The global economy is teetering on the brink of disaster. So, you might think the quants and smart-alecs of the financial industry would have a giant supercomputer humming away somewhere, processing zillions of what-if scenarios per second and sucking in all manner of real-time economic data from the global markets and …
Modelling the global economy isn't rocket science?
As I think Carmack said, "Rocket science isn't rocket science".
Modelling the global (or even national) economy I think would class as hard enough to be "rocket science".
It's not just a matter of buying a big enough computer. You have to, like, write a program and shit.
Economics as a predictive science is a bunch of utter toss compared to, say, Newtonian dynamics. You can't just "simulate" it.
So no, it isn't rocket science, it's about a million times harder.
BTW, if you really don't know what it means to take a loglinear approximation around the steady state and solve it in MatLab just STFU on this topic.
BTW2, if you read Beer you'll see that there might be ways of modelling an entire economy on a computer, which aren't rocket science, but which will also run happily on a Dell.
Playing the Devil's Advocate: given that these supercomputer simulations failed to predict the current mess we're in, why should we trust them to get us out? It seems the model Bernanke created actually works at some level.
Sometimes in these situations, less is more. If this workstation model *actually works* most of the time for short term predictions of some part of the economy, then it should be used.
In my experience, simulation methods often lack transparency and usually reflect the prejudices of the people who set the model up.
It's not rocket science, sure. But the reason the financial sector isn't modelled is becuase it's Trust Based. Any monkey with even a slight knowledge of banking knows this.
How one would model something as arbitrary as "trust" in a spreadsheet is utterly unknown to me. Never mind you'd have to add in the effects of a news story. I'm pretty sure the only HBOS went under was because of a few choice whispers in the right ears.
Quote of the "black <your_favourite>day"
"There are two sides on the balance sheet: Left and Right. In the current situation there's nothing right on the Left, and nothing left on the Right."
(from Professor Leif Sjöblom, IMD, Switzerland)
The real problem is that econometric modelling is based on assumptions not facts. It is not physics. So you run your model and a month later all the employment or gdp figures get revised by a few percentage points, this happens often the adjustment of macroeconomic indicators.
So rocket science stays the same this is like doing rocket science without the facts.
Thus the fashion for graphing predictions with a probability of error so it all looks good and then Russia invades Georgia and there goes your model like a blown up rocket.
So what redux
If the equations are right it does not f***ing matter if they are computed on a calculator or abacus (or its modern equivalent from Dell).
In fact, I am more inclined to believe a model that is using well known math than a model that is feeding numbers into a huge black box.
There is a fundamental reason to it - the type of mathematical problems that describe a stock market actually have the ugly habit of becoming unstable as you increase the "precision" of the calculation (usually the timing parameter). So the usual "improvement" you get from a supercomputer is actually one of the best ways to shoot yourself in the foot. Instead of getting a less precise, but realistic result you can get rubbish due to the problem going stiff while the supercomputer is trying to crunch at it using something that is essentially a non-stiff algorithm.
This isn't so bad, at least the economic accelerator isn't just a bunch of Excel Macros like most of the financial world is based on ;-)
It just goes to show that 1GHz and 1Gb RAM is more than enough for anyone.
The really clever bit is the maths (pencil and paper stuff) to approximate a model. Once you've got that worked out, the software model is relatively easy. Did you know you can do Mandelbrot Sets in Excel, you can do garden planning in Excel. You can even write books in Excel. It's only for stuff with huge datasets (like weather, finite element analysis, or oilfield simulations) that you need a supercomputer.
Heisenbergs Uncertainty Principle?
Economics modelling...... as soon as the model is considered reasonably good the output of the model is a factor in the behaviours the model is trying to model....as others try to game the model by exploiting its weaknesses.
There is no modelling..... the suits are doing what suits do... throwing money at broken things, only this time its suits throwing money around that caused the break.
Mines the one with the dusty Matlab CD in the pocket
Oh come on
Do you have any proof that more compute power would help anybody out in any way in this situation? Or a plausible explanation of why it might? Or an implausible explanation? Anything...? That would have been good to put in the article.
HPC side of things
Here's more on the subject:
Nah, digits are too high-tech...
We need a Moniac! http://en.wikipedia.org/wiki/MONIAC_Computer
10 PRINT "WE'RE SCREWED"
20 GOTO 10
"This seems like a reasonable assumption, right?" Wrong!
>"This seems like a reasonable assumption, right?"
Uh, no, it doesn't. It's transparently derived from the unproven and baseless (at least in terms of any evidence you have shown) premise that you /need/ a supercomputer for the job.
