Forums  > Trading  > Advice for enhancing quant strats  
     
Page 1 of 1
Display using:  

MacroPM


Total Posts: 3
Joined: Apr 2021
 
Posted: 2021-04-15 13:29
I used to be a frequent contributor to this site a long time ago but I lost my login and had to create a new one. Here is the situation - I'm currently a macro portfolio manager on a pretty conservative but relatively large shop. Prior to that, I spent many years on the sell-side on various roles - quant, trader, structurer, etc. All in fixed-income. A few years back I moved to the buy side initially managing credit/rates, but now broader macro where I run a book with rates, credit and equity risk. I'm now entertaining a few quant strategies to enhance my book and also to be more involved in this space. I already use some light quant stuff but mostly for asset allocation (ie to distribute risk amongst the varios risk factor that I run), but I wanted to move more into systematic strategies.

Due to the nature of the shop I work for, I have a number of constraints:

1) This will be a one-man show - me basically coding and running the strats. I do have though an execution desk
2) Data will be limited to what I can source from bloomberg and other common data sources, but nothing quant-dedicated as most quant shops have
3) I don't need a stellar Sharpe here, something like 1-1.2 after costs will do the trick provided it converges fast! (senior management not particularly patient)
4) Don't need a ton of capacity either - think US$100-200M

I have a somewhat decent background - PhD in EE, signal processing, stats etc and I still remember most of the stuff (or at least I like to think that I do!)

On the side, I have developed two low/med freq models that deliver good SR live (0.7-1) but I need to move to the next level in order to increase the risk on these strats. One model is simple TF with kelly betsizing, the other model is a factor based stat arb on a dozen broad index ETFs.

So my questions:

1) How do I improve sharpe given the constraints outlined above - I am not asking for a strategy obviously but perhaps pointers to things I could do to improve the SR if it is feasible
2) I currently use a mix of Matlab and R - can I get away with this or I have to learn python?
3) Are there other shops with quant managers with similar setup?

Any feedback greatly appreciated.
















rickyvic


Total Posts: 245
Joined: Jul 2013
 
Posted: 2021-04-15 13:49
PM if you need help

"amicus Plato sed magis amica Veritas"

doomanx


Total Posts: 113
Joined: Jul 2018
 
Posted: 2021-04-15 13:52
How many assets are you running the strats you developed on? Both are fairly general ideas so you should be able to use them on more products to get some easy sharpe.

For factor based stat arb you've got the estimation of the factors and fitting the regressions of the factor returns. On the regression side you could do all the usual model checking stuff to see if you can squeeze some more performance out there. On the factor side, are these statistical factors, hand crafted or some combination of both? There is a lot literature on improving the estimation of statistical factors (again depending on what approach you're using there, FA/PCA/etc, are the factors high dimensional) which may or may not help depending on your use case. Quite hard to say without more detail. If you're using any approach based a large covariance matrix for factor construction, you might like https://arxiv.org/abs/2012.05757.

No, you definitely don't need to use python.

did you use VWAP or triple-reinforced GAN execution?

rickyvic


Total Posts: 245
Joined: Jul 2013
 
Posted: 2021-04-15 13:58
I agree with doomanx about the language.
In broad terms matlab is connected to anything you might need, for mid frequency strategies is good enough and it is quick to deploy.

I have used it extensively and built a significant infrastructure on it.

"amicus Plato sed magis amica Veritas"

Jurassic


Total Posts: 415
Joined: Mar 2018
 
Posted: 2021-04-15 14:06
@doomanx always like your papers

ronin


Total Posts: 679
Joined: May 2006
 
Posted: 2021-04-15 14:30
Sharpe 0.7-1 for a low frequency strategy sounds reasonably stable. As a pair, they could be SR > 1 depending on the correlation.

Obvious things to look at would be subtracting all betas, and sizing.

I haven't looked at @doomanx' paper yet, but the fact that people are still writing papers about covariances in 2021 (2020?) says enough. I'd kill Kelly and go for simple fixed size.

"There is a SIX am?" -- Arthur

MacroPM


Total Posts: 3
Joined: Apr 2021
 
Posted: 2021-04-15 14:42
@doormax Thanks for the feedback.
@rickyvic - Thanks, will PM you in a bit.

The startarb I'm using statistical factors with FA or PCA, results are not materially different with either one. The trouble with increasing the number of assets is (1) I need to balance that with the turnover which is doable - I can for example do weekly rebalance without losing much SR, and (2) I need to have some degree of control on the execution side otherwise I can end up building a lot of slippage (execution team here you need to keep them honest all the time). That becomes harder the more assets you have to monitor. So currently I have 10 tickers but I think it should be doable to get to something like 30. What would you consider ideal?

Thanks for the paper, I will look it up for sure.




MacroPM


Total Posts: 3
Joined: Apr 2021
 
Posted: 2021-04-15 15:16
Thanks. Could you elaborate why would you not use Kelly?

On the TF strategy I run is a big part of the strat. Everything else I tried was not as good.

ronin


Total Posts: 679
Joined: May 2006
 
Posted: 2021-04-15 20:42
Are you absolutely sure this is a viable strategy?


"There is a SIX am?" -- Arthur

EuroYenDolla


Total Posts: 8
Joined: May 2021
 
Posted: 2021-05-12 03:01
Add more data, alternative data, economic releases and how to weigh their importance is a big project alone. Matlab is more than enough but python is much better for data cleaning so if you need to clean data u better start learning python. Look at your performance depending on the time of day you are executing (not sure about the time your trading so can't be sure) model how your orders effect the order book, don't leave limit orders out there just waiting. Again I can't be sure b/c i don't know what time scale you are working on but all of this stuff is a good place to do some basic optimization, after that I would say just find better strategies !

You going to carry that wieght?
Previous Thread :: Next Thread 
Page 1 of 1