Robert Taylor

Portfolio

• Projects / Global Data Plc

  › EMs: Designed, built, and deployed a production-grade weekly Economic Activity Indicator (EAI) framework from industrial production-oriented microdata, piloted on China, combining soft calendar adjustment, Kalman filtering, rolling statistical normalisation, and equal-weight aggregation within an explicitly causal, rule-based structure; engineered the full low-latency Python data pipeline end-to-end; including ingestion, orchestration, data-vintage management, model versioning, and subsequently extended the framework to a multi-country EM indicator suite, personally implementing the India, Brazil, and South Korea EAIs and extending to further EMs with the assistance of additional analysts.

  › Created a cross-country Forward EPS screener centred on earnings revisions, combining (out of sample) statistically significant short-term momentum and contrarian revision-exhaustion signals with a macro-oriented state-space (1 & 2 month multi-step) forecasting model for 12m forward EPS; with a feature matrix crafted by a novel feature-engineering pipeline covering hypothesis-driven channel construction, controlled feature expansion on primitives, and VIF/causal/information-theoretic pruning across activity, cost, financial-conditions, and external-demand channels. Conducted all research and experimentation end-to-end, delivered RMSE reductions of up to ~25–30% versus naïve benchmarks in multiple geographies, and contributed the resulting signals to TSLombard’s Global Asset Allocation publication.

  › Developed an unsupervised, state-space–based labour-market signal extraction model to fuse heterogeneous high-frequency employment indicators, replacing supervised MIDAS-style approaches with a Kalman-filter framework featuring precision-weighted observation matrices, lag-optimised feature selection, causal filtering, and post-estimation spline-based level reconciliation to annual benchmarks; conducted full methodological research, feature engineering, and empirical validation in Python, demonstrating robust signal fusion under biased or unreliable ground-truth labels and providing a production-viable alternative for real-time labour-market monitoring.

  › Creation of a comprehensive pipeline of daily parametric market event triggers for financial securities and using conversation embeddings to map and align client interactions with market and economic events for targeted outreach to buy-side clients.

  › Researched and developed a structurally identified multi-agent SVAR of USD dynamics relative to alternative reserve currencies (EUR, JPY), modelling global banks as balance-sheet-constrained intermediaries in the spirit of Gabaix–Maggiori and BIS literature; instrumented dealer capacity and institutional hedging channels using bank-level exposures constructed from Pillar 3 regulatory disclosures, cross-currency basis measures, and interest-rate differentials, stress-tested identification against alternative supply-side and funding shocks, and quantified ex-post USD responses via impulse-response analysis within a Python-based research workflow.

  › Created a continuous regime framework for modelling state-dependent CTA–equity return dynamics, replacing discrete regime classification with a smooth regime score inferred from orthogonal signal subspaces spanning cross-asset dispersion, volatility level and transition dynamics, and transforms of global term structures (slope innovations, wedges, trend-gap levels and accelerations); identified degradation in range-bound and reversal-dominated states versus early-stage trend regimes, and mapped the inferred state into a soft-gating indicator yielding continuous CTA exposure weights for portfolio construction in our GAA Publication

  Stack: Python (SKLearn, NumPy, Statsmodels, joblib, Polars, Pandas, PyTorch, Statsmodels, Trafilatura, BeautifulSoup), MySQL, Snowflake, Sourcetree, Salesforce, Datastream, CEIC

• Projects / ARC Research, SPDJI

  › Modelling market risk of the underlying securities for a with-profits, unit-linked fund on a semiannual basis. I assisted in the development and implementation of proprietary techniques to accurately measure ex-post returns and mark-to-market risk. This included bespoke relative performance reports, comparing the fund's performance against selected benchmarks and peer groups. These reports featured detailed performance attribution and visualisations of the fund's relative sperformance despite its otherwise smoothed with-profits nature.

  › Conducted fee report analysis, examining the fee dispersion (including total TER breakdowns) of UK-based PWM/DFM Investment Managers across different risk mandates. This analysis provided managers with critical information about their costs in the context of hundreds of other managers, ensuring compliance and competitiveness in light of the UK's consumer duty regulation.

  › Used LOOCV-based factor analysis to identify features for a closed MVO-style private tracker fund. Additionally, I was responsible for the daily rebalancing and validation of the portfolio, ensuring optimal performance and alignment with the strategic objectives of beneficiaries.

  › Contributed to the quantitative work of a FT-featured commentary that highlighted how wealth managers could selectively present performance data to appear in the top quartile. Our detailed report examined over 200k GBP-denominated private client portfolios, revealing the significant impact of meticulous reporting period use and benchmark selection to enhance perceived performance.

  Stack: Mathematica

• Research

  Evaluating Credit VIX (CDS IV) Prediction Methods with Incremental Batch Learning

  (Self-Published on SSRN/arXiv)

  My recent work presents a custom Attention-GRU Hybrid ANN which outperforms benchmarks and SV models in predicting the Credit VIX (CDS Implied Volatility levels). Tested on a highly volatile out-of-sample period in 2024, the research embraces incremental batch learning, contributing to improved financial risk model validation methods.