Olexandr Topchylo on Neural Trading Systems, Adaptation, and Risk
Summary
In this interview, Automated Trading Championship winner Olexandr Topchylo discusses automated trading, investor accounts, and lessons from competition. He describes using neural-network-based Expert Advisors, but cautions that neural networks are complex and not inherently superior to other approaches. He says adapting system parameters to recent price behavior can help a strategy endure changing markets, giving volatility-linked stop distances as an example. He also found little performance gain from neural-network committees in his own tests.
Topchylo sees extended live competition as a way to expose defects that short tests may miss, and suggests that studying participants’ statistics and logs can reveal strengths and mistakes. He discusses the difficulty of choosing position size when a contest rewards balance growth, since large lots can undermine an otherwise profitable system. The interview offers personal experience rather than controlled evidence: strategy behavior can change, parameter adjustments can be mistimed, and investor withdrawals after drawdowns are a practical challenge. Its observations should not be read as proof that neural methods outperform alternatives.
Key ideas
- Topchylo treats neural networks as one demanding approach to system design, not a guaranteed improvement over alternatives.
- He favors deriving many parameters from recent price history, such as scaling stop distances with volatility.
- He found no meaningful performance gain from neural-network committees in his own system tests.
- Longer live competition can expose weaknesses that short tests may miss, though it does not guarantee future performance.
- Position sizing creates a tradeoff between pursuing contest returns and avoiding damage from excessive risk.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.