How to create a model??

[napshnap]

…

I will try and simplify what I am asking so that it is easier to quantify and come back...

Just to clarify and see if I have understood your example correctly

If we keep things simple you could look at how much a price should move from its starting price as a function of time remaining (or metres if you have TPD) By modelling this you would then be able to put a number on the min/max price a runner should be trading at a specific distance / time etc.

.
Is this you "quantifing something normal under specific conditions", as a base to work from there?

TIA

I am glad it’s got you thinking, tbh sometimes that is all you need.

The answer is yes to all your questions, things repeat themselves in every market every day, the reality is it’s the same manual/drones/courtsiders/automated betters playing the game not many of us care about horse/jockey.

A main stream example of a model is the use of XG in football, although relatively simple it gives a much better idea of how a team is performing in a match compared to just the current score. This has the benefit of being useful inplay and after to give a more accurate price and thus find EV. Syndicates have hundreds of people to analyse each game, frame by frame, the use of live tracking is also changing things, this is all to improve their models and find value. The use of ML and AI have moved this to being more automated, with pysport / opencv you can do some cool stuff.

https://pysport.org/page/5/

But the game... it actually changes, politics changes, liquidity changes, mrkt players leave, evolve, even some fundamental stuff changes (~5-10% of uk gh races with less than 6 runners? ohhh thats cute BUT HOW ABOUT ~40% this summer!?). How do you stay afloat?

[Bobajob]

Hmm... "mmmodel"...

Collect shtload... like... centuries!! of "relevant" ta fa data, define dependent variable (the one that your "magic formula" will calc) and check multiple independent variables against it by using multiple linear (or nonlinear) regression or by using machine learning (which is embellished nonlinear regression, actually) that will give you the "magic formula/equation"/model that you then check on shtload of test data sets to "prove" its magicalness and then, with fanfare, you launch this ship (with "reasonable" stakesize, ofc) only to find out that your beautiful Taj Mahal was built on a quicksand, that the universe zoomed out revaling that it was just a microscopic part of some gods randomwalk...

I take it that you are still working on the Taj Mahal then

[napshnap]

Hmm... "mmmodel"...

Collect shtload... like... centuries!! of "relevant" ta fa data, define dependent variable (the one that your "magic formula" will calc) and check multiple independent variables against it by using multiple linear (or nonlinear) regression or by using machine learning (which is embellished nonlinear regression, actually) that will give you the "magic formula/equation"/model that you then check on shtload of test data sets to "prove" its magicalness and then, with fanfare, you launch this ship (with "reasonable" stakesize, ofc) only to find out that your beautiful Taj Mahal was built on a quicksand, that the universe zoomed out revaling that it was just a microscopic part of some gods randomwalk...

I take it that you are still working on your model then

You can call me Phil I guess.

[Kai]

you launch this ship (with "reasonable" stakesize, ofc) only to find out that your beautiful Taj Mahal was built on a quicksand, that the universe zoomed out revealing that it was just a microscopic part of some laughing gods randomwalk...

You basically said it there yourself. Lost edge at micro level? Zoom out to macro level. Works the other way around too.

A reasonable grasp at both levels should keep you going regardless of market changes.

In any case shouldn't underestimate quant finance etc, or sports scientists/data analysts... they make some crap football managers look very good.

[Euler]

I did my first model back in the 1980's to predict football matches. Every market I trade or bet on I model in some way.

But....

What you often find is the model just matches the market. Ultimatley there are loads of people doing this and lots cleverer than I.

But...

That's modelling is helpful as it allows you to look ahead and establish the risk you are taking. Create a model yourself allows you to put in parameters that others may not use.

So many models are public now that you can see where the gaps in understanding are.

But...

The big breakthrough for me came from understanding what people do and how.

I've often used a pseudonym to participate in modelling discussions and, especially for the very academic types, there is often a huge gap in understanding how to profit from a model.

When I'm stuck on a particular problem, I'll often wheel in a PhD to solve a specific problem without telling them specifically what it relates too.

