Fed Computer Models and GIGO

Yes, you bet!  As soon as I booked the loss on my experimental (and I knew it would be premature) short position, along has come my predicted “One last big down” before we finish at new all times highs in a few weeks (says my modeling) and then we will swan song into the fall.

In Peoplenomics Thursday, I went over some of the ideas embodied in the FRB/US econometric model that the Fed uses to study implication of policy changes and what-have-you’s.  I explained what they use as model inputs and some of their publicly disclosed thinking on point.

But that led a subscriber, who we’ll just call Bob, to ask a great question ab out “Garbage In, Garbage Out” – GIGO: “The Fed is still using information generated by those with an agenda unless they are generating all their own data. Ultimately won’t they be done in by the rigged data they use?”

(Continues below)

The answer, sadly, is both yes….and no…

We have to assume the Fed would not publish ALL of its fine print for the whole world to inspect.  And when you make up an incredibly complex model as dynamic stochastic general equilibrium (DSGE) models tend to be – far too big for practical home spreadsheet analysis, there are plenty of opportunities within such models to make them “adaptive.”

Let’s simplify the concept so you don’t fall asleep and do a face-plant into the coffee.

Let’s say I figure out a way to calculate where the economy (and to a lesser degree the market) is going based on  10 key government figures.  This is the GIGO part.

Still, the numbers are widely published so as a (not-really) government agency, the Fed can’t just go calling out BS at the very government that let’s it rent America it’s money….

So here’s what happens:  Gigantic economic models are published (the Europeans have one for their central banksters, too) and everyone look at the models with knowing nods.

However, just like in tournament poker, sometime there’s a way to get a peek at “the river card.”

In econometric modeling, you can make a model “adaptive” by measuring prediction versus performance.

When back-testing a model, you might load in data for 2007-2008, for example, and then compare the predicted outcomes with the actuals.

Let’s say you predicted GDP would grow 3.5% but in reality it came in 2.5%.  The model would then (adaptively) plug in various “fudge factors” to tune the model closer to reality.  It might conclude, for example, that the wages worked and average hours worked in the BLS Employment Situation report (EmpSit) were misstated by 0.5% on hours worked and 0.7% on average hourly wages.

While the model will run just fine with the actual raw data – and will present a rosy-enough forecast that no one’s going to bitch too loudly – in the back room somewhere, the adaptive model – which while everything else is running would be piling on the simultaneous equations such that inconsistencies in the source data would be “tuned out.”

People who do a lot of modeling don’t talk much about this adaptive angle.  But it doesn’t just figure into econometric modeling; it also show up in NWP – numerical weather prediction – models, as well.

But the resultant problem (in the weather arena) is not dissimilar from the economic issues.

You maybe haven’t thought too deeply on the matter previously, but Ure’s truly spends probably far too much headspace worrying about the boundary layer in linguistics and practicum.  (If you’re a programmer, you can think of linguistics as analogous to the algorithmic level of program design and the practicum as the application and/or transport layers.)

At the practicum level, understand that there are legit needs for periodic adjustments to instruments.  It doesn’t matter whether it’s a humidity sensor in the weather business, or an economic report from the Labor Department, adjustments will need to be made.

At the linguistics level, just “adjustments” are indeed “adaptive” but don’t necessarily bias the output…or so you’d hope.

A bit of sniffing around the web and you’ll find discussion  of this in papers like this little gem from the ECMWF – which everyone knows is the European Centre for Medium-range Weather Forecast (ECMWF) Reading, United Kingdom.

Anyway, they get into exactly the “making up fudge factors” discussion that sharp-thinking subscriber Bob was after.  As the weather bias-correction discussion puts it:

“The aim of bias correction is to remove the systematic errors corresponding to the observation, the radiative transfer and pre-processing steps. These errors are called observation bias though they rarely correspond to a real statistical bias. Different bias models have been developed to reproduce the shape and magnitude of the observation bias. Whatever the bias model, its parameters (or coefficients) are usually estimated intermittently (e.g. if a new radiative transfer model is introduced) and then held static for long periods. There are scientific and technical incentives to consider an adaptive bias correction, i.e. a correction with a bias model updated at each assimilation cycle.”  [source]

All of which gets us to the answer to Bob’s questions.

