This paper proposes a new shadow-rate model that incorporates stochastic jumps, macro-financial variables, and regime-dependent dynamics into the short-rate process. To this end, we extract the first principal component from daily zero-coupon yields in the United States, Euro Area, and United Kingdom and augment it with inflation, real GDP, unemployment, exchange rates, and sovereign CDS spreads. The latent short-rate component, augmented by these macro factors, follows a mean-reverting jump-diffusion process, and the final policy stance is constructed as a convex combination of the two pillars. We find that, relative to the standard benchmark in the literature, the resulting shadow rates exhibit smaller fitting errors and, in our forecasting exercises, provide more informative signals, particularly during periods of elevated macro-financial uncertainty. Robustness checks using an OLS VAR and a FAVAR framework broadly support these findings. Overall, our approach provides a flexible tool for policy analysis on monetary policy under extended episodes at or near the effective lower bound.
Abstract Forecast combination methods’ importance and usefulness in out-of-sample prediction of economic and financial variables are broadly acknowledged. Recent research highlights the importance of two components in improving point forecast accuracy: the initial selection of point forecasts to combine and the subsequent imposition of additional shrinkage. Motivated by empirical results that indicate the benefits of integrating these two procedures when combining point forecasts, this study contributes by comprehensively extending the dynamic forecast rotation method and proposing a forecast combination technique that implements point forecast selection and additional shrinkage in a parsimonious fashion. I apply the suggested technique to the well-known topic of forecasting aggregate equity return volatility out-of-sample by conditioning on economic variables. I demonstrate its superior performance both from a statistical and from an economic viewpoint.
Abstract We study comovement among major cryptocurrencies from a portfolio management perspective. To this end, we develop two new statistical tools. First, we propose a new measure called the portfolio-conditional correlation defined as the correlation conditional on the portfolio return being below or above a given threshold. Second, we develop a new multivariate model named the Common Autoregressive Jump Intensity Score-based (ComARJIS) model in which the time-varying intensity of a common jump in cryptocurrency returns is formulated under the Generalized Autoregressive Score (GAS) framework. Our main findings are as follows: First, we find an adverse downside correlation: the downside correlation is higher than the upside correlation. Second, the ComARJIS model successfully shows the correlation asymmetry of cryptocurrencies. Third, and most importantly, a market-timing strategy with the common jump intensity improves the Sharpe ratio. This result suggests that time diversification could be helpful for cryptocurrency investors even if asset diversification is impossible. Fourth, the meltdown risk represented by the common jump intensity is associated with the financial market stress in the U.S.
Abstract This study provides an empirical investigation of long-range dependence (LRD) in financial markets and evaluates the ability of deep generative models to reproduce such temporal structures. Using daily data from three sectors–equity (S&P 500, DAX, Nikkei 225), commodities (Wheat, Corn, Soybeans), and energy (UNG, USO, XLE)–we examine LRD through rescaled range (R/S) analysis, detrended fluctuation analysis (DFA), segmented multifractal analysis around the COVID-19 period, and an ARFIMA–FIGARCH model with Student’s t -distributed innovations. The evidence suggests that while mean returns exhibit limited persistence, pronounced long memory is observed in conditional volatility across most assets, and equity-market scaling properties change non-negligibly after 2020. Building on these findings, we assess whether Quant Generative Adversarial Networks (Quant GANs) can learn and reproduce these stylized temporal dependencies against econometric and resampling benchmarks. Although the generated series reproduce heavy-tailed return distributions and aspects of volatility clustering, they do not consistently capture the magnitude and persistence structure of LRD observed in real data. These results highlight an important limitation of deep generative architectures in modeling slow-decaying dependence structures and underscore the need for explicit long-memory mechanisms when synthetic financial data are intended for risk management or long-horizon forecasting.
This paper estimates a dynamic panel model with feedback in order to investigate the dynamic interdependence between county-level opioid dispensing and unemployment. The paper hypothesizes that opioid prescriptions have deleterious effects on local labor markets. But those negative labor market conditions, once manifested, then lead to increased opioid prescribing. That feedback loop results in an unfortunate cycle in which the two - opioid prescriptions and unemployment - reinforce each other. Estimates reveal the existence of such a reinforcing feedback loop, but the magnitudes involved appear to be too small to facilitate meaningful policy interventions. The paper also offers evidence that opioid dispensing at the county level is endogenous with respect to unemployment, with endogeneity most likely stemming from the presence of healthcare infrastructure that reduces unemployment but also provides easier access to opioids.
