Arnold Kling  

Earthquake Science

PRINT
Cash for Clunkers in Action... Do Normative Economists Need E...

Mark Thoma writes,


even though earthquakes cannot be predicted, at least not yet, it would be wrong to conclude that science has nothing to offer. First, understanding how earthquakes occur can help us design buildings and make other changes to limit the damage even if we don't know exactly when an earthquake will occur. Second, if an earthquake happens and, despite our best efforts to insulate against it there are still substantial consequences, science can help us to offset and limit the damage. To name just one example, the science surrounding disease transmission helps use to avoid contaminated water supplies after a disaster...

So even if we cannot predict earthquakes, and we can't, the models are still useful for understanding how earthquakes happen.

His point is that we should approach macroeconomic analysis similarly. Many people want to dismiss macro models because they do not predict particularly well. But perhaps they still have uses.

I wish I knew more about earthquake science. For example, have there been competing theories of earthquakes? How did scientists go about choosing among competing theories? In macroeconomics, what is frustrating is that people want to subject competing theories to forecasting contests (or backwards curve-fitting contests), but such contests turn out to be indecisive. It is like a Bayesian situation in which the evidence is weak and people have strong (and differing) priors, so the data do not have much impact. (To an earlier thread, that may explain the fascination and the frustration with the Great Depression. Here is a seemingly high-impact set of data points. Can't we move some people's priors with that data?)

Anyway, what I like about the earthquake analogy is that it gets you to focus on the propagation process. Just how is it that shocks in the financial sector produce unemployment?

For what it's worth, I have taken the view that in the 21st-century Great Recalculation, the fact that part of the shock took place in the financial sector (the other big part was in housing construction) is no big deal. In my view, the financial sector was too big to begin with, and the issue with shrinking it is to put the shrinkage behind us as quickly and cleanly as possible. I think that the economic contraction is due to the need to recalculate, not to the problems with inter-bank lending and repo markets. So, when I see international trade figures decline and international lending for trade decline, my instinct is to see the causality running from the decline in trade to the decline in lending. Other folks are inclined to see the exact opposite in terms of causality, and the policy implications of our divergent points of view are pretty stark.

Continuing with the earthquake analogy, the automobile sector was like a rickety building close to the fault line. The reason it toppled is that it did not take much to topple it.

Anyway, I think it is an interesting analogy, and much more could be done with it.


Comments and Sharing





COMMENTS (19 to date)
Barkley Rosser writes:

Oh, I cannot resist. And it is known that earthquake intensities are distributed as power law distributions, skewed with fat tails, not remotely Gaussian normal...

The Sheep Nazi writes:

Interesting metaphor. I've wondered at times if the recent run of policy, that which produced the Great Moderation, was not a rough equivalent to a technology for glueing fault lines together so as to prevent small earthquakes. I guess whoever came up with something like that would be a hero, for a while.

SydB writes:

I think one problem I have with your arguments regarding macro: You malign macro and mathematical economic models with words that have little or no predictive power, little or no means to assess their descriptive abilities, and they can be interpreted and re-interpreted willy-nilly.

That's not much of a proposal if you ask me.

Barkley Rosser writes:

Oh, and just to add to the ironies, there is now a post on marginal revolution about a good forecast by some econophysicists led by Didier Sornette. It does not say so in the linked article, but their methods are drawn on earthquake models, Sornette originally being a geogphysicist who was into exactly that.

Brian writes:

Arnold,

Doesn't the recalculation model of macro rely on an implicit hydraulic model? That is, it assumes that once the labor supply adjusts to the new sets of skills that are demanded in various labor markets, there will be enough demand for labor to achieve full employment - aggregate demand takes care of itself. I would tell a different story about what is happening: we're going through a structural readjustment AND an aggregate demand shock, so we need to take aggressive monetary/fiscal policy measures to stop the economy from spiraling downward, while recognizing that this won't lead to anything like full employment because structural changes will take a long time no matter what we do. What are your objections to that account of the recession?

