The problem with economic models
When students of statistics are introduced to creating and interpreting models, they are introduced to George Box’s maxim:
All models are wrong, some are useful.
It’s a call for humility and perspective, a reminder to superimpose the messy world on your clean lines.
If you’d like an essay-formatted version of this article to read or share, here’s a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:
https://pluralistic.net/2023/04/03/all-models-are-wrong/#some-are-useful
Even with this benediction, modeling is forever prone to the cardinal sin of insisting that complex reality can be reduced to “a perfectly spherical cow of uniform density on a frictionless plane.” Partially that’s down to human frailty, our shared inability to tell when we’re simplifying and when we’re oversimplifying.
But complex mathematics are also a very powerful smokescreen: because so few of us are able to interpret mathematical models, much less interrogate their assumptions, models can be used as “empirical facewash,” in which bias and ideology are embedded in equations and declared to be neutral, because “math can’t be racist.”
The problems with models have come into increasing focus, as machine learning models have increasingly been used to replace human judgment in areas from bail assessment to welfare eligibility to child protective services interventions:
https://memex.craphound.com/2018/01/31/automating-inequality-using-algorithms-to-create-a-modern-digital-poor-house/
But even amidst this increasing critical interrogation of models in new domains, there is one domain where modeling is all but unquestioned: economics, specifically, macroeconomics, that is, the economics of national government budgets.
This is part of a long-run, political project to “get politics out of budgeting” -a project as absurd as “getting wet out of water.” Government budgeting is intrinsically, irreducibly political, and there is nothing more political than insisting that your own preferences and assumptions are “empirical” while anyone who questions them is “doing politics.”
This model-first pretense of neutrality is a key component of neoliberalism, which saw a vast ballooning of economists in government service — FDR employed 5,000 economists, while Reagan relied on 16,000 of them. As the jargon and methods of economics crowded out the language of politics, this ideology-that-insisted-it-wasn’t got a name: economism.
Economism’s core method is reducing human interaction to “incentives,” to the exclusion of morals or ethics — think of Margaret Thatcher’s insistence that “there is no such thing as society.” Economism reduces its subjects to homo economicus, a “rational,” “utility-maximizing” automaton responding robotically to its “perfect information” about the market.
Economism also insists that power has no place in predictions about how policies will play out. This is how the Chicago School economists were able to praise monopolies as “efficient” systems for maximizing “consumer welfare” by lowering prices without “wasteful competition.”
This pretense of mathematical perfection through monopoly ignores the problem that anti-monopoly laws seek to address, namely, the corrupting influence of monopolists, who wield power to control markets and legislatures alike. As Sen John Sherman famously said in arguing for the Sherman Act: “If we will not endure a King as a political power we should not endure a King over the production, transportation, and sale of the necessaries of life.”
https://marker.medium.com/we-should-not-endure-a-king-dfef34628153
Economism says that we can allow monopolies to form and harness them to do only good, enforcing against them when they abuse their market dominance to hike prices. But once a monopoly forms, it’s too late to enforce against them, because monopolies are both too big to fail and too big to jail:
https://doctorow.medium.com/small-government-fd5870a9462e
Today, economism is helpless to do anything about inflation, because it is ideologically incapable of recognizing the inflation is really excuseflation, in which monopolists blame pandemic supply shocks, Russian military belligerence and supposedly overgenerous covid relief programs for their own greedy profiteering:
https://pluralistic.net/2023/03/11/price-over-volume/#pepsi-pricing-power
Mathematics operates on discrete quantities like prices, while power is a quality that does not readily slot into an equation. That doesn’t mean that we can safely discard power for the convenience of a neat model. Incinerating the qualitative and doing arithmetic with the dubious quantitative residue that remains is no way to understand the world, much less run it:
https://locusmag.com/2021/05/cory-doctorow-qualia/
Economism is famously detached from the real world. As Ely Devons quipped, “If economists wished to study the horse, they wouldn’t go and look at horses. They’d sit in their studies and say to themselves, ‘What would I do if I were a horse?’”
