Peter Kostovčík is a quantitative trader and researcher with a background in mathematics, machine learning, systematic equities trading, and the development of live trading systems.
In this episode, Peter joins us for a practical conversation about machine learning in finance, feature engineering, research pipelines, trading-system design, portfolio construction, position sizing, liquidity, and the growing role of AI agents in quantitative research.
Peter explains why the model itself is often the least important part of a machine-learning strategy. While researchers may focus on increasingly sophisticated algorithms, he argues that most of the real work lies in understanding the data, developing meaningful features, defining the correct objective, and forming ideas that reflect how markets actually operate.
We discuss when machine learning is useful and when it may introduce unnecessary complexity. Peter explains why it can be effective when ranking or classifying thousands of equities, but far more vulnerable to overfitting when applied to a single asset with only one historical time series.
The conversation explores how Peter uses classification, probability distributions, clustering, and other machine-learning methods to narrow large investment universes into manageable groups of potential trades. Rather than expecting a model to predict an exact return, he focuses on identifying relative opportunities that can be combined within a portfolio.
Peter also breaks down his research process, beginning with an idea and a simple statistical test before moving into feature engineering or model development. We discuss in-sample analysis, out-of-sample validation, protected backtest periods, and why repeatedly examining test results can quietly transform supposedly unseen data into another source of overfitting.
The discussion then moves from research into live implementation. Peter explains why liquidity, spreads, slippage, short-selling restrictions, turnover, market capacity, and execution constraints must be considered from the beginning. A strategy may look attractive statistically while remaining impossible to trade at meaningful size.
We also examine the role of AI agents in quantitative research and software development. Peter shares how AI has accelerated his ability to build research infrastructure and test ideas, while emphasizing that it still requires experienced supervision. Without clear constraints and market knowledge, an AI-generated system may reproduce unrealistic assumptions found in academic papers, online examples, and historical training data.
Peter discusses his preference for equities, where large and diverse universes provide more opportunities for systematic research. He also shares lessons from expanding strategies into international markets, where a valid signal can still fail because the available liquidity is insufficient.
Finally, we explore portfolio construction and position sizing. Peter explains why he begins with equal weighting before adding sector, liquidity, volatility, and concentration constraints. We compare equal weighting with probability-based sizing, discuss the limitations of universal formulas such as the Kelly criterion, and examine why the appropriate solution depends on the specific strategy, holding period, and market environment.
A practical and detailed conversation on machine learning, feature engineering, systematic trading, research validation, AI agents, liquidity, portfolio construction, and the challenges of turning a promising model into a robust live trading system.
*DISCLAIMER*
The information shared on this podcast is for educational and informational purposes only and reflects the personal opinions of the hosts and guests at the time of recording. Nothing in this podcast constitutes financial, investment, legal, tax, or trading advice, and nothing should be interpreted as a recommendation to buy, sell, or hold any security, cryptocurrency, derivative, or financial product.
Trading and investing involve substantial risk, including the possible loss of all or part of your capital. You are solely responsible for your own decisions, and you should consult a qualified professional before making financial decisions. By listening to this podcast, you agree that the hosts, guests, and producers are not liable for any losses or damages arising from the use of any information discussed.
