This is one of these “don’t bother reading through everything unless you already know what I am talking about” posts.
Scott Sumner has a long rebuttal to o1 on monetary theory, offering many criticisms. He does not like the quality of the answer given by the AI. But I view the AI more positively here, at least relative to the current state of the research literature, which admittedly is not so satisfying. I think Scott, a’la Kasparov, is being spooked by “the machine,” and his usual clarity of thought is not always present in this exchange. Here are a few points in response to Scott:
1. Nominal relationships seemed much clearer in the age of Milton Friedman and now they are extremely murky (and yes economists may have been wrong in Friedman’s time about this, but that is not the point). Scott seems to deny this, and I am not sure why. The Monetary History persuaded a significant batch of highly intelligent economists that there was a pretty stable relationship between the monetary aggregates and nominal income. Hardly anyone holds this view today. o1’s portrait of this change seems to me more accurate than Scott’s insistence that inflation has become easier to predict over time.
2. Scott in his criticisms is focused mainly on whether inflation prediction has become harder over time, but mainly o1 is answering why it is so hard today to explain inflation dynamics. (Revisit the question: “Please write an essay on how current macroeconomists find inflation dynamics so very difficult to predict, and why that has made them reject various forms of monetarism, even as approximations of what is going on behind price level behavior.”) So he is grading it on the wrong issues.
3. Scott writes “It [o1] mentions a bunch of irrelevant stuff like QE, and misses the key point that the payment of interest on reserves and the zero lower bound problem have made the money multiplier far more unstable.” Inflation forecasting has been a problem before and after the ZLB. And the payment of interest on reserves was a huge one-time problem for forecasting, but the literature rightly ignores or downplays this as a general issue over time, because it isn’t. So here o1 is closer to the research consensus than Scott is.
4. Most of all, Scott goes out of his way to avoid presenting or even citing a better approach to inflation dynamics. You might look at this 2011 post from Scott on inflation dynamics. It endorses the quantity theory, which I think is true sometimes, but doesn’t go into why the evidence has gone so badly in the other direction.
4b. More seriously, Scott seems to dismiss the price level concept altogether. For instance he once wrote: “In the past, I’ve frequently argued that inflation is an almost meaningless and useless concept. I’m not even aware of any coherent definitions of the concept.” I don’t think this is a defensible point of view, and you have to compare Scott’s criticisms of the o1 model to his own approach, which is fairly nihilistic. And I think wrong. If inflation were higher and someone offered Scott an inflation-indexed contract to sign, would he be unable to evaluate such a transaction? Obviously not.
5. As a side note, I think o1 also did better than the varied observations by Krugman on inflation dynamics over the years. Most recently I recall Krugman arguing that we didn’t have a recession the year before because the initial inflation was almost entirely about supply side shocks. That view has been refuted by a number of recent research papers, some of those cited on MR, showing it was both a supply side and demand side phenomenon.
More generally, here is one recent model of price level dynamics, you can read through the model and results. Real wages matter in many of these investigations, which Scott dismisses and o1 endorses. Here is an attempt to forecast price inflation for 2024, again you can look at the model, which is not so simple and I would also say not super-impressive (I intend no criticism of the authors here, the questions are hard). It is still a puzzle why inflation rates were not even lower during the 2008-2010 period. In other words, interest on reserves may have mattered less than models would suggest.
I understand full well that “kitchen sink” approaches are unsatisfying to many economists. Yet when there is in fact a clear theoretical answer to an economics problem, usually o1 gives it to you.
If you ask o1 Scott’s oil and price theory question, it gets the right answer. The second sentence is: “Whether that quantity is higher or lower than before depends on why the price rose.” In other words, it does not reason from a price change.
So I think Scott is seriously underrating o1 as a reflection of what the profession believes on inflation dynamics. Scott has a right to disagree with that consensus, but I don’t see he has put up the evidence to establish a better view. In any case, on these issues o1 beats both Sumner and Krugman, noting that each is putting forward a fairly extreme point of view. Notably, o1 pro, a yet more advanced model, comes up with a better answer yet. Or you can ask it to spend at least 5000 logic tokens answering the question. Yes people it is worth $2500 a year.
Arnold Kling comments. And Kasparov did eventually come around.
The post o1 is still doing well on monetary economics appeared first on Marginal REVOLUTION.