https://pluralistic.net/2022/10/27/economism/#what-would-i-do-if-i-were-a-horse
But this disconnection isn’t merely the result of head-in-the-clouds academics who refuse to dirty their hands by venturing into the real world. Asking yourself “What would I do if I were a horse?” (or any other thing that economists are usually not, like “a poor person” or “a young mother” or “a refugee”) allows you to empiricism-wash your biases. Your prejudices can be undetectably laundered if you first render them as an equation whose details can only be understood by your co-religionists.
Two of these if-I-were-a-horse models reign invisibly and totally over our daily lives: the Congressional Budget Office model and the Penn Wharton Budget model. Every piece of proposed government policy is processed through these models, and woe betide the policy that the model condemns. Thus our entire government is conducted as a giant, semi-secret game of Computer Says No.
This week, The American Prospect is conducting a deep, critical dive into these two models, and into the enterprise of modeling itself. The series kicks off today with a pair superb pieces, one from Nobel economics laureate Joseph Stiglitz, the other from Prospect editor-in-chief David Dayen and Rakeen Mabud, chief economist for the Groundwork Collaborative.
Let’s start with the Stiglitz piece, “How Models Get the Economy Wrong,” which highlights specific ways in which the hidden assumptions of models have led us to sideline good policy (like increasing spending during recessions) and make bad policy (like cutting taxes on the rich):
https://prospect.org/economy/2023-04-03-how-models-get-economy-wrong/
First, Stiglitz sets out a general critique of the assumptions in neoclassical models, starting with the “efficient market” hypothesis, that holds that the market is already making efficient use of all our national resources, so any government spending will “crowd out” efficient private sector activity and make us all poorer.
There are trivially obvious ways in which this is untrue: every unemployed person who wants a job is not being used by the market. The government can step in — say, with a federal jobs guarantee — and employ everyone who wants a job but isn’t offered one by the public sector, and by definition, this will not crowd out private sector activity.
Less obvious — but still true — is that the private sector is riddled with inefficiencies. The idea that Google and Facebook make “efficient” use of capital when they burn billions of dollars to increase their surveillance dragnets is absurd on its face. Then there’s the billions Facebook set on fire to build a creepy dead mall it calls “the metaverse”:
https://www.youtube.com/watch?v=EiZhdpLXZ8Q
Then we come to some of the bias in the models themselves, which consistently undervalue the long-run benefits of infrastructure spending. Public investments of this kind “yield very high returns,” which means that even if a public sector project reduces private sector investment, the private investments that remain produce a higher yield, thanks to public investment in a skilled workforce and efficient ports, roads and trains.
A commonplace among model users is that we must make “The Big Tradeoff” — we can either reduce inequality, or we can increase prosperity, but not both, because reducing inequality means taking resources away from the business leaders who would otherwise build the corporations whose products would make us all better off.
Despite the fact that organizations from the OECD to the IMF have recognized that inequality is itself a brake on economic growth, fostering destructive “rent seeking” (seen today online in the form of enshittification), the most common macroeconomic models continue to presume that an unequal society will be as efficient as a pluralistic one. Indeed, model-makers treat attention to inequality as an error bordering on a mortal sin — the sin of caring about “distributional outcomes” (that is, who gets which slice of the pie) rather than “growth” (whether the pie is getting bigger).
Stiglitz says that model makers have gotten a little better in recent years, formally disavowing Herbert Hoover’s idea of expansionary austerity, which is the idea that we should cut public spending when the economy is shrinking. Common sense tells us that this will make it shrink faster, but expansionary austerity (incorrectly) predicts that governments that cut spending will produce “investor confidence” and trigger more private investment.
This reliance on what Paul Krugman calls the “Confidence Fairy” is tragically misplaced. Hoover’s cutbacks made the Great Depression worse. So did IMF cutbacks in “East Asia, Greece, Spain, Portugal, and Ireland.”
