Australian (ASX) Stock Market Forum

Trend trading as anti-pair trading

Jump to new
P

PrudentInvestor

Joined
24 February 2010
Posts
23
Reactions
0
Having played around with mean-reverting portfolios (not just pairs) for ages, I recently came across the idea of studying trending portfolios as anti-mean-reverting ones. The process of estimating the best trending portfolio out of a given universe of assets is a mirror image of the process of constructing the best mean-reverting portfolio. Has anyone tested this kind of a strategy?

If you do not know what I'm talking about, go and get historical security data from Yahoo! Finance from 2003 onwards for the following portfolio:

LONG 330 CME (CME Group on Nasdaq)
LONG 440 M (Macy's on NYSE)
LONG 840 WHR (Whirlpool on NYSE)

This portfolio trends immensely well - almost any simple trend strategy would have worked.
 
Pairs Trader said:
Hindsight is a wonderful thing.
Click to expand...

What is this supposed to mean? You don't think it is possible estimate the portfolio weights so that their time-stability is also taken into account?

I merely provided an example to illustrate that it is very much possible to find strongly trending combinations of assets even though the individual assets would not show significant trending.

Now, do you also have an opinion on the strategy itself?
 
If you do not know what I'm talking about, go and get historical security data from Yahoo! Finance from 2003 onwards
Click to expand...

Ahh what wonderful history....

'Pairs Trader' got it in one. If you don't know what he/she is talking about, you need to do the research

brty
 
brty said:
Ahh what wonderful history....

'Pairs Trader' got it in one. If you don't know what he/she is talking about, you need to do the research

brty
Click to expand...

Come on. He/she thinks I'm one of those curve fit fanatics anxious to trade everything that has been optimized. In my answer to him/her I indicated that there are ways to estimate the portfolio weights so that the results are (relatively) stable across the potential trade duration. Needless to say, this does not apply to all porfolios, and I have not made such claims regarding the sample portfolio at all. To reiterate: The sample portfolio of my first post aims to show that it is possible to construct portfolios that trend much better than the individual portfolio members. To ensure that the portfolio is tradable, one has to ensure statistically that the portfolio weights remain constant, or, more generally, that they are predictable to a sufficient degree.
 
brty said:
What is your initial/underlying premise??

brty
Click to expand...

That even the simplest trend trading strategies do work once the traded instrument (in this case the portfolio) is chosen properly.

I must admit that I've been focusing on the opposite side of the coin - i.e. mean-reverting strategies. But since almost the same techniques that can be used to find mean-reverting portfolios also apply here, I see no reason why this premise would not be feasible.
 
Pairs Trader said:
Prudent, what have your back testing results using your criteria on over 1,000 historical trades shown?
Click to expand...

Nothing, I have not backtested this idea yet. I guess I'm just (being lazy and) asking if someone has already done it and found a dead end. As with relative value strategies - including pair trading - the catch is to also back test the portfolio construction algorithm dynamically. This requires some programming effort.

By the way, it might actually be possible to perform an online estimation of the weights of the trending portfolio using a Kalman approach. This should be an interesting excercise for the mathematically oriented. Any volunteers? ;)

I guess there are no excuses - I have to do it myself anyway.
 

Similar threads

Dividend
Ultimate Acronym List (UAL)
Replies
0
Views
5K
Dividend
Dividend
Top