DISQUS

Jim_Satterfield · 11 months ago

It reminds me of a friend of mine that critiques climate models because of the problems of economic modeling, which he is familiar with. Somehow he can't acknowledge that the basis of those two model systems are vastly different because while climate modeling involves huge amounts of data from different disciplines and a certain level of assumptions the basic facts are physical ones concerning physics and chemistry. The models can be adjusted for these factors if you want to assume different success levels in reducing the things that people do to change how much we affect the environment. Economic modeling almost never tries to accurately forecast the psychological factors, both individual and "mob" that are in reality the core of economic transactions. And let's be honest in admitting that they probably couldn't succeed in accounting for those factors if they did try.

In addition I notice that too many think that economics is similar to physics or chemistry in that the "laws" or "rules" they think they've discovered are immune to being affected by societal or technological changes even though the truth about economics is that it is a "soft" or "social" science. They think that the economics of the 18th century are the same as the 21st. Sorry. I just don't buy it. To get into some science geekiness, it's like expecting a physicist to believe that the behavior of light in vacuum and in a Bose-Einstein condensate is identical. But somehow certain kinds of economists buy it.

mikkel · 11 months ago

Yes, there was a snide comment about climate models on the Naked Capitalism thread and I just shook my head about the ignorance. I do have to say that I cautiously disagree with you that they couldn't "succeed" with newer models of economic behavior. While I agree with you about your past tirades against the Austrian School thinking that there are basic fundamental principles about human behavior that you can magically derive all economic behavior from, I do think there are certain dynamics that can be modeled much more accurately and fully (I mean the current models don't even account for resource utilization for chrissakes) and if done from a systems theory perspective they'd give good rules of thumb at least.

Which I mean honestly I believe that's all the climate models can do as well. Even though those are physically oriented -- and I've worked with people that are making them -- things are too insanely complex to have that accurate of an estimate on. I think the models are going to prove to be ill equipped for "predictions" a long way out but well equipped for determining whether our basic ideas about feedback loops, etc. are valid and talking about general climactic trends. With those too, it's a constant struggle because the nature of the feedback loops will change as concentrations change. Of course this critique applies to all chaotic systems...but it doesn't mean that it's not worthwhile since it gives a framework for understanding. I also think that if anything the climate models are drastically understating the potential for warming, based on real time observations that huge positive feedback contributors are quickly ramping up (i.e. methane already pouring out of permafrost and now Canada just recently determined its forests are a net contributor to greenhouse gases because they are too warm and dried out) and biochemical markers such as ocean pH are where they were predicted to be in another decade or two. Unfortunately, that, like this, gives no satisfaction in "I told you so."

There is another perspective about these financial models, and that is that they weren't changed "on purpose." This view was articulated by some commenters in that link plus I've read other insider accounts that wonder. The idea there is that everyone knew they were wrong but since they understated risk it pulled more external money into the system and it allowed people to rip off the system. The most famous critic, Nicholas Nassim Taleb runs a hedge fund that makes money literally just by exploiting flaws in the models.....and while he's honest about it and argues against it, there are growing suspicions that other people just took advantage and propagated the flaws either willingly or through willful ignorance. After all, now that they messed up what happened? The government just bails them out, or even if not, they go home with hundreds of millions of dollars.

This sad episode shows what happens when the government doesn't have enough scientific and mathematical experts in regulatory positions, so they can't actually critique whether things were any good.

Jim_Satterfield · 11 months ago

I also think that if anything the climate models are drastically understating the potential for warming, based on real time observations that huge positive feedback contributors are quickly ramping up (i.e. methane already pouring out of permafrost and now Canada just recently determined its forests are a net contributor to greenhouse gases because they are too warm and dried out) and biochemical markers such as ocean pH are where they were predicted to be in another decade or two. Unfortunately, that, like this, gives no satisfaction in "I told you so."

And that is a statement that I can agree with you wholeheartedly on. That is one of the things I can't get the so-called skeptics to understand. That it is just as likely given discoveries like the ones you cite that I'd also read about that the models were skewed towards understating the problems.

Rudi · 11 months ago

Mathimatical models are only as good as the date used and who it's analysed. While artificial neural networks almost works with credit fraud and stock analysis, their use in the GWOT in data mining is a joke. Stocks and credit haVE EXTENSIVE data histories and limited scope, but monitoring calls for terror yield raids on pizza deliveries. Rudy J is laughing all the way to the bank...

mikkel · 11 months ago

I don't think it's just a matter of the amount of data collected but the number and nature of relationships, as well as the incidence rate. Terror data mining is the worst of all worlds as everyone is highly connected and there is extremely low incidence. Even if you could get a mere 0.1% false positive rate, that is still tens if not hundreds of thousands of falsely accused terrorists. I wish I could find this post on this counter intelligence blog that I saw a few years ago...they showed mathematically that even with perfect knowledge and each person only having 20-30 connections, it wouldn't give us anything valuable because of being drowned out by false positives. I assume that credit fraud works better because the fingerprint is much stronger, with fewer relationships needed, am I right?

Also I think it's important to distinguish data mining from data modeling. Data mining doesn't necessarily mean that we understand dynamics, so just because we can mine doesn't mean we can make predictions or interventions.