What is modelling?

A place to discuss anything.
Post Reply
sa7med
Posts: 800
Joined: Thu May 18, 2017 8:01 am

I've been backtesting now for over a year. After testing thousands of different ideas, I have only had limited success but im seeking to take things to another level now. I've seen Peter mention modelling before as the way to go but I confess I only have a basic understanding of what this means. Would love to see some light shed on this area.

I guess what would help most is an example of what a question to model would be. Are we modelling say the true odds of a tennis player winning vs another tennis player? Or are we modelling the likelihood of a price falling further after already falling x %? or both? What re the statistical techniques associated? I need to learn what to learn. I remember Peter shared a book on statistical techniques a while back but can't find it. Sorry if this seems a bit misguided but I am! Any resources are welcome.
foxwood
Posts: 390
Joined: Mon Jul 23, 2012 2:54 pm

My thoughts on modelling v backtesting and a relevant thread viewtopic.php?f=54&t=17789&p=173827&hilit=model#p173827
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

I'm a bit confused myself, but isn't it just a synonym of strategy development?

peter's latest vid was quality; struck me as very very genuine: https://www.youtube.com/watch?v=lNjydODC0f4

& I thought this was an interesting article: https://medium.com/causal-data-science/ ... 4f20d26438

love this bit:

"Consider your commute to work. We have an intuitive understanding that traffic will cause you to be late for work. We also know that if your alarm doesn’t go off, it will cause you to be late to work. This picture is a great start, but these are really just two of the most common causes of being late for work. Others could include your car not starting, forgetting to make the kids lunch, getting distracted by the news, etc. etc. How can our picture incorporate all of these little things that we don’t include? Can we ever hope to get a reasonable picture of the world when we can’t possibly measure all of the causes and effects?

The answer turns out to be relatively simple. Our model just needs to include the most common, large effects on our trip to work. If we omit a large number of small, independent effects, we can just treat them as “noise”. We stop talking about things as being completely determined by the causes we take into account. Instead, we talk about a cause as increasing the chances of its effect. You go from intuitions like “my alarm not going off causes me to be late” to intuitions like “my alarm not going off causes me to be much more likely to be late

We can build a much more comprehensive picture by chaining together causes and effects. What are the causes of traffic? What are the causes of the alarm not going off?"

Blimey.. that'll keep people up late! :)

For me, simply coming up with "variables" (i.e. alarm clock) is difficult, especially with pre-race. But I am a fan of things like "how's it moved between two points" (but simply defining "moved" & "two points" is a nightmare!
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

foxwood wrote:
Mon Mar 25, 2019 6:10 pm
My thoughts on modelling v backtesting and a relevant thread viewtopic.php?f=54&t=17789&p=173827&hilit=model#p173827
Because yeah, I see this:

"To me, I would model it by thinking what would happen to the market if the certain winner suddenly looks like losing."

..as an enhancement of price - but that doesn't violate the usefulness of backtesting

<1.05 (1 variable) = lay

vs.

<1.05 & the runner looks like losing (2 variables) = lay

We've sacrificed sample size, for the hope of an edge/stronger edge. Of which the "hope" can be turned into "expected" by backtesting the idea.
User avatar
Derek27
Posts: 23469
Joined: Wed Aug 30, 2017 11:44 am
Location: UK

I did a quick search of modeling hoping to post a link to some attractive girls, but I found this instead.

What is modelling
User avatar
Euler
Posts: 24700
Joined: Wed Nov 10, 2010 1:39 pm
Location: Bet Angel HQ

Backtesting is looking at Stonehenge and realising that it predicts where the sun is going to rise and fall at critical times of the year.

Modelling is predicting the next solar eclipse on Stonehenge but also building a probe that explores the sun and planets of the solar system because you know how it works and can re-run your models endless to make sure.
User avatar
ruthlessimon
Posts: 2094
Joined: Wed Mar 23, 2016 3:54 pm

Euler wrote:
Mon Mar 25, 2019 7:39 pm
Modelling is predicting the next solar eclipse on Stonehenge but also building a probe that explores the sun and planets of the solar system because you know how it works and can re-run your models endless to make sure.
So assuming we had a model to work out a solar eclipse on Earth. A "good model", means that logic will still work on Mars - but the "inputs" will have to be adjusted slightly?
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

It's no coincidence that backtesting is also called hindcasting.

