95% ASX100 success rate

[RobertoHood]

Hello,

Over the past 4 years i have been working to develop a system that can predict weekly moves tracking all companies listed in the ASX100.

I have refined this now to the point where i get on average an accuracy rate of 95.65%

With percentage gains averaging 1.62% per company per week.

Tonight i will list the companies which have triggered BUY in the system based on last weeks date.

The only rule of the system is to buy on the open of monday morning, and either sell at the close of friday , or sell when your stock gain reaches a level where you want to take profits.


Of the 23 buy signals i got last week, only 1 was not successful.

Sorry everyone , this was meant to state 23 BUY signals over the 12 week period not over the week.

Thank you rnr for picking this up

 

[rnr]

Sorry everyone , this was meant to state 23 BUY signals over the 12 week period not over the week.

Thank you rnr for picking this up

OK, I see where your coming from and confirmed by the 95.65% (22/23).

 

[tech/a]

Of the 23 buy signals i got last week, only 1 was not successful.

Would that not have to read.

"Of the 23 buy signals over the last 12 weeks----etc etc''

 

[RobertoHood]

Would that not have to read.

"Of the 23 buy signals over the last 12 weeks----etc etc''

Yes that is correct tech, my sincere apologies

 

[tech/a]

Yes that is correct tech, my sincere apologies 

All good.

So each WEEK the companies are re set?

Could be the same or different companies but the buy sell timeframe is a week.
Of 27 different signals over 12 weeks 95% on a week to week basis finished higher.

Just making sure I understand the result.

 

[RobertoHood]

All good.

So each WEEK the companies are re set?

Could be the same or different companies but the buy sell timeframe is a week.
Of 27 different signals over 12 weeks 95% on a week to week basis finished higher.

Just making sure I understand the result.

Yes, if you do not sell at close of Friday or before, you are in unmeasured waters my friend.

Scan can be performed for the next week week once Fridays data is available. I myself perform the scan Sunday night.

And yes ,of the 23 BUY signals over the 12 week period , 22 finished higher.

 

[sinner]

I was thinking about ideas for anyone who wanted to track the selections, and realised I needed to ask this first.

RobertoHood, without giving away your system...would you describe it as being based on trend/strength/continuation etc? Or would you describe it as mean reversion/short-term weakness etc?

Reason I ask is that it might help those who want to track it with different sell criteria.

I'll bet dollars to donuts that it's a machine learning system and that therefore this question is unanswerable.

Basically the algorithm will pick up on whichever edge has shown the most likelihood for a positive weekly return over the duration of the training data, i.e. basically the training data will pick up on whichever direction weekly autocorrelation is pointing and therefore sometimes predict weakness leads to strength while other times predicting that strength will lead to strength.

Roberto, if my bet is correct, just so you know, there are some ML algos out there that let you do regression mode rather than classification (i.e. both sign and magnitude rather than just sign). I like Support Vector Machines, my favourite implementation is SVM-Light: http://svmlight.joachims.org/

Another free note for you: I think you'll find the success of this system will be hugely influenced by the volatility regime the market is currently in (one of: high and declining, high and increasing, low and declining, low and increasing). My guess is you'll need to completely shut this beast off during at least 2 of those volatility regimes to avoid it crashing or flatlining your equity curve.

EDIT: To all the naysayers: so long as Roberto keeps moving his 3 month window of training data he can very effectively capture whatever the current autocorrelation trend is - as long as the trends in autocorrelation display momentum behaviours, which they do tend to.

 

[RobertoHood]

I was thinking about ideas for anyone who wanted to track the selections, and realised I needed to ask this first.

RobertoHood, without giving away your system...would you describe it as being based on trend/strength/continuation etc? Or would you describe it as mean reversion/short-term weakness etc?

Reason I ask is that it might help those who want to track it with different sell criteria.

For example, if it's based on current weakness, someone might like to try the, "first profitable close" exit (nicked from Rob Hanna, Cesar Alvarez, Nick Radge).

But if it's trend following / continuation in nature, that is not going to be a good exit to use! Here, you'd obviously want to do something to cut the losers and let the profits run.

I'm assuming, based on the strike rate, that you would describe it as weakness or reversion based.

Hey system,

I am not avoiding this question , but i really dont know how to answer it without giving away what im doing.

But let me tell you a few things i can say

- it does not require a chart ( no trend following, no reversion trading)
- therefore oscillators and any other form of indicator are not used

- i do use excel
- i do not use any fundamentals what so ever

- its not a "machine" well i guess my laptops a machine

- the calculations i do , my 11 year old brother could do (im bad at math)

- all that matter is that i have the information from the previous week to copy and paste into my spread sheet which is embedded with formulas , which have shown patterns for stocks which tend go up the following week.

