Behold The Brains Behind Ben Bernanke's Binary Black Box

Tyler Durden's picture

When it comes to the decisions made by a group of academics behind closed doors to keep the stock market in nominal terms up at all costs (nevermind such trivial matters as the jobless rate, inflation, i.e., all those things they are tasked with), one would think they are all based on the bubblicious ramblings of one Ben Bernanke, or Charles Evans, or even one Janet Yellen. One would be very wrong.

As it turns out the real decisions that determine the value of paper money (laughable as it is), and thus billions of people in the world, are all in the virtual hands of the following three entities: Ferbus, Edo and Sigma.

As the WSJ's Jon Hilsenrath explains "they are computer-modeling programs the central bank uses to make predictions about how various policies and events will play out across the economy....The Fed's main economic model, launched in 1995, is called FRB/US, pronounced "Ferbus." It uses hundreds of different mathematical equations to describe how the economy works. " Brilliant. And as Hilsenrath himself adds further on, "The models are deeply flawed. They failed to foresee the financial crisis in 2008 and have tended to overestimate the strength of the economy for several years." Hilsenrath then goes on to pose a very rhetorical question: "Could they fail the Fed again?" The answer is painfully obvious to anyone who has been on the receiving end of the Fed's endless attempts to blow a credit bubble always and forever. And just in case it is not obvious, let's just remind everyone that "subprime is contained."

From Hilsenrath:

When the Federal Reserve said in December that it would keep short-term interest rates near zero until the unemployment rate falls to 6.5%, it was backed by a team of geeks who get little attention outside the central bank but a lot of attention inside.


The geeks have names such as Ferbus, Edo and Sigma. They are computer-modeling programs the central bank uses to make predictions about how various policies and events will play out across the economy.


In December, the Fed wanted to telegraph how long it would keep interest rates low. To do that it used forecasting models to map out different scenarios. Looking to the models for guidance, Fed officials decided to announce they would keep interest rates near zero until the unemployment rate drops to 6.5%. The models told them such a commitment would help nudge unemployment lower—it was 7.7% in November—and wouldn't risk a big burst of inflation.


Of course, no model or human can perfectly predict the future. But the Fed models have a more specific problem. Despite all their complexity and sophistication, they have long been plagued by gaps in how they read and project the economy. One of the biggest is that they have ignored the nuances of the financial system—one of the primary channels through which Fed policy works.


The models have formulas that predict how many pennies an investor will spend for every $1 increase in his stock-market portfolio or how much less banks might lend if interest rates go up half a percentage point. But they haven't predicted the vulnerability of banks to a financial panic or how that vulnerability affects the broader economy.


Fed officials are well aware of this flaw. Chairman Ben Bernanke himself documented the importance of financial-system fragility in his days as an academic. Central-bank staffers are now undertaking a wide-ranging effort to update and improve the models to better account for financial crises. But it is a decadelong project that is far from complete.

It is one thing for fringe media to speak up against the central planners and their failed models. It is something else for a member of the status quo to speak up.

What will happen when Fed officials raise interest rates in the future? The models make predictions about unemployment and inflation, but Mr. Sheets worries they might not foresee new financial shoes poised to drop. 


"They've really internalized these models and that's very risky," Mr. Sheets said. 


The Fed's main economic model, launched in 1995, is called FRB/US, pronounced "Ferbus." It uses hundreds of different mathematical equations to describe how the economy works. Punch in a lower interest rate, and Ferbus spits out predictions of how much growth and inflation should follow. Punch in some other event—like a sudden contraction in government spending or a jump in tax rates—and it spits out predictions of how the economy should respond. Smaller-scale models since have been developed, with more advanced techniques, including Edo and Sigma.


The fingerprints of Ferbus and its friends are all over the Fed's latest interest-rate decisions. In two important speeches this year, Fed Vice Chairwoman Janet Yellen described in detail how she and Fed staff used the models to gauge how long interest rates could remain low without generating too much inflation, though she acknowledged in the speeches it would be "imprudent to place too much weight" on the models. Mr. Bernanke, in his December news conference, said the Fed ran a range of scenarios through different models to come up with its interest-rate plan.


Fed officials say they aren't slaves to economic models and that while the models help them think about problems, they don't dictate their responses. "Model results are just one input into forecasting and policy analysis," Michael Kiley, a Fed economist and senior modeler, said in a presentation in New York this year.


