Too many trades? Mean variance portfolio optimization and rebalancing?

M

mizhael

Hi all,

How does one handle the too-many-trades problem in mean variance portfolio analysis?

Let's say I have 10 trading strategies, each one is on a basket of instruments(say 100, such as those big ones in SP500).

For each trading strategy, I run a mean-variance weighted portfolio on the 100 instruments. The 10 trading strategies each have a different optimal "rebalance" period.

For the 10 trading strategies, again I run a mean-variance weighted portfolio on these 10 trading strategies, again with some optimal "rebalance" period.

Now there are 1 big periodic rebalance on top of 10 individual periodic rebalances.

All together, it terms out that I am trading almost every day.

On some days, it's strategy1 that's rebalancing(within strategy1's basket); on some other days, it's strategy 2 that's rebalancing; ...

On some big days, it's the whole 10 strategies/baskets got completely reshuffled.

Overall, it terms out that I am trading almost every day.

Is this a big problem? Any pointers? Please shed some lights on me.

Thanks a lot
 
Quote from mizhael:
....mean variance portfolio analysis?....10 trading strategies....**** SP500****.....mean-variance weighted portfolio.....optimal "rebalance" period.....mean-variance weighted portfolio.....optimal "rebalance" period.....1 big periodic rebalance......10 individual periodic rebalances.....I am trading almost every day......strategy1.....rebalancing(within strategy1's basket).....strategy 2......rebalancing.......10 strategies/baskets.......reshuffled......I am trading almost every day......Please shed some lights on me.
More...
1) Balance, balance, balance, re-balance!
2) Your broker(s) must really love you.
2) Put all of your money into an S&P-500 index fund and leave it there.
4) You must have mistook this place for wilmott. :D
 
A more intelligent version of Mr Nazzdack's post would be: How does your model balance transactions costs against the alpha-gain from rebalancing?
 
Read Rish Narang's new book - "inside the black box"

http://www.amazon.com/Inside-Black-Box-Quantitative-Trading/dp/0470432063/ref=ntt_at_ep_dpi_1

Specifically - the transaction cost model is discussed.

Probably at any given time, for the most important trades, very few stocks in the basket underperform or outperform the basket itself.

I'd run all the strategies but filter out the majority of trades where the expected return is the least.

Also, strategies can turn into filters by giving the strategies a value, which is summed +- with the other concurrent strategies. This way you only get decent signals, and avoid overtrading. It doesn't have to be 0 or 1, it could be a cumulative indicator. .1+.2-.3+.5 signal threshold is 1 for long -1 for short...

If you're looking for a short-term anomaly, transaction costs get very important.
 
Quote from mizhael:

For the 10 trading strategies, again I run a mean-variance weighted portfolio on these 10 trading strategies, again with some optimal "rebalance" period.
More...

Based on what data? Hypothetical from backtesting? Your results will be as hypothetical as the future performance of your strategies.

It is different to use allocation for major classes of assets, like beteen stocks, bonds, futures, forex and cash than to use it within same class and further use hypothetical performance to get results.

It is my opinion that you are misusing the concept and of course it has been misused by others too, even more educated than you are, I pressume. Key point: correlations between assets in same classes are not stationary and thus the method is rendered either useless or too expensive or both. Get it?
 
Quote from intradaybill:

It is different to use allocation for major classes of assets, like beteen stocks, bonds, futures, forex and cash than to use it within same class and further use hypothetical performance to get results.

It is my opinion that you are misusing the concept and of course it has been misused by others too, even more educated than you are, I pressume. Key point: correlations between assets in same classes are not stationary and thus the method is rendered either useless or too expensive or both. Get it?
More...

Correlations <i>between</i> asset <i>classes</i> aren't stationary, either. But in any case, what would you propose to use instead of "useless" mean/variance?
 
Quote from Rodney King:

Correlations <i>between</i> asset <i>classes</i> aren't stationary, either. But in any case, what would you propose to use instead of "useless" mean/variance?
More...

Good point. They are not stationary either but to a lesser degree than those within an asset class.

I can propose something if the problem is made more specific. I made reference to the fundamental problems of the method. Now, specific proposals should depend on trading objectives. What is the trader trying to accomplish, how much capital must be invested, what is the trading method used?
 
Quote from intradaybill:

Good point. They are not stationary either but to a lesser degree than those within an asset class.
More...

Not sure I would agree with this. The corr(stocks,bonds) or corr(stocks,$) will swing from roughly +50 to -50 over time, whereas (for instance) largecap stocks' corrs stay consistently large and positive.
 
Quote from Rodney King:

Not sure I would agree with this. The corr(stocks,bonds) or corr(stocks,$) will swing from roughly +50 to -50 over time, whereas (for instance) largecap stocks' corrs stay consistently large and positive.
More...

The problem with stocks is those negative correlations that turn positive not with those that remain large and positive.

The correlation of major asset classes may not be stationary but there are other benefits from diversifying there, including minimizing unsystematic risk, credit default risk, foreigh exchange risk, etc. You can still use (Stocks, cash) or (bonds, cash) or (futures, cash) to be safe. Many funds do that, either stocks and cash, or bonds and cash, just to be on the safe side of correlations.

This is why I always say, not too many understand math, very few understand how to apply them and even fewer understand when and where to apply them. Math, and especially complex optimization, is useless and even a dangerous tool in the hands of those who do not have a deeper understanding of finance theory.
 
You must log in or register to reply here.
Back
Top