בינה מלאכותית וסטטיסטיקה בשוק ההון

At first glance, it may sound like this, but this course is not primarily about building trading strategies. It is a course about statistical thinking and how such thinking relates to developing trading strategies. About what happens to familiar machine learning methodology when it meets a domain that violates many of the assumptions that make ML successful elsewhere. Financial data typically offer low signal-to-noise ratios, a single non-repeatable history rather than i.i.d. samples, distributions that shift over time, and a feedback loop that is rare in other domains. Finance can therefore be understood as a stress test for machine learning practice. Techniques that are routine in vision or NLP, like cross-validation, hyperparameter search, and evaluating on a held-out set, can often fail here when applied without specific care. Moreover, results may still look excellent. A central skill this course develops is the ability to distinguish genuine predictive power from artifacts and to statistically reason about information leakage, multiple testing, selection bias, and overfitting. In the course of addressing these issues, we will also define and understand market-related statistical characteristics and how to interpret them.