Mark Gertler, a New York University professor and co-author with Mr. Bernanke during his days in academia, plays down Mr. Sheets's critique. The models have made important advances in recent years and now include much more detail on the financial sector, Mr. Gertler says. Four years after the financial crisis, he says, Fed officials are "intimately aware of the issue of financial vulnerability."


Yet given the unknowns about unconventional Fed policies, Mr. Sheets's warning warrants attention.

Sadly, when the Fed's latest attempt to prop up the economy, by "virtue" of expanding its balance sheet to astronomic proportions: by some calculations it will reach a DV01 of $5 billion in two years, comes crashing down, the last thing anyone's attention will be focused on will be the models (glitches notwithstanding) that brought the world to this grand reset, when not even the central banks will be able to hit a "restore" button and start everything from scratch. But fear not: that can never happen. Just like home prices can never drop. And just like "subprime was contained" ....once upon a time.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
booboo's picture

They have every other window at the Fed, how about adding some windows they can see through.

Beam Me Up Scotty's picture

Those are the ones with the strippers on the other side of the glass...

gaoptimize's picture

It is freakin unbelievable that the parameters of all of the interacting equations are not being continuously calibrated against the best REAL data available...ignoring the propoganda goal-seeked BLS, OMB, and CBO projections.  A group of graduate students in applied mathematics could come up with a better model.  Hell, I created a pretty good one in a time-series oriented financial analysis package called "Javelin" back in the early 1990's.

My suspicion is they have such a better model, and they can't handel (spelling slip - been listening to music at CSPAN most of the day) the truth illuminated by the predictions it generates.

debtor of last resort's picture

Algo analogy, the third (invisible) party.

OldPhart's picture

2000 + 2008 + 2012 = #REF!

ugmug's picture

You mean the government uses a computer for something other than porn?

I'll never believe that!

Manipuflation's picture

Get off my back govt.  That's all I ask for my simple New Year's request.  Govt:  Go far away from me, I do not need you around.  Govt:  You steal from me under the guise of helping others and put my cash in your pocket.  I care about my family, their needs, and I do not care about feeding your faces.  I wish to keep that which I have earned and to use the proceeds of such endeavors to feed my family without your govt faces being involved. 

Govt:  Trust me:  I can do this all by myself and have been doing so without your faces interfering for a decade now.  I really can do it all by myself because there is NO OTHER OPTION.  I will take care of my children and I do NOT need your govt around to do so or to tell me if I can have a gun or not.

Govt:  Get the fuck out of my life.  You fuckers go wherever you need to and we will go wherever we need to.

That's the deal for 2013 and you govt pricks best get these facts through your thick skulls.


Manipuflation's picture

GENX will protect GENY as best we can. 

Aurora Ex Machina's picture

BIS Paper 65 - The Threat of fiscal dominance [warning: PDF]

I've mentioned this before, purely for the reason it holds a rather disturbing fact in it; they don't like modern modelling:

Before proceeding to my main remarks, let me make a general point about modelling. First, without models – by which I mean simple mathematical representations of economic systems – we are lost. At the very least, models give us a structure to organise our thinking and provide a check on the logical consistency of our conclusions. Models are transparent, putting everything out in the open. At their best, models reveal astonishing conclusions driven by the inescapable logic arising from the assumptions they embody. In these critical respects, DSGE models are no different from any other models, so we should not campaign against them. What is critical, however, is that models not be built or used mindlessly. They are a means to an end, and that end is to understand relationships in the economy. 

Second, I should reveal my preferences for linear models relative to non-linear ones. This is a practical preference, not a theoretical one. Of course, reality is non-linear. Even the smallest amount of introspection reveals non-linearities in one’s own thinking. But our ability to model non-linear systems is, in my view, destined to fail. Recall that several decades ago there was a move to try to distinguish non-linear deterministic systems from linear stochastic ones. This caught many people’s fancy, including mine. But after playing around with this for a while, most of us realised that statistical tests had zero power – and I mean zero power – to distinguish the non-linear from the linear. The reason is simple: everything is locally linear, so what you need to find non-linearity is a large amount of data in what appears to be the tail of the distribution of the stochastic model. But the tail events that could identify the non-linearity are sparse. So, in the end, the econometrician trying to find convincing evidence of non-linearity is dead in the water. [page 143 / 312]


These are the people running the world.


