The speaker discusses the use of M1 data with a 20-year range for backtesting, stating that it is not overkill for precision and can cover various market phases, such as gold price spikes and consolidation periods. However, the practical side of using M1 data over 20 years can be challenging, and the main concern is whether the strategy becomes overfitted. The speaker suggests splitting the data into 15 years for training and the last five years for validation, allowing for a more accurate understanding of market conditions. They also mention the use of genetic evolution on M1 data but suggest keeping it focused on key areas like exits and position sizing. The speaker emphasizes the importance of testing the strategy live before fully committing, as it should be robust and adaptable to today’s conditions.
Key Takeaways:
- M1 Data and 20 Years: Great for precision and long-term viability.
- Segment Testing: Split into in-sample (15 years) and out-of-sample (5 years).
- Walk-Forward Optimization: Ensure adaptability across evolving market conditions.
- Focused Genetic Evolution: Optimize key areas without overwhelming the algorithm.
- Live Testing: Validate the strategy in real conditions to account for slippage and latency.