To Arnold Kling,

You said: "Many people want to dismiss macro models because they do not predict particularly well. But perhaps they still have uses."

And: "In my view, the financial sector was too big to begin with, and the issue with shrinking it is to put the shrinkage behind us as quickly and cleanly as possible."

First, if a "science" does not predict well, then it is useful only for spinning stories after the fact. Mark Thoma comments that earthquate science is useful for designing buildings to resist earthquakes. So, observation of earthquakes leads to better buildings, and that part of the science gives useful predictions, of how buildings will react. But, that is mostly "construction science" not "earthquake science".

Thoma says that the science of disease transmission helps after earthquakes. That is disease science, not earthquake science.

About Macro Economics. You think the "financial sector" was too big. What part of ME Theory predicted that one fact? How can you tell if a sector is "too big"? This would seem like a money maker for future economic events.

Real science starts with knowing more about what you don't know. It seems uproductive to gloss over non-predictions, to substitute personal hunches, and hope that something will someday be useful.

The huge cost of bad ME theory is that it is promoted as valid science, today. This allows governments to claim that they "know", while the ME scientists don't know.

It would be better if Macro Economists all loudly said that "we just don't know, but we are working on it", rather than give cover to politicians who are quite happy to claim that they are following valid ME advice while they play economic Bet-The-Society.

Jim Glass writes:

I've never understood why anyone would expect high predictive power in econ models given:

1) The huge lag delays in data (just one year ago oil was near $140, "oil speculators" were the hot topic, and many were warning the Fed was being slack in controlling inflation and popping the commodities bubble -- not realizing they had already been in a recession for eight months.

Casey Stengel used to say it was hard to make predictions, especially about the future. In econ it's hard to make predictions about the present.

2) The need to predict human behavior. (Physicists with their models should have this problem.)

3) The huge amount of chance and unknowns involved -- creating one whole lot of real and effective randomness.

I am continually impressed how #3 is so relentlessly understated or denied.

But consider: I think I mentioned before that NFL football game outcomes are determined 50% by chance. That's not just "close games", 50% includes a lot of one-sided games too. (This is not against "the spread", just W-L.) That 50% random outcome creates a pretty fat error band for game predictors. And even physicists who brag of having the tightest error bands in the world can't reduce it! (Any who could would quit physics, move to Vegas, and own a casino.)

Now consider: For purposes of modeling and making predictions, for which do we all have much less information about all the data and processes involved: an NFL football game or the world economy? (In football at least you know the current score in real time -- in economics you may not know it until eight months later!)

If NFL games are 50% unpredictable, and we have so much less complete information about the economy than any football game, is it really surprising to find a big error margin in economic predictions?

What's surprising to me is that in spite of overwhelming data so many people succeed at denying the big degree of randomness in both. You never hear talk of the random element in either. (A lot of football fans will punch you out if start talking about "luck" turning a big game.)

Any psychologist will tell you human beings are systematically blind to most randomness in life and see causation everywhere it isn't, for reasons of natural selection that should be apparent.

But there's a great amount of self-interest involved too. No high-priced TV football analyst is ever going to say, "Well, these games looked like a couple random outcomes." By making up just-so stories, er, analytical explanations after the fact that sound gripping and persuasive, they get raises and power. Matt Millen got to be GM of the Detroit Lions dynasty!

In economics, need I explain how much more is at stake and the powerful disincentives for any journalist, pundit, economist, regulator or politician ever to say: "We'll, nobody could've seen that coming!" Even when it is obviously true, because nobody did. You know, afterwards instead it's always "I did see it coming!" or "You should have!"

But for anyone who thinks econ forecasts are really so bad, here's an observation by Scott Sumner:

The future is determined by factors both known and unknown. Thus, if economic shocks always hit as a complete surprise, out of the blue, as a "major wrong prediction", this can be read as a good thing -- policy makers successfully adjust for all known risks, and the economy is hit only by unknowns. Which is the best possible result.

