Deadline June

This is a question to MR and also to you Adam, how many degrees of freedom are in your systems that you run at the moment? Just interested.... ;-)

I run two systems, a short term using 4hr bars and a medium term using daily bars.

Degrees of freedom are

1. stake size
2. distance to initial stop in ATR
3. channel days for entry
4. channel days for exit
5. short MA
6. long MA

that's it I think.

then of course there are the markets I trade, which represent another degree of freedom altogether.

Now my problem with defining degrees of freedom is that you have so many degrees of freedom to do so - boom boom! No but really, how can "channel days for entry" be one degree of freedom? Surely a channel has a definition requiring at least one more degree of freedom than the number of days?

Hmmm, possibly. It's highest high or lowest low for x number of bars.

So x is changeable, as is the length of a bar. So maybe it counts as two, not one.

OK here goes with degrees of freedom in the FrankenDucks.

I use fixed fraction money management so that's not a variable in the system.

(1) MA length for 3 Ducks.
(2) Time frame of 3 Ducks - I'm adding this in here because I've just started playing around with 10mins, 60mins, 240mins instead of 5,60,240.
(3) Trading long or trading short - I'm not sure if this counts but I optimise seperately
(4) Initial target in number of ATR (3rd Duck's ATR)
(5) Initial stop in number of ATR (ditto)
(6) Second exit mechanism, e.g. long trades exits when closes below 2nd Duck

OK admittedly my (1) could be 3 in itself but I don't optimise them seperately. That would overfit everything straight away.

I disagree, regarding (1). This is something that I don't think would cause over-fitting, it would just change the types of moves you capture (which can be a good thing). I think (3) is sliding more into over-fitting territory, unless you think markets move down drastically differently than they move up?

Sorry had to send that message without finishing it.

I have another degree of freedom - a support and resistance indicator which has 2 inputs.

So that is actually (7) and (8).

I think the markets can move differently up compared to down. The prime example is stock indices, and I guess the forex markets are where the difference is the weakest. I guess worst case scenario - I am twice as likely to overfit if optimising once for long and once for short trading.

Glad you disagree about (1). Interesting that you don't consider it overfitting but it would actually add another 2 degrees of freedom to my list. Making 10.

All this kack I've been doing to keep my trading system going this last month and checking out alternative brokers and providers and platforms has got me out of practice on the system building front. Never could keep that kind of stuff in my head. I'm going to work my way back into it this weekend and perhaps note down the lists of Do's and Don'ts that I keep forgetting, and the thresholds and limits to stop polishing turds and start somewhere else and all that knowledge.

Rule No (1): don't curve fit
Rule No (2):

I am not sure if you are using standard optimisation tools but the ones In Tradestation work like this:

1 - You tell them which parameters can be optimized.
2 - You tell them the boundaries for each parameter being optimized.
3 - You choose either 'brute force' or 'genetic' optimization.
4 - You give a date range to optimize.

Brute force will try every combination and genetic will attempt to be more savvy in terms of testing things that look promising.

So - this is all well and good. You could use these techniques for years and never question them. The problem as I see it is that this is a very one dimensional view of how optimization should work.

At the start of each day, I like to look at the prior days high, low, volume and range:



I just get a picture over time of how the market has been behaving. Over time, you can see the market get thinner/thicker and ranges expand/contract.

If we can believe that the markets might hold onto a set of behaviours for a few days/weeks, then any optimisation run over a 10 year period would not be optimal for the next few days/weeks.

Maybe the best optimisation is one that is run every day for the past 5 days. In fact, optimising over years as Tradestation allows makes a lot less sense that an optimisation algorithm that you tell to run for 10 years but which changes the parameters every week based on the parameters that would have worked best in the prior week. So - you could run for 10 years but use 2 week optimisation, 1 month optimisation etc so you try to be a bit more in tune to the current market conditions.

For some reason the tools don't have this option. For some reason no-one questions that and so people use the tools available without considering that the tools limit you to only one kind of optimisation which gives you the most generic settings.