All Events

We study whether machine learning forecast combinations can improve on the equal-weight mean of survey forecasts from Slovak commercial banks for GDP growth and inflation. Using monthly bank forecasts with full histories (2007–2024; 216 observations per series) and real-time data vintages that respect publication lags, we run a rolling out-of-sample evaluation. We compare linear and penalised regressions (OLS, ridge, lasso, elastic net), nonlinear methods (trees, random forests, XGBoost, kNN, SVR), an ARIMA benchmark, and stacked ensembles estimated via non-negative least squares. Performance is measured by relative MSE versus the bank mean. For inflation, regularised linear combinations deliver consistent gains (RelMSE at around 0.81–0.82), while nonlinear and stacked approaches underperform, suggesting overfitting in small panels. For GDP, several nonlinear models and a restricted stack improve materially on equal weights.

This event is both online and in person. Join the Teams meeting

Despite years of transparency reforms, their impact on civic participation is not well understood. We study whether government performance information from one of the largest transparency initiatives can motivate citizens to participate and whether changes in beliefs are the mechanisms driving this effect. The initiative tracks 112 of India’s 806 most underperforming districts across key development indicators and posts performance information online for public accountability. We conducted a randomized field experiment with 2,107 adults in a district targeted by the initiative. Over 90% in our setting were unaware of the initiative, and 75% overestimated their district’s performance. Our low-cost, scalable video intervention informed citizens that their district was included, revealed its declining performance relative to neighboring districts, and listed formal grievance channels. Citizens randomly assigned to watch the video were 5 pp more likely to sign the petitions and 8.4 pp more likely to attend workshops on complaint filing portals. They updated beliefs about their district’s low rank and became more aware of the initiative. Citizens who initially overestimated their district’s rank updated their beliefs and increased participation, while we observed no effects among those who had not overestimated prior to treatment. Beliefs about citizens’ influence over service delivery remained unchanged. One month later, beliefs about the district’s rank converged, yet citizens in the intervention group continued to meet civic officials and file complaints. Those who initially overestimated their district’s rank remained significantly more likely to participate. To explain why citizens continue participating despite low perceived influence, we develop a model where participation depends on expected performance gaps relative to aspirational service levels and perceived influence over government. Larger perceived gaps alone can sustain participation through an accountability channel. These findings have implications for 250 million Indians in most-underperforming districts and highlight how prior beliefs mediate effectiveness of transparency programs.

This event is both online and in person. Join the Teams meeting

We assess the welfare impact of the introduction of universal daycare services in Qu´ebec in 1997. Unlike the standard sufficient-statistic metric, which assumes marginal changes in fiscal policy, our approach accounts for the non-marginal nature of the program and quantifies nonpecuniary benefits. Through a structural model of childcare demand, we estimate substantial welfare gains from the policy, yielding a Marginal Value of Public Funds (MVPF) above 3.5. Using the sufficient-statistic approach underestimates welfare gains by half. Counterfactual simulations and a difference-in-differences analysis suggest that increasing availability, rather than solely improving affordability, is crucial for the effective design of universal programs.

This paper studies the relationship between patient cost-sharing and the demand for care after hospital stays (post-acute care), and how this relationship is mediated by non-medical custodial care needs and the presence of the family. Using plausibly exogenous variation in the number of free skilled nursing facility (SNF) days at hospital discharge, I first show that higher cost-sharing for SNFs leads to substitution away from SNFs to home health and routine discharges with no care. Moreover, for those who are already in a SNF, discharges rise sharply when cost-sharing begins. I further document that people with family are less likely to use SNFs, while people with custodial care needs are more likely to use SNFs. I use these findings to motivate a dynamic structural model that begins at hospital discharge and follows patients through further discharge decisions in different post-acute settings. Using the model, I explore the implications of alternative cost-sharing structures. I find that eliminating the non-linearity in SNF copay can generate large Medicare savings at small costs to the patients. Medicare subsidizing custodial care at home enhances patient welfare substantially for individuals without family, at no additional cost to Medicare. For patients with family, the same policy, while still beneficial to the patient, will increase Medicare costs substantially, resulting in a net zero effect overall.

