Nikhil Sunder
Quantitative Economics & Finance · Open-Source Scientific Software
Building open-source infrastructure for computational economics — from raw data acquisition to model-ready analysis.
I am an undergraduate in Quantitative Economics & Finance at the University of Miami Herbert Business School, a Research Fellow at the Intelligent Computer Systems Research Institute, and founder of toros-dev. My work sits at the intersection of econometrics, machine learning, and research-grade software engineering: I build the data infrastructure that quantitative research depends on, and I use it to study non-stationary dynamics and monetary policy.
The through-line is a single stack. Acquisition libraries pull economic and disclosure data from primary sources. Representation layers turn that data into typed, validated structures. Modeling libraries — linear and neural — sit on top. Each piece is released as a durable, documented, independently useful artifact rather than as research code.
Software
Production Python packages on PyPI and conda-forge, built to a consistent standard: full type coverage, strict sync/async parity, defensive validation, OpenSSF Best Practices certification, and archived releases with citable DOIs.
| Project | Role in the stack | Status |
|---|---|---|
fedfred |
FRED / ALFRED / GeoFRED / FRASER client | Released |
edgar-sec |
SEC EDGAR REST API client | Released |
toros |
Self-validating financial data representation | In development |
cultivars |
Bayesian, time-varying, and structural VAR | In development |
ns-sdn |
Neural spectral state-space architecture | In development |
Research
Current work centers on state-space models, spectral decomposition, and reinforcement learning for monetary policy.
- NS-SDN — a neural state-space architecture that decomposes non-stationary time series into a time-varying trend plus adaptive spectral components, published at FLAIRS-39 with a follow-up on adaptive spectral emission heads under review.
- Autonomous Fed — reframing monetary policy as a control problem, benchmarking an RL agent against Hinterlang & Tänzer (Bundesbank DP 51/2021) across linear TVP-SVAR and nonlinear NSSM environments.
Affiliations
ICSRI · toros-dev → · Press & coverage → · About me →
