About the Role This is a quantitative research role in the buy-side sense of the word. You will be responsible for the alpha content of a live market-signals product: deciding what constitutes a real, tradable signal, proving it with statistics that would survive a due-diligence meeting, and standing behind the numbers when a sophisticated financial client asks how they were produced.
What You Will Do Own the signal set
- Make the promote, hold or deprecate decision on every candidate signal produced by the discovery process. A full run evaluates thousands of candidates across taxonomy groupings, markets and horizons - your judgement is the gate between a backtest and a published claim.
- Interrogate promotion evidence rather than accepting it: rank information coefficient, AUC, directional hit rate, precision at K, temporal stability, and false-discovery-rate-adjusted significance against minimum observation counts.
- Confirm every promoted signal survives a locked out-of-sample holdout and a placebo battery (shuffled dates, shuffled labels, future-shifted timestamps) before it reaches clients.
- Separate genuine inverse relationships - negative IC is common and legitimate in risk and geopolitical themes - from artefacts, and confirm sign handling is correct at prediction time.
Backtesting and research design - Own the walk-forward backtesting framework in practice: expanding folds, purge and embargo gaps to prevent look-ahead, per-fold aggregation, and combination of evidence across folds. Challenge the design where it is too permissive or too conservative.
- Design and test compound research hypotheses - multi-factor combinations, sentiment-conditioned filters, geographic constraints, and volatility-regime conditioning.
- Own the promotion threshold policy. Recommend evidence-backed changes and quantify the false-discovery cost of loosening any criterion.
- Guard against the classic failure modes: multiple comparisons, survivorship, data leakage from enrichment, and regime-specific overfitting.
Prediction quality and calibration - Measure live forecast quality honestly - headline accuracy, accuracy by market and by horizon, Brier score, and reliability curves with expected calibration error.
- Own the accuracy-versus-coverage trade-off. A high accuracy figure only means something on a defined high-confidence slice; determine and defend the confidence threshold at which the target holds, with the coverage cost stated explicitly.
- Set and tune the abstention policy - when the model should decline to call a market - balancing selectivity against commercial usefulness.
- Benchmark against naive baselines (always-neutral, always-long) and refuse to report an edge that does not beat them.
Monitoring and decay - Track promoted signals for decay using rolling IC, Z-scores, changepoint detection and slope-change diagnostics; confirm or override automatic deprecations.
- Maintain the health of the resolution pipeline that converts forecasts into realized outcomes, since every accuracy metric depends on it.
Required - Quantitative research experience. 5+ years in a quantitative research, quantitative analyst, systematic strategy or financial data science seat - at a hedge fund, asset manager, proprietary trading firm, bank quant desk, or a financial data or alternative-data provider. You must have owned signal or factor research, not solely implemented someone else's model.
- Financial markets fluency. Genuine comfort with equity index and ETF return series, forward-return construction, trading horizons, volatility regimes, and macro context. You should be able to look at a signal and form a view on whether the economic story behind it is plausible.
- Statistical rigour. Working command of hypothesis testing, multiple-comparisons correction (Benjamini-Hochberg or equivalent), rank correlation, ROC/AUC, calibration and proper scoring rules. You should be able to explain what a q-value guarantees that a p-value does not.
- Time-series discipline. Hands-on experience with walk-forward and purged cross-validation, look-ahead bias prevention, holdout design, and regime-dependent performance.
- Python as a research tool. Fluent with pandas, NumPy, SciPy, statsmodels and scikit-learn - enough to reproduce, modify and extend research code independently. You are not expected to build production services.
- SQL. Able to write non-trivial analytical SQL to interrogate signals, forecasts and coverage without waiting on an engineer.
- Intellectual honesty. A demonstrable track record of killing your own results. This role exists to prevent the publication of a false edge; scepticism has to be a reflex, and it has to survive commercial pressure.
- Communication. Able to write methodology that stands up to a buy-side reader and explain it verbally to a non-quantitative executive audience.
Preferred - Experience with news, sentiment, filings or other alternative-data signals and their particular failure modes.
- Familiarity with gradient-boosted ensembles (LightGBM, CatBoost, XGBoost), stacking with out-of-fold predictions, and isotonic or Platt calibration.
- Exposure to conformal prediction, selective-prediction or abstention frameworks, or cost-sensitive decision thresholds.
- Prior work on a commercial data product where the methodology was client-visible and contractually relevant.
- Working knowledge of a cloud analytics environment (GCP BigQuery / Vertex AI or equivalent).
- Graduate degree in statistics, financial engineering, econometrics, mathematics, physics or a comparable quantitative discipline. CFA, CQF or FRM is a plus but not a substitute for research experience.
Benefits: 
