Ivan Shchapov
29 posts
@ShchapovEcon
On the Job Market 2025. PhD candidate @CrestUmr | Monetary macro, macro-finance, monetary-fiscal interactions | @OxfordEconDept alumnus.
Real question. Is there any central bank in an advanced economy that has shrunk its balance sheet back to pre-2008 levels after launching large-scale asset purchases?
📢#CallForPapers - 5th ASB/Banco Central de Chile/CEPR/ERSA Workshop on Macroeconomic Policy in Emerging Markets 📆14-15 Jan 2027 |📍 @UPTuks, Pretoria, South Africa ⌛Deadline: 15 Sep 2026 Sofia Bauducco @bcentralchile, @RefetGurkaynak, Özer Karagedikli @ASBedu_official, @ricco_giovanni, Nicola Viegi @UPTuks ow.ly/SSgH50YVqJw
Real question. Is there any central bank in an advanced economy that has shrunk its balance sheet back to pre-2008 levels after launching large-scale asset purchases?
I have an important update about this: we are looking to fill out 5 senior slots and 3 junior slots. Please, apply and reach out if interested! Or help me to spread out the news! @HannoLustig @luigi_bocola @JonSteinsson @IvanWerning @glviolante @JesusFerna7026 @LudvigsonSydney x.com/Francesco_Bia/…
The Fed just cut rates by 25bp on October 29, but was this decision already baked into the Fed's own communications? Markets seemed to have priced it, yet a key question remains: What would you have expected if you only read the Fed's pre-meeting documents? In my Job Market Paper, I tackle this question by developing a Multi-Agent System of Large Language Models that extracts conditional expectations directly from Beige Books and FOMC Minutes, creating a novel series of monetary policy surprises. Let's zoom in on last week's example: Reading only pre-meeting Fed documents, my system assigned: • 65% probability to a 25bp cut • 35% probability to no change • Expected cut: 16.25bp The Fed delivered the full 25bp cut, resulting in a small 8.75bp dovish surprise. ❗ Therefore, the decision was mostly expected by reading the official documents that were available before the meeting. How it works (and why it matters) Four agents work together on a common task: computing a monetary policy surprise for an upcoming FOMC meeting. • Agents IA and IM read the Beige Book (for this meeting) and the Minutes (for the previous meeting), respectively. • Agent II builds the expectations. • Agent III computes the surprise. In this way, I extract expectations from Beige Books and FOMC Minutes and compare them with the actual decision to compute the surprise. This approach: • Bridges narrative and high-frequency identification: Combines narrative approach with high-frequency measures' shock identification • Builds the first multi-agent LLM system for monetary policy analysis: Synthesizes heterogeneous Fed communications (Beige Books, Minutes, Statements) to extract ex ante probability distributions. • Enables direct extraction without ex post cleaning: No econometric orthogonalization, regression residuals, or filtering, just strict pre-meeting information cutoffs. • Uses LLMs for survey-based belief elicitation: LLMs extract probabilistic expectations from text, like surveying a Fed expert who has read all pre-meeting documents. This gives "New Hope" to monetary policy shock identification. The multi-agent architecture could be augmented with additional agents that read non-Fed information (macroeconomic releases, financial conditions, etc.) to further refine expectations. You can read more on: SSRN: papers.ssrn.com/sol3/papers.cf… My webpage: rubenfernandezfuertes.com #EconTwitter #MonetaryPolicy #LLM #JobMarket #MacroFinance
