Last weekend I read "Hubbert's Peak" by Deffeyes when I was in Canada. This book explains the prediction by King Hubbert that world oil production would peak around 2003 because oil discovery and oil production can be fitted to a logistic bell curve. I don't have the book with me any more but a fair amount of the material from the book is also available online through the previous link. I also discussed it with my Dad who is also a petroleum engineer but who isn't afraid of economics.

Deffeyes is a fine petroleum geologist and an entertaining writer, too. He does a great job of explaining how oil discovery has matured in the last century, and why previous predictions of oil production peaks were wrong -- a natural history of oil, as David Isenberg puts it in his review. For example, around 1910 people were predicting the US would run out of oil in 5-10 years. However, petroleum geologists found new ways to identify where oil could be found, drilling technology was improved, and US production made it through that "crisis" and several more. Deffeyes explanations of oil discovery and drilling technology are quite interesting.

In the late 50s, King Hubbert noticed that conventional oil discoveries in continental US had peaked, and he calculated that since oil production followed oil discoveries by about 11 years, that conventional oil production would peak in 1970 or shortly thereafter. He was right, and conventional US oil production does follow a logistic bell curve. Then Hubbert applied this analysis to world oil production and predicted a peak of I think 1999. Deffeyes uses Hubbert's analysis with additional information on offshore reserves (which Hubbert could not have predicted since offshore drilling wasn't technically feasible then) and predicts that world oil production peaks around 2003.

Hubbert's peak analysis seems basically to be sound, but a naive understanding of the analysis leads to very poor forecasts and planning.

1. The justification for choosing a logistic curve is still a little weak. Ideally, to justify this model you'd like to have a number of natural resources for which production peaked and descended back toward zero while demand still existed. That's very hard to find, because most resources (gold, silver, copper) have not peaked in production. Other resources have peaked in production (coal in some areas) but only because the resource was replaced by more effective resources and demand massively declined.
2. One of the earliest and best examples is Pennsylvania coal production. This peaked and declined far enough in the past that we can see the roughly symmetrical bell-shaped production curve quite easily. This is shown in Deffeyes' book with numbers, but I sketched a rough example.

   

   Some other historical peaks:

   • US continental conventional oil production
   • Canadian continental (Alberta) conventional oil production
   • UK sector North Sea oil production
   However, note that we don't yet have the full tails for most of these curves and we can't yet say they're symmetrical. Possibly the Pennsylvania coal production curve was so nicely symmetrical by accident - other coal reserves were found in other areas, transportation became cheaper, and/or demand was reduced in such a way to make the tail neatly symmetrical.
3. The symmetrical tail seems to show most clearly only when the geographical production region and the type of oil are closely defined. When one kind of oil and one location is cheaper than other kinds of oil from other locations, clearly we pump that oil until we run out and need to move to other types and other regions.
4. Even if we agree that an unfettered locally-limited carefully-defined oil production curve would be a normal distribution, oil production isn't unfettered -- it's currently limited by cartel quotas and regional unrest. That would imply a flattened peak, which doesn't tell you where the mid-point of the peak is unless you know how much it has flattened.
   
5. Technological advances can only have one effect - to slow the decline if not reverse it. Improvements in extracting oil from reserves that had previously been considered tapped (which Deffeyes explains has happened several times) may make the curve look more like this.
   
   This slow decline being borne out by data as we get more examples of peaking production.

So although it's tempting to make plans based on a 2003 world oil production peak, don't do anything drastic. Canada/Alberta offers an example why. Conventional oil production from Alberta peaked in a classic Hubbert curve in about 1976. Heavy oil production took over, however, and that kind of oil peaked in 1984. National oil production dropped 5 to 6% from its peak in 1984 but since then has gone up again because tar sands oil extraction is now economically feasible. So Canada experienced a drastic and sudden increase in economical oil reserves when tar sands entered the picture: an increase of 315 billion barrels, or roughly one sixth of total world conventional oil reserves. If you add these curves together they look more like this.

Deffeyes puts some of this into perspective in Hubbert's Peak. He does acknowledge the future of natural gas and does a good job of comparing that energy source to oil given today's technology. However, he doesn't look at the big picture the way Julian Simon did and Bjorn Lomborg does now. Even if you disagree with Simon and Lomborg, we should at least agree that what's important is the service provided by energy, more so than an individual energy source. I say "service" because I mean that if we want our homes heated, it doesn't matter what energy source provides that heat or how efficient it is as long as we can afford it (including externalities). So, will world energy services peak when oil production peaks? That's extremely doubtful, given that we have many other energy sources and are getting increasingly efficient at using them.

The next part of the bigger picture, again discussed by Simon and Lomborg, is technological improvements. You can't predict where or how the real technological improvements will change the situation. Deffeyes believes that tar sands and oil shales don't count as oil reserves because it's too costly to extract the useful oil from them. Yet the kind of technology my father works on is bringing tar sands extraction costs to a level that is nearly competitive at today's oil prices. Twenty years ago we suddenly had a lot more oil in existing "reserves" because we discovered that we could shoot steam into existing unproductive wells to force more oil up the well. I don't know what's next, and neither does Deffeyes.

A final note on Deffeyes is his typical disregard for economists. He pooh-poohs the idea that oil discovery responds to economical incentives (higher oil prices), arguing that you can't wring oil from a stone no matter how incentivized you are. However, that's taking too limited a view. Development responds to incentives in one way or another, sometimes in unexpected ways. While you can't always predict quite how people follow incentives you have a pretty good idea what the result will be. Who knows -- maybe someday we will be able to wring oil from stones.

Update: I had to rewrite this today after talking to my Dad and getting a more nuanced view. If you saw this earlier Sunday a.m, it's a fair bit different now.