Glenn Loury and John McWhorter are at their best when they disagree, and I enjoyed their discussion about the disastrous exchange between Trump, Vance, and Zelensky at the White House. Both agree that the U.S. is right to push for a negotiated settlement, which involves pressuring Ukraine to acknowledge its precarious position. However, they diverge on how this pressure was communicated and its potential repercussions. Glenn argues that Trump and the vice president rightly prioritized American interests by applying pressure on Zelensky, while John takes the opposite stance, framing his argument within a broader critique of the U.S. president and his administration.
Read MoreThe BIS has a Bulletin out on the usefulness of AI and large language models. They’re not terribly impressed.
Read MoreWhen posed with a logical puzzle that demands reasoning about the knowledge of others and about counterfactuals, large language models (LLMs) display a distinctive and revealing pattern of failure.
The LLM performs flawlessly when presented with the original wording of the puzzle available on the internet but performs poorly when incidental details are changed, suggestive of a lack of true understanding of the underlying logic.
Our findings do not detract from the considerable progress in central bank applications of machine learning to data management, macro analysis and regulation/supervision. They do, however, suggest that caution should be exercised in deploying LLMs in contexts that demand rigorous reasoning in economic analysis.
This is my third use case for how to do quantitative analysis with Chat GPT 4. The two others, on Eurozone inflation and times series regression with macro data, can be found here and here. I started in the industry as Head of Research for Variant Perception, a research shop that specialises, among other things, in quantitative trading models, asset allocation tools, and trade signalling analysis. One tool that came up again and again in my analyses was binary signals to identify turning points in asset classes, stocks or economic data series. The idea is simple. First, you create a binary indicator which takes the value of 1, if a certain threshold in the data is breached to the upside or downside, and zero otherwise. Secondly, you investigate what happens after such a signal has gone off, either in the original data set or mapped to a separate data set. You can combine signals across datasets to get a rolling series of signals, which can be compared to asset prices or economic data. You can see an example of such an analysis with the Nasdaq here.
Read MoreThe following chart is one of hundreds that I use in my day-job as Chief Eurozone Economist at Pantheon Macroeconomics. It plots a normalised Z-score index of surveyed new manufacturing orders in Germany alongside year-over-year growth in factory orders, ex-major orders. It’s worthwhile spelling out the meaning of this chart in the world of economic research and forecasting. The factory orders numbers are so-called hard data, which in this case means that they’re official numbers of real activity reported by the statistical office. The PM new orders index, by contrast, is my home-cooked index of so-called soft data. Specifically, these are survey data, compiled by the likes of the EU Commission, IFO, S&P, and national statistical offices. We’re only interested in these numbers to the extent that they tell us something about the official/hard new orders data, which in turn could help us pin down trends in industrial production, exports, GDP growth, employment and so on. From simply eye-balling the chart, the two series look coincident, but note that the surveys are released ahead of the official data, so that we always have survey numbers that are one-to-two months ahead of the official data. In other words, when it comes time to forecast new orders for the month of December, we will already have survey data for that month. This should, in theory, help us to better forecast the official real new orders data.
Read More