Expansionary austerity is politics dressed up as economics. Indeed, the political ideology subsumed into our bedrock models has caused governments to fail to anticipate crisis after crisis, including the 2008 Great Financial Crisis.
The politics in modeling are especially obvious in the process running up to the Trump tax cuts (as is often the case with Trump, he draws with a fisted crayon where others delicately shade with a fine pencil, making it easier to see the work for what it is) (see also: E. Musk).
Axiomatic to model-building is the idea that if you tax something, you’ll get less of it (“incentives matter”). The theory of corporate tax cuts goes like this: “if we tax corporations for the money they might otherwise use to build new plant and hire new workers, they will do less of those things.”
That’s a reasonable assumption — which is why we don’t tax companies on capital investments and their payrolls. These expenses are deducted from a company’s profits before it calculates its taxes. Corporate taxes are levied on profits, net of spending on labor and plant.
But when the CBO modeled the Trump cuts, it operated on the assumption that the existing tax system was punishing companies for hiring people and expanding operations, and thus concluded the reducing taxes would lead to more of these activities. On that basis, the tax cuts were declared to be expansionary, a means of driving new private sector activity. In reality, all they did was create more profits, which rich people used to bid up the prices of assets, creating a dangerous asset bubble — not investment in productive capacity.
In “Hidden in Plain Sight,” the other Prospect piece that dropped today, Dayen and Mabud tell us just how wrong the models were about the Trump cuts:
https://prospect.org/economy/2023-04-03-hidden-in-plain-sight/
The CBO predicted that the cuts would drive a 0.7% increase in GDP over a decade, while Penn Wharton predicted 0.6–1.1% growth. Both were very, very wrong:
https://www.npr.org/2019/12/20/789540931/2-years-later-trump-tax-cuts-have-failed-to-deliver-on-gops-promises
Despite the manifest defects of these models, we still let them imprison our politics. When Elizabeth Warren proposed a 2% wealth tax on assets over $50m, she asserted that this would reduce billionaires’ fortunes by $3.75T over 10 years, but the Penn Wharton model knocked $1T off it, and declared that the real impact of the policy would be a reduction in investment, depressing long-run growth. The politics of a wealth tax are sound — the kind of politics that wins elections and restores faith in democracy — but the economism of models sweeps the proposal off the table and into the dustbin of history.
The Penn Wharton model simply refuses to factor in absolutely key aspects of a wealth tax plan, from the impact of increased enforcement to the economic benefits of universal child care, increased education funding, student debt cancellation and other programs that could be enacted with the fiscal space opened up by reducing billionaires’ spending power.
The Warren policy is rare because we got to hear about it — through a national election campaign — before it was strangled by the model-makers. More often, proposals like this are quietly snuffed out even before they’re introduced to the legislature, when they are run through the model and told Computer Says No.
Modeling isn’t intrinsically bad, but “all models are wrong” and what determines whether a model is useful are the politics of its assumptions. Economism insists that there are no politics in model-making, which creates unfixable flaws in its models.
One core political assumption in economism’s models is that government shouldn’t exercise power to produce outcomes — rather, it should “nudge” markets with incentives (which, we are constantly reminded, “matter”). This means that we can’t ban pollution — we can only offer “cap and trade” systems to incentivize companies to pollute less. It means we can’t do Medicare For All, we can only “bend the cost-curve” with minor interventions like forcing hospitals to publish their rate-cards.
Economism — and its institutions, like the CBO — are “short-run Keynesian and long-run classical” — that is, they only consider the benefits of public spending over the shortest of timespans, and assume that these evaporate over long time-scales. That’s exactly backwards, as anyone who’s ever traveled on a federal highway or visited a national park can attest:
https://prospect.org/politics/congress-biggest-obstacle-congressional-budget-office/
All of this is worsened by politicians, who exploit the primacy of economism to attack their adversaries. When the CBO or Penn Wharton release a report on a policy, they often wrap their conclusions with caveats about uncertainties and ranges — but these cautions are jettisoned by opportunistic politicians who seize a single headline figure and use it as a club against their opponents.
