AI in Investments 2400-MIKRO-AIBF-AII
The course introduces participants to the complete process of building, testing and deploying AI-driven investment strategies, structured around the pipeline of a modern systematic investment firm: Data → Features → Signals → Forecasts → Portfolio → Risk → Execution → Performance → AI Agents. The central question is: how can a quantitative researcher build an AI-driven investment strategy that survives contact with noisy data, transaction costs, risk constraints and execution reality? The course is delivered in one day as four 90-minute sessions, each combining academic research context with practical demonstrations based on runnable Python notebooks. Case studies are anchored in peer-reviewed research by Robert Ślepaczuk and co-authors.
Session 1 — From Efficient Markets to Algorithmic Strategies: the Efficient Market Hypothesis and its empirical violations (momentum, calendar effects, overnight anomaly); anatomy of an algorithmic strategy; data sources and quality traps (survivorship bias, adjustment errors, time zone mismatch, point-in-time bias); walk-forward optimization (WFO) and double out-of-sample validation; performance metrics (ARC, ASD, MD, SR, IR*, IR**, IR***); position sizing (fixed-fractional, volatility targeting, Kelly criterion); MADL and GMADL loss functions; the S&P 500 momentum case study; transaction costs and the statistical significance of backtests.
Session 2 — Machine Learning and Deep Learning: classical models (ARIMA, GARCH) as baselines; machine learning models (Random Forest, XGBoost, LightGBM); deep learning sequence models (MLP, LSTM, GRU, TCN); transformer-style architectures (Transformer, TFT, Informer, PatchTST, Chronos); the LSTM-ARIMA hybrid case study; volatility forecasting (VIX, GARCH and LSTM ensembles); explainable AI (SHAP, feature importance) as a governance requirement.
Session 3 — Deep Reinforcement Learning, Portfolio Optimization and AI Agents: Markowitz portfolio construction with ML-forecasted returns; systemic risk indicators; deep reinforcement learning for portfolio management (MDP formulation, PPO, SAC, reward shaping); systematic index option writing (Black-Scholes-Merton and Variance-Gamma models); large language models for financial documents; retrieval-augmented generation (RAG); multi-agent investment architectures (data, forecast, risk, execution and portfolio-manager agents with veto logic); final project briefing.
Session 4 — Informer for High-Frequency Bitcoin Trading: a deep-dive into Stefaniuk & Ślepaczuk (2026, Expert Systems with Applications): research questions and BTC/USDT high-frequency data (5/15/30-min); WFO methodology with rolling windows; the evolution from feed-forward networks through RNN/LSTM and Transformer to Informer (ProbSparse self-attention, O(L log L) complexity); the GMADL loss function and its parameters; comparative results across frequencies; extensive sensitivity analyses (validation window size, number of windows, data source, transaction cost levels, market regimes); feature and step importance. Key finding: the GMADL-trained Informer at 5-minute frequency achieves ARC +116% and IR** 7.55 versus 0.04 for Buy-and-Hold.
Participants receive slides, runnable notebooks, case studies and reusable Python modules. Demonstrations use public or synthetic data; nothing in the course constitutes investment advice.
The project ‘Application of Artificial Intelligence in Business and Finance’ (No. BPI/SPI/2024/1/00078) is implemented by the University of Warsaw as part of the Spinaker programme organised by the National Agency for Academic Exchange (within the project ‘Wsparcie tworzenia i realizacji międzynarodowych programów kształcenia’, funded by the European Funds for Social Development 2021–2027).
Course coordinators
Micro-credential certificate
Learning outcomes
Upon completing the course, the participant:
Knowledge:
1. explains the three forms of the Efficient Market Hypothesis and empirically documented anomalies (momentum, calendar effects, overnight anomaly);
2. describes the full pipeline of an algorithmic investment strategy: data, features, signal generation, forecasts, portfolio construction, risk management, execution and performance evaluation;
3. identifies the main reasons why AI strategies fail: overfitting, look-ahead bias, survivorship bias, data leakage, non-stationarity and ignored transaction costs;
4. characterizes the main model classes used in quantitative investing: classical (ARIMA, GARCH), machine learning (Random Forest, XGBoost, LightGBM), deep learning (MLP, LSTM, GRU, TCN), transformer-style (Transformer, TFT, Informer, PatchTST, Chronos), deep reinforcement learning (PPO, SAC) and LLM/RAG-based multi-agent architectures;
5. defines the performance metrics ARC, ASD, MD, SR, IR*, IR**, IR*** and the trading-aligned loss functions MADL and GMADL.
Skills:
6. designs walk-forward optimization (WFO) and the double out-of-sample protocol, avoiding look-ahead bias and data snooping;
7. calculates and interprets risk-adjusted performance metrics, including their sensitivity to transaction costs;
8. applies MADL/GMADL loss functions to align model training with trading profitability;
9. interprets SHAP-based feature-importance and step-importance analyses of financial forecasting models;
10. prototypes ML-based portfolio optimization and a deep-reinforcement-learning portfolio agent using the provided Python notebooks;
11. critically evaluates the empirical results of AI strategies (statistical significance, sensitivity analyses, market regimes).
Social competences:
12. works in a team designing a complete AI investment agent (final project);
13. communicates model limitations and risk controls to both technical and non-technical audiences;
14. understands the ethical and governance constraints of AI in investing and recognizes that the course content does not constitute investment advice.
Assessment criteria
Learning outcomes are verified with a written post-test (closed questions covering the material of all four sessions), administered at the end of the course and compared with a pre-test to measure the learning gain, in line with the SPINAKER program methodology. Completion criteria: attendance at the sessions and completion of the written post-test.