Using daily data for major global equity indices over 2010-2025, this study decomposes close-to-close price changes into an overnight opening gap (diffrate) and an within-session return (intradayrate), and quantifies their multiscale structure using Multifractal Detrended Moving Average analysis (MFDMA) and Multifractal Detrended Cross-Correlation Analysis (MF-DCCA). We document three robust patterns. First, the overnight component is typically more persistent and more heterogeneous: MFDMA estimates show higher h(2) and larger multifractality proxies for diffrate in most markets, consistent with information accumulation and cross-market spillovers being compressed into a single opening price-discovery event. Second, intraday and overnight dynamics are nontrivially coupled: within-index MF-DCCA yields cross-Hurst exponents centered around h(xy)(2) approximate to 0.49, and the cross-exponent declines monotonically as q increases, indicating tighter coupling in calm regimes than during extreme co-movements. Third, international comovement is stronger in opening gaps than in intraday returns, suggesting that global information is synchronized primarily at the open, while continuous trading reflects more local microstructure and liquidity conditions. Clustering based on multifractal signatures further reveals richer regime heterogeneity overnight than intraday.
We employ a novel machine learning framework to assess the evolving transmission of U.S. monetary policy since inflation targeting. Estimating time-varying parameter local projections via ridge regression, our approach captures gradual shifts in macroeconomic relationships and accounts for heteroskedasticity across parameters. Our analysis reveals that the time-varying effects of monetary policy on inflation have strengthened, while the Phillips curve has flattened over much of the pre-pandemic period. The results further indicate a diminished pass-through from short- to long-horizon inflation expectations, consistent with more firmly anchored expectations and improved policy credibility. We also document a temporary steepening of the Phillips curve during the post-pandemic inflation surge, suggesting that Phillips curve dynamics are indeed regime dependent.
The demand for green energy has become a critical priority in today's world, where ethanol plays an important role. However, global factors contribute to significant price fluctuations in both ethanol and agricultural markets, leading to extreme risks. This study examines the bidirectional extreme risk spillover effect between ethanol and four key agricultural markets: corn, wheat, soybeans, and sugar. Extreme risk is quantified using CVaR, with dynamic CVaR time series generated through the long-memory FIAPARCH model. The spillover effect is assessed using an innovative robust linear quantile regression method. Our findings show that corn and soybeans exert the strongest extreme risk impact on ethanol, particularly during periods of high volatility. In contrast, the impact from the wheat market is considerably weaker, and no spillover effect is observed from the sugar market. When looking at the reverse relationship, agricultural markets are only minimally affected by extreme risk from the ethanol market, with no spillover detected in the wheat and sugar markets. A complementary portfolio analysis highlights that soybeans provide the best risk reduction for ethanol, as they are the least volatile agricultural commodity.
This paper examines whether the output elasticity of energy varies across emissions states andwhether aggregate emissions or emissions intensity provides the more informative conditioning variable. Afixed-effects panel threshold model is estimated for 110 countries over 2006-2022. Logged total greenhouse-gas emissions are first used to select regimes endogenously. Under this benchmark specification, energy raisesoutput in both regimes, and the point elasticity is larger in the high-emissions state, but the cross-regime differ-ence is not statistically distinct under inference robust to cross-sectional dependence, while bootstrap evidenceprovides only limited support for a global threshold based on aggregate emissions. The empirical picture issharper when emissions intensity is used as the conditioning state. In that case, the energy-output elasticity islower in high-intensity regimes, and the contrast becomes more visible in interaction-based specifications andafter 2015. Source decomposition shows that both renewable and non-renewable energy are less productivein high-intensity regimes, with the attenuation especially clear for non-renewables and with renewables turn-ing negative in the high-intensity state. Overall, aggregate emissions mainly sort countries by scale, whereasemissions intensity more clearly captures economically meaningful heterogeneity in the productivity of energy.
In this paper, we propose a method based on distance correlation theory to measure and test nonlinear dependence between a set-valued random variable and a random vector. This distance-based measure of dependence takes a value of zero if and only if the set-valued random variable and the random vector are independent. The proposed method thus effectively extends the scope of distance correlation from real-valued random vectors to set-valued random variables. This extension can have many potential applications in economics and finance. We then apply the proposed method to measure and test for the association between three salient cryptocurrency characteristics - namely, market capitalization, liquidity, and investor attention - and future returns and volatility, which are summarized by a random interval, in the cross section of over one thousand cryptocurrencies. We find that this association is very strong and it tends to be persistent over time in our sample.
