Markus' Academy

Follow the link for the full summary: https://markusacademy.substack.com/p/us-treasury-yieldsLink to sign up for the webinar series: https://markusacademy.substack.com/Watch on Youtube: https://youtu.be/ZseTYM_A-60Listen to part 1: https://open.spotify.com/episode/6HXuoDrSOO0dEmCxDyj0qf?si=49aaf2a925d346f0Bill Dudley and Jonathan Payne joined Markus’ Academy for a two-part conversation on US Treasury yields. This is part 2. Dudley is a Senior Advisor to Princeton’s Griswold Center and a former President of the Federal Reserve Bank of New York. Payne is an Assistant Professor at Princeton. A summary in three bullets:● The expectations hypothesis holds in general throughout US history except in the period of 1965-1990, so that the risk premium on government debt was time-varying only then.● The fact that Microsoft’s yields have widened with respect to other AAA corporates or sovereigns suggests growing default risks around AI, not that the government may be crowding out AI investment● R* has drifted up, from zero after the financial crisis to 1.2% today (as projected by the Fed)Timestamps:[0:00] R* has moved up, and we only learn it through its works[8:15] The expectations hypothesis and the stock-bond correlation[23:12] Is the government crowding out AI investment?[27:25] Hope is not a strategyThe views expressed by Jonathan's coauthors in the papers discussed are those of the authors and do not necessarily reflect the views of the Board of Governors of the Federal Reserve System.

Follow the link for the full summary: https://markusacademy.substack.com/p/us-treasury-yieldsLink to sign up for the webinar series: https://markusacademy.substack.com/Watch on Youtube: https://youtu.be/SAUrOFLlK-kLink to LIsten to part 2: https://open.spotify.com/episode/4aexS87sDsu6x2Fj8diXuM?si=kqTUF8zDQ-CzuTdOyCfyhwBill Dudley and Jonathan Payne joined Markus’ Academy for a two-part conversation on US Treasury yields. This part covered the history of American debt starting in minute 21:09. Before then it started with a refresher on basic concepts to study yield curves.Dudley is a Senior Advisor to Princeton’s Griswold Center and a former President of the Federal Reserve Bank of New York. Payne is an Assistant Professor at Princeton. A summary in three bullets:● There is no Phillips-style curve between debt-to-GDP ratios (safe asset scarcity) and the government funding advantage. The simple relationship disappears when adjusting our prior measures of the funding advantage for the tax treatment of the government’s long-term debt during the Great Inflation● The US government’s funding advantage was largest in the 19th century during the national banking era, not after WWII or Bretton Woods.● Governments face financing trilemma, having to pick two among: (1) a large funding advantage, (2) a solvent banking sector, (3) a regime that inflates the debt away. The funding advantage is a reward for prudenceTimestamps:[00:00] Today’s yields are not high by historical standards[11:52] Fiscal-monetary interactions[22:38] The history of America’s debt[34:30] The national banking era was a stablecoin regime[41:15] The funding advantage is a reward for prudenceThe views expressed by Jonathan's coauthors in the papers discussed are those of the authors and do not necessarily reflect the views of the Board of Governors of the Federal Reserve System.

Follow the link for the full summary:https://markusacademy.substack.com/p/ai-for-economic-theorists-and-mathematiciansLink to sign up for the webinar series:https://markusacademy.substack.com/Pietro Ortoleva and Fedor Sandomirskiy joined Markus' Academy for a mini-series on AI for economic theorists and mathematicians. This is episode 4. Both are economic theorists at Princeton University.The value has shifted from prompt engineering to context engineering: specify the task. Write a lazy two-line prompt, ask the same model that will do the work to expand it. Never let one session grade its own work, but run a separate verifier and, if the two disagree, a third instance as a judge. Fedor dissected the prompt OpenAI published for the cycle double cover proof - 64 agents, running at least eight hours, a supervisor eliminating agents that converge on the same route. And in his work adds tricks of his own.Timestamps:[0:00] Is prompt engineering still a thing?[3:40] Prompt expansion: let the model write the prompt[9:45] Prover versus verifier, LaTeX not PDF, and when to restart[16:43] Agent swarms, and AI proofreading

Follow the link for the full summary:https://markusacademy.substack.com/p/ai-for-economic-theorists-and-mathematiciansLink to sign up for the webinar series:https://markusacademy.substack.com/Pietro Ortoleva and Fedor Sandomirskiy joined Markus' Academy for a mini-series on AI for economic theorists and mathematicians. This is episode 3. Both are economic theorists at Princeton University.An AI's intelligence and stamina are substitutes. A slightly shallower model that grinds all day often gets further than a deeper one you can only afford to run for twenty minutes. Pietro compared the models available today: their preference, narrowly, is GPT-5.6, since it has more stamina than Fable. Pietro also argues that agents are fundamental in empirical work, but that for theory the browser is still enough for brainstorming.Timestamps:[0:00] The frontier since June: Fable 5, GPT-5.6, Opus 5[4:13] Intelligence and stamina are substitutes[8:53] The verdict, and what to run on a Pro, $20, or free budget[11:10] Browser or agents?

Follow the link for the full summary:https://markusacademy.substack.com/p/ai-for-economic-theorists-and-mathematiciansLink to sign up for the webinar series:https://markusacademy.substack.com/Pietro Ortoleva and Fedor Sandomirskiy joined Markus' Academy for a mini-series on AI for economic theorists and mathematicians. This is episode 2. Both are economic theorists at Princeton University.In this video, Fedor asks whether models can actually generate new ideas. In economic theory this is hard to settle because a model's quality is subjective, while in math a proof either holds or it does not. AI is poor at attribution, but superb at aggregation of knowledge.Timestamps:[0:00] Can models be creative, and why the evidence comes from math[2:20] The ten Erdős problems episode, and Terence Tao's ledger[5:19] The unit distance conjecture falls[8:44] Jacobian, cycle double cover, and what it means for theory