What this covers

Lars Peter Hansen, a Nobel laureate in economics, sits with Russ Roberts to examine the paradox at the heart of economic quantification: models are necessarily wrong, yet abandoning them entirely is worse. Hansen defends mathematical modeling while insisting on intellectual humility about its scope. The conversation spans the 2008 financial crisis, the measurement of systemic risk, and the gap between how economists talk about their work in public and what their formal methods can actually support. Roberts challenges Hansen on whether the discipline should retreat toward history-like caution rather than claim scientific precision, and the two probe the political incentives that push economists toward overconfidence.

The discussion moves through three domains of failure and repair. First is the structure of uncertainty itself: Hansen articulates Three Components of Uncertainty—risk, ambiguity, and model misspecification—and argues that All Models Are Wrong only becomes operationally useful when you stop treating that fact as an excuse and instead build models that acknowledge what you cannot know. Second is financial regulation after Dodd-Frank: both speakers worry that designating systemically important institutions creates moral hazard and politicizes rescues, though they part company on whether the Penn Central case proves systemic risk was overblown. Third is the relationship between complexity and policy: Hansen pushes back against the idea that complicated problems warrant complicated solutions, and sketches an alternative approach to fiscal stimulus using pre-vetted infrastructure projects with timing flexibility. Throughout, the speakers return to a recurring tension—whether to quantify despite persistent uncertainty or retreat to qualified historical narrative—and Hansen consistently argues the former is the only defensible path if done honestly.

Sharpest takeaway

Hansen argues that economic models are always wrong yet remain indispensable, so the right path is not to abandon quantification but to build models that explicitly incorporate uncertainty, ambiguity, and misspecification—especially when measuring elusive concepts like systemic risk.

  • All models are simplifications and therefore wrong, but discarding quantification entirely is too extreme and harms decision-making
  • Uncertainty has three components—risk, ambiguity, and model misspecification—that must be confronted rather than ignored
  • Overconfidence in models is dangerous, and complicated problems under limited knowledge do not justify complicated policy solutions

The argument · threads6 threads · 28 claims
0.85

The productive response to uncertainty is building models that incorporate it credibly rather than abandoning quantitative methods altogether.

5 pointscentrality 5/5
  • Rather than abandoning quantitative modeling because of pervasive uncertainty (Hayek's concern about overconfidence), the productive path is to build models that incorporate uncertainty in credible ways, which helps us understand systematically when evidence is more versus less informative and serves as a sensible guide to policy.

    “just because there's lots of uncertainty out there doesn't mean that models can't be useful guides. So, I want to incorporate uncertainties inside models, in credible ways. And that's what I view as the productive step forward.”

  • Throwing away quantification completely is far too extreme and discards useful parameters for decision-making; sensible management of any firm requires some quantitative analysis, and the real challenge is to do it better and expand the tools rather than reject mathematical methods because they are not always right.

    “I think throwing away quantification completely--to me I think it's far too extreme and leads you to theory, not--almost throwing out useful parameter[?] for decision-making.”

  • A more sensible approach to fiscal stimulus would be to pre-position infrastructure projects that have passed cost-benefit analysis and have timing flexibility, executing them in downturns when labor costs are lower; this differs fundamentally from standard fiscal stimulus that simply puts money out there to get the economy going, which is a flimsy lever.

    “Suppose that government has in place a set of projects, infrastructure projects that they've done cost-benefit analysis and that they think are really important to be done eventually... maybe in a time of economic downturn... if labor costs are down... That's a different kind of perspective than people pushing for standard fiscal stimulus--let's just get the economy going by putting money out there. That lever, I agree, can be quite flimsy.”

  • Uncertainty has three distinct components: risk (a fully specified model assigns probabilities to all future events), ambiguity (uncertainty about how much weight or confidence to place across multiple competing models or unknown parameters), and model misspecification (all models are wrong, requiring robustness to forms of misspecification you cannot fully identify).

    “I like to think about there being three different components to the concept of uncertainty... once you've got the model... I like to think of that as risk... that process of assigning probabilities across models--there I think of that as a potential source of ambiguity... all the models are in some sense wrong. How do I use models in sensible ways, in ways that are in some sense robust to different forms of misspecification?”

  • Past economic analysis treated risk effects as having a second-order nature, but a critical and underexplored research area is recognizing that uncertainty may be a first-order rather than second-order impact in economic analysis, with the additional challenge of bringing uncertainty into policy discussions despite political incentives against it.

