Re: [amibroker] Re: testing multiple systems simultaneously, AmiBroker Email List Archive

Hi guys,

The optimal portfolio is a fascinating topic.

I gave been using some of the Vince ideas for my futures trading
and one of the biggest advantages it gives me is a way of keeping
myself sane - knowing how to size a position when you are getting
crushed or elated is important. It's great having a number to tell
you how greedy or scared to be!

To give you an idea of where I find optimal f useful:
    - I stay well short of the optimal amount.
    - Where it does come in useful is at the portfolio level.
    - I trade a number of systems in futures:
         - intraday Index and Bonds - Asian session
         - intraday Bonds - European session bonds
        - EOD Curve ( 3Yr Bond / 10 year Bond )
        - EOD Bill Spread ( 90 day treasuries ) (This has been only
            recent so no stats yet - trades go for weeks )

Not sure if you would consider these as uncorrelated, but there
seems to be a measurable advantage so far in optimising at the
portfolio level. Not sure if this is an artifact or something that I
can use. I've been doing it in spreadsheets and it is a lot of work
to get it right. I also suspect that the number of intraday trades which
are needed to be significant is way more than other time frames.

The other thing I've noticed about optimal f ( at the itnraday level at
least ) is that it is moves around more than I expected it to. Once
again this could be due to what i think is greater noise at the intraday level.
To give you an example, after 50 backtested trades I get on optimal
f of 0.55. Vince suggests that 30 trades should be sufficient - but he
does acknowledge that opt f does move around. Anyhow, now I'm
trading, and I get to
- trade 80    ( 50 backtest, 30 actual ) opt f is 0.60
- trade 109 ( 50 bt, 59 actual ) opt f is 0.36
- trade 191 ( 50 bt, 141 actual ) f is 0.45 and finally seems to be
getting a bit more stable.

What is really dangerous ( or seems to be in my limited experience ) is
that the largest losing trade can take you out when trading at optimal f.
If you are trading a non-discretionary trading system at full f, then
you MUST pull the plug if you see the biggest losing trade coming your
way. I can't remember if Vince mentions it, but the backtesting/calcuation
of optimal f assumes that the biggest loss is in the past. Experience seems
to indicate other wise!

I've put none of this into AB yet. I've still got the trial version! As soon
as I can tear my self away from exploring it I will buy it!

Z

From: brian_z111 <brian_z111@xxxxxxxxx>
To: amibroker@xxxxxxxxxxxxxxx
Sent: Tuesday, 5 May, 2009 12:08:28 PM
Subject: [amibroker] Re: testing multiple systems simultaneously

Have you guys read Vince .... the first book, chapter 6, "The Total Portfolio Approach"?

He gives a very good account of how to construct an optimal portfolio using system diversification ... it would be quite easy to implement in AB..... his mathematical modelling is quite straightforward. .... basically he nets the (special case) period returns (as growth factor) of each system .... net (portfolio) geometric mean == reward and variance of the net (portfolio) geometric mean == risk.

He also discusses Markowitz and the Capital Asset Pricing model and briefly shows why the 'geometric mean portfolio strategy' is a valid extension of those models.

I'm not a Vince, or a portfolio, guru and I don't have the time for an in depth project at this stage ..... just some quick comments ... scanning your posts it seems that an understanding of the theory, or some portfolio theory, needs to precede the implementation ... IMO:

- each system should be optimized independently first
- systems should be standardised to a common time frame e.g. daily returns
- correlation of the period returns of the already optimised systems should be measured (how are you all doing that in AB?)
- the only optimisation that is done at the portfolio level is allocation of capital (this is done on an iterative basis in Vinces model).
- I find that Money Management combined with equity curve analysis (of any kind) is logically flawed because Money Management varies the outcomes e.g.

one trade series .... +10%,-10%;

start $100 -> 110,99;// a 1% loss of capital
start $100 -> 110, add 890 capital -> 900;//approx 10% loss of capital
- an account can never be fully invested unless we find the HolyGrail of systems?

Optimal f was developed specifically to minimise the drag on equity recovery caused by asymmetrical leverage i.e. it calculates the optimum staking to return equity to a postive state after the max loss has been encountered.

- if you want to reopt then you will need to recalc 'all of the above again'.

Out of curiosity ... are you finding it easy to come up with noncorrelated systems?

--- In amibroker@xxxxxxxxx ps.com, "bh.hicks" <bh.hicks@xx .> wrote:
>
> I am basically looking for a way to have AmiBroker run multiple systems concurrently in order to examine how trading multiple non-correlated strategies affect drawdowns. I think if there was a way to "name" an entry condition so that stops and position sizing rules could be applied to a particular entry criteria, it would be possible to do without too many changes to AB architecture.
>
> I can already do this in excel using exported equity curves but it would be nice to be able to do this internally so that the optimizer engine could be exploited.
>
> Is anyone aware of a technique to do this and if not, is this something others would find useful if integrated into a future version?
>
> As always - thank you.
>