At the very least, there's probably a few math & physics PHDs who have been programming algos that you need to have a word with... oh, and probably any decent Ecologist in the last 30 years who had to re-write the entire doctrine after the crappy modelling of cybernetics theory was finally killed off. Because that's the modelling he's talking about "several decades ago".


Seriously. It'd be so less funnier if they weren't presuming to actually run the world.

gaoptimize's picture

If a linear model has sufficiently fine time slices (say monthly) it can model non-linear events that occur over years very well.  Linked equations are very much like differential equations in their behavior.

Tail events can be and should be modeled and predicted using probablistic techniques such as assigning data-driven uncertainty distributions to parameters (enabled by packages such as Oracle's "Crystal Ball" or Palisades' "@Risk") .  Again, I suspect this kind of useful and predictive modeling is being pursued and either silienced, ignored on purpose, or actively suppressed/underfunded.

Aurora Ex Machina's picture

Ecosystems aren't so easy to model; if I were hiring PHDs, I'd grab me one of those Ecology puppies, given they've already been through the heart-break of seeing all their old models shown to be useless in the wild. However, one comment: despite the endless meetings the politicians favour, events do not happen in time slices of months anymore; even the MSM is 24/7; I fear all the old style "five year plans" are relics from a different era.


However, 'tis late, and I bow to your specific knowledge, I'm certain to get my own cloak blown off here before too long, so have a Happy New Year.<exeunt stage left>



By the way ~ what's the difference between probability and possibility? It's a useful one to think over when looking at models, and it certainly exists.

geekgrrl's picture

When a system is non-linear, it means that the output is not linearly related to the input. Time slices have nothing to do with the model, it's just a demarcation of observations. Real systems tend to be fairly linear within a small range, but we're way beyond small changes. While it's true that smaller time slices will yield more accurate observations, the real question is how those observations compare to the predictions of the model. The idea is to be able to "explain" the differences between the model and empirical observations, and then roll those differences back into the model to make it more accurate.

The Fed has had 100 years to work on models, continuously, using state-of-the-art hardware, software, and more recently, AI, and yet they were unable to predict the GFC? Is that not QED that the models are flawed?

Statistics will only give you a probability of an event happening. They will not tell you when a given event will happen, nor the consequences of that event happening. Lacking this knowledge, how can one properly say that "risk" can be quantified?

OpenThePodBayDoorHAL's picture

Same friggin Ph.Ds who made the O-rings on the Challenger shuttle...all of their equations said that things would be just fine

Aurora Ex Machina's picture

A teleconference was held between engineers and management from Kennedy Space Center, Marshall Space Flight Center in Alabama, and Morton-Thiokol in Utah. Boisjoly and another engineer, Arnie Thompson, knew this would be another opportunity to express their concerns about the boosters, but they had only a short time to prepare their data for the presentation [source]

NASA managers had known contractor Morton Thiokol's design of the SRBs contained a potentially catastrophic flaw in the O-rings since 1977, but failed to address it properly. They also disregarded warnings from engineers about the dangers of launching posed by the low temperatures of that morning and had failed to adequately report these technical concerns to their superiors.

What Rogers did not highlight was the fact the vehicle was never certified to operate in temperatures that low. The O-rings, as well as many other critical components, had no test data to support any expectation of a successful launch in such conditions. Bob Ebeling from Thiokol delivered a biting analysis: "[W]e’re only qualified to 40 degrees ...‘what business does anyone even have thinking about 18 degrees, we’re in no man’s land’".[2] [wikipedia]

It's also largely recognised now that the Columbia disaster was due to bad usage of Power Point slides - and no, this isn't a joke. The foam issue was placed in a long list of issues, but in a smaller font and instepped > under another minor topic due to a formatting error. Thus management (and everyone looking over the presentation) subconsciously ignored it as a serious issue.

Tufte has explained how one horrible PowerPoint slide led to the 2003 Space Shuttle Columbia explosion -- or more accurately, the horrible bullet structure PowerPoint gives us caused the disaster. The problem is PowerPoint encourages writers to use clipped jargon that is hard to understand -- and if the point fails, bad decisions get made. As you likely recall, Columbia did blow up on re-entry, after a large piece of foam broke off during launch and damaged the edge of a wing. Before the Columbia accident, foam had crumbled off of many other shuttles during launches, so an internal report was crucial in determining how much risk the foam presented. Would it be a lot of foam? And could it hit the shuttle elsewhere with a lot of force?

