Valuing corporate bonds in systemic economies is challenging due to intricate webs of inter-institutional exposures. When a bank defaults, cascading losses propagate through the network, with payments determined by a system of fixed-point equations lacking closed-form solutions. Standard Monte Carlo methods cannot capture rare yet critical default events, while existing rare-event simulation techniques fail to account for higher-order network effects and scale poorly with network size. To overcome these challenges, we propose a novel approach -- Bi-Level Importance Sampling with Splitting -- and characterize individual bank defaults by decoupling them from the network's complex fixed-point dynamics. This separation enables a two-stage estimation process that directly generates samples from the banks' default events. We demonstrate theoretically that the method is both scalable and asymptotically optimal, and validate its effectiveness through numerical studies on empirically observed networks.

Data on marriage flows are not available in most developing countries, making marriage market imbalance difficult to measure. Existing measures use crude fertility rates and do not account for early-life mortality, overstating the number of births surviving to marriageable ages. This paper develops the Surplus Groom Index to quantify marriage market imbalance under monogamy using census age structure, vital registration of births and deaths, and marriage timing data. The index incorporates effective fertility-total births adjusted for under-five mortality - to reflect actual cohort progression from birth to marriageable ages. This adjustment matters in settings where child mortality shapes the supply of marriage partners. Using India's 2011 Census data, we find that eleven percent of men aged 15-54 cannot marry due to bride shortage, approximately 39 million men. Marriage imbalance is widespread rather than regionally concentrated. Punjab records the highest deficit at 33 percent, but states considered demographically progressive show substantial imbalance: Kerala 18 percent, West Bengal 14 percent, Karnataka and Tamil Nadu 11 percent each. Declining fertility has produced smaller female cohorts unable to absorb male-heavy cohorts from earlier birth years. Balanced sex ratios at birth do not ensure marriage market equilibrium once fertility declines and marriage is delayed.

Recessions are periods in which the least productive firms in the economy exit, and as the economy recovers, they are replaced by new and more productive entrants. These cleansing effects improve the average firm productivity. At the same time, recessions induce a loss of varieties. In an economy with Homothetic Single Aggregator technology, we show that their long-run welfare effects trade off these two forces. This trade-off is governed by love-of-variety and the elasticity of substitution in aggregate production. If industry output is aggregated using the standard CES aggregator, recessions do not improve long-run GDP or welfare. If the economy features more love-of-variety than CES, the social planner optimally subsidizes economic activity both in steady state and even more so in recessions to avoid firm exit. We use the model and quasi-exogenous variation in demand to estimate love-of-variety. We find it to be significantly higher than implied by CES aggregation, suggesting that even the long-run effects of recessions are negative. Finally, we quantitatively characterize the optimal policy response both along the transition and in the steady state.

The use of reinforcement learning algorithms in financial trading is becoming increasingly prevalent. However, the autonomous nature of these algorithms can lead to unexpected outcomes that deviate from traditional game-theoretical predictions and may even destabilize markets. In this study, we examine a scenario in which two autonomous agents, modelled with Double Deep Q-Learning, learn to liquidate the same asset optimally in the presence of market impact, under the Almgren-Chriss (2000) framework. We show that the strategies learned by the agents deviate significantly from the Nash equilibrium of the corresponding market impact game. Notably, the learned strategies exhibit supra-competitive solution, {which might be compatible with a tacit collusive behaviour}, closely aligning with the Pareto-optimal solution. We further explore how different levels of market volatility influence the agents' performance and the equilibria they discover, including scenarios where volatility differs between the training and testing phases.

In this work, we study statistical arbitrage strategies in international crude oil futures markets. We analyse strategies that extend classical pairs trading strategies, considering the two benchmark crude oil futures (Brent and WTI) together with the newly introduced Shanghai crude oil futures. We document that the time series of these three futures prices are cointegrated and we model the resulting cointegration spread by a mean-reverting regime-switching process modulated by a hidden Markov chain. By relying on our stochastic model and applying online filter-based parameter estimators, we implement and test a number of statistical arbitrage strategies. Our analysis reveals that statistical arbitrage strategies involving the Shanghai crude oil futures are profitable even under conservative levels of transaction costs and over different time periods. On the contrary, statistical arbitrage strategies involving the three traditional crude oil futures (Brent, WTI, Dubai) do not yield profitable investment opportunities. Our findings suggest that the Shanghai futures, which has already become the benchmark for the Chinese domestic crude oil market, can be a valuable asset for international investors.

We study the dynamic portfolio selection of an investor who uses deep learning methods to forecast stock market excess returns. In a two-asset allocation problem, deep neural networks -- both feedforward and long short-term memory (LSTM) recurrent architectures -- deliver economically significant gains in terms of certainty equivalent returns and Sharpe ratios relative to linear predictive regressions. These gains are robust to alternative performance measures, the inclusion of transaction costs, borrowing and short-selling constraints, different rebalancing horizons, and subsample splits, and are particularly pronounced during NBER recessions and periods with large return swings. Within the class of neural networks we consider, economic performance is broadly similar across architectures, with the recurrent LSTM specification providing incremental benefits with more frequent rebalancing. Overall, our evidence suggests that exploiting the time-series structure of standard predictor variables via deep learning can generate meaningful portfolio improvements for investors beyond those obtained from linear models.

Using a representative survey of Korean workers, we provide evidence on the adoption of Generative AI (GenAI) and how GenAI reallocates time at work. We find that 51.8\% of workers use GenAI for work and GenAI reduces working time by 3.8\%. However, these gains may not materialize in aggregate productivity statistics yet: the correlation between time savings and output changes is near zero. We show this disconnect arises because workers capture efficiency gains primarily as on-the-job leisure, rather than increasing their output. These findings suggest that standard productivity measures may understate AI's impact by missing non-pecuniary welfare channels.

Conditional Value-at-Risk (CoVaR) quantifies systemic financial risk by measuring the loss quantile of one asset, conditional on another asset experiencing distress. We develop a Transformer-based methodology that integrates financial news articles directly with market data to improve CoVaR estimates. Unlike approaches that use predefined sentiment scores, our method incorporates raw text embeddings generated by a large language model (LLM). We prove explicit error bounds for our Transformer CoVaR estimator, showing that accurate CoVaR learning is possible even with small datasets. Using U.S. market returns and Reuters news items from 2006--2013, our out-of-sample results show that textual information impacts the CoVaR forecasts. With better predictive performance, we identify a pronounced negative dip during market stress periods across several equity assets when comparing the Transformer-based CoVaR to both the CoVaR without text and the CoVaR using traditional sentiment measures. Our results show that textual data can be used to effectively model systemic risk without requiring prohibitively large data sets.

This paper asks whether regulatory monitoring exhibits nonlinear capacity limits as the scale and complexity of the regulated environment increase. Using a county--year panel of U.S. Food and Drug Administration (FDA) inspections merged with local establishment counts, we identify a sharp breakpoint: beyond a threshold scale, severe inspection findings rise while inspection effort per establishment flattens or declines. The threshold and the post-break deterioration vary across food-related industry groups and shift with proxies for local density and connectedness, consistent with monitoring becoming ``too big to monitor" in more interconnected production environments rather than driven by simple reallocation or delay. Methodologically, we provide a portable breakpoint selection and piecewise-estimation framework that can be applied to other enforcement settings.