True predictive modelling (aka the useful one out of the many types of modelling) is beyond the intellectual capabilities of everyone I've seen on this forum, except one person who only appeared here very fleetingly. I don't think anyone here even has a decent maths degree, unless they'd like to admit to it now...and be swamped with requests for help.

But on a more helpful note, if you wanted to look at predictive modelling seriously, I suggest going down to your local University and putting a card on the notice board asking for (paid) help. There's bound to be someone who can give you a steer, but I'd brush up on your basic calculus first, otherwise you might not get much from it.

Round here you'll only get riddles and if you're very lucky, some dubious o-level maths, what you won't get are postings that either withstand mathematical rigour or even have enough detail to begin that process.
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

Incidentally I'm not deriding the talk about analysis on here, I take part in that myself. I'm just managing your expectations and saying are these things are often made to sound far easier than they actually are. I think even PW has said he's hired statisticians previously and that's an excellent idea.
sa7med
Posts: 800
Joined: Thu May 18, 2017 8:01 am

ruthlessimon wrote:
Mon Mar 25, 2019 6:21 pm


For me, simply coming up with "variables" (i.e. alarm clock) is difficult, especially with pre-race. But I am a fan of things like "how's it moved between two points" (but simply defining "moved" & "two points" is a nightmare!
This is what ive been doing for a year to no avail...well, some avail. Perhaps we should join forces :P
Last edited by sa7med on Tue Mar 26, 2019 8:31 am, edited 1 time in total.
sa7med
Posts: 800
Joined: Thu May 18, 2017 8:01 am

ShaunWhite wrote:
Mon Mar 25, 2019 9:05 pm
It's no coincidence that backtesting is also called hindcasting.

True predictive modelling (aka the useful one out of the many types of modelling) is beyond the intellectual capabilities of everyone I've seen on this forum, except one person who only appeared here very fleetingly. I don't think anyone here even has a decent maths degree, unless they'd like to admit to it now...and be swamped with requests for help.

But on a more helpful note, if you wanted to look at predictive modelling seriously, I suggest going down to your local University and putting a card on the notice board asking for (paid) help. There's bound to be someone who can give you a steer, but I'd brush up on your basic calculus first, otherwise you might not get much from it.

Round here you'll only get riddles and if you're very lucky, some dubious o-level maths, what you won't get are postings that either withstand mathematical rigour or even have enough detail to begin that process.
Thanks for your reply Shaun. Yes indeed, Im prepared to do some real learning to master this, but like I said I need to learn what to learn! I was a math minor in college but virtually no statistics in there. A lot of good that's done me! It seems a lot easier to do now a days with all these python libraries. Still, you have to understand what the hell you're doing.
sa7med
Posts: 800
Joined: Thu May 18, 2017 8:01 am

Euler wrote:
Mon Mar 25, 2019 7:39 pm
Backtesting is looking at Stonehenge and realising that it predicts where the sun is going to rise and fall at critical times of the year.

Modelling is predicting the next solar eclipse on Stonehenge but also building a probe that explores the sun and planets of the solar system because you know how it works and can re-run your models endless to make sure.
Peter do you remember the textbook you recommended a while back? It was called something like: Statistics for everyday applications or something like that
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

sa7med wrote:
Tue Mar 26, 2019 8:26 am
but like I said I need to learn what to learn!
That is THE problem with this stuff, I totally sympathise and have the same problems.
On the one hand it's vague and fluffy concepts, and then when you try to do some study, the maths gets pretty crazy pretty soon.

The starting point is defining the problem to which you need an answer, easier said than done.

Even doing 101 stuff like plotting a trend isn't as easy as people imagine, I vaguely remember the problem of Anscombe's Quartet (my maths exams were nearly 40yrs ago and I failed). It's 4 sets of data and using a std linear regression, they all plot the same trend...which is obvioulsy less than ideal. The exam question was to suggest a more appropriate type of regression analysis for each one. I'm no wiser 40yrs on, #1 & #2 are fine but I still can't do #3 or #4.
Screenshot_2.png
You do not have the required permissions to view the files attached to this post.
Post Reply

Return to “General discussion”