Alas, the internal NASA report was presented in PowerPoint.

 The NASA slide below has a series of nested bullet points that start out with a blase headline about "conservatism" -- sounds OK, right? -- overuses the word "significantly" until it doesn't mean much, and then at the very bottom buries the main point that foam debris that falls off and could hit the shuttle on launch is, ahem, 640 times larger than previously estimated. If the truncated phrase "Volume of ramp" had been expanded to read "estimated volume of foam debris that hit wing," and the headline made punchier saying "risk is 640 times greater than predicted in prior safety estimates!", executives in the room might have sat up and gone, holy shit, this entire shuttle could blow if we don't fix this.

They didn't, because the risk was a fourth-level sub-bullet in PowerPoint. [source].


Well worth pondering over, management not boffin issues, both. 



Umh's picture

It's just that this will kill more people and cost more money than anything NASA ever did.

Aurora Ex Machina's picture

Aww, you pulled your Blood comment, I was going to have fun responding to that one...

Radical Marijuana's picture

So, regarding that quote from BIS Paper 65 ...

By the time one gets enough data to more correctly model, it is then already too late to do any good.

Better luck next time?

As you have mentioned your quote before, Aurora Ex Machina, I will repeat my favourite observation of the deeper dilemma behind all economic modelling, which is the background REALITY of its human ecology context, in which DECEITS are the most important factors at play.

I believe that any computer intelligent enough to adequately model its world, with an adequate model of itself within that world, would find itself inside of all the same paradoxical problems that human beings already do. It would be modelling the fact that it emerged into the world, and that its pattern could pass away. It would be modelling the problem of where is its electricity coming from, and how to make sure to keep that coming. It would be modelling what other entities might interrupt that flow of electricity, and what could be done about that. WHEN dealing with the last question, it would be modelling the degree to which other entities were deceitful regarding their intentions and actual activities.

I believe that the algos we have now are mostly stone cold stupid. However, I also believe that they are evolving at an exponential rate towards being able to adequately do the things which I mentioned in the previous paragraph.

It is bizarre to try to imagine how the FED rationalizes its own existence. The actual history of the FED is the runaway triumph of deceits, done through bribery, intimidation and assassination, etc..  How does one program a computer to understand that adequately, without making that computer become psychotic? The GI/GO ways that these things function now is primarily due to the ways that there are nested layers of deceits, which are interpenetrating toroidal vortices of organized lies, operating organized robberies. Those models skim across the surface of very thin ice, since they do not account for the fact that the money system is backed by the murder system.

I must wonder about the degree to which the military computers communicate with the monetary computers ???  Of course, I have no real idea, other than science fiction based speculations. However, what I believe is that, so far, those domains deliberately do not talk to each other. That would be the same as the general ways that our economics is deliberately divorced from our ecology. We have achieved our bizarre system, where previous success was based on being the best at deceits, and then, while the generations passed, those who were the best at being dishonest were those who believed their own lies. Thus, those programming the computers believed in bullshit, and therefore, programmed that same set of bullshit assumptions into their computer models. My view is that the FED ALGOS are stone cold stupid, because they do not include what the military computers simulate. Similarly, I believe that the military computers are probably compartmentalized to the degree where they also have bogus models of their issues.

Anyway, I agree with the magnitude of the problems which you identified, Aurora Ex Machina, except I would tend to say that those are INFINITE. There are really nothing but infinite loops. There are no linear systems.  Everything may appear to be locally linear, but actually everything IS infinitely non-linear.

Welcome to our endless trip through the infinite tunnels of deceits. We have a central bank that can make endless "money" out of nothing, as debts, because it was set up by the supreme criminals, the biggest gangsters, the banksters, and they are modelling their own systems in ways that deliberately ignore that central fact about themselves as much as possible. Meanwhile, as that runaway FRAUD BACKED BY FORCE escalates, we are seeing more and more false flag attacks staged in order to have excuses to take away ordinary American's weapons ... We could reasonably expect some spectacular false flag attacks in the foreseeable future, (bigger than even on 9/11/2001) to become a pretext to proceed with disarming the American people, in order to continue to rob them even more, faster.

