Fully automated futures trading
- Thread starter globalarbtrader
- Start date
I've heard that some systems delay prices by a day, which is what made me float the idea. Can you elaborate on what you mean by "this"?
This is a classic path to overfitting... which I call "fighting the last battle". And don't worry, it's endemic to both retail and institutional investors, as the latter are always under pressure from investors to "improve" their system and make it more "resiliant" to whatever went wrong last week..
What is the systematic way to approach this? You'd need to fit some kind of non linear response to volatility changes. You'd need to fit this out of sample, and on a market specific (or perhaps across asset classes) basis, since you think this effect only occurs on the volatility indices. That means a lot of parameters. It's highly unlikely this will be a successful enterprise. At best you might get a small, non statistically significant, improvement in performance - in exchange for a huge of complexity.
But if you want to pursue this, be my guest. But I strongly advocate that you don't just introduce a discretionary fiddle into your system in the way you suggest.
GAT
Good points, GAT. However, I believe that your system does not allow a positive forecast on the volatility indices. Is this correct? If so, could one argue that that represents overfitting based on past data, since you allow positive forecasts on all other instruments?
No I think you've misunderstood something (your signs are also the wrong way round, but that isn't important). I have a 'no-rule' rule forecast just for volatility, which has a fixed value of negative 10.The effect of this is to make positive forecasts less likely, but not impossible.
However the forecast weight on this rule is fitted on a rolling out of sample basis. Where I could be accused of overfitting is that this rule only applies in volatility indices. Where I to include it as a possible rule for say stocks and bonds I'd see an allocation to it come through the optimisation process, since these assets have had a secular up trend throughout the back test. But my larger long only portfolio, which the optimisation process doesn't know about, already includes plenty of stocks and bonds, so I don't allow the use of the long only rule here.
The rule is specifically a short volatility rule, since this is a well known risk /return factor which I don't get exposure to otherwise.
GAT
I see, so you combine this fixed forecast of -10, with the variable forecasts you get from EWMAC, carry, and any other rules you have set up, to result in a short bias to volatility indices?
Imho you're being too gentle there GAT. N@D has taken a serious risk with his/her trading account and managed to get away with a minor injury by pure good fortune. The initial suggested remedy (a kind of over-fitting to recent benign markets) was deeply inappropriate and suggestive of almost no understanding of what he/she is doing as well. N@D needs to take a good hard look at their approach and risk-management before it's too late.
I find that on the internet, and on ET especially, the harsher you are the more defensive people are - and less likely to listen.
GAT