In the coming week, the Prospect will run deep dives into the defects of CBO and Penn Wharton, along with other commentary. It’s very important work, throwing open the doors to the inner sanctum of economism’s sacred temple. I’ll be following it eagerly.
Have you ever wanted to say thank you for these posts? Here’s how you can: I’m kickstarting the audiobook for my next novel, a post-cyberpunk anti-finance finance thriller about Silicon Valley scams called Red Team Blues. Amazon’s Audible refuses to carry my audiobooks because they’re DRM free, but crowdfunding makes them possible.
Image:
bert knottenbeld (modified)
https://www.flickr.com/photos/bertknot/8375267645/
CC BY-SA 2.0
https://creativecommons.org/licenses/by-sa/2.0/
[[Image ID: A Tron-like plane of glowing grid-squares. Two spherical cows roll about on the plane, chased by motion lines. The gridlines are decorated with complex equations from the Penn-Wharton Budget Model.]]
82 notes
·
View notes
This day in history
Monday (October 2), I'll be in Boise to host an event with VE Schwab. On October 7–8, I'm in Milan to keynote Wired Nextfest.
#10yrsago Asteroid named after Randall “XKCD” Munroe https://blog.xkcd.com/2013/09/30/asteroid-4942-munroe/
#10yrsago HOWTO make $200,000 off of the above-inflation USPS postage hike https://qz.com/128329/a-step-by-step-guide-to-profiting-off-the-3-cent-hike-on-us-postage-stamps
#10yrsago Talking about Ron Howard’s Haunted Mansion album with the Comedy on Vinyl podcast https://stolendress.com/comedyonvinyl/podcast_episodes/COV_278.mp3
#10yrsago The Big Lie: how polygraph companies convinced the US government to use pseudoscience on job applicants https://www.wired.com/story/inside-polygraph-job-screening-black-mirror/
#5yrsago California bans all-male corporate boards https://www.cnn.com/2018/09/30/business/california-requires-women-board-of-directors/index.html
#5yrsago The history of a Zorklike programming interpreter is a tale of language, art, code and literature http://www.emshort.com/ifmu/inform.html
#5yrsago Brazilian electoral upset possible as fascist loses ground to surging leftist https://www.theguardian.com/world/2018/sep/30/huge-protests-in-brazil-as-far-right-presidential-hopeful-jair-bolsonaro-returns-home
#5yrsago Chinese students, made to study Communism, are rising up for workers’ rights https://www.nytimes.com/2018/09/28/world/asia/china-maoists-xi-protests.html
#5yrsago “Like Lord of the Flies”: working at the TSA really sucks https://federalnewsnetwork.com/congress/2018/09/report-blasts-tsa-leadership-for-toxic-culture-blames-it-for-high-attrition-poor-morale/
#5yrsago Large scale psych study identifies “homo economicus” as the source of all evil in the world https://psycnet.apa.org/doiLanding?doi=10.1037%2Frev0000111
#5yrsago Justin Trudeau’s NAFTA concessions include 20 year copyright extension https://www.michaelgeist.ca/2018/10/from-copyright-term-to-super-bowl-commercials-breaking-down-the-digital-nafta-deal/
#5yrsago All levels of UK government have been paralysed by Brexit https://www.bloomberg.com/news/articles/2018-09-19/brexit-has-brought-britain-to-a-standstill#xj4y7vzkg
#5yrsago California’s Net Neutrality bill is now law https://arstechnica.com/tech-policy/2018/09/california-governor-signs-net-neutrality-rules-into-law/
#1yrago How Palantir will steal the NHS https://pluralistic.net/2022/10/01/the-palantir-will-see-you-now/#public-private-partnership
3 notes
·
View notes