Your assumption that blindly shovelling money into a ludicrously over-specified machine for the task at hand will somehow get it done "better" , or "faster", betrays a lack of engineering experience. Matlab on a workstation is probably a perfectly good fit to the model at hand, but you wave around the word "supercomputer" like some kind of magic talisman. This is a fine example of cargo-cult programming. I must remind you that just because it takes a woman nine months to make a baby does not imply that a team of nine women can make one in a single month.
Are you a manager by any chance?
The predictions are the reason we're in the mess we're in. Simulations are based on idealised economic models. They don't pay attention to anything so irrelevant as real life trader action. They're great when everything's shiny but irrelevant as soon as something unexpected happens.
scary thing is...
that this is actually a relief to hear after what I read last week.
'In fact, some of the most basic details, including the $700 billion figure Treasury would use to buy up bad debt, are fuzzy. "It's not based on any particular data point," a Treasury spokeswoman told Forbes.com Tuesday. "We just wanted to choose a really large number."'
Predictive economic models don't exist - especially for unusual and desperate situations with no transparency like we're in right now.
So if your predictions are going to be crap, why not produce the crap at low cost on a Dell? Those supercomputer thingies aren't cheap you know.
How the fake US economy
Now why did I read the title as "How the fake US economy ..."
Fake because these merchant bankers have been gambling using other people's money. Worse. Other people's money that they hadn't yet earned.
No, this isn't rocket science: it's much harder than that. Rocket science is based on the application of some well-known physical laws to engineering systems. Depending on the laws and the systems, that may be the sort of thing that you can usefully throw a supercomputer at.
In the case of economic modelling no one actually knows the laws, and in fact there is no real evidence that there *are* any laws in the sense a physicist would like, because the game is changing all the time. No one knows the boundary conditions either. And the whole system is (almost certainly) nonlinear, with resulting SDIC and chaotic behaviour. Finally, there are these awkward "people" who behave in ways which are often very hard to predict.
There just is no evidence that throwing computational power at a problem like this will help you at all. After all there were some organisations who had very deep pockets - plenty deep enough for quite significant computing facilities - and would have gained considerable advantage from simulations, if those simulations were any good. I'm sure they ran them: somehow the results they got failed to save them from bankruptcy.
It wasn't $700 billion
That was just the bailout figure, because the Fed took $630+ billion and gave it away earlier this week when the bailout was turned down.
So that makes a total of $1.4 trillion or thereabouts.
Maybe they are not telling us what the computers are showing, I can't imagine the results would be very nice...I can't see this bailout being anything but a band-aid
Half of the problem with the market was that they were relying on models which were not complete (and never will be).
The economy is mathematically chaotic - you might be able to get a model which can give you trends, especially during the more stable times, but they fall apart randomly, especially at tipping points and during market turmoil.
If you are going to try to base your trading on models, then only the simplest ones that you can nail to actual events should be used. Simple GENERAL models like "conflict near oil fields means higher oil prices" or "lending money to people who can't afford to pay it back is dumb". If you think any economic model can be more accurate than rules as simple as them, then you're fooling yourself.
Bigger, in this case, actually *might* be better...
"maybe the thing to do is get Gilchrist a few more grad students, a baby supercomputer *like* *he* *says* he will need to be able to pump real-time data into the model, and get us to a point where we can do some more sophisticated "what-if" modelling on economic policy. Put an earmark for the project right into the bill as it passes from the House to the Senate.
"I studied rocket science, and I can tell you - this ain't rocket science."
See, the Good Professor does think that more cycles and more data input monkeys will allow the model to be more accurate and timely.
What is not rocket science is that there is a model which appears to make sound predictions and it would be sensible to actually feed and juice it to its potential to check whether adding another trillion to the national debt is actually going to have the desired effect.
That's why Paulson didn't want any questions
He was obviously planning to use the $700bn for a PC upgrade.
Yes you could do any one of the following things and get the required results;
1) Get a thousand chimpanzees in a room to sit at desks with pencil and paper and watch what happens.
2) Do it yourself on an abacus
3) Use a Dell workstation
4) Use a supercomputer
5) Use a PS3 grid network
IMO it depends how many things you want to calculate to achieve a given result and how fast you want it all done by.
Nobel + reality = not much
Economics is the most inexact of sciences. In what other field can you be wrong 100 percent of the time, get the Nobel prize and be utterly irrelevant to the vast majority of living beings on the planet? (Oh, I just described Al Gore, didn't I.. Oops!)
Seriously, economics makes sociology look like the hardest of sciences by comparison. I'd love it if the current environment elevated the field and caused investment (ha!) in quantitative methods and allowed accountability to be introduced into economic models.
Economics is a soft science...
Economics is closer to group psychology or sociology than it is to physics.
The only place where economics generates enough hard data to require any kind of serious computing power is the analysis of historic data to generate patterns from which we may be able to derive rules or laws from.
There is a school of thought that if you can analyze those historic patterns well enough you can begin to spot them in their infancy - whether for profit, or to avert economic disaster.