Bibliography
Core readings — research underlying the course:
– Stefaniuk F., Ślepaczuk R. (2026). A neural network informer in algorithmic investment strategies on high-frequency bitcoin data. Expert Systems with Applications, 317, 131599. https://doi.org/10.1016/j.eswa.2026.131599
– Michańków J., Sakowski P., Ślepaczuk R. (2024). Mean Absolute Directional Loss as a new loss function for machine learning problems in algorithmic investment strategies. Journal of Computational Science, 81, 102375. https://doi.org/10.1016/j.jocs.2024.102375
– Michańków J., Sakowski P., Ślepaczuk R. (2024). Generalized Mean Absolute Directional Loss as a Solution to Overfitting and High Transaction Costs in Machine Learning Models Used in High-Frequency Algorithmic Investment Strategies. arXiv:2412.18405. https://doi.org/10.48550/arXiv.2412.18405
– Michańków J., Sakowski P., Ślepaczuk R. (2026). Alternative Loss Function in Evaluation of Algorithmic Investing Strategies Based on Transformer Models. Pattern Recognition Letters, 203, 170–176. https://doi.org/10.1016/j.patrec.2026.03.007
– Mroziewicz T., Ślepaczuk R. (2026). A Novel Approach to Trading Strategy Parameter Optimization, Using Double Out-of-Sample Data and Walk-Forward Techniques. arXiv:2602.10785. https://doi.org/10.48550/arXiv.2602.10785
– Kashif K., Ślepaczuk R. (2025). LSTM-ARIMA as a hybrid approach in algorithmic investment strategies. Knowledge-Based Systems, 320, 113563. https://doi.org/10.1016/j.knosys.2025.113563
– Roszyk N., Ślepaczuk R. (2024). The Hybrid Forecast of S&P 500 Volatility Ensembled from VIX, GARCH and LSTM Models. arXiv:2407.16780. https://doi.org/10.48550/arXiv.2407.16780
– Grudniewicz J., Ślepaczuk R. (2023). Application of machine learning in quantitative investment strategies on global stock markets. Research in International Business and Finance, 66, 102052. https://doi.org/10.1016/j.ribaf.2023.102052
– Kość K., Sakowski P., Ślepaczuk R. (2019). Momentum and Contrarian Effects on the Cryptocurrency Market. Physica A: Statistical Mechanics and its Applications, 523, 691–701. https://doi.org/10.1016/j.physa.2019.02.057
– Ślusarczyk D., Ślepaczuk R. (2025). Optimal Markowitz portfolio using returns forecasted with time series and machine learning models. Journal of Big Data, 12(1), Article 127. https://doi.org/10.1186/s40537-025-01164-z
– Sakowski P., Sieradzki R., Ślepaczuk R. (2025). Systemic risk indicator based on implied and realized volatility. Quality & Quantity, 60(2), 6389–6427. https://doi.org/10.1007/s11135-025-02515-x
– Wysocki M., Ślepaczuk R. (2025). Systematic index option-writing strategies with Black-Scholes-Merton and Variance-Gamma Models. Economic Modelling, 152, 107234. https://doi.org/10.1016/j.econmod.2025.107234
– Kashif K., Ślepaczuk R. (2026). Deep Reinforcement Learning Framework for Diversified Portfolio Management Across Global Equity Markets. arXiv. https://doi.org/10.48550/arXiv.2605.17307
– Lebiedź D., Ślepaczuk R. (2026). Dynamic Multi-Pair Trading Strategy in Cryptocurrency Markets with Deep Reinforcement Learning. arXiv. https://doi.org/10.48550/arXiv.2606.04574
– Płachta K., Ślepaczuk R. (2025). Machine Learning for Daily Return Direction Forecasting: A Comparative Study with Explainable AI Insights. SSRN:5272156. https://ssrn.com/abstract=5272156
– Bieganowski B., Ślepaczuk R. (2025). Supervised Autoencoder MLP for Financial Time Series Forecasting. Journal of Big Data, 12(1), Article 207. https://doi.org/10.1186/s40537-025-01267-7
– Łaniewski S., Ślepaczuk R. (2025). Evaluating the Chronos Foundation Model for Daily Stock Index Forecasting. Proceedings of the 33rd International Conference on Information Systems Development. https://doi.org/10.62036/ISD.2025.48
– Lis S., Ślepaczuk R., Sakowski P. (2026). Overreaction as an indicator for momentum in algorithmic trading. arXiv:2602.18912. https://doi.org/10.48550/arXiv.2602.18912
Foundational references:
– Fama E.F. (1970). Efficient capital markets: A review of theory and empirical work. Journal of Finance, 25(2), 383–417.
– Jegadeesh N., Titman S. (1993). Returns to buying winners and selling losers. Journal of Finance, 48(1), 65–91.
– Lou D., Polk C., Skouras S. (2019). A tug of war: Overnight versus intraday expected returns. Journal of Financial Economics, 134(1), 192–213.
– Harvey C.R., Liu Y. (2015). Backtesting. Journal of Portfolio Management, 42(1), 17–28.
– DeMiguel V., Garlappi L., Uppal R. (2009). Optimal Versus Naive Diversification: How Inefficient Is the 1/N Portfolio Strategy? Review of Financial Studies, 22(5), 1915–1953.
– Madan D.B., Carr P.P., Chang E.C. (1998). The Variance Gamma Process and Option Pricing. European Finance Review, 2(1), 79–105.
– Moody J., Saffell M. (2001). Learning to Trade via Direct Reinforcement. IEEE Transactions on Neural Networks, 12(4), 875–889.
– Vaswani A. et al. (2017). Attention Is All You Need. Advances in Neural Information Processing Systems (NeurIPS). arXiv:1706.03762.
– Zhou H. et al. (2021). Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting. Proceedings of the AAAI Conference on Artificial Intelligence, 35(12), 11106–11115.