Portfolio optimization of a basket of high frequency and low frequency strategies

mizhael · Jun 28, 2011

Hi all,

Lets say you have a basket of strategies: A, B, C, D

A is really high frequency, based on real-time tick data;
B is mid frequency, based on 30min bars;
C is low frequency, and daily, every day, at End of Day, you might or might not have some trades;
D is ultra low frequency, somewhat weekly, and indeed it's event-driven. So if you look at daily pnls, this strategy has a lot of zeros as daily pnls.

How do you calculate the correlations among these strategies?

Furthermore, how do you do portfolio optimization?

I am trying to renovate the traditional "mean-variance" portfolio optimization ... but didn't get any meaningful results at all.

Your thoughts are highly appreciated!

Thanks!

intradaybill · Jun 29, 2011

You have it all wrong but I am not going this time around to try to explain anything to you because you haven't shown enough appreciation in the past. I think you are asking questions at a level where you should be hiring a consultant. It is clear you are involved with some hedge fund because normal traders do not care about this stuff. Tell your boss to hire a consultant. I knew it was you who posted this right from the title.

bone · Jun 30, 2011

I knew who started this post before I opened the topic tab for the first time. Amazing powers of deduction.

bone · Jun 30, 2011

Quote from intradaybill:

Tell your boss to hire a consultant.

Maybe boss should have paid up for a Quant with some chops for the position. This is where saving $50K is really the height of stupidity if indeed intradaybill has sniffed this one out and flushed it into the open for a clean shot.

Do you at least have a Bloomberg terminal and MatLab with the Bloomberg interface module ?

Mike805 · Jun 30, 2011

Quote from bone:

Maybe boss should have paid up for a Quant with some chops for the position. This is where saving $50K is really the height of stupidity if indeed intradaybill has sniffed this one out and flushed it into the open for a clean shot.

Hehe... thing is, the questions the OP is asking aren't really that hard. I've always assumed this issue was a fairly standard concept found in traditional portfolio theory, something anyone who has taken a basic graduate level finance class should be aware of....

So each strategy has a return profile (read: pnl/exposure curve) over time, right? Just like any equity, future, option, bond, swap etc etc does, right?

So if one constructs a portfolio of products (or *strategies*) then one can study correlation scenarios in the portfolio based on intra and inter-strategy allocations as a percent of the portfolio, right?

As always, the devil is in the details...

mizhael · Jun 30, 2011

Quote from intradaybill:

You have it all wrong but I am not going this time around to try to explain anything to you because you haven't shown enough appreciation in the past. I think you are asking questions at a level where you should be hiring a consultant. It is clear you are involved with some hedge fund because normal traders do not care about this stuff. Tell your boss to hire a consultant. I knew it was you who posted this right from the title.

My sincere apologies IntradayBill. Please let me know where I did wrong things to you in the past... It's definitely not my intent.

mizhael · Jun 30, 2011

Quote from Mike805:

Hehe... thing is, the questions the OP is asking aren't really that hard. I've always assumed this issue was a fairly standard concept found in traditional portfolio theory, something anyone who has taken a basic graduate level finance class should be aware of....

So each strategy has a return profile (read: pnl/exposure curve) over time, right? Just like any equity, future, option, bond, swap etc etc does, right?

So if one constructs a portfolio of products (or *strategies*) then one can study correlation scenarios in the portfolio based on intra and inter-strategy allocations as a percent of the portfolio, right?

As always, the devil is in the details...

What assumptions do we make when deriving the traditional portfolio theory?

Mike805 · Jun 30, 2011

Quote from mizhael:

What assumptions do we make when deriving the traditional portfolio theory?

It's a starting point... and, the derivation is useless in a practical sense. The main assumption is that the assets are not correlated, and by combining assets and optimizing single asset allocation, one can increase risk adjusted returns.

That's a loaded assumption when it comes to combining strategies because when shit hits the fan; *everything* moves in one direction.

So here's the million dollar question: how correlated are your strategies?

Do you know how to apply some fundamental reasoning to you portfolio of strats?

I'll give you a hint: the most important factor is not that you've found an optimal per-strategy allocation; its that (if) you're able to identify your portfolio's worse case scenario. Even then, designing a method to deal with that worst case scenario is a greater challenge on its own right.

savagemp5 · Jul 2, 2011

Quote from mizhael:

Hi all,

Lets say you have a basket of strategies: A, B, C, D

A is really high frequency, based on real-time tick data;
B is mid frequency, based on 30min bars;
C is low frequency, and daily, every day, at End of Day, you might or might not have some trades;
D is ultra low frequency, somewhat weekly, and indeed it's event-driven. So if you look at daily pnls, this strategy has a lot of zeros as daily pnls.

How do you calculate the correlations among these strategies?

Furthermore, how do you do portfolio optimization?

I am trying to renovate the traditional "mean-variance" portfolio optimization ... but didn't get any meaningful results at all.

Your thoughts are highly appreciated!

Thanks!

Hi mizhael, after hearing others commenting that you work for a prop firm or some small cap fund, asking these qns, i wondered why u are hired ?

I run a portfolio of strategies, but from the way you described it, it seems to be a portfolio of time-framed strategies, rather than strategies of different natural (eg. managed futures, derivatives arbing, long/short etc...)

How is it possible to find... correlation between a trade that lasts, as mentioned 1 seconds VS an event driven news eg. Greece default. VS a synthesis split strike on derivatives options.

I mean seriously, you ask stupid questions, you get stupid answers. No wonder no one could answer that.

savagemp5 · Jul 2, 2011

Quote from Mike805:

It's a starting point... and, the derivation is useless in a practical sense. The main assumption is that the assets are not correlated, and by combining assets and optimizing single asset allocation, one can increase risk adjusted returns.

That's a loaded assumption when it comes to combining strategies because when shit hits the fan; *everything* moves in one direction.

So here's the million dollar question: how correlated are your strategies?

Do you know how to apply some fundamental reasoning to you portfolio of strats?

I'll give you a hint: the most important factor is not that you've found an optimal per-strategy allocation; its that (if) you're able to identify your portfolio's worse case scenario. Even then, designing a method to deal with that worst case scenario is a greater challenge on its own right.

And mizhael, this is pretty obvious from Mike805. Correlation of strategies and time frame are 2 different tracks. You can have all similar time frames strategies but have different correlations. But if all strats are short frames, the shorter they are, the lesser correlated they are, irregardless if they are all on the same asset classes, agree ?

That's why people do index arb for a reason.