The Problem With Predictive Modeling

MD

or "Why We Can't Know, But Soon Will"

One particular segment of a predictive analysis at The Oil Drum caused me to pause in a momemt of great epiphany.

My great "eureka moment" follows. Let's see how it stands up to scrutiny.

The predictive models I have seen to date look at discovery and production history and attempt to use various mathematical models to predict future decline rates. The above referenced analysis does the same, but the segment that caused me to pause was right here:

"Here's a table that give the average growth rates during each of those intervals (as computed from the slopes of the linear fits).

1860-1891 13.9%
1891-1929 7.9%
1929-1942 3.9%
1942-1973 7.4%
1973-1979 2.1%
1979-1983 -4.0%
1983-2004 1.5%

Many of these dates have some economic or political significance."
(emphasis mine)

Thus my conclusion:

Production and Consumption are driven primarily by economic and polical factors as demonstrated by steady state growth(decline) rates that follow each cusp event.

Therefore:

Future production and consumption cannot be predicted by application of mathematical models based on discovery and production history!

To support my conclusion, I point to the 1.5% growth from 1983 to date. Given that growth has been driven by economics and politics, then the current fluctuation in energy markets is simply explained by the economic machine hitting up against production limits; which until recently have been met through energy providers mad scramble to eke out next year's 1.5%.

In simple terms, welcome to the peak of oil production my friend!

So what does that mean next?

I don't know, except that we are clearly at a cusp where a new growth(decline) rate is taking shape that is being driven more by geological factors than economic or political factors.

If it were possible to integrate politics and economics into predictive modeling, maybe our crystal balls would work better.
_________________
"Don't ever become a pessimist... a pessimist is correct oftener than an optimist, but an optimist has more fun, and neither can stop the march of events."
Robert A. Heinlein

md@peakoil.com

netfind · Coal Joined: Nov 13, 2005 Posts: 11 Location: The Ozarks

The economics factors will either overproduce oil fields above the math curve or delay production below the math curve. But if you superimpose the math curve on all this, you get some predictive value when you consider that the final area under the actual production curve with all the deviations has to wind up being the same as the area under the math curve. So the over and under deviation areas have to be about equal.

MicroHydro · Light Sweet Crude Joined: Apr 10, 2005 Posts: 1286

AirlinePilot

And then we have the unknown reserve situation in Kuwait coupled with an even hotter fire in Nigeria to contend with. I hold little hope that things will remain on an even enough keel for smooth sailing.

I'm actually not as worried about Iran at the moment as I am with the seemingly increasing terrorist threat to infrastructure. I'm talking short term but just tonight there was video of something ablaze in Iraq(CNN). God forbid some extremist group figures out how to block up the Straight of Hormuz or take out a refinery or two.

pup55 · Expert Joined: May 26, 2004 Posts: 3681

MD

backstop · Last edited by backstop on Sun Feb 19, 2006 12:40 pm; edited 2 times in total

MD,

While I'd concur with MicroHydro's agreements, I rather think that the term "Predictive Modelling" is a misnomer.

What a model can do effectively is to project probabilities from highly constrained information;
it cannot define, and thus predict, the outcome, as there are a swathe of unpredictable factors, some known and some never before seen,
that are certainly going to affect the experience of depletion.

Given that these unpredictables are a constant possibility, they perhaps deserve to be classed as K factors,
not least because that was Katrina's initial.

In this light with regard to Pup55's question:

"Can the argument be made that the production of one well can be predicted by mathematical modeling ?"

I'd differ with your (qualified) proposal that it can, since to predict production from a given well
there would be a need for various K factor data including the geographic -

i.e. if the well in S. Iraq or in Hampshire UK, how stable will conditions be during the production period,

and ideolgical -
e.g. concerning both the discount rates to be applied to define the rate of well-forcing,

(for those without my mere graduate level economics the discout rate is effectively the rate at which future income is discounted compared to present income,
and as such it forms a unique measure of our society's present cultural depravity and ongoing impoverishment)

and also the international price per tonne of CO2 emissions-permits under the operational UNFCCC treaty of the day,
which is likely to be a prime factor in dating a well's loss of economic viability.

So, while I'd well agree that the probabilities of a given well's production can, by excluding K factors, be projected by maths modelling,
predicting that production can not.

And this is not to knock Projective Modelling in the slightest, as its is plainly a critically important tool.

Regards,

Backstop
_________________
"The best of conservation . . . is written not with a pen but with an axe."
(from "A Sand County Almanac" by Aldo Leopold, 1948.

pup55 · Expert Joined: May 26, 2004 Posts: 3681

MD

netfind · Coal Joined: Nov 13, 2005 Posts: 11 Location: The Ozarks

A single well can't be modelled by any Hubbert type math. Even a massive field of wells can't be accurrately done by a Hubbert bell curve. This is clear if you just look over a collection of production histories of the major mature fields. They hardly ever follow any smooth bell curve. A lot of Hubbert's detractors use this as evidence that he was a complete screwball. But Hubbert himself wrote in his early work that single fields should not be expected to be modelled by his process because of individual field rock characteristics. But he contended that a large aggregate of geographically mixed fields typically cancells out these rock differences, as for example the U.S. lower 48, which did, in fact, form a pretty smooth logistic math curve per his model.

backstop

pup55 · Expert Joined: May 26, 2004 Posts: 3681

Thinking kind of creatively at the moment...

You can call this the "as bad as things ever got" model. Examples: Iran, after the revolution, still managed to produce a milliion or so barrels per day. Saudi maybe it's 5 million. For the Soviet Union, maybe it's what they were producing after their revolution, 4-5 million. Mexico and Indonesia probably have similar curves. In other words, at some point in time, not necessarily the same point, this was as bad as things ever got.

So the base model would be some gaussian/logistics simulation of the production of these countries, using these points of "minimal production under stress" as a point on the curve somehow, then sum the curves to determine the global "worst case" curve of minimum produciton.

Everything over and above that might be "swing capacity" subject to being shut off at a given point, and this can also be modeled: sine wave, random peaks and troughs, spikes of magnitude X , steady state, etc. until the reserves run out.

MD

ReserveGrowthRulz · permanently banned Joined: Dec 30, 2005 Posts: 818

MD

My initial conclusion statement may have been too broad. It may be beter stated as:

"Future production and consumption cannot be predicted by fitting curves to global production history"
_________________
"Don't ever become a pessimist... a pessimist is correct oftener than an optimist, but an optimist has more fun, and neither can stop the march of events."
Robert A. Heinlein

md@peakoil.com