Forums  > Trading  > portfolio optimization and alpha decay  
     
Page 1 of 1
Display using:  

ahgt_123


Total Posts: 25
Joined: May 2020
 
Posted: 2021-02-07 20:31
Is running portfolio optimizer at regular intervals a potential partial solution to alpha decay problem.
If yes , whar are the caveats of it that i am missing?

Also, is there any literature/rule of thumb for rebalance period selection (1 mo ,3 mo,6mo,....) say for a (daily,weekly,monthly) rebalance strategy.




ronin


Total Posts: 676
Joined: May 2006
 
Posted: 2021-02-09 14:48
> Is running portfolio optimizer at regular intervals a potential partial solution to alpha decay problem.

No.

Portfolio optimization is about getting the most out of the alpha you have. It can't help wou with the alpha you don't have.

"There is a SIX am?" -- Arthur

ahgt_123


Total Posts: 25
Joined: May 2020
 
Posted: 2021-02-11 08:33
there are multiple assumptions below but still theoretically

if an alpha is "decaying" the optimizer will assign lesser weight to it in the next iteration
and after a couple of iterations the alpha may have very less weight in the portfolio.

also,Are there any other quantitative solution to this problem?

ronin


Total Posts: 676
Joined: May 2006
 
Posted: 2021-02-11 09:39
Well, yes.

And all sources of alpha continue to decay. After enough time, you have a portfolio of strategies that only produce volatility, but no alpha.

You haven't 'solved' alpha decay. You have just been making the most of the alpha you had as it was decaying.

"There is a SIX am?" -- Arthur

nikol


Total Posts: 1351
Joined: Jun 2005
 
Posted: 2021-02-11 09:41
@ahgt_123

To explain the problem outlined by ronin, assume extreme:
- All your sub-strategies deliver zero alpha.
- You have "best in the world" portfolio optimizer.

Q: What is the alpha of optimized portfolio?

... What is a man
If his chief good and market of his time
Be but to sleep and feed? (c)

ahgt_123


Total Posts: 25
Joined: May 2020
 
Posted: 2021-02-11 10:11
@ronin I agree with you, but lets say there's a portfolio where you constantly add few new alphas, isn't this approach better than say removing the alphas completely which are underperforming.

ronin


Total Posts: 676
Joined: May 2006
 
Posted: 2021-02-11 10:42
Sure. Low Sharpe or even negative Sharpe strategies can still be useful in a portfolio, if they provide enough diversification.

But it's a fairly limited set of scenarios in which that can happen. The rest of the portfolio needs to have a strong common factor, so a little bit of diversification of that common factor has high impact.

"There is a SIX am?" -- Arthur

doomanx


Total Posts: 113
Joined: Jul 2018
 
Posted: 2021-02-11 10:48
@aght_123 you can improve the geometric growth rate by diversifying more, even if some of the bets have poor odds (look at the investment wheel in Luenberger's Investment Science). But if all the strategies are independent it's unlikely to do a lot - if they have a lot of joint exposure and the weak alpha is uncorrelated then it may be more useful.

edit: ronin is too fast ;)

did you use VWAP or triple-reinforced GAN execution?

nikol


Total Posts: 1351
Joined: Jun 2005
 
Posted: 2021-02-11 11:00
@ronin
> even negative Sharpe strategies can still be useful in a portfolio

How can it be?

... What is a man
If his chief good and market of his time
Be but to sleep and feed? (c)

ronin


Total Posts: 676
Joined: May 2006
 
Posted: 2021-02-11 11:22
Say you have strategy A that does something really smart with spx. It is correlated with spx, but it has higher Sharpe than spx.

Then you add strategy B, which is just plain short spx. Spx has positive Sharpe, so strategy B is negative Sharpe. But it annihilates your spx factor and leaves you with the pure alpha of the first strategy.

A+B has higher Sharpe than just A.

"There is a SIX am?" -- Arthur

doomanx


Total Posts: 113
Joined: Jul 2018
 
Posted: 2021-02-11 11:23
@nikol when you're maximising growth rate, the killer is sample paths that take the wealth to zero (as you cannot recover). Diversification helps kill off these sample paths. If the strategy has negative sharpe but is independent, then it can improve the growth rate with the right allocation.

did you use VWAP or triple-reinforced GAN execution?

nikol


Total Posts: 1351
Joined: Jun 2005
 
Posted: 2021-02-11 11:48
I agree, it make sense.

