Strategy Development: Modelling

[ShaunWhite]

I really do feel like the slow kid at the back of the class

I always have to come up with a reason why it would work.

It? It being somethng you found in the data or a gut feeling ?

then when it's deploying

Deploying being the key word? Because if 'it' was something you found in the old data, then finding the opposite in the old data would be self fulfilling, pointless? So you look forthe "anti-it" also occuring in the newest data being collected first, not just more 'it'. Because doing so tests that the opposite is evident asap, rather than immediately trying to benefit from an assertion you haven't confirmed?

I immediately use that assumption to test the opposite.

I must say this is where I struggle.

So if "it" was a strategy to back then lay, you'd lay then back ? To see if would lose ? or if you were backing in races <1m you'd try backing in races >1m.

So it's the process of trying to disprove a theory that proves it.

Therefore, proving 'not it' is false, as confirmation of your theory that 'it' is true? Double confirmation.

....I feel like I'm panning for gold.

If I never make trading pay 'well' I'll be utterly crushed because I've never worked or thought about something so hard, as I have in this last year. Must have put in about 3000hrs but still feel like I've only reached the six-chair-challenge

[spreadbetting]

I remember back in the 1990's there was a relay agent called Keith Nicholson who used to sell on all the premium rate tips and systems from the small ads in the Racing Post/Sporting Life. Think I paid around £10 and got a massive bundle of photocopied systems which generally consisted of backfitted systems for the weeks big races where someone had managed to get a 80% profile for the previous winners and sold it on as a winning system for the weekend. Whenever I see posts with people trawling thru reams of data looking for edges it always reminds of that £10 I wasted

[ruthlessimon]

I really do feel like the slow kid at the back of the class

If I never make trading pay 'well' I'll be utterly crushed because I've never worked or thought about something so hard, as I have in this last year. Must have put in about 3000hrs but still feel like I've only reached the six-chair-challenge

+1 Really hope Peter expands

[Euler]

There was another thread somewhere where I was hinting that prices are often out of kilter. Plenty of counter arguments on the thread as well.

So to test that (in 2011!!!) I started looking at opening prices, closing prices and variation with a view I could probably pick where something was no longer value. Once I'd refined how I was going to do that I started laying with small amounts to test what happened. That seems to work, so I adopted the anti strategy and that lost money convincingly so I was satisfied I could raise stakes. I started at £50 and worked my way up to £1k stakes. But have refined the selection process significantly along the way.

[Euler]

I remember back in the 1990's there was a relay agent called Keith Nicholson who used to sell on all the premium rate tips and systems from the small ads in the Racing Post/Sporting Life. Think I paid around £10 and got a massive bundle of photocopied systems which generally consisted of backfitted systems for the weeks big races where someone had managed to get a 80% profile for the previous winners and sold it on as a winning system for the weekend. Whenever I see posts with people trawling thru reams of data looking for edges it always reminds of that £10 I wasted

I remember that well and have also been there.

[Euler]

On greyhounds I had a stab at what I thought would work, unknown to me at the time I pretty much hit the nail on the head. Which is probably an experience thing.

However, it's often just luck. So to test it I created a super refined version of the trigger which looked the 'perfect' scenario and a really loose trigger which threw away all the refinements. Remarkably it was the loose one that performed better, so I nudged my strategy in that direction. I've varied it about 30+ times since I created it, all with the view of seeing how a variation works.

I really got into the theories of Darwinism once and buried myself in that work and it taught me to keep testing and evolving constantly and eventually the best system wins out. Sometimes though you to an flipping giant meteor and wipe out a successfully strategy, but as it's not an earth ending event, so you generally re-find it again in the future.

https://en.wikipedia.org/wiki/Evolution

I've found so many successful strategies this way.

[ruthlessimon]

So to test that (in 2011!!!) I started looking at opening prices, closing prices and variation with a view I could probably pick where something was no longer value. Once I'd refined how I was going to do that I started laying with small amounts to test what happened. That seems to work, so I adopted the anti strategy and that lost money convincingly so I was satisfied I could raise stakes. I started at £50 and worked my way up to £1k stakes. But have refined the selection process significantly along the way.

Out of interest. Did switish & Korattt do this analysis? If not, could this be the reason for their failure? Not down to psychology - but down to being given pure refined edge too early?

A primary school pupil won't be taught calculus. If a primary school pupil was taught calculus, I'd have that would stunt their growth & they would fail their SAT maths. Basics first, refinements later.

[Euler]

They were both at opposite ends of the spectrum in terms of the way they worked IMHO.

[ruthlessimon]

They were both at opposite ends of the spectrum in terms of the way they worked IMHO.

Again that's a really interesting insight. Although the connotations that gives Korattt are pretty harsh!

From a newbies perspective. The traders who fail, & lose their interest in trading, are absolutely golden mentors. They need to know why they failed - what links them - so that they don't make the same mistakes. What do the 90% have in common etc.

[ShaunWhite]

Great stuff Peter (and everyone actually), thanks for all that. I'll go away and digest it.

(btw when i said "3000 hrs" above, i actually meant "every waking hour")

[ruthlessimon]

(btw when i said "3000 hrs" above, i actually meant "every waking hour")

For us Shaun it's every hr after 1am

[ShaunWhite]

(btw when i said "3000 hrs" above, i actually meant "every waking hour")

For us Shaun it's every hr after 1am

My sleep's shocking atm. A log would look like a perm 5 from 24 matrix.

[Euler]

Try trading at random. When I first started trading I did that and hardly lost anything. So I then developed a strategy that didn't trade at random and noted where the big losses came from. I then started working on striking them out.

[Cards37]

Try trading at random. When I first started trading I did that and hardly lost anything. So I then developed a strategy that didn't trade at random and noted where the big losses came from. I then started working on striking them out.

Late to the party

Peter two questions if I may:

(1) What do you actually mean by trading at random? Seems a silly question but I assume you mean come up with a basic idea and then apply it in random markets?
(2) Typically how many trades do you think is statistically valid "per evolution" to make judgements and start nudging in a particular direction? This would seem fundamental to have a high enough confidence factor that those big losses were themselves not random and indeed generated by the system parameters.

[ruthlessimon]

(2) Typically how many trades do you think is statistically valid "per evolution" to make judgements and start nudging in a particular direction? This would seem fundamental to have a high enough confidence factor that those big losses were themselves not random and indeed generated by the system parameters.

Really like the question, as I've been trying to answer a similar "thought experiment". One which Shaun also alluded to in a previous post

When it's valid to say, "I've learned something new.", "this edge no longer works" etc

Let's say we have a day of 10 straight losing markets: Is that simply the expected probability of a longterm edge? (i.e. edge shouldn't be changed, nothing was learned, this was expected) Or was there a bias on the day? (i.e. edge needs to be adjusted to incorporate this new info). How do we tell the difference? Analysing every turn of the market, & you risk never seeing the bigger picture. At the same time, a huge dataset, will be extremely slow adjusting to genuine changes in the market. There has to be a balance