Cynthia Cen Chen

In my previous post, I wrote about using LLM's for extracting structured features from natural language. I recently came across a course on GenAI for Quantitative Finance helped clarify an important nuance: 1) LLMs aren't/shouldn't really be used directly in predicting markets 2) LLM's are better used for reducing semantic "noise" LLM's cannot/should not be used to: 1) Judge signal stability 2) Infer lag vs lead of a signal 3) Detect regime dependence LLM's should be used for: 1) Converting noisy articles into consistent concepts 2) Extracting direction, tone, and context 3) Turning text into features at scale These statistical properties are only learned by testing features against market data over time, which is where Traditional ML comes into picture. I am no trading buff but rather than looking at a model which "predicts" price or even direction, I would look at it as a Probability of commodity/stock going up given these features/conditions/regimes. P(Commodity/stock goes up over period k | features or conditions or regimes at time t). The last mile is human! Certificate: https://lnkd.in/ggJK9CCi #Fractal #GenerativeAI #LLMs #MachineLearning #QuantitativeFinance #FinancialML #DataScience #AIinFinance

2

Excited to share: Modern Causal Inference—Methods and Applications is officially published! Causal inference has become one of the most influential areas in data science, and this book aims to capture that momentum. We call it a "modern" guide because it brings together the latest advancements in the field—from machine‑learning–based estimators like Causal Trees, Causal Forests, and Double Machine Learning to emerging frameworks for automated causal discovery. The book also highlights the growing role of Large Language Models (LLMs) in supporting causal analysis, particularly in identifying confounders, refining domain knowledge, and proposing credible causal structures. I’m thrilled to make this contribution to the community, and I hope it becomes a valuable resource for practitioners, researchers, and students alike. You can also explore Chris Kuo’s other popular titles below. Each one is carefully curated to guide and inspire you as you master the subject. Purchase the paperback edition and enjoy a complimentary digital version on your preferred device—yours to keep for personal use. The print edition adopts the classic Springer font and layout for a pleasing reading experience, while the digital edition features a beautifully typeset LaTeX layout for optimal readability. (1) Modern Causal Inference (includes free e‑copy): Methods and Applications: https://a.co/d/0g0GY1VT (2) Introduction to Statistics with Python (includes free e‑copy): https://a.co/d/esAL5P1 (3) Modern Time Series Forecasting Techniques For Predictive Analytics and Anomaly Detection (includes free e‑copy): https://a.co/d/5pL7DxQ (4) Handbook of Anomaly Detection (includes free e‑copy): https://a.co/d/9ED6049 (5) Handbook of NLP with Gensim (no e-copy): https://a.co/d/bXH3Nrg https://a.co/d/0g0GY1VT

140

5 Comments

If you're a data science learner who wants to expand their fundamental analytical skills, follow along Junior Jumbong's detailed and accessible tutorial on credit-scoring EDA with Python.

14

Targeted maximum likelihood estimation (TMLE) is a great tool for causal inference. However, using it when there is missing data can be tricky. Why? What's the best approach? Dashti et al. (2024) explore just this! The authors explain the 8 missing-data methods and scenarios they explored in detail. They also included directed acyclic graphs (DAGs)! Highly recommend this one if you're working with TMLE! [Link: https://lnkd.in/eedkTUCQ] #CausallyCurious #RealWorldEvidence #CausalInference #RealWorldData

72

2 Comments