(Of course, the popular reaction will be to interpret it rather the opposite: "These idiots never see the problem coming...." So the best possible result will get those who produce it castigated and fired. But nobody said life is fair.)

Robert Speirs writes:

"...policy makers successfully adjust for all known risks ... "

Damning with faint praise, that is. Don't policy makers need to insist that unknown risks be made into known ones, if they're important to outcomes? And if they can't be quantified, shouldn't policy makers say so?

Predicting easy, unimportant risks is worse than useless if it leads to complacency and insufficient effort to assess the important risks.

Jeremy, Alabama writes:

Earthquake science is not politicized. It is *actual science*, recognized as full of unknowns, the subject of friendly but competitive research etc etc.

The Pioneer trajectory anomaly is science, too:
http://en.wikipedia.org/wiki/Pioneer_anomaly

Compare this web page with ANYTHING on global warming. 99% of global warming "research" is not science, but is political speech intended to either justify or dispute trillions in carbon taxes.

Unfortunately, the macroeconomics of which you speak is not science either. It is very hard to separate the true science from political posturing that justifies or disputes massive government intervention in the economy.

Anyway, few are interested in the true science of macroeconomics. People want the answers that current macro is unable to deliver, so macro degrades into a political shouting match about whether, say, a trillion of "stimulus spending" is good or bad.

fundamentalist writes:

Thoma commits a subtle but serious fallacy by using the earthquake analogy. Earthquake science, like all natural science, is about events external to human beings, events we must study intensely or we know nothing about them. Economics is about human behavior; it's about us. We know about us by simply thinking about us. Therefore, we should know a lot more about economics than we know about earthquakes.

And that's why behavioral economics is so important. It will eventually lead economics to the same conclusions drawn by Austrian economists a century ago.

On forecasting, here is Hayek from his Nobel Prize speech: "It has, of course, to be readily admitted that the kind of theory which I regard as the true explanation of unemployment is a theory of somewhat limited content because it allows us to make only very general predictions of the kind of events which we must expect in a given situation. But the effects on policy of the more ambitious constructions have not been very fortunate and I confess that I prefer true but imperfect knowledge, even if it leaves much indetermined and unpredictable, to a pretence of exact knowledge that is likely to be false."

In other words, the issue isn't between accurate forecasts or no forecasts at all, but between very specific but false forecasts and general but accurate forecasts.

Brian: "...we're going through a structural readjustment AND an aggregate demand shock, so we need to take aggressive monetary/fiscal policy measures to stop the economy from spiraling downward..."

Here's Hayek again: "In fact, in the case discussed, the very measures which the dominant "macro-economic" theory has recommended as a remedy for unemployment, namely the increase of aggregate demand, have become a cause of a very extensive misallocation of resources which is likely to make later large-scale unemployment inevitable. The continuous injection of additional amounts of money at points of the economic system where it creates a temporary demand which must cease when the increase of the quantity of money stops or slows down, together with the expectation of a continuing rise of prices, draws labour and other resources into employments which can last only so long as the increase of the quantity of money continues at the same rate - or perhaps even only so long as it continues to accelerate at a given rate. What this policy has produced is not so much a level of employment that could not have been brought about in other ways, as a distribution of employment which cannot be indefinitely maintained and which after some time can be maintained only by a rate of inflation which would rapidly lead to a disorganisation of all economic activity."

Hayek's speech is available at Mises.org

SydB writes:

This statement is false: "We know about us by simply thinking about us."

And this statement is dogmatic: "[behavioral economics] will eventually lead economics to the same conclusions drawn by Austrian economists a century ago."

Let behavioral economics lead where it may.

joe writes:

Read Mark Buchanan's book Ubiquity: Why disasters happen. Good layman's review of earthquake science. Very applicable book to understanding financial market disasters from a natural science point of view

fundamentalist writes:

SydB: "This statement is false: "We know about us by simply thinking about us."

