Farmer-Andofatto debate, but the chartblogging explanation is that Roger Farmer is testing to run across whether at that topographic point is a vogue for existent gross domestic product (shown every bit a log) to revert towards a fourth dimension trend, pushed past times an unknown "reversion force". (A linear fourth dimension vogue of the log of gross domestic product is equivalent to steady exponential growth.) He demonstrates using statistical tests that at that topographic point is no such vogue (under this hypothesis).

As a cute application of how i mightiness role statistical analysis - if nosotros know zip nearly how economies travel - Professor Farmer's analysis is reasonable enough. But the employment is that nosotros do know something nearly how the economic scheme works.

The nautical chart higher upward shows nominal gross domestic product growth for the U.S. of A. of America for the same fourth dimension interval used past times Professor Farmer. There is an obvious structural intermission inwards behavior betwixt the recent era (the so-called "Great Moderation") too the before catamenia which saw "stop too go" policy action, too large inventory corrections. The recent behavior shows a vogue towards relatively steady nominal gross domestic product growth inwards an expansion, punctuated amongst recessions. This form of behavior is easily produced inwards a stock-flow consistent (SFC) model, every bit nominal incomes are stabilised past times the welfare state. (There are a broad issue of models that could travel used; I am non interested inwards pinning downwards which i here.) The differing growth rates for dissimilar periods would travel explained past times changes inwards the financial opinion too income distributions, every bit good every bit the outcome of a housing bubble (which is inherently difficult to model; every bit discussed next).

The distinction betwixt this model too the uncomplicated model proposed past times Professor Farmer is that he is assuming that the stabilisation acts on real GDP, too that the reversion is towards a constantly growing vogue line. Since nosotros know that the financial opinion too income shares have got changed inwards a dramatic fashion since the early on 1980s, at that topographic point is no ground to assume that nominal growth rates would travel constant. Therefore, the failure of the statistical tests is essentially meaningless; nosotros demand to seek a model that ameliorate captures the behavior of the economy.

Professor Lars P. Syll frequently emphasises this point; run across the article "As I discussed inwards an before article, these models contain productivity shocks, which is an unfortunate legacy of Real Business Cycle models. Productivity is pro-cyclical; if activity falls, workers are less productive, every bit "fixed costs" growth to "variable costs". Therefore, the deep insight that these models provide is that falling economical activity "explains" recessions. It is no surprise that mathematical models cannot forecast recessions, but if a modelling technique cannot provide non-trivial backcasts, it is difficult to conclude that it is anything other than useless.

The Natural Rate Of Interest

I have got attempted to ignore the debate sparked Ben Bernanke's blogging nearly the natural charge per unit of measurement of interest. This is because the entire tidings is silly, which is a strong claim, too I attain non have got fourth dimension to railroad train the arguments inwards detail.

But the summary version is this: arguing that deadening growth ("secular stagnation") is the outcome of the natural existent charge per unit of measurement existence also depression is equivalent to proverb that deadening growth is existence caused past times gremlins.

The gremlin theory industrial plant similar this: although nosotros cannot run across the gremlins, nosotros tin stair out their outcome on the economy.

1. Use a smoothing technique (the Hodrick-Prescott filter) to generate an justice of potential GDP.
2. We too hence role "maximum likelihood techniques" to generate model parameters, for a model inwards which gross domestic product growth reverts to potential every bit good every bit existence influenced past times the issue of gremlins.
3. This model ignores whatever other economical variables, most notably financial policy (which acts inwards fashion which is directionally stabilising almost all of the time).
4. We too hence role a Kalman filter to create upward one's hear the fourth dimension serial of the issue of gremlins. If the economic scheme is growing slower than usual, it way that at that topographic point is an unusually high issue of gremlins.

Influenza A virus subtype H5N1 reasonable soul would ask: tin nosotros non create applied scientific discipline to stair out the issue of gremlins, or at to the lowest degree predict the issue of gremlins using other economical variables? And the response is: nosotros reject to await at that question. One mightiness too hence laid out to recollect that this model is alone circular too non-falsifiable.

Returning to the "Natural Rate" theory, all you lot demand to attain is supersede "gremlins" amongst "deviation of observed existent rates from the natural existent rate" too you lot larn the Laubach/Williams newspaper "the version published inwards 2003 is to a greater extent than frequently than non cited every bit the best way of calculating it; I have got non looked at the 2003 version to run across if at that topographic point are difference betwixt it too the 2001 version). The fact that they are looking at the difference betwixt an unmeasured too measured variable creates a difference betwixt the natural charge per unit of measurement estimation technique too the "gremlin model", but my initial thought is that this is non significant.* It would accept considerable infinite to fully explicate why I believe the ii techniques are effectively equivalent (and I am unaware of whatever enquiry that I could cite).

Since the natural charge per unit of measurement of involvement appears to travel a tautology, it is real difficult to unwrap anything constructive to say nearly a debate which revolves around its implications.

Footnote:

* The model tally is real pitiful inwards the Laubach/Williams model, too the measured existent charge per unit of measurement is relatively stable, hence that at that topographic point should non travel a important outcome on model dynamics if nosotros supersede the difference betwixt a measured too non-measured variable amongst a unmarried non-measured variable. Since the Kalman filter is a linear technique, too nosotros presumably cannot have got a negative issue of gremlins, i mightiness demand to supersede the "number of gremlins" inwards my text amongst the "deviation of the issue of gremlins from the natural issue of gremlins".

(c) Brian Romanchuk 2015