Advanced Econometrics and Deep Learning for Financial Time Series

Learning objectives:

• Master the transition from classical statistical models (ARIMA, GARCH) to advanced deep learning architectures (LSTM, Autoencoders) for financial forecasting and classification.
• Analyze market behavior through the lens of volatility clustering and heavy-tailed distributions.
• Implement end-to-end algorithmic trading pipelines and automated portfolio management systems in Python.
• Evaluate model uncertainty and regime detection using sequential learning techniques.

After completing this course, participants will be able to:

1. Build and tune ARIMA and GARCH models to estimate financial risk and volatility.
2. Design and implement LSTM-based architectures for multi-step price forecasting.
3. Use sequential learning to identify shifts in market regimes and volatility clusters.
4. Develop automated trading strategies and optimize portfolios using Deep Learning.
5. Integrate alternative data (Sentiment, News) into financial decision-making processes.

Week 1: Financial Markets Zoo and Paper Trading

• Introduction to financialmarkets: purpose, asset classes, andmarket jargon.
• Overview of securities: stocks, bonds, options, futures andforwardscontracts.
• Understandingpayoffsand price relationships.
• Portfolio basics andmarketindices (S&P 500, ETFs). (Video lecture on Market Indices Zoo)
• Arbitrageconceptsand transaction costs.
• Basics of TechnicalandFundamental Analysis. (Video lecture)

Lab:

• Setting up a paper tradingaccount.
• Placingorders, monitoringpositions, and understanding commissions.
• Creatingandevaluating simple trade ideas.

Week 2: Time Series Analysis and Forecasting

• Principles of time series forecasting.
• Introduction to stationaryand non-stationary processes.
• Autocorrelation, seasonality, andtrends in financial data.
• Econometric models: AR, MA, ARIMA.
• Modelingvolatilitywith ARCH and GARCH. (Video lecture on Volatility)

Lab:

• Pricedirectionalityand forecasting
• Translating model outputs into simple tradeideas.
• Usingforecastsand volatility estimates to guide paper trading decisions.

Week 3: Neural Networks for Time Series and Strategy Design

• Introduction to neural networks for forecastingfinancial data. (Videolecture)
• LSTM networks for sequential data.
• Autoencodersfor featureextraction and anomaly detection.
• Trainingandevaluating neural networks on time series.
• Comparing neural network forecastswithtraditional time series models.
• Practicalconsiderations: cross-validation for sequence data, overfitting, data scaling, andinterpretability. (Video-lecture)

Lab:

• Evaluatethemodel’s output in a trading environment.
• Autoencoders for frauddetection

Week 4: Advanced Applications – Portfolio Theory, and Algorithmic Trading

• Introduction to portfoliotheory: risk, return, anddiversification.
• Constructingandevaluating portfolios using risk-adjusted metrics.
• Alternative data, sentiment analysis (Take-home readings, videolecture)
• Basics of algorithmictradingand automated execution.
• Combiningforecasts, volatilityinto actionable trade signals
• Evaluatingstrategy performance andrisk metrics
• Practicalconsiderations: overfitting, transaction costs, andmarket impact

Lab:

Portfolio rebalancing strategies, backtesting

Algorithmic trading with alternative data

Argimiro Arratia

Ariel Duarte López

 Argimiro Arratia. Researcher and Associate Professor at the Polytechnical University of Catalonia (UPC), professional educator and private Fintech consultant.  PhD in Mathematics from the University of Wisconsin – Madison, and MSc in Computer Science, Master in Mathematics from Wisconsin-Madison. My scientific career comprises two stages (so far): I began doing research on Finite Model Theory and Computational Complexity (i.e. Descriptive Complexity), but since 2010 I have focused my research and publication efforts in Financial Time Series Analysis, Optimization Heuristics and Mathematics of Finance with an emphasis on their algorithmic and numerical aspects (i.e. Computational Finance).

Ariel Duarte López. Associate Lecturer in Statistics at the Polytechnic University of Catalonia (UPC). He holds a PhD in Statistics and Operations Research from UPC, with research on extending the Zipf distribution to model real network structures, as well as a Master’s in Informatics and a Bachelor’s in Computer Engineering. With over a decade of experience in data science and applied mathematics, his work spans statistical modeling, machine learning, deep learning, network modeling, and sentiment analysis in finance.

The financial industry is undergoing a paradigm shift where traditional econometric models are being augmented by Deep Learning to capture non-linearities and process massive datasets. This course bridges the gap between these two worlds.

Participants will begin with an essential “Financial Market Zoo” to understand different types of financial instruments, basic financial laws, and classical trading methods used for traders and market investors. We then dive into Time Series Analysis, mastering stationary processes and Gaussian forecasters. The core of the course focuses on Deep Learning applications: using LSTMs for return forecasting, Autoencoders for risk metrics, and sequential learning for market regime detection. Finally, the course culminates in practical applications of Algorithmic Trading and Portfolio Theory, incorporating alternative data such as financial text-based sentiment indicators and technical indicators.

Prerequisites (knowledge, skills)

• Comfortable programming in Python (variables, functions, basic libraries)
• Basic knowledge of linear algebra (vectors, matrices)
• Introductory understanding of probability and statistics
• Ability to work with CSV/tabular data and Jupyter notebooks is highly recommended

Certificate of Achievement

Required readings or materials

• Argimiro Arratia (2014) Computational Finance: An Introductory Course with R (Atlantis Press-Springer). Available online for UPC members.
• E. Zivot & J. Wang (2006) Modelling Financial Time Series with S-PLUS. https://faculty.washington.edu/ezivot/econ589/manual.pdf

• James et al., An Introduction to Statistical Learning. https://www.statlearning.com
• Deep Learning for Finance (Selected papers on LSTM for sequence modeling, Autoencoders).
• Scikit-learn user guide: sections on supervised learning, unsupervised learning and model evaluation

Technical setup or resources needed:

• Google Colab (Python 3.x).
• Libraries: numpy, pandas, matplotlib, statsmodels, arch, tensorflow, scikit-learn
• Computer with a minimum of 8 GB of RAM. 16GB recommended.
• Stable internet connectionwith the ability to video conference