TradeWrecker's LRA

The Long Road Ahead...

Not going to bore anyone with my experience or my goals. I've been around long enough to be suspicious of just about every claim, have an opinion on just about every statement, to be more than a little jaded, and genuinely disappointed with the status quo of almost anything associated with “retail”.

With that in mind I'm here mostly to hear/watch myself think, and document what I'm working on and the road ahead. A lot of my creativity with my own work has been tied into writing about what I'm doing. This is a good way for me to keep the thought process flowing... Maybe others will take something useful from it.

If I was a betting man, I'd say I'm three to five years away from where I once was. It's a sobering thought, made more so by the much needed, but harder to implement and more costly compliance. I'll need to be thinking a lot more about operational risk and impacts – but I don't have to focus on that today.

Now I'm catching up on some old research, planning the initial testing and optimization and deciding how to structure the data, what time series to use etc. All that knowing there is very little interest in the quant models as of late and that probably won't change in the future... So one might ask “what's the point”. And I'd answer... I don't know. I believe you do what you know, work to find a niche and let the cards fall where they may... What else can you do?

TradeWrecker

 

I think you should do so with your own thread in the "Journal" sub-forum and not here in "Career Trader". Or perhaps ask the moderators to move this for you.

 

I'm a professional trader and licensed CTA... so this seemed appropriate. That said, I could care where the thread is...

 

I've always tried to be thorough in my testing. But like anything there is always room for improvement. The trick is to find the balance between spending a lifetime researching/testing and rushing a model to market.

I've been reviewing some of the initial research I based the first trade model on (Using genetic algorithms to find technical trading rules – Allen/Karjalainen, 1999) and was reminded of some ideas I had, that were never fully vetted.

Once I get through the doc again, I'm going to run some regressions with some of these other ideas in place and see how they compare to KA's work and original approach.

If you haven't read their piece yet it's worth spending some time with. Even with my own initial model I had to expand on the concepts and while it did okay, there was still a lot of volatility in the returns. I think this has become the major issue with quant style models, like a lot of things they've not really lived up to their promise – but the market is an efficient thing and quick to adjust when too many people are on the same train of thought...

 

Whenever I'm conducting research I know I need to keep the belief bias in check and this requires looking for good research that also contradicts my work. It's been a while since I originally downloaded the paper mentioned above (Using Genetic Algorithms... KA 1999) so I spent some time using the title as a key word search this morning.

I found an interesting paper written by Christoper J. Neely; “Using Genetic Algorithms to Find Technical Trading Rules: A Comment on Risk Adjustment” 1999. Neely findings weren't as positive as the original paper and notes the importance of adjusting the returns for risk vs the buy and hold method.

Some of this has I have already addressed in my early work, but not to the extent presented in this paper. I'll be adding some of the other risk measurements in this round and will see how they compare. Because I'm working in alternative investments, buy and hold equity benchmarks may not be the best measurement. On the other hand, using a cta index where there is a high likelihood that the method I'm using is well represented within the index is a mixed bag of tricks.

The approach I took in the past was to determine first, what I wanted the model to be, and then worked to reduce the instances where the rule sets were highly correlated to a specific type of approach. If I wanted a model with a low correlation to trend models than I eliminated the rule sets that mirrored those technical approaches. Then when I had what I felt were good rule sets, I could compare them to a Trend Index of sorts. This is different than the KA approach as they often ended up with rule sets that were technically close to buy and hold. I'm not looking for more complexity with high correlation to simplicity...

Neely's article is worth the time.