I’d like to answer some questions.

canoe said:
can you elaborate on the whole 'ES is not a solo thing, it's a team effort'?

are you trying to say that the ES market has become too efficient and alpha has eroded? b/c that has absolutely not been the case in the past 3 years of wild fluctuations we saw with the ES market.

perhaps you're referring more to the 2010-2017 period?

the ES and esp the NQ during the past 3 years were a gift from heaven in terms of price inefficiencies. i'd almost agree with you if were talking strictly daytrading the ES but there was plenty of alpha to be generated with swinging the ES from 2018-2021.
More...


If you walk into any trading firm I know there’s no ‘ES guy’ like there might be for ags or metals. ES and interest rates are usually large teams. For different reasons.
I’ve seen better luck ‘swing trading’ literally anything other than ES but that’s good for you if you’ve developed a feel for it. I never have.
 
How would you go about estimating the parameters for a Kalman filter if you have used them before?
Have you used HMM before and what do you think of them? Would you estimate the parameters or would you choose to define them using intuition?
Lastly, how do we find out more about colocation fees and setup for different asset classes?
 
garachen said:
the feel of a mean version vs momentum strategy is very different. It’s harder to feel the edge in a reversion trade because it’s harder to feel blowouts that go against you.
More...

Another question if you don't mind... what do you use to determine regime change, i.e. from mean reversion to momentum?

Commonly cited in research papers are always the Markov, Garch or Hurst models but as always, the PhD folk get lost in their own navel gazing without articulating anything tangible.
 
traider said:
How would you go about estimating the parameters for a Kalman filter if you have used them before?
Have you used HMM before and what do you think of them? Would you estimate the parameters or would you choose to define them using intuition?
Lastly, how do we find out more about colocation fees and setup for different asset classes?
More...

Didn't see this before posting as I quoted from a few pages back. Might be able to merge a reply..
 
traider said:
How would you go about estimating the parameters for a Kalman filter if you have used them before?
Have you used HMM before and what do you think of them? Would you estimate the parameters or would you choose to define them using intuition?
Lastly, how do we find out more about colocation fees and setup for different asset classes?
More...

the Kalman filter question I can’t answer succinctly without a lot of assumed knowledge. So.. assuming the knowledge.. kalman filters tend to work best on trades that are otherwise not great - where you are creating correlated baskets. In algorithmic trading there’s usually a ‘fitting’ process and an optimization process. If the KF is to be combined with other alphas I’d run a pre-optimization set of sims to set them and then add the resultant model into the fitting. Otherwise, you’d solve for them during the optimization process in conjunction with the other strategy parameters. The optimization algorithm itself is quite important. Most people have moved to the same one but it’s not quite industry standard yet so I’d rather not say.

HMM. Other than academically no. Every once in a while someone out of school wants to try. And it might come up during an interview but I’ve never seen one functional.

colocation. I’ve argued about this definition on this forum before. Inside the industry colocation means one specific thing. But retail vendors seem to have corrupted this meaning with misleading marketing. Direct from the exchange you are usually buying a full or half rack and then paying additional to manage each server. I’ve always gone through a 3rd party. Probably pay too much but there’s a lot of things they solve. Colocation is to solve a very specific problem- latency. But it introduces a host of new problems that all very hard and expensive to solve. I won’t say who we use now but searching for exchange colocation some vendors should appear. You’d call them and ask for a quote. Anything that consolidates the feed for you or is listed as ‘proximity hosting’ is not true colocation. Maybe you’d use them for reliability but by the time you get your data, everyone doing true colocation has already acted and taken a good long nap.
 
InAllHonesty said:
Another question if you don't mind... what do you use to determine regime change, i.e. from mean reversion to momentum?

Commonly cited in research papers are always the Markov, Garch or Hurst models but as always, the PhD folk get lost in their own navel gazing without articulating anything tangible.
More...


I’ve most commonly seen a NN deployed here.
 
garachen said:
the Kalman filter question I can’t answer succinctly without a lot of assumed knowledge. So.. assuming the knowledge.. kalman filters tend to work best on trades that are otherwise not great - where you are creating correlated baskets. In algorithmic trading there’s usually a ‘fitting’ process and an optimization process. If the KF is to be combined with other alphas I’d run a pre-optimization set of sims to set them and then add the resultant model into the fitting. Otherwise, you’d solve for them during the optimization process in conjunction with the other strategy parameters. The optimization algorithm itself is quite important. Most people have moved to the same one but it’s not quite industry standard yet so I’d rather not say.

HMM. Other than academically no. Every once in a while someone out of school wants to try. And it might come up during an interview but I’ve never seen one functional.

colocation. I’ve argued about this definition on this forum before. Inside the industry colocation means one specific thing. But retail vendors seem to have corrupted this meaning with misleading marketing. Direct from the exchange you are usually buying a full or half rack and then paying additional to manage each server. I’ve always gone through a 3rd party. Probably pay too much but there’s a lot of things they solve. Colocation is to solve a very specific problem- latency. But it introduces a host of new problems that all very hard and expensive to solve. I won’t say who we use now but searching for exchange colocation some vendors should appear. You’d call them and ask for a quote. Anything that consolidates the feed for you or is listed as ‘proximity hosting’ is not true colocation. Maybe you’d use them for reliability but by the time you get your data, everyone doing true colocation has already acted and taken a good long nap.
More...

What is the difference between 'fitting' and 'optimizing' ?
 
You must log in or register to reply here.
Back
Top