DATA-DRIVEN MODELS FOR INVESTMENT

Antonio Simeone, Elio Stocchi

Instructional goals

•Learn how to use alternative data like satellite and geospatial data, web data and news/media with LLM and Generative AI to improve investment decisions •Explore Machine Learning and Decision-Making methods with a focus on investment applications. •Implement a live trading strategy following the most valuable Hedge Fund techniques •Build a systematic trading strategy with both a qualitative and a quantitative AI approach

Prerequisites

The students should know basis of Python language, even if it is not essential. No previous knowledge of finance is required.

Intended learning outcomes

Qualitative AI: Understand the principles of Generative AI and LLM, with a focus on investments Explore advanced data synthesis techniques Analyze alternative data with Generative AI Models Apply Generative AI in classification and prediction of financial investments Quantitative AI: Integrate Fuzzy Logic and Neural Network to predict financial time series Analyze and predict real financial data with AI Develop a Quantitative AI systematic trading strategy

Course Contents

Use of alternative data (web, satellite, social, news & media) Natural Language Processing Systematic Trading Large Language Models Fine – tuning of AI Models Real & Complex Fuzzy Logic Adaptive Fuzzy Neural Networks Prediction of financial phenomena

Reference Books

The course materials will be based on collection of top scientific articles and book reports. Articles and other materials (cases) will be available through the Luiss Learn platform.

Teaching Methods

Real time coding sessions Teamwork Interactive trading session To successfully complete the course, students must attend at least 60% of classes.

Assessment Method

1 case study for the first qualitative AI part, not mandatory, evaluated up to 3 points. 1 case study for the second quantitative AI part, not mandatory, evaluated up to 3 points. Final project regarding the creation of a systematic trading strategy integrating qualitative and quantitative AI approaches, evaluated up to 15 points. Oral exam evaluated up to 15 points. Each vote greater than 30 will be 30L.

Thesis assignment criteria

The thesis will be assigned (upon specific request to the instructor) to students who demonstrate a serious and motivated interest to the course subjects.

Week 1

Collection and Analysis of Alternative Data Identifying and sourcing relevant alternative data (e.g.,satellite and geospatial data, ESG data) Initial preprocessing and data cleaning techniques

Week 2

From Natural Language Processing to Large Language Models Architecture of Transformer models Self – attention mechanism and fine tuning

Week 3

Training Large Language Models Fine – tuning of LLM Iterative testing and refinement of LLM

Week 4

Text-based Data Processing with LLMs Techniques for extracting financial insights from alternative data (e.g.,satellite and geospatial data, ESG data) Analyzing news articles and reports for market impact using LLMs

Week 5

Retrieval Augmented Generation (RAG) Optimization of an LLM with RAG RAG to fine – tune analysis of alternative data

Week 6

Enhancing Trading Signals with Generative Models Generating synthetic financial time series data for model training Techniques for augmenting trading signals using generated data

Week 7

Creation of a concrete trading strategy Learning and application real techniques deployed in one of the most important UK based Hedge Fund Insights from the Head of Capital Markets at this Hedge Fund

Week 8

Real Fuzzy Logic Introduction to Fuzzy Logic and Membership functions

Week 9

Fuzzy Neural Networks Metaheuristic optimization models Adaptive Fuzzy Neural Networks

Week 10

Complex Fuzzy Logic Complex Numbers Introduction to Complex Fuzzy Logic

Week 11

Complex Fuzzy Neural Networks Complex Neuro Fuzzy Logic ANCFIS & Hybrid Learning

Week 12

Systematic Trading strategy Use Fuzzy Logic and Neural Networks to predict financial time series