However (sorry for being tedious) shorting helps only up to some point as it increases leverage and hence ruin rate, isn't it? Perhaps, fraction of short position must be optimized too.

... What is a man
If his chief good and market of his time
Be but to sleep and feed? (c)

ronin


Total Posts: 676
Joined: May 2006
 
Posted: 2021-02-11 12:15
> However (sorry for being tedious) shorting helps only up to some point as it increases leverage


It doesn't have to be a short, it can be just a diversifier.

Say strategy A has vol 10%, return 10%, so Sharpe 1. Strategy B has vol 10%, return 5%, so Sharpe 0.5. Correlation is 0.

A + B has vol 10%*sqrt(2) ~ 14%, return 10%+5%=15%, so Sharpe ~ 1.1.

And then there are other things it can do for you, like bring down the overall funding cost, reduce the beta etc.

"There is a SIX am?" -- Arthur

ahgt_123


Total Posts: 25
Joined: May 2020
 
Posted: 2021-02-11 22:04
Can you please suggest some papers and "modern" methods on portfolio optimization.
Something that may properly beat
"a third in land, a third in merchandise, and a third ready to hand" .

ronin


Total Posts: 676
Joined: May 2006
 
Posted: 2021-02-12 08:57
Oh man - that's a tall order. Portfolio optimization can't beat poetry.

In all seriousness, optimizing a portfolio of Gaussian random walks with known drifts and correlations doesn't require some great skill. It's taught to undergraduates.

The problem is that things aren't really Gaussian, the word "drift" is used as a fancy way to say "hope", and correlations are, for lack of a better word, wrong.

Here is one paper from Avellaneda from 2019 that tries to find ways around correlations:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3467712


"There is a SIX am?" -- Arthur

doomanx


Total Posts: 113
Joined: Jul 2018
 
Posted: 2021-02-12 10:37
Less directly applicable, but you may also be interested in https://projecteuclid.org/euclid.aos/1342625460

did you use VWAP or triple-reinforced GAN execution?

bandi_np


Total Posts: 11
Joined: Nov 2020
 
Posted: 2021-02-12 12:52
Here is a another paper that uses hierarchy idea to build a portfolio https://arxiv.org/abs/2005.08703

doomanx


Total Posts: 113
Joined: Jul 2018
 
Posted: 2021-02-12 13:10
Just remembered the authors of the paper I linked also used the method to construct portfolios in https://www.tandfonline.com/doi/full/10.1080/07350015.2017.1345683. The idea is they use their covariance estimation method to handle the cross-sectional noise and use some kind of econometrics filter I don't know much about to handle the dynamic correlations (although it appears to be based on GARCH, which has it's own stack of issues). Broadly has the right idea dealing with both types of estimation error though.

did you use VWAP or triple-reinforced GAN execution?

doomanx


Total Posts: 113
Joined: Jul 2018
 
Posted: 2021-02-12 13:40
' Can you please suggest some papers and "modern" methods on portfolio optimization.'

The first thing you should read (and this isn't modern but needs to be read by every single student immediately after they learn about mean-variance allocation) is https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2387669. This should give you some food for thought and put the other papers referenced here into context.

did you use VWAP or triple-reinforced GAN execution?

rickyvic


Total Posts: 244
Joined: Jul 2013
 
Posted: 2021-02-16 09:36
I have been messing with portfolio optimisation quite a bit over time. Pardon me if I repeat concepts already explained.

1. risk parity and vol/risk targeting is hard to beat, and so meanvariance (which if you think about and similar without weird constraints).
2. you can run portfolio optimisation on a daily basis, below that probably not worth it and also tedious to code
3. covariance estimation is b... . It makes sense for highly correlated assets, say 2 calendar spreads on same instrument, other than that highly unpredictable
4. there are some interesting constraints to add that do make sense. Say you want to have a maximum risk per single instrument or bet, or add transaction costs to the utility function, etc...
All very useful tools but obviously things get complicated. I think it is worth it in the long run especially from a risk management perspective (probably not if you just want to maximise your sharpe)
5. max sharpe, sortino, etc is not really what you want to do especially if you look at recent history, unless you have to select models with some strong logic. Meaning you are putting intelligence in what you do.
6. dont invest too much time in portfolio optimisation compared to alpha generation and add risk management to each model/strategy 's logic rather than in a portfolio model. It is easier and cheaper.

"amicus Plato sed magis amica Veritas"
Previous Thread :: Next Thread 
Page 1 of 1