Hi all
I've heard Peter say several times in videos and on the forum that if you trade at random you should break even over the long run but lose because of commission.
This got me thinking ... how does this apply with the crossover points. If I lay a series of runners at 2 with a 10 tick greening offset and a 10 tick stop loss, then is the long term scenario either:
(a) the price will trade in either direction 50% of the time thus making more in the long run due to the crossover; or
(b) the market factors in the price compression and will trade 5 ticks high / 10 ticks low resulting in a breakeven result.
I'm thinking it must be (b) otherwise it would be simple just to trade crossovers and random and profit.
Thoughts?
Trading at Random Question
- ruthlessimon
- Posts: 2094
- Joined: Wed Mar 23, 2016 3:54 pm
I'm actually analysing this atm, specifically for breaches coming from above 2.0.
What I will say is, why the fixation on 10 ticks? Why not follow Peter's advice & try 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13...
P.S. Peter might have meant something completely different, by the multiple strategies quote - but that was my personal interpretation
What I will say is, why the fixation on 10 ticks? Why not follow Peter's advice & try 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13...
P.S. Peter might have meant something completely different, by the multiple strategies quote - but that was my personal interpretation
- ruthlessimon
- Posts: 2094
- Joined: Wed Mar 23, 2016 3:54 pm
It's a fascinating concept, & if you have time well worth looking into.
My mantra for delving into those statistics was: "Peter would have that data, so there's no excuse for me not to have that data"
My mantra for delving into those statistics was: "Peter would have that data, so there's no excuse for me not to have that data"
I have asked myself the same questionsCards37 wrote: ↑Tue May 29, 2018 9:33 pmHi all
I've heard Peter say several times in videos and on the forum that if you trade at random you should break even over the long run but lose because of commission.
This got me thinking ... how does this apply with the crossover points. If I lay a series of runners at 2 with a 10 tick greening offset and a 10 tick stop loss, then is the long term scenario either:
(a) the price will trade in either direction 50% of the time thus making more in the long run due to the crossover; or
(b) the market factors in the price compression and will trade 5 ticks high / 10 ticks low resulting in a breakeven result.
I'm thinking it must be (b) otherwise it would be simple just to trade crossovers and random and profit.
Thoughts?
This old thread may shine some sunlight on the subject of random trading for you if you read through it all.
viewtopic.php?f=2&t=196
- ruthlessimon
- Posts: 2094
- Joined: Wed Mar 23, 2016 3:54 pm
The really fascinating (edit:nasty) problem when it comes to random trading, is that we all assume a random strategy must be "just under breakeven". But this isn't true.
A lot of random strategies diverge massively from the 0 expectancy threshold - but they are still random - which makes it incredibly difficult to tell if it's a real inefficiency
This is what a random equity curve looks like:
A lot of random strategies diverge massively from the 0 expectancy threshold - but they are still random - which makes it incredibly difficult to tell if it's a real inefficiency
This is what a random equity curve looks like:
Great read thanks. Didn’t directly answer the question but definitely shed some light.
Emmson wrote: ↑Tue May 29, 2018 11:43 pmI have asked myself the same questionsCards37 wrote: ↑Tue May 29, 2018 9:33 pmHi all
I've heard Peter say several times in videos and on the forum that if you trade at random you should break even over the long run but lose because of commission.
This got me thinking ... how does this apply with the crossover points. If I lay a series of runners at 2 with a 10 tick greening offset and a 10 tick stop loss, then is the long term scenario either:
(a) the price will trade in either direction 50% of the time thus making more in the long run due to the crossover; or
(b) the market factors in the price compression and will trade 5 ticks high / 10 ticks low resulting in a breakeven result.
I'm thinking it must be (b) otherwise it would be simple just to trade crossovers and random and profit.
Thoughts?
This old thread may shine some sunlight on the subject of random trading for you if you read through it all.
viewtopic.php?f=2&t=196
So I think I will run my own experiment on this, look at the wins and losses and see if I can tilt the strike right slightly.
Interesting though a good point on variance. Let’s say I make a simple change where instead of trading at random and green/redding up just before the off, I have a stop loss built in. How many samples do I need to run before I know my alteration is actually positive and not just the result of variance?
Interesting though a good point on variance. Let’s say I make a simple change where instead of trading at random and green/redding up just before the off, I have a stop loss built in. How many samples do I need to run before I know my alteration is actually positive and not just the result of variance?
- ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
This is because the random move up/down is the implied% rather than a random #ticks?
The real world works on chance/confidence, ie nice and linear. On BF we have bumps in the road. Ticks on BF are a bloody nuisance really.
Just for the range 1.9 - 5.0:
Blue line - Implied% (not straight)
Orange line - Change in Implied% for each tick
You do not have the required permissions to view the files attached to this post.
So I tried the following overnight:
At 5min to post, place $2 back bet on favourite at best market price, 50 tick offset, 10 tick trailing stop loss.
33 races, overall result $-2.09.
Again the biggest question that comes to mind initially is how many samples really do you need before beginning to look for tweak conditions. Seems to me like I could run this 400 times and variance would still be a factor.
At 5min to post, place $2 back bet on favourite at best market price, 50 tick offset, 10 tick trailing stop loss.
33 races, overall result $-2.09.
Again the biggest question that comes to mind initially is how many samples really do you need before beginning to look for tweak conditions. Seems to me like I could run this 400 times and variance would still be a factor.
- ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
If you stick your results into excel you can use the STEYX function to find the standard error and CORREL for the correlation to trend, from there you can find the error significance. <5% is a start but <2% and you're getting somewhere.
For example this trend shows and error significance of just 6.7%....looking promising eh ?
...nope, it's a coin flip, no edge, just a random 50/50 on 1000 flips.
For example this trend shows and error significance of just 6.7%....looking promising eh ?
...nope, it's a coin flip, no edge, just a random 50/50 on 1000 flips.
You do not have the required permissions to view the files attached to this post.
- ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
That error significance is something I made up, it's the proportion the StdErr is of the total pl, ie 1-(pl-StdErr/pl). So a £2 stderr on a £500 pl would be 1-(498/500) = 0.04%
The others are just normal Excel things..
StdErr = STEYX(Values,SampleNumber)
Correlation ( R ) =ABS(CORREL(Values,SampleNumber))
R squared =R^2
I'm a loooong way from knowing much about stats though so don't quiz me too much on what that tells us.
After 1000 flips it could go anywhere. All it says is that the previous 1000 flips when that way. I had to hit F9 about 20 times to get a random test that produced what looked like a strong trend. It was just to show how easy it is to see trends where there isn't really an edge. I guess that's why Peter is always so keen to try and understand why it might be the case rather than just showing it is in the data.
Agreed on the last point. I think the challenge when starting out is learning- or even approximating - the value of variance and not seeing something that isn’t there. Confirmation bias has to be a real danger in this analysis.
- ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
Absolutely, the whole area is a minefield.
This 'finding the reason why' is so easy to say and so hard to demonstrate unless it's something that's physically happening at the course. The confirmation could easily be made to fit the results if you're not careful.
It's as though you need a double blind test, one looks at the results and guesses a reason. The other starts with that reason and tries to find it in the results. I might have a dim memory of hearing Peter say he's given stuff to statasticians to look at, maybe that's the type of thing he's been doing?