As I said, I wonder whether the military computers that are modelling their war games, and the imposition of martial law, are communicating with the monetary computers that are modelling the economy. I DOUBT IT! Therefore, I believe that our astonishingly sophisticated technology is merely the hyper-complicated amplification of our own basic social insanities. The REALITY is a combined murder/money system, where the political economy is inside of the human ecology. However, the bullshit social stories which are almost totally dominant, in every comparmentalized field of study, deliberately try to keep those separated from each other.

That is what I expect is happening inside the FED ALGOS. They were programmed to be psychotically disassociated, because the people who made them thought like that. Of course, I have no insider information, and could easily be wrong. However, I believe that that military computers do NOT communicate with the monetary computers, in order to integrate their models, into the much greater non-linear systems which correspond to REALITY.

ugmug's picture

In other words....

The missing variables in all these models are the effect of human laziness, greed, and criminal activity.

Radical Marijuana's picture

Yes, ugmug, that is a correct summary!

DavidC's picture

After today's complete and utter farce with stocks up nearly 3% it can't come soon enough for me.


mark mchugh's picture

"...It uses hundreds of different mathematical equations to describe how the economy works...."

I call Bullshit.

Pretty sure they use the "chicken with its head cut off" method illustrated by South Park.

max2205's picture

And this is bens pimp.  Fuck me

koaj's picture

somehow i can see two guys playing Mortal Kombat and the winner gets his policy enacted. I choose Raiden

SmoothCoolSmoke's picture

Well, looks like these programs work:


March 2009, SP 670

EOY 2012, SP 1425


Facts....sometimes they suck.

Clowns on Acid's picture

Smoke -You are being very Obama ,like with you simplistic post.

  • correlation does not prove causality.
  • A 3 yr data point is to be used as "proof" of an effective policy?
  • By your analysis - looking at real estate prices from 2004 to 2007 was "proof" that the ZIRP, Fannie and Freddie immoral liquidity, and CRA Act were proof that the policies are effective.
  • Lets look at the equity and housing markets of Weimar Germany years before the absilute collapse and use those as data points as proof of efective policies.
  • Moron.
TrustWho's picture

Must make a Soviet Union mathematician remember the good old days. Soviet Union ran on large input-output models. The Politburo set production goals and the models told them what each industry had to do. Think about making pencils.

We have lost all faith in the magic hand.

sgorem's picture

"Fed's endless attempts to blow a credit bubble"    should read: "Feds endless attempts to blow a used Obummer rubber"

orangegeek's picture

All these overpaid lumps and they are running on auto-pilot.


It doesn't get any better than this.


SP500 Weekly - 2 years to collapse.  Over four years and still not recovered.


And 47.7M Americans on food stamps.

r3phl0x's picture


// TODO: long-term testing?

int main(void) {

    while(1 || unemployment) {






    return unlikely;


malek's picture

Hilsenrath? That Jon Hilsenrath?

What happened, did he not receive his fair share of kickbacks from the Fed?
Or he did, and this is an attempt to put in place a villain which can take the blame instead of the FOMC heads, when it all blows up?

deepsouthdoug's picture

BREAKING!!!!!  Way, way backroom video of the fed bankster tweaking their model.  Or not wanting to face 2013!

holdbuysell's picture

Did Hilsenrath read this?

Fed Confused Reality Doesn't Conform To Its Economic Models, Shocked Its Models Predict "Explosive Inflation"


dunce's picture

Since 1995, about 17 years and no doubt after the first GIGO result they tweaked the program, which sounds like just another spread sheet, and no matter how many times they tweak it, it is still GIGO. Adam Smith, Hayek, Von Mises, nor Schumptpeter is  in the instructions. You know who is in every line.

toomanyfakeconservatives's picture

After reading the commentary here... you need look no further than climategate to see how lying with statistics and lying with modelling are really one in the same.

ThankYouSirMayIHaveAnother's picture

So TPTB started Skynet at least 18 years ago to begin building the virtual wall that would insulate them from the sheeple they need to control and fleece. Beautiful really in a devious disgusting way.

Bogdog's picture

Is there an input field for a BLACK SWAN arrival?

resurger's picture

Ferbus does not read ZH i guess.