Are you saying that we don't know anything about humanity and can't know anything about us? Or are you saying that thinking is no help.


More from Hayek: "Let me illustrate this by a brief sketch of what I regard as the chief actual cause of extensive unemployment - an account which will also explain why such unemployment cannot be lastingly cured by the inflationary policies recommended by the now fashionable theory. This correct explanation appears to me to be the existence of discrepancies between the distribution of demand among the different goods and services and the allocation of labour and other resources among the production of those outputs."

SydB writes:

I'm saying we think about and observe physical system and we think about and observe human systems. No big difference.

We can't just think about ourselves and come up with truth. Descartes tried it. Failed big time. (Even though he thought he'd proved the existence of god--just by thinking about it. The nerve of that guy.)

Jim Glass writes:

First, if a "science" does not predict well, then it is useful only for spinning stories after the fact.

First for me, I don't care a whit whether anyone considers economics a "science" or not. What matters is if it is effective in accomplishing its purpose. Whether or not it is a science is linquistic debate over the definition of "science", and trivially resolved by the definition one adopts. Such issues are pursued only where people have time to waste on such things -- in the halls of academia, and on the Internet.

But beyond that, if the claim is "economics is not a science because it does not make accurate preditions", the claim is ridiculous.

Economics makes a great multitude of highly accurate predictions, from the basic to the elegant and complex. Start with "lower price, increase quantity demanded" which one sees empirically confirmed every time one sees a retail store with a "SALE!" sign in its window ... and "lower price, reduce quantity supplied", which one sees empirically confirmed in the abandonment of half the rental housing in the Bronx after the city imposed rent control, and move on from there.

Saying economics can't make a particular prediction you want, such as when and how this recession will end and the next one will start -- because it is beyond the limits of computational power and is highly dependent on random and currently undetermined future events -- is like saying "Meteorology isn't a science because it can't forecast when and where the next hurricane will hit the US coast, and whether I should up my homeowners insurance".

Dr. T writes:

Earthquake science suffers from the same problem as climate science: inability to perform experiments. The scientists observe, hypothesize, and then create models in lieu of experiments. In both sciences, the hypotheses fail, the models do not work, and there is no ability to predict.

A direct correlate of the above is that no one can blame a climatologist or a seismologist for not doing well. Therefore, pseudoscientists or poor scientists can enter these fields and make a living (as the recent influx of government and eco-organization funds to global warming-supporting climatologists has shown).

I suspect that some branches of economics have similar problems, but I don't know which branches are like climatology and which are like chemistry.

Drewfuss writes:

Jim Glass writes:
"Saying economics can't make a particular prediction you want, such as when and how this recession will end and the next one will start -- because it is beyond the limits of computational power and is highly dependent on random and currently undetermined future events..."

Fine. But could current macroeconomic models be used to predict the GFC retrospectively?
Or the Great Depresssion? Or stagflationary episodes? Or the non-depression at the end of WW2?

Lets mentally seperate the data required before a prediction can be made, which is always late and changing, with the 'retro-predictive' power of the models, once that late data is available. Do they even give the right answers after the event(s)?

Jim Glass writes:

Fine. But could current macroeconomic models be used to predict ... the Great Depresssion? [etc.]

Lets mentally seperate the data required before a prediction can be made, which is always late and changing, with the 'retro-predictive' power of the models, once that late data is available. Do they even give the right answers after the event(s)?

Fine. Let's consider some of what's involved here:

1) A top econometric model running on a top computer using the data we have now about the Great Depression correctly predicted its course until 1931, when it predicted strong recovery in '31 and '32, but at which time the Fed chose to sharply increase interest rates to stop a gold outflow -- which many econ historians now consider maybe the Fed's most calamitous single decision ever. (Imagine if Bernanke had sharply raised interest rates after the Lehman failure).