Unfortunately economic events are rather unpredictable, with a vast number of factors affecting the outcome (from consumer confidence to the weather in Florida) - so any current computer simulation is going to be a very vague and inaccurate guess. This is borne out by the less than spectacular results that the quant funds produce, despite recruiting some of the world's smartest maths and physics boffins, not to mention having the multi-million dollar supercomputers.
So the only reason to throw a multi-million dollar supercomputer at the figures is to get Bernanke's Matlab toy to run a bit faster - and a wrong guess produced twice as fast is still just as wrong.
It sounds like the Financial Accelerator is more of a basic tool to test assumptions before applying brainpower to them, or to demonstrate to other people what you're thinking. This makes it little more than a very specialized variant of Powerpoint than a serious analytical tool.
You do need a computer but only to work out the planetary orbits
"Although astrology is generally regarded in mainstream Western thought as having no logical basis, advocates of the efficient market hypothesis would make the same claim with respect to most forms of fundamental analysis and technical analysis."
I assumed they still used a MONIAC machine?
Put punched holes in its pipes
These days, with AMD's Stream Computing, or Nvidia's CUDA, supercomputer-like power is potentially within the reach of people with a moderately high-end gaming machine including the right video card. Take, say, an ASUS M3A motherboard with a PCI-e 16x slot, and an ATI 4850 from any number of manufacturers...
Re:EconoSim 1.0 @AC
Brings back memories :-)
read almost any book on chaotic systems. then you might see that gigamaflops has sweet FA to do with it.
How about open-sourcing the project
Wouldn't it be nice if they would publish what they are using?
Of course, it really doesn't matter when, as posted above, they are using arbitrarily big numbers for grins and giggles, and the plan consists of lots of money and no oversight or bookkeeping.
Ride of one particular valkyrie
There goes a very curious fact about Bernanke. How uncanny that the Fed should select him to replace Greenspan. Almost as if the Fed were expecting to need his specific expertise down the line. But I wonder if they read his CV and noticed a predilection for riding in helicopters and throwing lots and lots of bits of paper out the window (not confetti).
Which recent history tells us that his model must be junk because any input data available today is hyperreal, if not actually surreal. Which is exactly where the problem is. You cannot regulate a derivatives market which is called over-the-counter but has all the transparency of in-the-back-alley transactions conducted after dark.
The begging on bended knee of $700bn is a clear indicator that no clean-up of derivatives is going to take place.
So we are really not much further on than in 1949!
See http://en.wikipedia.org/wiki/MONIAC_Computer, for the true power of a REAL (very analogue) computer model of the economy, the MONIAC (Monetary National Income Analogue Computer) "also known as the Phillips Hydraulic Computer and the Financephalograph" Created in 1949 by the New Zealand economist Bill Phillips to model the national economic processes of the United Kingdom, and built in his landlady's garage, the MONIAC used tanks and pipes to channel the flow of water (and, in a later version I believe , honey) representing the flows of capital and income round the economy. Perhaps the original could be resurrected from the Science Museum to save the Brirish economy in the absence of any serious computer modelling to extract us from the current crisis.
The city doesnt model the economy - they'd realise they're parasites if they did.
Or perhaps they do and dont care.
I have been led to believe that the treasury used three different models to predict the economy in the early 80's. Two contained keynsian thinking so , despite their greater accuracy, Thatcher had them thrown out.
Not that anyone needed a computer model to guess that what has just happened would happen.
Positive feedback 101 "its uncontrollable but we can bleed the desperate for a while"
Economics, a science?
How generous of you lot...
It's just poor man's math for now, I'm afraid. It might become a science, maybe one day, if we learn enough about it. Or not.
Excellent suggestion - a supercomputer running really mind-bogglingly complex models in real-time is clearly what we need here. If only we had such tools available to model the risks inherent in complex financial derivatives in the first place we probably wouldn't be in this mess. And maybe get some rocket scientists, too. And maths PhDs. Lots of 'em. Yeah! The more the better. In fact, we should get thousands of very clever people who are good with sums but not so hot on statistics but who think that maths and stats are the same thing, and we should get them to build a simulation that is so complex no one can possibly hope to understand it (making sure it's based on a number of poorly understood and probably wrong assumptions of course) that will give us a simple set of numbers we can plug into our economy repair machine with total confidence to fix the current crisis and save the world.
After all, that got is into this mess so surely it will get us out, right? It's like so totally obvious, but clearly everyone but me is an idiot because they can't work this out and I can. Man, I'm so awesome!
What an awesome article!!!
Seriously, though, I hadn't realised that I had wandered on to slashdot by mistake. I thought this was El Reg! But since I'm here - Macs rule and Linux freetards are losers and Windoze is beneath my notice.