1.  Yes, garbage in, garbage out is real.
2. Within tight bounds, instrumentation must be calibrated for high resolution data.
3. However, the biasing/adaptive limits MUST be carefully monitored (and audited) lest models be corrupted deliberately.
4. This is why, when I look at complex numerical models, it doesn’t matter a whit whether the numbers are economic or weather models, the distortion (and loss of integrity) via biasing really is one of the new “grand loopholes” that would-be social engineers are using to control the public mind on topics as diverse and the Economy and Climate.

Or to sum up:  Who’s correcting the Correctors?

Even more nefarious when the nature of the Correcters Correcting is as “in Ure face” as the climate-jiggering being done via smartcards in Australia or via “tweeks” to economic models at the U.S. economic policy level.

There, bet you feel better knowing how little actual effort it takes to spin your thinking, don’t you?

Add a fudge factor, drop half a degree, add or subtract average hours worked, and your whole way of viewing the world is wrecked.

Auuuummmmmm.

Pricing Extinction

Great Bloomberg article over here on how does one price a potential extinction-level event?

Be sure and watch the video on the page as it goes over NK plans to launch missiles and have them land 30-40 KM from Guam.

Kid Krazy in NK might as well put on a jersey with a big bull’s eye on it.  Not especially bright, to our way of thinking.

Producer Prices

With the Dow set to open down another 60, or so, based on global Korea jitters, we would be remiss if we didn’t mention Producer Prices just out from Labor:

“The Producer Price Index for final demand declined 0.1 percent in July, seasonally adjusted, the U.S. Bureau of Labor Statistics reported today. Final demand prices inched up 0.1 percent in June and were unchanged in May. (See table A.) On an unadjusted basis, the final demand index
increased 1.9 percent for the 12 months ended in July.

Over 80 percent of the July decrease in final demand prices is attributable to the index for final demand services, which fell 0.2 percent. Prices for final demand goods edged down 0.1 percent.

The index for final demand less foods, energy, and trade services was unchanged in July following a 0.2-percent advance in June. For the 12 months ended in July, prices for final demand less foods, energy, and trade services rose 1.9 percent.

Final Demand

Final demand services: The index for final demand services moved down 0.2 percent in July, the first decline since a 0.3-percent drop in February. Most of the July decrease can be traced to margins for final demand trade services, which fell 0.5 percent. (Trade indexes measure changes in margins received by wholesalers and retailers.) Additionally, prices for final demand transportation and warehousing services declined 0.8 percent. In contrast, the index for final demand services less trade,
transportation, and warehousing advanced 0.2 percent.

Meantime, the clock is running toward the next inflation report due out tomorrow – Consumer Prices.

Bitcoins this morning convert to about $3,391 each, as we told you to expect when the “Great Hesitation” was over.

Fleeing the Coastlines

No, not some ridiculous global subsidence talk or coming within range of NK missiles.

This is something more basic (and should we say tangible?):  Economics.

“Home Affordability Continued to Shape Migration Currents as Homebuyers Looked to Leave Expensive Coastal Cities in the Second Quarter.”

This could be read as “Good news” out here in fly-over country.

 Viral Idiocy

Shame on you for actually thinking!

Why aren’t you watching the BIG stories like “Cockroach lands on reporter’s boobs before live broadcast“?

How’s Fishing?

We’re wondering because the Mueller fishing expedition is in full-swing: FBI conducted predawn raid of former Trump campaign chairman Manafort’s home.

Since beer and fish are synonymous we’ll bet one as follows:

“Care to bet the judge issuing the search warrant was an Obama appointee?”

My sense in that Manafort’s computer would likely be taken and searched….I mean, that’s just a guess where the fish might be biting.

Here’s an interesting idea:  Could Trump pardon Manafort preemptively?

Captions We Wonder About

See the story here: https://www.thesun.co.uk/news/4214321/russian-isis-fanatic-arrest-turkey-drone-plane-crash/

See the photo caption? “An F-15 fighter jet lands at Turkey’s Incirlik Air Force Base (file image)”?

As a pilot, this kind of caption bothers me:  Don’t you think the aircraft would have its landing gear down?

Judging by the attitude and heat signature, sure looks more like a takeoff or low pass to me…and yes, I have landed on a parallel runway with an F-15 out in Tucson a few years back…I do know what they look like.

It’s the little things like this that jump out at us…useless, of course.