With the interest rate hike, the model predicted a plunge in the economy as occurred. (See: "Forecasting the Depression: Harvard versus Yale," by Dominguez, Fair and Shapiro, _American Economic Review_ 78:5)

2) Looks good for the model -- but one can never be entirely sure, because the Depression can't be re-run 1,000 times to see of the model accurately predicted, or was lucky, or was designed after the fact to fit the Depression.

3) If a model using real-time data through 1931 redicts strong growth in 1932 and after, is it really "wrong" if unforseen and/or unforseeable random factors intervene to change the scenario?

Your question still implies the elimination of major random factors and unforseen human interventions from the process. But that's just wrong.

Again, NFL games outcomes are 50% random. Throws of the dice are all random.

Suppose the world's top scientists with the latest Cray supercomputers "model" NFL game outcomes to "predict" them with 100% accuracy, retrospectively, after the fact, using post-game data. They actually aren't "predicting" anything, they are only explaining past outcomes after all random outcomes are known -- which is a very different thing -- as proven by the fact that they still won't be able to predict W-L outcomes going forward with any more than 75% accuracy (50% correct + another 25% correct by "coin-flip")

Same with dice throws. They may be able to 100% explain throw outcomes after the fact if they get enough data on the throw from the throw. But they'll never be able to predict future throw outcomes better than by the standard dice odds. Is this a failure of physicists' ability to model dice throws?

People have to get it out of their heads that the economy is fully deterministic and thus should be fully predictible. I've explained the sources of this widespread powerful belief in what is obviously false before.

Hey, AdvancedNFLstats has an in-game, real time, "win probability calculator". Check it out and look at the huge swings that occur play-by-play. What do we have vastly more complete information about, football games we are watching or the economy?

Do you really believe there aren't major insufficiently known and truly random factors moving the economy, from "butterfly effects" to things much bigger -- like heads of state and central bankers leaders of big financial institutions getting out of bed in the morning and deciding "Oh, maybe I ought to do X today after all ..." Suppose the British regulator had allowed the buyout of Lehman?

4) Again, your question assumes that it is a valid "test" of economics that somebody be able to accurately predict the course of the economy in some detail, in spite of all the above. (As if economics does not make a myriad of other accurate predictions all the time)

Is it a valid "test" of meteorology that it be able to predict where and when and how many hurricanes will hit the US coast next season?

Is it a valid "test" of NFL team GMs that they be able to actually predict specific game outcomes and season records? With 50% chance at work they can't do it better than anyone else. Their job is to improve probable outcomes. And that's also the job of central bankers et. al.

5) And again consider Sumner's point: Economic progression is not a determisitic physical process, human intervention is involved. Central bankers et al have models of the economy, when the models indicate the economy will move "off course" their job is to intervene to take offsetting action. But their models' effectiveness is limited by incomplete knowledge and truly random factors.

If they do their job well they will gain incrementally greater knowledge over time. Thus, their best possible outcome is increasing their probability of offsetting "shocks" -- shocks should occur with declining frequency, and when they do occur they should come as a complete surprise (due entirely to unknown and random factors). That is success.

There actually is a long-term record of declining frequency of shocks. And this recession certainly arrived as a big surprise -- in spite of all the pundits who just like their NFL bretheren claim to know everything about it after the fact. Six months into it, they didn't know it had started!

Before the fact, NOBODY predicted it and its course. Not the Fed, Bank of England, ECB, Russians, Citibank, Bear Stearns, Lehman ... Anyone who predicted it could have made a *fortune* -- but who did? Buffett? Soros? Nope. And all these parties had plenty of "skin in the game" giving them incentive to predict it.

If *nobody* predicts an event like this that affects so many, giving so many such strong reason to predict it, it's a fair bet the event was, in fact, unpredictible.

Drewfuss writes:

Jim,
i don't believe NFL statistics have any relevance for macroeconomic theory or forecasting.

Comments for this entry have been closed
Return to top