Household wealth from the stock market is unevenly concentrated among higher-income groups, and its distribution may be affected by stock market volatility. This paper investigates the relationship between stock market volatility and consumer spending through the lens of wealth distribution, with particular attention to different monetary policy environments. Using quarterly U.S. data from 1989 to 2024 within a state-dependent econometric framework, the study finds that stock market volatility significantly increases wealth inequality and lowers real per capita spending on durable goods, nondurable goods, and services during periods of easy monetary policy. Conversely, no significant effects are found under tight monetary policy, highlighting the state-dependent nature of the volatility-consumption relationship. This state dependency indicates a nonlinear dynamic relationship between stock market volatility and consumer spending. Additionally, results show that stock market volatility has a temporary effect on economic uncertainty and a smaller impact on consumption behavior than consumer sentiment. Policy recommendations include stabilization of stock market volatility and implementing confidence-building measures to sustain consumer spending and promote steady economic growth.
We propose generative artificial intelligence to measure systemic risk in the global markets of sovereign debt and foreign exchange. Through a comparative analysis, we explore three novel models to the economics literature and integrate them with traditional factor models. These models are: Time Variational Autoencoders, Time Generative Adversarial Networks, and Transformer-based Time-series Generative Adversarial Networks. Our empirical results provide evidence in support of the Variational Autoencoder. Results here indicate that both the Credit Default Swaps and foreign exchange markets are susceptible to systemic risk, with a historically high probability of distress observed by the end of 2022, as measured by both the Joint Probability of Distress and the Expected Proportion of Markets in Distress. Our results provide insights for governments in both developed and developing countries, since the realistic counterfactual scenarios generated by the AI, yet to occur in global markets, underscore the potential worst-case scenarios that may unfold if systemic risk materializes. Considering such scenarios is crucial for designing macroprudential policies aimed at preserving financial stability and assessing the effectiveness of the implemented measures.
We study the impact of temperature on aggregate and sector-specific growth in gross value added using regional EU data observed over 1980-2019 using a mixed frequency dataset. The regional data on economic growth are annual and, since there is wide consensus on projections of a temperature rise more concentrated during fall-winter than spring-summer, we focus on seasonal region-specific temperature shocks and on their interaction also with global seasonal temperature of a given year. The empirical analysis, based on panel local projections, suggests no evidence of adaptation and projections of large losses during hot seasons under a scenario of no implementation of climate change mitigation policies. Investments are important transmission channels driving a negative effect on growth due to a one-degree Celsius increase in the level of seasonal temperature. Moreover, construction is the most vulnerable sector, and the regions with a higher income per capita or competitiveness are the most resilient to temperature shocks.
This paper measures media sentiment for 25 commodities and their dynamic connections with commodity market groups (real assets price indices) and macro-financial shocks, such as inflation, economic policy uncertainty, financial conditions index, supply chain pressure index, geopolitical risk, and industrial production. By applying the time-varying parameter vector autoregressive model, we find that commodities and their sentiments and industrial production are predominantly shock receivers, while economic policy uncertainty and geopolitical risk are shock transmitters. Also, the network approach reveals two main mechanisms by which macro-financial indicators spill over shocks to the commodity markets, either directly or indirectly through the commodity sentiments. We find an asymmetric impact where positive sentiment about commodities receives shocks from macro-financial shocks, while negative sentiment spills over shocks to global inflation. Our results play a significant role in investment decisions and policy formulation by showing how global macro-financial shocks interact and influence commodity markets.
This study examines Brent crude oil price dynamics using an integrated framework of bootstrap sequential break detection and Asymmetric Exponential Smooth Transition Autoregressive (AESTAR) modeling. We demonstrate that oil prices follow an AESTAR process where structural breaks emerge endogenously through dual transition functions, reconciling previously competing explanations in the literature. Analysis of monthly data (1985-2023) identifies major structural shifts coinciding with critical economic events, while revealing these breaks emerge automatically through regime-dependent means. Enhanced testing confirms embedded LSTAR-dominant dynamics with ESTAR components, while skeleton analysis validates the dual equilibrium framework with balanced regime distribution. Generalized Impulse Response Function analysis reveals distinct shock transmission patterns: Tier 1 extreme events (delta max > 1.8) exhibit persistent deviations requiring sustained policy intervention, while Tier 2 events demonstrate mean reversion properties suitable for conventional responses. The framework provides observable threshold levels ($53.62, $37.39) enabling real-time policy intervention, supporting regime-contingent monetary policy and strategic petroleum reserve management protocols. This approach offers policymakers actionable tools for managing oil price volatility through empirically validated intervention strategies.
