Seeking feedback on my rules-based system

[morganpbrown]

So I implemented an exit strategy whereby I sell the security if the signal that caused me to buy the security is no longer active. It seems to work well. For most of the securities I'm trading, it adds a couple percentage points to annualized ROR.

For securities like TVIX, it works like gangbusters! Not a lot of people are buying and holding VIX for 30 days, so the new exit strategy is more attuned to the characteristic price action of TVIX.

New profit curve for XOP:

Old profit curve for XOP:

Quite similar, but the new strategy has lower volatility and a higher ROR.

 

[nooby_mcnoob]

stdin

as plot title. Sounds about right for the F90 programmer

 

[morganpbrown]

stdin

as plot title. Sounds about right for the F90 programmer

Ha, yeah, this is not an interface for the masses. At least I have a color screen. And it's not a CRT!!!

 

[digitalnomad]

I'm using a past "states" to predict the future behavior of the target security. The past states could be 30-day past S&P return, VIX over or under 200 d MA, etc. For each predictor, say 30-day past S&P return, I loop over all possible states (e.g., 0-3%, 3-6%, etc.) and backtest.

Nice little toolbox you’ve built. I love crude shit like this. Looks like you’ve caught on to some form of multi-variable analysis. IMO, this could yield some pay dirt if you can filter and optimize the states properly. Good luck young fellow

 

[dreamer2019]

The strategy seems making sense. It uses MA, past return to discover macro relationships between securities.
How did you use 30-day past return ?

 

[morganpbrown]

How did you use 30-day past return ?

The 30-day past return is one of many indicators that I use. I have 14 indicator securities. For each indicator security, I compute 30/10/5-day past return, 200/50/10-day MA, and 200/50/10-day MA crossover. That's a total of 126 indicators. I backtest each indicator to find the "sweet spots", for instance 30-day past return of -5% to -2%. I trade the sweet spots.

 

[nooby_mcnoob]

The 30-day past return is one of many indicators that I use. I have 14 indicator securities. For each indicator security, I compute 30/10/5-day past return, 200/50/10-day MA, and 200/50/10-day MA crossover. That's a total of 126 indicators. I backtest each indicator to find the "sweet spots", for instance 30-day past return of -5% to -2%. I trade the sweet spots.

The problem is these kinds of approaches have the "sweet" spots change over time. Find out how/why/when they change (this is my biggest blocker).

 

[morganpbrown]

The problem is these kinds of approaches have the "sweet" spots change over time. Find out how/why/when they change (this is my biggest blocker).

That's a good point. I guess you could run the backtesting and remove a year (or more, or less) of historical quotes, and see how "stable" the sweet spots are.

One problem I have is that the profit curves for many of the sweet spots are very similar, implying that the indicators are obviously highly correlated! I need to figure out a way to cull redundant strategies and only trade the uncorrelated strategies.

 

[nooby_mcnoob]

This is kind of what I'm leaning towards:

1. Identify intellectually where a sweet spot could exist
2. Research sweet spot (oh look it exists)
3. ??? figure out how to predict the parameters for the sweet spot
4. Profit

Stuck on step 3. Last N days is a good approximation but it's weak.

 

[nooby_mcnoob]

That being said, I think certain sweet spots can persist. You probably found one.