Thats it, and thats all it is

Hope that helps

 

[RobertoHood]

I'll bet dollars to donuts that it's a machine learning system and that therefore this question is unanswerable.

Basically the algorithm will pick up on whichever edge has shown the most likelihood for a positive weekly return over the duration of the training data, i.e. basically the training data will pick up on whichever direction weekly autocorrelation is pointing and therefore sometimes predict weakness leads to strength while other times predicting that strength will lead to strength.

Roberto, if my bet is correct, just so you know, there are some ML algos out there that let you do regression mode rather than classification (i.e. both sign and magnitude rather than just sign). I like Support Vector Machines, my favourite implementation is SVM-Light: http://svmlight.joachims.org/

Another free note for you: I think you'll find the success of this system will be hugely influenced by the volatility regime the market is currently in (one of: high and declining, high and increasing, low and declining, low and increasing). My guess is you'll need to completely shut this beast off during at least 2 of those volatility regimes to avoid it crashing or flatlining your equity curve.

EDIT: To all the naysayers: so long as Roberto keeps moving his 3 month window of training data he can very effectively capture whatever the current autocorrelation trend is - as long as the trends in autocorrelation display momentum behaviours, which they do tend to.

Im sorry but i really dont understand this message , can you break it down into more simple terms.

 

[sinner]

Im sorry but i really dont understand this message , can you break it down into more simple terms.

Guess I owe somebody some dollars and/or donuts...

 

[RobertoHood]

Guess I owe somebody some dollars and/or donuts...

Just dollars please , no donuts (bodys a temple)

 

[RobertoHood]

OK so far for anyone whose following so far :-

ASX100 : -0.07% (Based on Mondays open to today's close)

6 Stock Picks : .696%

Now i know that's not a very impressive return (although for anyone whose been watching some good profits could have been taken, all 6 were up from Mondays open this morning around 10-10:30am)

But in regards to this "system" following the index, i can assure you all that it doesn't , but time will tell. And the only way to prove this is to show you!

Hope everyone had a good day on the markets!

 

[RobertoHood]

Wowzers! What a day

 

[ThingyMajiggy]

Wowzers! What a day

Yep, this should certainly test out your strategy of a higher close on Friday!

 

[RobertoHood]

Yep, this should certainly test out your strategy of a higher close on Friday!

Haha yes it should Thingy

 

[So_Cynical]

I suppose that even a 95% success rate for 9 weeks in every 10 would still be a viable outcome...as long as you didn't lose to much in that one week.

 

[systematic]

tech/a, did this make sense?

 

[tech/a]

Yes of course.

Not a system I could live with.
I was fully aware how it could occur.

But still think that with a method which is in a market
Which you can trade long OR short that to actually design
A system as you suggest and have it lose 100 or so ticks
While not having a winning trade ---- would be very difficult.

If you've done it that's pretty interesting---to me.

 

[systematic]

Yes of course.

Not a system I could live with.
I was fully aware how it could occur.

I think you've got me confused with another poster.

You seemed incredulous toward cynic on page 2 of this thread, so I gave you an example of how it could happen. Otherwise, why be surprised? Anyway...

But still think that with a method which is in a market
Which you can trade long OR short that to actually design
A system as you suggest and have it lose 100 or so ticks

...In my example, it was a 14 pip loss, over 100 trades. Not sure where you've got 100 from.

...to actually design A system as you suggest and have it lose 100 or so ticks While not having a winning trade

Where in my example was there not a winning trade? My example had 97% winning trades!

Based on the above highlighted comments, I think you've misunderstood entirely, but will leave it there. All good; just offering an example. Might have been insightful for a newbie reading anyway.

 

[tech/a]

I think you've got me confused with another poster.

You seemed incredulous toward cynic on page 2 of this thread, so I gave you an example of how it could happen. Otherwise, why be surprised? Anyway...





...In my example, it was a 14 pip loss, over 100 trades. Not sure where you've got 100 from.




Where in my example was there not a winning trade? My example had 97% winning trades!

Based on the above highlighted comments, I think you've misunderstood entirely, but will leave it there. All good; just offering an example. Might have been insightful for a newbie reading anyway.

Not understanding the pedantic nature of the to and fro ings.
Aware it can be done.
If you have a single loss of 100 ticks in 3% of your trades then you would have to go for an extended no of ticks
without a winning trade.

Anyway---