    “The so-called 'risk model' in which lots of these risk effects in economic analyses kind of take on a second-order nature to them--that's been the past perspective on this. I think thinking through why uncertainty might be much more of a first order than second order impact in economic analysis is really critical.”

0.79

All economic models are simplifications that can work well in normal times yet fail catastrophically in tail events when evidence is thinnest.

4 pointscentrality 5/5
  • All economic models are simplifications and abstractions and are therefore wrong; the essential challenge is assessing whether they are wrong in ways that matter for the conclusions, while still recognizing that quantitative models guided by empirical evidence are valuable for policy.

    “This is true of all models. Models are always wrong. It seems kind of strange to hear that initially, but there is a sense in which models are simplifications; they are abstractions. And they are wrong.”

  • Value at Risk models examine the tails of distributions where empirical evidence is thin and sparse, so they can work very well through normal times and then completely miss exactly when needed; excessive confidence built during normal times is precisely what gets you burned in low-frequency tail events.

    “Value at Risk is looking at what people call the tails of distributions. And the tails of distributions are places where the amount of empirical evidence we have is often fairly thin and sparse. And a Value at Risk model can work very, very well through normal times, and then all of a sudden just completely miss.”

  • Many who claim to have predicted the financial crisis did not predict its quantitative magnitude; the scale of the response surprised people and exposed gaps in macro models, which at the time treated the financial sector as fairly passive, prompting a rethink of the interplay between financial markets and the macroeconomy.

    “did they really predict the quantitative impact of the financial crisis?... at that point in time, a lot of the macro models had a fairly passive financial sector inside the models. And now we are trying to rethink all the models”

  • Lord Kelvin challenged Darwin's theory of evolution by calculating the age of the sun as far too young, but his models had a critical flaw—a key energy source was missing—making his confident quantitative conclusion off by orders of magnitude, illustrating the danger of too much confidence in one's own models.

    “Lord Kelvin was in this debate with Charles Darwin. He was challenging Darwin's theory of evolution, based on his own calculations... a key energy source is missing from his models, and his models actually had a mistake, or a flaw, or a gap in them that was critical”

0.65

Understanding economic patterns requires imposing formal economic structure on data, not letting data speak without coherent theory guiding interpretation.

4 pointscentrality 4/5
  • Accepting the principle of no arbitrage is something, but by itself it doesn't have a huge amount of content; the real challenge in finance is to understand the actual patterns in data—why financial markets fluctuate and why markets look more risk averse in some periods than others—through fully specified economic models.

    “Accepting the principle of no arbitrage is some--but by itself doesn't have a huge amount of content. The challenge in Finance, for me anyway, is to understand the actual patterns in the data through a fully specified economic model.”

  • Descriptive data collection like Burns and Mitchell's business cycle work can be useful for characterizing patterns, but to do something with it—understand mechanisms, conduct policy analysis, and assess the quality of evidence—you must impose more formal economic structure, since letting data 'speak' to specific questions requires that structure.

    “Once the data start speaking the specific questions then that requires more structure to it. So once you want to get beyond the fact, well, here's some pattern in the data that looks intriguing... to really understand relevant policies, macroeconomic policy analysis... that requires more formal economic analysis”

  • When empirical evidence looks anomalous (e.g., Wal-Mart driving wages down only in large cities), the right response is not to discard economic theory and 'let the data speak,' but to use an economic model grounded in incentives and market interactions to understand the disconnect, because without a coherent theory you do not know how to use the evidence and risk being fooled by spurious patterns from extensive fishing.

    “if I've got this empirical evidence and it looks anomalous, I want to go out and understand it. And I want to understand it from the guise of what I think of as an economic model that puts on place some incentives that I expect to be out there”

  • A class of climate models are highly non-linear, driven by unknown initial conditions rather than the stochastic shocks economists use, and while they produce rich output when solved, it is not easy to quantify the uncertainty coming out of them—which is critical for designing economic policy that must weigh acting now (when costs are lower) against waiting to learn.

    “There's a whole class of climate models out there that are kind of very non-linear models of a different type than economists are used to because they don't really have models of these kind of shocks hitting systems; it's all kind of unknown initial conditions... they are not models that it's really easy for me to think about how to quantify the uncertainty coming out of them.”

0.59

Formal mathematical elegance in models creates false confidence; formality adds clarity to assumptions but does not warrant trust in conclusions.

