ChatGPT and monetary policy

Article by Gonzalo de Cadenas-Santiago, Executive Director of MAPFRE Economics

Yesterday I decided to ask the artificial intelligence chatbot how high rates will be set by the European Central Bank and the Federal Reserve in a key week. And the answer it gave me was: “I cannot predict the future with certainty. A central bank's interest rate policy is determined by a variety of economic and financial factors...It is advisable to closely follow the institution's statements and decisions.” And the same applies to the Fed.

As of today, in trying to understand how the model underlying the ultra-chat-bot works, it occurred to me that, given that such models, known as large language models, essentially create texts that are conditioned distributions of word sequences that reflect concepts collected in the infinite maps of meaning (“maps of meaning” as per Piaget) that exist online, the answer they give will also be the most plausible. In other words: what is expected and not at all improbable, orthogonal to the main idea or related to levels other than the pure semantic surface that links question and answer.

Therefore, having verified that my job and that of others in my profession, and even that of other humans, is not in danger – at least for now – I will dare to predict what will happen with monetary policy on both sides of the Atlantic.

Lagarde's latest statements reaffirm that the ECB will maintain its hardline language, thus signaling a tightening cycle, still with some way to go and on a more long-lasting basis, in order to bring the inflation outlook under control, and this despite the stagflation scenario and the risks of recession. At the same time, it is crucial to insist on the need to close the current fiscal gap given the opposite direction of the current monetary policy.

Consequently, and given the ECB's announcement of a balance sheet reduction starting in March, the possibility of triggering divergent offsetting movements in financing costs is increasing, while the effectiveness of the “anti-fragmentation” mechanism has a tailwind before it has even become effective. In sum, and although the details of the asset sale program are not yet known (let’s wait for this week), the starting point is uneven growth under divergent inflation rates and asymmetric fiscal expansion-austerity, adding to the imbalances of balances already accumulated in TARGET2, governed by German law.

On the Fed side, macro data continues to surprise us on the upside, as was the case with last week's GDP data, thus justifying its hard line in the face of a market consensus that is still expecting earlier cuts and a more benign final interest rate. This growth figure offers some additional room to tighten the message, pull back employment, and ease aggregate demand, thus refuting the FOMC's view, given the endogenous and differential overheating factor present in the U.S. economy. Strong data supports the Fed's view of higher rates for longer, and soft data (PMI's, ISM and LEA) are continuing to show some weakness and are leading the markets to anticipate a more benign final rate.

On the inflation front, the latest data revealed further moderation in both the general and core indexes, suggesting that the worst is likely to become visible in retrospect. However, on the one hand, the reading of both indicators continues to be historically high and somewhat far from the objective of the Federal Reserve, and, on the other, given the greater resilience of the economy, the anchoring of core inflation could show similar behavior supported by: (i) persistent components, as both housing and rental inflation, as well as services, remain unmoderated; (ii) the reluctant momentum of wages (real wages fell for the twentieth consecutive month) in a market that, according to official data, remains robust, and (iii) the effect of the reopening of China, a priori neutral, due to the inflationary impact on raw materials versus the deflationary effect due to the improvement of supply chains, but with basic supplies exerting more immediate pressure due to its impact as the first link in the production chain.

In short, some decoupling can still be seen with respect to the implicit expectations of futures markets and what central banks are actually going to do. In the US, we expect a 25bp hike in February and another 25bp in March, up to 5% from the current 4.5%. The combination of strong data (GDP, employment, etc.) is favorable for further upside movement, even if high-frequency soft data indicate some weakness, an appropriate stance to pause until June with a new macro picture and the cumulative effect already making itself felt. In Europe, we expect 50bp in February and March, to 3.5%. Strong and soft data, albeit cautiously, have been surprising on the upside; inflation, in turn, is not pointing to such a clear path as in the case of the US; and in March, as noted, it is introducing the 10-15 b/m QT and turning on autopilot with the Fed until June.

Lastly, let’s return to ChatGPT.  The Turing test on ChatGPT and, therefore, to see if the text you are reading has been created by our OpenAI friend or something similar, is to see if there is an ounce of creativity and orthogonal thinking. If there were none, you would certainly be talking to a robot. This test will make you realize how, little by little, if this technology comes into widespread use in an uncontrolled way, it will effectively replace some of us writers and will fill up texts online with predictable semantic structures, with ideas that will increasingly (if not always) be the expected ones, and little by little you will no longer be surprised by what you read. However, perhaps out of prudence, it is also not the job of AI to answer questions such as those relating to monetary policy.