Forums  > Basics  > Sources of Return  
     
Page 1 of 1
Display using:  

riskPremium


Total Posts: 5
Joined: Nov 2018
 
Posted: 2018-11-22 17:55
I am newly registered but I have been surfing this phorum for quite some time and have benefited greatly from it.

I am a independant trader and only started trading in Jan 2018 (Sharpe 1.8, return ~ 15% YTD). The way I did it is by trading well known risk premium (momentum, carry, etc) with some model tweak. Result is ok but somehow I feel these strategies will decay quickly as everyone knows it. I wonder what other systematic traders do. Are they trading more short-term alphas with some sort of "ensemble" models. Is that the next step for a systematic trader like me?

Thanks!

ronin


Total Posts: 385
Joined: May 2006
 
Posted: 2018-11-27 12:43
All strategies decay. Make the most of it while it lasts.

Systematic traders try to keep innovating. But most happen to find one, two, maybe three good strategies in their careers. When those are spent, the career is over.

Short term alphas with smart allocations is what dying hedge funds do when all their strategies have died, but they still have some investors left to milk. It will help you flog a dead horse for a bit longer. Nothing more.

"There is a SIX am?" -- Arthur

riskPremium


Total Posts: 5
Joined: Nov 2018
 
Posted: 2018-11-28 12:37
Hi Ronin,

Thank u for your reply. Not sure what you mean by this one:

"Short term alphas with smart allocations is what dying hedge funds do when all their strategies have died, but they still have some investors left to milk. It will help you flog a dead horse for a bit longer. Nothing more."

If you are saying garbage in garbage out then thats for sure. But I was under the impression that people are trading more and more "weak" signals and only ensemble them together so they are tradable? Is that the case? Forgive me for asking this because i am newly graduated and have never worked in a hedge fund.

doomanx


Total Posts: 17
Joined: Jul 2018
 
Posted: 2018-11-28 13:52
He is not talking about ensembling weak signals, he is talking about allocating different amounts of capital to different alphas in an optimal way based on some objective (you may have heard of mean-variance or other such criteria). So you might have two strategies, one is following trend and one is selling puts. You have some notion of the expected return and risk of these strategies based on current market conditions, and you want to decide how much of your cash should be allocated to each strategy. Ronin is saying that if your alpha is garbage, no amount of smart allocation will prop up a set of strategies with no edge remaining for an extended period of time.

In terms of ensembling weaker signals, the answer is this can be done but again suffers from all the usual financial estimation problems. You are dealing with a process that is mostly random and so all methods you pick should be very robust to noise in the input variable. There is a very large literature on these sort of methods. I'm assuming you are talking about ensembling in a machine learning context as it is mentioned a lot in that space - most machine learning literature is targeting very high signal to noise ratio problems (see image recognition) and so porting these ideas over generally doesn't work out of the box.

ronin


Total Posts: 385
Joined: May 2006
 
Posted: 2018-11-28 16:39
what @doomanx said.

Historically, any attempt at smart allocation has turned out to be a lot dumber than the dumbest dumb allocation. Especially for weak signals.

"There is a SIX am?" -- Arthur

riskPremium


Total Posts: 5
Joined: Nov 2018
 
Posted: 2018-12-05 12:58
Doomanx,

Gotcha. As for ensemble learning methods, I am having some progress in intraday but find it quite difficult on lower frequency strategies...


riskPremium


Total Posts: 5
Joined: Nov 2018
 
Posted: 2018-12-05 13:00
Ronin,

I agree. Dumb allocation can be quite robust.

Strange


Total Posts: 1475
Joined: Jun 2004
 
Posted: 2018-12-05 17:49
> But I was under the impression that people are trading more and more "weak" signals and only ensemble them together so they are tradable?

There are garbage alphas (i.e. stuff that stopped working) and there are "thin" alphas (i.e. stuff that has strong statistical significance but is hard to monetize). Putting a bunch of thin signals together allows you to trade them, but it does not really change the nature of these alphas.

I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'

NeroTulip


Total Posts: 1016
Joined: May 2004
 
Posted: 2018-12-06 01:57
To build on what Strange just said:

Say you have a bunch of thin alphas e.g. 2 SR before transaction costs, 0 SR after. It would not make sense to trade any of them individually. However, if they often trade against each other, you can trade the net signal and save transaction costs, making the net signal worth trading. Note that you do not need to use very sophisticated ensemble methods for this to work though.

"Earth: some bacteria and basic life forms, no sign of intelligent life" (Message from a type III civilization probe sent to the solar system circa 2016)
Previous Thread :: Next Thread 
Page 1 of 1