Research
Research
Sovereign debt pricing with shifting long-run growth expectations, Online Appendix, [SSRN], with Pei Kuang
European Economic Review, Volume 186, June 2026, 105310
Overview: The paper presents new evidence of systematic patterns in real-time estimates of long-run output growth rates and, importantly, reveals a negative, nonlinear relationship between these estimates and sovereign debt spreads during the Eurozone debt crisis of the 2010s. To study the implications of these beliefs, we develop a sovereign default model in which agents infer trend growth from aggregate output and from noisy signals about the trend. The model reproduces these empirical patterns in the trend growth estimates and their negative and nonlinear relationship with spreads, unlike a comparable full-information model. Overoptimism about trend growth during booms encourages excessive borrowing, leading to persistently elevated spreads thereafter.
Transfer-Induced Debt Dynamics in Sovereign Default, [SSRN]
Journal of International Money and Finance, Volume 167, August 2026, 103623
Overview: During the 2010s, the Greek sovereign debt crisis necessitated unprecedented EU financial aid accompanied by austerity conditions. Despite reforms to government spending and taxation, limited consolidation of social transfers led to unexpected expansions in transfer payments. This paper constructs a strategic sovereign default model, calibrated to Greek data, to examine the effects of transfer shocks on sovereign debt spreads. Under high financial stress (e.g., the elevated spreads at the onset of the 2010 Economic Adjustment Program), positive transfer shocks exacerbate already excessive absorption and significantly raise spreads. Under milder stress, these shocks initially produce negligible increases in spreads but lead to persistently higher spreads over the longer term. Stricter transfer-side austerity can mitigate crises and may avert default.
Recovering Annual Dynamics from Lower-Frequency Survey Data: A Discrete-Time Framework for Dynamic Projection Models, corresponding author, with Justin van de Ven, Hrushikesh Kalakandra and Matteo Richiardi
Overview: Dynamic projection models, including agent-based models and dynamic microsimulations, require behavioural and transition rules at their updating frequency, yet the data used to estimate those rules often record outcomes less frequently. Using relationships estimated from lower-frequency data as higher-frequency updating rules, interpolating intermediate outcomes, or ignoring preceding states can distort projected paths. We develop direct estimators for binary and multistate transitions and conditional continuous outcomes, with analytical recovery as a benchmark. The direct estimators parameterise the one-step process and fit its implied multi-step probabilities or densities to observed endpoints. Applications to personal-pension contributions and four-state unsecured debt in the UK Wealth and Assets Survey show improved prediction relative to models that omit the preceding state, treat observed-gap relationships as one-step updating rules, or linearly interpolate intermediate outcomes. Recursive pension projections show that alternative treatments of observation timing produce markedly different individual participation histories, turnover and cumulative contributions. Monte Carlo experiments confirm accurate recovery and the advantage of imposing admissibility when observation gaps vary.
Parallelized Dynamic Programming for Nonlinear Macroeconomic Models: A MATLAB MEX-CUDA Approach, corresponding author, with Alessandro Di Nola, Jan 2026, GitHub link
Overview (new version): We present a MATLAB-centered approach that compiles nonlinear macroeconomic model code into C and CUDA MEX executables. The method reduces runtime and the coding burden of moving from MATLAB prototypes to compiled execution. Using a canonical one-period sovereign default model as the main benchmark, we show that MEX-CUDA delivers roughly an order-of-magnitude speedup relative to MATLAB-parfor while preserving the same MATLAB-centered workow. The method remains eective in richer sovereign default and heterogeneousagent applications, delivering full-solution speedups ranging from 10 to 30 relative to MATLAB-serial. The incremental gains relative to MEX-parfor are smaller because serial equilibrium components remain outside the compiled kernel. MEX-CUDA also remains competitive with CUDA Fortran and PyTorch-CUDA. The results show that MEX methods can relax computational constraints in applied macroeconomic research, while performance gains depend on the parallelizable structure of the underlying algorithm.
Fiscal Austerity and State-Dependent Response of Sovereign Spreads, June 2024 (Previously named "Fiscal Austerity, Investment and Sovereign Default"); revising
Global Financial Cycle, Cross-border Capital Flow and Exchange Rate Regime Choice, with Shaochen Han, Bing Gong and Shiqi Yang, May 2023， Shanghai Journal of Economics, DOI: 10.19626/j.cnki.cn31-1163/f.2023.05.010