My big lesson was that you need two levels of intelligence to participate profitably in the market. Intellectual and emotional.

If you get the mix right it opens the door to a lot of things. Lean too much on either and you end up achieving little.

That's my experience.

[Kai]

My big lesson was that you need two levels of intelligence to participate profitably in the market. Intellectual and emotional.

Absolutely. You do sort of need to be a bit clever about things, but not necessarily in a traditional sense.

Admittedly, I underestimated the importance of EQ at first, but now generally value it higher than IQ, at least for a primarily manual skill set.

Finding a healthy balance is the trading journey itself I feel, and both sides are important, the IQ side for the problem solving and the EQ side for managing emotions which directly relates to managing positions.

And I suppose If day trading was a regular day job then social intelligence (SQ) would be a bigger factor towards finding success.

[Brovashift]

Cheers guys, plenty to get stuck into here...

Ye this Captin Jack was were I started at previously, possibly from that very thread on here. But dabbled in a few of his videos and followed along with this Monte Carlo Model, decided it was a bit too above my pay grade at the time and so left it there. I will definitely revisit this as things always tend to make more sense second time round. Thanks (actually that link to his video has opened up exactly where I left it, at 10:08 / 16:58 )

I could keep things simple like you say Peter... or, I will rephrase that, I 'will' keep things simple, initially to create a simple model. But, and correct me if I'm going off track here, I was reading about Regression Analysis which sounded like it would be ideal to factor in all those variables in my OP.

Let me try this, relating to dobbing (or any price shortening);
E.g.
For each horse:
1) I think my dependent variable would have to be 'Price'? as thats the thing Im trying to predict/forcast, well 'price movement' anyway.
2) Keeping it simple, my independent variable(s) could be 'recent form'? But could I repeat this by replacing the independent variable with each of my variables from the original post, and any/ all things likely to factor whether a horse is likely to shorten in price, or not?

So I end up with several scatter charts, one for each independent analysis. Then could I use these results to create one Master model??

So I would be learning to create a simple model that has legs. What do you guys think, am I on track?

TIA

[LinusP]

That doesn’t sound simple, it sounds like you are trying to do everything.

[Brovashift]

That doesn’t sound simple, it sounds like you are trying to do everything.

Not all at once. I was thinking if I start with just the two metrics "Price and recent form" as suggested, familiarise myself with the process of analysing and using the data to create models, then in the future, e.g. months from now I could expand on the project to analyse how other variables also effect the price. Or do you still think no?

In terms of picking a project, I kinda like the area of 'shortening prices', or value, if we can call it that? I had a project a few years ago where I tried to double some crypto I had and doing the magic penny thing, doubling 30 times. It didn't work of course, or took forever to reach 100%... So I have some additional interest in dobbing, if it can be done with higher probability.

Is (Multiple Linear) Regression Analysis too hard to start with, am I jumping in the deep end? Looking at types of models the RA sounded ideal by examining the relationship between 2 or more variables and how they influence eachother.

[Brovashift]

I've been thinking about these models and regression models, and in the words of 'Captain Jack Andrews', there are many way to profit from sports trading and modeling isn't for everyone...

Having said that, I think the regression analysis could be quite useful for an application I've been working on to analyse daily racecards, as I was having difficulty in creating a scoring system to properly indicate the strong and weak runners in a race based on race conditions and recent performance. I tried using +/- for each positive/negative racecard variable, but this did not give a true indication of a selections chances because nothing was weighted by importance/relevance.
Im thinking I might be able to implement some type of regression analysis in a separate 'class' thats performed on each selection, by using the analysed results of a seleceted racecard to output individual scores, on the fly... Im not entirely sure if that would work but I feel like all the parts are there, not necessarily in the right order lol.

I should have almost finished it by now but had a family bereavement in September and lost all momentum, with everything. Hence the last 9th Aug date... Thats why I was looking at getting a real job . Slowly getting my mojo back now.