We investigate whether the transmission of monetary shocks in Poland depends on the level of economic slack. To this end, we estimate smooth transition panel local projections using Poland's regional data and analyze how monetary shocks affect unemployment and prices in regimes of high and low unemployment. Our key finding aligns with economic intuition: the response of unemployment to monetary policy shocks is stronger when economic slack is high, compared to when it is low. Conversely, the adjustment of prices to monetary innovations is more pronounced when idle resources in the economy are scarce, compared to when they are abundant. Our main conclusion is further supported by evidence showing that the difference in the strength of the employment response to monetary shocks, depending on the unemployment level, is more pronounced in sectors producing non-tradable goods than in those manufacturing tradable goods. Moreover, comparing our model with its linear counterpart confirms that monetary transmission in Poland indeed exhibits state-dependence, while the analysis of monetary shock distributions under low and high unemployment shows that our results are not driven by the presence of a regime-dependent pattern in monetary disturbances.
We examine the effect that higher natural disaster frequency has on economic outcomes. Even if there is clear evidence that natural disaster incidents are not only going to be more frequent but will also start affecting a wider pool of countries, research has not yet analyzed the economic impact of the interaction between climate change and more frequent extreme rare events. With this study, we try to unveil the mechanisms through which natural disasters and climate change are interconnected, as well as provide policy insights regarding the adoption of greener inputs, in the form of green capital. Our findings suggest that rising temperatures are expected to negatively affect consumption as well as increase debt. We also show that under green technology adaptation, countries are projected to achieve higher levels of consumption and welfare.
A key limitation of traditional efficiency measurement is the inability of typical functional forms to capture non-linearities. This study proposes an alternative framework for efficiency analysis by integrating a Bayesian neural network with a generalized true random-effects Stochastic Frontier Analysis (SFA) model. In doing so, complex functional forms are better approximated, while both time-varying and time-invariant inefficiencies are considered for. The neural network SFA model is empirically compared with the conventional SFA model using a Cobb-Douglas specification. The results highlight that the neural network model better captures underlying non-linearities and provides a more accurate assessment of inefficiency. Bayes factors provide strong evidence in favor of the neural network model. These initial findings signal the potential of neural networks to enhance the precision and flexibility of efficiency analysis.
This paper investigates global equity market interdependence and contagion during the COVID-19 pandemic using a high-dimensional vector autoregressive (VAR) framework with LASSO regularization. We estimate dynamic structural interconnections among 19 G20 markets by applying a rolling-window sparse VAR model to equity returns purged of global factors. Connectedness is then quantified using generalized forecast error variance decomposition. To detect excess co-movement beyond structural transmission, we conduct residual-based contagion tests comparing pre- and post-pandemic dependence across multiple co-moment channels, including correlation, co-skewness, co-kurtosis, and co-volatility. The results reveal a sharp increase in market interdependence during early 2020, with advanced economies such as the US becoming key sources of return spillovers. Contagion tests uncover significant rises in higher-order dependencies - particularly in asymmetric and tail-related metrics - indicating not only a rise in the intensity of market co-movements, but also a transformation in their structure. These findings highlight the importance of disentangling structural linkages from crisis-induced contagion to better understand systemic risk in global equity markets.
Uncertainty can affect monetary policy through its influence on macroeconomic variables. In this paper, we examine the extent to which economic policy uncertainty influences the effectiveness of monetary policy in the 1965:1-2023:12 period for the U.S. economy. Using a threshold regression model, we find evidence of threshold effects where a threshold is estimated at the 62nd percentile of the economic policy uncertainty variable distribution, which defines two regimes: high and low uncertainty. By estimating a Structural Vector Autoregression (SVAR) model with sign and zero restrictions in each uncertainty regime, we find that the monetary policy is effective during low-uncertainty periods but loses its effectiveness during high-uncertainty ones. These results are robust to the addition of further constraints and other specifications.