3 pointscentrality 4/5
  • Writing a model down as a formal, elegant mathematical structure that generates beautiful output adds clarity in mapping assumptions to conclusions, but it does not make the model right or warrant 100% confidence; one should not trust a model simply because it is formal.

    “just because you've written it down as a formal mathematical model... That alone doesn't mean that it's right. And that alone doesn't mean that you should have 100% confidence in it. And you shouldn't just look to this formal apparatus to say, See, it's nice and formal, therefore we should have confidence in it.”

  • Winning the Nobel Prize doesn't change the laureate as a person but suddenly grants more attention and deference, as if one's IQ jumped 40 points; Hansen sees this as an opportunity to become a more effective advocate for promising lines of research and to nurture young scholars.

    “I tell people, I'm the same person I was 8 months ago. But somehow you're treated differently, as if your IQ just jumped by 40 points or something.”

  • Politicians like to embrace economists who express their views with great confidence, which is problematic because that confidence is often not premised on solid evidence or analysis but is merely opinion stated confidently—creating a danger of false confidence in the policy arena.

    “politicians like to embrace economists who express their views with incredible confidence. That can be problematic... It is opinion. It could be stated with great confidence, but if it's opinion, it's also good to ask, are there other opinions”

0.44

Policy solutions should be simple and durable rather than complex, and major decisions like bailouts and stimulus require counterfactuals we cannot run.

5 pointscentrality 2/5
  • We never ran the counterfactual of what would have happened without the 2008 bailouts and interventions, so whether those policies were justified remains a genuinely open question; by analogy, it took a very long time to reach firm understanding of the Great Depression's causes and recovery.

    “Somehow we didn't run the counterfactual about what if we didn't do x, didn't take these steps, what would have happened?... I think it's a tremendously important question and it's one that remains open.”

  • Dodd-Frank's designation of systemically important financial institutions risks politicization, because designation carries implicit government guarantees that bad things won't happen to those firms; this undermines the market-disciplining device of letting firms fear failure, and the challenge is enabling firms to fail without so much fear attached.

    “as soon as these financial firms get designated as being systemically important, associated with that is some type of political government guarantees to not let bad things happen to them. And the incentive effects of that look to me to be quite problematic... Part of the ways to get enterprises to behave better is to at least let them think about the risk of failure.”

  • Even with meager knowledge, it may be sensible to start acting on climate change now rather than waiting for knowledge to increase, precisely because of the possibility that there could be very big or dramatic consequences—though Hansen would prefer to be able to quantify that tradeoff systematically.

    “there could be very big consequences. So maybe it's sensible to start doing things now just because of the possibility that there could be very dramatic consequences. I find that personally of some appeal, but would I love to quantify that... The answer is yes. But, I don't know how to do it yet.”

  • Even if a policy problem is genuinely complicated and poorly understood, it does not follow that the policy solution should be complicated; a complicated solution can add additional uncertainty, be counterproductive, and overreact to false knowledge, causing harm.

    “complicated problems in the face of limited knowledge don't obviously lead to complicated solution. Because the solution itself can add additional uncertainty; it can be counterproductive; it can be overreacting to false knowledge. And that can be harmful.”

  • There is nothing more durable than a short-term increase in a government program—once such programs start they are hard to stop—which is why deferring long-term budgetary consequences with the promise to fix them later is dangerous, since 'it's always later.'

    “'There's nothing more durable than a short-term increase in a government program.' Or, once you start these things they are hard to stop.”

0.40

Systemic risk measurement remains primitive and risks becoming undefined buzzword jargon; citation counts reflect fads rather than durability of research.

2 pointscentrality 3/5
  • Economics has genuine research fads driven by hot-topic chasing—if you write on the topic du jour everyone cites each other—so citation counts are a poor measure of importance; the more revealing question is how much staying power and durability a literature has 5-8 years later.

    “if you write a paper on the topic du jour, then everyone jumps in and they all cite each other. But then you have to ask: Where is this literature 5, 6, 7, or 8 years later? How much staying power and durability does it have?”

  • The measurement and quantification of systemic risk is at very primitive stages; the term was not on people's radar before the 2008 crisis and risks becoming a buzzword—'you know it when you see it'—that was invoked to justify regulation despite barely scratching the surface of understanding it.

    “The term 'systemic risk' really was not on people's radar screen prior to the financial crisis. And it only became a topic of conversation among academics and policy makers prominently, after the financial crisis.”