Older workers are often viewed as obstacles to innovation, suggesting that their exit allows firms to reallocate resources toward new capital and technology. I argue instead that older, experienced workers support both the continuity of current production and the capacity to integrate new technologies into existing operations. The key empirical challenge is that when retirements are anticipated, firms have time to transfer knowledge internally, making the productivity value of older workers difficult to observe. I address this by studying a 2014 German pension reform that unexpectedly lowered the early retirement age for experienced workers by up to 29 months, inducing a sudden and unanticipated loss of long-tenured employees. Firms exposed to the reform reduce capital accumulation, delay technology adoption, and experience subsequent declines in revenue and value added, consistent with the erosion of firm-specific human capital. To interpret these findings, I develop a stylized model in which older workers transfer uncodified, firm-specific knowledge that is essential for maintaining legacy capital and integrating new technologies into firms’ operations. The model predicts, and the data confirm, that unexpected retirements weaken firms’ ability to sustain production and slow the pace of technological upgrading.

This event is both online and in person. Join the Teams meeting

Using household-level survey data on 22 European countries collected in four waves between 2010 and 2022, we document the distributional effects of borrower-based macroprudential measures (BBM). We find that the tightening of the BBM significantly reduces the availability of mortgages for lower-income households, leading to increased reliance on renting. Regulation works not only through the extensive margin by cutting off risky borrowers from the mortgage market, but also through the intensive margin by reducing the average loan amount of households with lower incomes. This reduction in risk has the positive side effect of lower borrowing costs, benefiting mainly households with above-average incomes. The easing of BBM, observed in several countries, does not yet fully reverse these effects.

TBA

Inaccurate inference about economic policy uncertainty affects international investment returns, and investment decisions made with imperfect information result in financial losses for all market participants. The classic widely used Baker et al. (2016) index of economic policy uncertainty (EPU) suffers from methodological imperfections in construction due to (1) neglecting semantics, (2) data quality fluctuations, and (3) ignoring tonality/emotion in the data. In this paper, we introduce a novel emotion-based economic policy uncertainty index that (i) improves the semantic identification of economic policy by extracting a large set of tokens from RoBERTa, (ii) uses a vast CC-News media dataset, and (iii) identifies uncertainty by a finetuned emotion recognition model.

This study examines the long-term health consequences of reduced instructional time during compulsory schooling. I exploit a German reform of the school year schedule from the 1960s, which substantially shortened the length of two school years without adjusting the core curriculum. As a result, learning shifted out of the classroom, primarily through increased homework. Applying a difference-in-differences design to biographical administrative records, I estimate the impact of this policy on severe health issues, as measured by the number of long-term sick leave taken throughout individuals' careers. The results reveal a significant increase in long-term sick leave taking, suggesting that educational disruptions can have lasting effects on adult health.

Germany was the first country worldwide to introduce a structured reimbursement pathway for digital therapeutics (DiGA), effectively allowing physicians to prescribe smartphone apps covered by statutory health insurance. In this seminar, Simon Reif presents recent research evaluating whether this pioneering regulation successfully incentivized innovation in the digital health market.  Using high-frequency data from the Apple App Store and synthetic control methods, the study analyzes the market entry of health apps following the policy's introduction. The findings reveal a significant increase in the number of German-language digital therapeutics; however, this quantity did not translate into a broader diversity of medical conditions treated. Furthermore, the surge was largely driven by apps that monetize patient data for advertising rather than high-quality applications backed by scientific evidence. The presentation will discuss the implications of these results for designing policy that balances market incentives with patient privacy and care quality.

Teaser: Simon will be holding a separate, short workshop on the reproducible research tools used at his institute. Join us to learn how to implement these transparent workflows in your own analysis. 

Motherhood is associated with persistent penalties in both labour market outcomes and health. However, the role of family policies in mitigating these effects, particularly with respect to maternal mental health, remains unclear. This paper examines the impact of extending paternity leave on maternal mental health. Using administrative health records covering all publicly funded births in Catalonia (Spain) between 2006 and 2024, we implement a local difference-in-differences design around the April 1st, 2019 reform, which introduced partially mandatory paternity leave for fathers. The analysis focuses on first time mothers giving birth within a three-month window around the reform cutoff and compares them to a control cohort from 2016. Maternal mental health is measured using diagnoses and prescription records for stress-related, mood-related, and broader mental health conditions drawn from primary care and pharmacy data. We find an increase in stress-related diagnoses and in the use of antidepressants and anxiolytics from the second year after childbirth among mothers exposed to the reform. These effects are concentrated among older mothers (above the median age of 31) and among those who do not have a second child within four years after birth. We find suggestive evidence that these patterns may be related to increased healthcare use and to stress-related mechanisms, including delayed fertility, changes in labour market attachment, and relationship instability. Ongoing work will extend the analysis to additional mental health outcomes and to subsequent paternity leave reforms