I recently came across a strategy described by Ernie Chan in his book, Quantitative Trading: How to Build Your Own Algorithmic Trading Business. The section of interest is on p. 122 (p. 144 of the pdf).
Quantitative Trading: How to Build Your Own Algorithmic Trading Business
If I'm reading it correctly, the simple strategy he describes revolves around identifying large 1-day moves on GS stock and then optimizing the amount of the position held in subsequent days. So unlike more typical optimizations, such as identifying indicator parameters, he's fixing the entry criteria and optimizing an element of the execution of the strategy.
I've read about using genetic programming to determine exact strategy parameters. I use Python for creating algorithms, currently only stocks and crypto, but I'd love to be able to add FX (I know that the FX community is better about getting into the details of trading strategies so I thought this would be a good place to ask). If I designed my own optimization framework, there wouldn't be any limitations to what could be optimized. This would represent one end of the spectrum, a completely brute force approach to creating a strategy. I think it might also be possible to optimize any parameter in MetaTrader (as opposed to just indicator parameters), but that would also be a brute force approach. What I'm looking for are more narrow strategy aspects that could be optimized on a rolling basis.
I've spent countless hours reading about machine learning for trading, but I can't seem to find much about rolling optimization other than simple guidance that frequently updating strategy parameters is a best practice. It seems that there has to be more.
Some optimization ideas:
- Holding period/percentage
- Stop-loss/take-profit
- Market regime/indicator reliability (e.g., which indicators of a candidate set are working at a given time and may be more likely to work in the near future)
- Statistical model parameters such as changepoint detection algorithms or Kalman Filters (this isn't much different than optimizing indicators, but these models have greater adaptability than most traditional indicators)
I'm really just curious if anyone can suggest anything or any reading on the topic.
Quantitative Trading: How to Build Your Own Algorithmic Trading Business
If I'm reading it correctly, the simple strategy he describes revolves around identifying large 1-day moves on GS stock and then optimizing the amount of the position held in subsequent days. So unlike more typical optimizations, such as identifying indicator parameters, he's fixing the entry criteria and optimizing an element of the execution of the strategy.
I've read about using genetic programming to determine exact strategy parameters. I use Python for creating algorithms, currently only stocks and crypto, but I'd love to be able to add FX (I know that the FX community is better about getting into the details of trading strategies so I thought this would be a good place to ask). If I designed my own optimization framework, there wouldn't be any limitations to what could be optimized. This would represent one end of the spectrum, a completely brute force approach to creating a strategy. I think it might also be possible to optimize any parameter in MetaTrader (as opposed to just indicator parameters), but that would also be a brute force approach. What I'm looking for are more narrow strategy aspects that could be optimized on a rolling basis.
I've spent countless hours reading about machine learning for trading, but I can't seem to find much about rolling optimization other than simple guidance that frequently updating strategy parameters is a best practice. It seems that there has to be more.
Some optimization ideas:
- Holding period/percentage
- Stop-loss/take-profit
- Market regime/indicator reliability (e.g., which indicators of a candidate set are working at a given time and may be more likely to work in the near future)
- Statistical model parameters such as changepoint detection algorithms or Kalman Filters (this isn't much different than optimizing indicators, but these models have greater adaptability than most traditional indicators)
I'm really just curious if anyone can suggest anything or any reading on the topic.