Chris and Tyler talk about curve fitting and over-optimization of trading strategies.
This episode is a lot shorter than normal, we touch on a single subject vs. covering a wide variety of subjects. If you like these short and pointed episodes and want more, let us know! If you prefer long form drop us a line. Tell us what you want!
In this episode, Chris starts off talking about what he looks for in a trading strategy:
- Predictability - the ability to predict a trade is forthcoming, not predicting the results.
- Repeatable - does this setup repeat over time?
- Definable - can you clearly define the parameters?
Tyler and Chris then go on to discuss the following:
What does over-fitting look like in trading systems?
Why seeking high win rate systems leads us down the path of over-fitting.
"Finding" setups in random walk price data. How our brains are easily fooled by randomness.
Testing logic on in-sample and out-of-sample data (training and test datasets). A model that works well on both sets is best.
Why resilience is our true goal for a trading system. If it isn't fragile, it has more potential to last over time.
How to deal with black swan risk. (Risk that doesn't show up in any past datasets.)
Why robust systems are uncomfortable to sit through.
Why building a system that fits your temperament is the key to having success.
Finally, we hit on why we all need to practice the basics of a system over and over. Get those reps in!
Editor's Note: The summary bullets for this article were chosen by Seeking Alpha editors.