Strategy Development: Modelling

A place to discuss anything.
Post Reply
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

To simplify the above:

Image
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

i'll think about that later. Insanely busy here. I never realised how much MrsW does, dog needs walking, I don't have a clean mug, and I can't find a lighter and I'm hungry.... & i'm trying to test new excel stuff I was writing till 6am which is chaos when time moves so fast, plus people to email, check on mum etc etc the list is endless!

I'm a shell after just 24hrs :|
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

Just taken it a step further - separated the winners & losers from this:

Image

And just had a look at the movement prior to the entry signal

Winners (Backing (58%):
41% Steamed to entry
53% Drifted to entry :? :shock:
6% Flat on entry

Losers (Backing) (36%):
48%: Steamed to entry
49%: Drifted to entry
2%: Flat

Need to check those figures, but it would suggest the bulk of the winners are caused by a reversal. The drifters are failing (mean reversion).

& here's my problem, this could keep on fractaling as the refinement improves
Last edited by ruthlessimon on Thu Jun 21, 2018 5:23 pm, edited 1 time in total.
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

ShaunWhite wrote:
Thu Jun 21, 2018 4:53 pm
& i'm trying to test new excel stuff I was writing till 6am
I find that more & more these days, sleeping simply becomes a waste of time :lol:
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

ruthlessimon wrote:
Thu Jun 21, 2018 5:09 pm
ShaunWhite wrote:
Thu Jun 21, 2018 4:53 pm
& i'm trying to test new excel stuff I was writing till 6am
I find that more & more these days, sleeping simply becomes a waste of time :lol:
It's a big inconvenience, I've always resented it.
Same planned tonight. When SVs appeared I made a rough prototype of something to control a bot from excel (you could do it before actually but they made it more flexible & easier). I need to sort that out and plug it into my new excel stuff if it's fast enough so that BA can do more of the trading donkey work and position management, my current excel code for placing trades is a disaster waiting to happen.

Theoretical testing doesn't amount to a hill o' beans for some things, I need a way to get some real trades through rather than assuming I'm getting fills. Time to dip into the R&D budget and bear some testing losses I suspect :roll:

i LOVE the way BA is so open for input and outputs, a trading front end and a transaction processing engine! It's the dog's doodahs.

(BA merch to the usual address pls whoever is in marketing & PR)
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

ShaunWhite wrote:
Thu Jun 21, 2018 4:53 pm
Theoretical testing doesn't amount to a hill o' beans for some things, I need a way to get some real trades through rather than assuming I'm getting fills. Time to dip into the R&D budget and bear some testing losses I suspect
Why not just go MATCH("theoretical price","trading range",0) (with some extra checks in case of the price spiking, & which direction the trade takes place)

I prefer having the raw numbers as a sandbox - then you can literally make thousands of strategies using RANDBETWEEN for the strategy input variables. Ironically the problem I face is not enough computing power for the stuff I need to test. Most of my time is spent optimising formulas to avoid lag! :roll:

When are the quantum computers coming out? :D
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

The end result is then ending up with sheets that look like this - giving me the time of the theoretical touch - which can then be checked via a graph etc.

Image
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

ruthlessimon wrote:
Thu Jun 21, 2018 8:03 pm
Why not just go MATCH("theoretical price","trading range",0) (with some extra checks in case of the price spiking, & which direction the trade takes place)
Been there, it's theoretical matching. I need to quantify the 'f-it factor', where your money isn't fully matched even though it theoreticaly should be, and the frequency trades aren't fast enough to get what's there. Even with all my vol & time & price info, because PIQ being is meaninless, it's impossible to run a real simulation.

They're not the sort of thing that balances out. The 'wrong' side always gets filled better then the right one.

You see a lay entry at 4.1/4.2, you lay it at reverse, or even at best, what's your fill sr? One thing's certain, if the price goes the wrong way that sr is 100%, and the ones you want just sit there untouched. The problem is similar at the close. Paper edges are tough enough to find, now try getting that 4.3 tick avg upswing, and if you're lucky you can keep the downswing to 4 when it should be 3.

If you haven't had any execution psych issues yet it's because you haven't tried to turn those stats into clicks :)
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

ruthlessimon wrote:
Thu Jun 21, 2018 8:53 pm
The end result is then ending up with sheets that look like this
And as a 3d chart it shows where the peaks of the moutains break through the sea water :)
Layering those price histories like plywood and slicing through it at angles makes some pretty pictures.
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

ShaunWhite wrote:
Thu Jun 21, 2018 8:54 pm
You see a lay entry at 4.1/4.2, you lay it at reverse, or even at best, what's your fill sr? One thing's certain, if the price goes the wrong way that sr is 100%, and the ones you want just sit there untouched.
That is a danger.

The compensating factor I'd say is an edge that is up many hundreds of ticks should be at less of a risk to slight misses
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

Speaking of slicing plywood at angles, I made this a while ago for peanuts out of a block of scraps I glued up.

I did it after playing with excel, or was it the other way round :)

It's not on topic but if chewing gum on sticks is, then inspiration for slicing through layers of data is too!
You do not have the required permissions to view the files attached to this post.
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

ruthlessimon wrote:
Thu Jun 21, 2018 9:03 pm
The compensating factor I'd say is an edge that is up many hundreds of ticks should be at less of a risk to slight misses
Many hundereds of 0.94 of a tick actually :)
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

Actually, that'd be interesting to know.

Whether it's better to actually build something that can be traded (real stakes).
Or be built "theoretically tradable", to highlight general biases

My issue is you'd be limited by using real stakes - to what you "assume" will be profitable. Excel you can do literally millions of combinations, but the results will have a slight error component.
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

Imagine 2months in & having a new trade idea. It's gonna have to start from scratch if I'm trading it live, then a couple of months to tell if it even works. But then that data is "dirty" & cannot be used for any other experiments.
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

ruthlessimon wrote:
Thu Jun 21, 2018 9:22 pm
Whether it's better to actually build something that can be traded (real stakes).
Or be built "theoretically tradable", to highlight general biases.
Find the theoretical.
Implement practically.
Ditch or scale up and add to portfolio.
Repeat.

.... Meanwhile something that did work isn't anymore, is it real change, seasonal change or varience?

It's bot development 101 but you're the bot, which adds huge benefits as well as human weaknesses

Or light the inscense and wing it :D (by the way TITZ is less interesting to read than Nuts even though you'd image it would be the opposite)
Post Reply

Return to “General discussion”