Introduction to Financial Data Science

Investment firms now work with data far larger and messier than price histories and financial statements, and with tools that learn patterns from that data or generate text on request. This module introduces the vocabulary of financial data science: what makes data 'big', how machine learning models are trained and evaluated, what large language models can and cannot do, and how these tools are used, and governed, in fintech and investment management.

Start this lesson free Practice questions

What you will learn

1. Describe the characteristics and sources of big data, including alternative data, and the steps needed to prepare it for analysis.

Four Vs: volume, velocity, variety and veracity (plus value).
Structured data fit rows and columns; unstructured data include text, images, audio and video.
Alternative data come from individuals, business processes and sensors.
Preparing data: capture, curation (cleaning), storage, search and transfer; text needs tokenizing and normalizing.
Check privacy, licensing and the risk of material non-public information before using alternative data.

2. Describe the main types of machine learning, explain how models are trained and validated to avoid overfitting, and calculate and interpret basic measures of classification performance.

Supervised learning uses labelled data (regression, classification); unsupervised finds structure (clustering, dimension reduction).
Deep learning uses multi-layer neural networks; reinforcement learning learns from rewards.
Overfitting: strong in-sample, weak out-of-sample; underfitting: too simple to capture the pattern.
Use training, validation and test sets, or k-fold cross-validation, to judge out-of-sample performance.
Precision = TP / (TP + FP); recall = TP / (TP + FN); accuracy misleads when classes are imbalanced.

3. Describe artificial intelligence, natural language processing and large language models, their uses in investment management and their main limitations.

AI: systems performing tasks needing human intelligence; generative AI creates new content.
NLP turns text into usable information; LLMs are transformer networks trained to predict the next token.
Fine-tuning and retrieval-augmented generation adapt LLMs to specific tasks and documents.
Uses: summarizing filings and calls, extracting data, drafting reports, sentiment analysis, coding.
Limitations: hallucinations, knowledge cut-offs, bias, poor explainability, confidentiality risk; outputs must be verified.

4. Describe how big data, machine learning and AI are applied in fintech and investment management, and the risks and governance these applications require.

Text analytics, alternative data and ML support research; net sentiment = (positive - negative) / (positive + negative).
Robo-advisers automate portfolio construction and rebalancing at low cost.
Algorithmic trading reduces market impact; high-frequency trading exploits speed.
ML supports risk monitoring, fraud detection, credit scoring and compliance.
Risks: model and data risk, crowding, flash events, bias and cyber risk; governance needs oversight, validation and controls.

Module at a glance

Big data: volume, velocity, variety and veracity; structured versus unstructured.
Alternative data come from individuals, business processes and sensors, and must be curated and checked legally.
Supervised learning predicts labelled targets; unsupervised learning finds structure; deep learning uses multi-layer networks.
Guard against overfitting with training, validation and test sets and cross-validation.
Precision = TP / (TP + FP); recall = TP / (TP + FN); accuracy misleads for rare events.
LLMs predict tokens and can summarize, extract and draft, but hallucinate; verify outputs.
Applications span research, robo-advice, trading and risk; they require model governance and human oversight.

This free module has 4 lessons, 12 practice questions with full explanations.

Start the lesson