The Finance Association
The Quant Research Track is at the heart of TFA. It operates on the basis of a set of clearly defined methodological principles designed to produce credible and reproducible results. The Quant Research Track is managed by a team of experienced researchers who are responsible for designing and implementing research projects.
The Quant Research Track is the core activity of TFA. Teams work in small groups across full research cycles, producing a documented research paper reviewed internally before any external communication. Results are presented in a public showcase at the end of the cycle.
The objective of the Quant Research Track is not rapid trading or the maximisation of short-term returns.
The aim is to build a structured, reproducible and credible process — a framework in which every analytical decision is documented, every assumption is explicit, and every result is subject to critical scrutiny.
This prepares members for industry, where process quality consistently takes precedence over the appeal of isolated results.
TFA’s research pipeline follows a structured eight-step framework designed to ensure rigor, reproducibility, and methodological credibility.
Formulate a testable investment hypothesis
Define methodology and data sources
Validate the strategy on historical data
Submit to rigorous stress testing
Each stage of the pipeline addresses a specific requirement in the construction of a credible deliverable:
Formulation of an investable hypothesis grounded in explicit economic rationale. The idea must be testable and refutable.
Selection of data sources, cleaning, structuring, and definition of the methodological framework. Documentation of choices and justifications.
Implementation and testing on historical data. Analysis with attention to potential biases (look-ahead, survivorship, overfitting).
Assessment of stability across market conditions, parameter settings, and time periods. Identification of points of fragility.
Formal documentation of context, methodology, results, risks, and limitations in a structured research format.
Peer-review within the association to verify methodological consistency, documentation quality, and transparency of assumptions.
Public presentation of results before members, mentors, and industry professionals. Analytical choices must be defended and discussed.
Post-publication tracking under controlled educational conditions to study implementation constraints and signal degradation.
Discover the latest research papers, trading strategies, and financial insights from our talented members. New publications are added weekly.
By Haewon Jeong, Alexandre Goumaz, Philipp Glaser
Combines technical indicators, sentiment analysis, and momentum trading into a unified deep learning investment framework for NYSE stocks.
By Michel Saliba
Hybrid CNN-LSTM architecture for high-frequency cryptocurrency forecasting, demonstrating out-of-sample predictive performance on XRP.
By Jens Cancio, Youssef Chelaifa, Szymon Ścibior
This project aims to build a Reinforcement Learning agent that learns optimal dynamic hedging strategies for derivatives.
By Michel Bassil, Chandrasekhara Devarakonda
This study evaluates whether fill probability alone is sufficient for market making profitability. Using high-frequency Binance data, it shows that while fill prediction is strong, profitability only emerges when optimizing directly for realized PnL.
By Mert Ülgüner, Linze Li
This project studies whether dispersion in tone during earnings call Q&A sections contains predictive trading signals. Using NLP, it analyzes heterogeneity in analyst–management interactions to capture uncertainty and disagreement beyond average sentiment.
By Pablo Habib, David Attias
This project explores a physics-inspired Marketron framework for equity markets, introducing a nonlinear latent-state model with regime shifts, memory effects, and metastability to improve equity timing beyond classical GBM assumptions.
By Karl Saliba, Michel Saliba
We test five NLP models (TextBlob, CryptoBERT, FinBERT, Qwen2.5-7B-Instruct) on 2021–2023 crypto news for agreement and predictive value against market data. Bitcoin resists sentiment-based prediction, but Qwen2.5-7B-Instruct predicts Ethereum/altcoin direction at 53% accuracy, mostly above chance. Open-weight models appear viable for crypto sentiment analysis without retraining.
By François Goybet, Gavriil Kharkhordine, Haocong Li
We use cross-sectional learning-to-rank (XGBoost) on Open Source Asset Pricing characteristics to rank stocks monthly rather than predict returns. This yields significant long–short spreads, monotonic deciles, and strong out-of-sample performance, benchmarked against a transaction-cost-aware RL baseline.
By Aryan Pandey, Rayan Harfouche
This project models market activity following macroeconomic shocks, drawing on prior research (Lillo & Mantegna) that shows extreme price movements decay in a pattern similar to earthquake aftershocks. It tests whether that post-shock volatility relaxation is predictable enough to exploit in systematic trading strategies.
By Gaël Saugy, Balázs Peisz, Axel Turin-Plessia
Systematic trading strategy adapting to market regimes, using signal selection and asset universe adjustments.
By Moutie Mohamed Ali, Guillen Steulet, Jules Broglin
Market-neutral statistical arbitrage strategy applied to US and Swiss equities, extending vanilla pairs trading with calibrated methods.