Maths to markets. Systematic from day one.
I started in mathematics and statistics, then spent time in industry building data pipelines and analytics at scale. That work sharpened an interest in machine learning for adversarial settings, which led to an MSc in Financial Engineering at Imperial and the development of systematic trading systems for live markets.
Tevvis grew out of that work in 2023 as an independent research vehicle for systematic equity strategies deployed on real capital. The name means 23 in Marathi.
Current research extends this into game-theoretic multi-agent reinforcement learning for computer vision, through a funded PhD supervised by Shaheer at the Barts Cancer Institute.
A regime-conditional ML strategy deployed on real capital via Trading212. Fully systematic with no discretionary overrides. The strategy identifies market regimes, generates signals within each, and constructs portfolios of liquid equities.
Performance represents live, real-capital results from a funded portfolio managed since January 2023. Past performance is not indicative of future results. Tevvis is a research operation, not an investment adviser.
Working papers on regime-conditional equity strategies.
Investigates the distinction between features used for screening and features used for clustering in systematic equity strategies, arguing that conflating the two introduces systematic bias into portfolio construction.
Read on SSRNExamines how short-horizon excess returns in liquid equities vary across market regimes, demonstrating that a regime-conditional programme captures alpha that unconditional models miss.
Read on SSRNUses bootstrap methods on a century of US equity data to disentangle whether systematic strategy performance arises from genuine return predictability or from implicit risk timing through regime detection.
Read on SSRNDecomposes the CDS-bond basis through a Markov-switching framework, modelling friction dynamics and nonlinear amplification mechanisms across credit market regimes.
Read on SSRNExamines currency factor combination through shrinkage convergence and alternative methods, identifying volatility risk as a distinct priced factor in cross-sectional currency returns.
Read on SSRNProposes clustering-based portfolio construction methods as alternatives to traditional mean-variance optimisation, showing improved out-of-sample stability and reduced sensitivity to estimation error.
Read on SSRNApplies probabilistic and machine learning methods to predictive modelling in alternative data settings, with applications to sports analytics as a domain for out-of-sample validation.
Read on SSRN