And then you have an idiot screw it up
Well the bill just passed so we will see where it goes. The following actions are why the models will never work:
Insurers dive on Reid's 'bankrupt' quote
Paris because she really knows about screwing things.
Economic Modeling/Science? Bollocks!
To have a science means you need theories( so far so good) those theories then need to make predictions which are consistently borne out. You now have a theory that appears to work and will do so until a better theory comes along or it fails to make a prediction that is in agreement with the evidence.
It is not possible to do that with psychology, sociology or economics, because in these disciplines there exists no rigidity, any prediction you may care to make will be based on a set of conditions that have probably changed by the time you have finished describing them, so don't bother.
The problem with mathematicians is they depend on maths to answer every problem, and it won't.
Stick with the Dell or better still, a few honest businessmen with intuition and acumen, don't use accountants, managers, bankers or politicians, they are there to help or hinder businessmen but they are not businessmen.
Did you not make the connection to weather?
Hello, is this thing on? We don't know every where every butterfly will flap it's wings, but we can make useful climate models. Is it a little voodoo? Yes. Is it perfect? No. Does it get better with more compute time? Yes. He stated it takes hours to run a simulation. Do it faster and you can do it more times, get a better average. Or try altering a different variable, or set of variables. More data = better match historical record = better predict future = more stability for us.
Unfortunately economic events are rather unpredictable...
I have a good random number generator you can use.
This is a job for ... MONIAC!
It's even got pumps to implement feedback loops!
"Playing the Devil's Advocate: given that these supercomputer simulations failed to predict the current mess we're in, why should we trust them to get us out? It seems the model Bernanke created actually works at some level."
What happened was that they finished the program a couple of years back, ran it on the supercomputer and then got their management to take a look at the results, he said...
- "What the hell is this? Economic meltdown, credit crunch, massive forclosures, established banks going under - Go back and redo it, this is total garbage"
The rest is history.
It's not the size, it's how you use it
In neuroscience--not too far off economics in complexity-- I've done both computation-heavy simulations and also much simpler ones. But much of the heavy computational horsepower is wasted unless you are very, very confident of *all* the parameters and equations you're simulating, which probably isn't the case for macroeconomic models.
In fact, there is a disadvantage to using supercomputers on approximate models: the supercomputer lends credibility where it isn't deserved, and laypeople--nay, even journalists--might be fooled into believing the results.
model the market?
How the heck can you expect someone to model the market - a system based completely on irrationality and human behaviour?!? If someone had done that they'd be pretty rich/screwed the stock market by now. It's the Physics grads head-hunted by the hedge-funds with no sense of economics or moral consequences that got us into this mess.
The first person to try and treat economics as a science was Marx. Now it's just mirco economics they look at as a science. Not the big picture. Not only is it pretty impossible to model the global economy but they wouldn't like what they saw if they got anyway close.
Skull & Crossbones because finance capitalism is piracy.
"Playing the Devil's Advocate: given that these supercomputer simulations failed to predict the current mess we're in, why should we trust them to get us out? It seems the model Bernanke created actually works at some level."
There's the story about some shipwreck survivors on a deserted island that discovered a crate of canned food washed up on the beach.
After their intial elation over the discovery, came the grim realization that they had nothing to use as a canopener.
While they discussed various possible methods to open the cans, one fellow, an economist, kept snickering louder and louder until he was roaring with laughter.
"What's up with you?" aked one of the others, "Why aren't you helping instead of laughing at us?
The economist cleared the tears from his eyes and answered,"Because the answer to the problem is so simple and none of you see it. All you need to do is assume there's a canopener!"
And that is why we're in the mess we're in today. If the projections don't meet the economist's goals, he'll just plug in assumptions until they do, and plow blithely on.
(Two words from the past: "Voodoo Economics" It's kind of a redundancy.)
And where is that WTF? icon?
Didn't they read 'Chaos'?
James Gleick's 'Chaos' was all the rage back when I was an undergrad, in er, 1989. I think he said that global ecomony was as complex as weather...
So why are these economists using Dells? Because Mercedes-fucking-Benz logos on your executive platinum cigar-butt bins cost big dollars, baby!
Bunch of bloody B-Ark colonists if ever there were some.
Burn, because once we get them all in one building, that's what we'll do to them.
@Did you not make the connection to weather?
Quite right, the modelling similarities are many-fold, and there is a case to be made for throwing computing-power at some of the problems.
However "The weather" is subject to limits imposed by immutables like the laws of thermodynamics, etc. In economics they can and do remove these limits at will - like say, abandoning the gold standard and imposing fiat currency, or invoking an arbitrary "utility function" when results go out of band for example. It's not "science", it's voodoo, and it is a reasonable baseline position to regard anyone who describes economics as a "science" as a fucking charlatan.