Any system traders here?

[avano]

Hi everyone! I wonder if there are individuals here who actually develop and trade automated futures and stock system. I know many talk about trading system but few use them and even fewer make money with them. I am researching new ideas about developing systems. Please note that I am not interested in forex and in intraday trading of any kind.
Suggestions about trading system development will be appreciated. I have used many platforms and have been developing systems for 15 years so I am willing to talk about my findings in exchange. Here is one to start with: the number of trades from backtesting should neither be too low or too high. Both very low and very high number of trades indicate fitting to the data.

 

[Liquid validity]

I'd stick to quantnet and wilmott TBH.
T2W does not have a strong focus in this area.
Better yet, f**k forums altogether

 

[avano]

Thanks but quantnet has very limited participation and wilmott is more for investment banking quants. I am looking for retail trading experience with systems. What do you think about ET?

 

[the hare]

There's nowhere that does this particularly well. The forums provided by the charting application providers are probably about as good as its going to get, tradestation, wealthlab, and even metaquotes have the odd post worth reading, but its mostly nonsense.

Alternatively publish some stuff on a blog or website, just to prove you have something to contribute, and the right sort of people will usually find you (unfortunately it'll attrack a lot of ******s too !)

 

[Liquid validity]

Thanks but quantnet has very limited participation and wilmott is more for investment banking quants. I am looking for retail trading experience with systems. What do you think about ET?

Yeah ET is probably more in depth for this kind of stuff.
As the Hare said though, not really anywhere that is going to go
into heavy detail. People are generally reluctant to openly share
complete ideas for obvious reasons. If it works, keep it to yourself
Its no coincidence that the major quant outfits are also very secretive.

I'm not knocking T2W, but the top 6 main forum categories don't help to promote
anything other than forex and intraday trading (nothing wrong with that).
Big mikes trading forum does cover algo trading a bit as well.
Generally though, from what I've seen, most of what is posted in forums is garbage
curve fitted to backtests of 6-12 month datsets with high win rates that make
50% a month until they blowup

 

[Hotch]

I have used many platforms and have been developing systems for 15 years so I am willing to talk about my findings in exchange. Here is one to start with: the number of trades from backtesting should neither be too low or too high. Both very low and very high number of trades indicate fitting to the data.

That's like saying you shouldn't drink too much water or to little water.

There are very few areas which haven't already been comprehensively explored, especially when ignoring intraday. I can only think of one and it doesn't lend itself to sharing and caring.

If you have 15 years experience I think you should work on being content with the edge you've already discovered.

 

[Tayto_Crisp]

The overall knowledge of people on this forum is almost non existent. Most are content to advertise themselves as super traders, super studs, super men and so on. Learn what institutional entries are, learn how to spont them on a chart then choose say a moving average. Back and forward test some of these for a year, taking notes of entries and exits, then you know where you might have an entry and exit equation that consistently gets you in and out of the markets when the money is moving. Can't say if this will make you any money or not but give it a shot. Either way you will never get a trader to sign up for it but there are several dumbingens that will take an interest.

 

[avano]

If you have 15 years experience I think you should work on being content with the edge you've already discovered.

I need at least one new edge every year to survive. Does that make sense to you?

 

[tony1979]

Gloves off?

 

[alphadude]

I have used many platforms and have been developing systems for 15 years so I am willing to talk about my findings in exchange..

in exchange for what?

how much are you willing to sell your secret nazo box?

 

[avano]

in exchange for what?

how much are you willing to sell your secret nazo box?

Listen dude, I'm not selling anything. You have a very bad attitude for this forum and you are basically one of the reasons that traders avoid them. I suggest you abandon your bad attitude or you will live the rest of your life in misery.

Now to the subject: I have found out that some systems work best in real trading if one uses the full price history to set their parameters rather than setting them in an in-sample and then testing for out-of-sample performance. This makes sense to me and I tend to think that out-of-sample testing handicaps the good systems although it is of help for screening out bad systems.

What are your opinions/experiences on this one?

 

[rsh01]

What are your opinions/experiences on this one?

I think you are talking absolutely crap. Forward test whatever system you have, provide a few live calls if you wish, and stop posting twaddle.

 

[the hare]

Now to the subject: I have found out that some systems work best in real trading if one uses the full price history to set their parameters rather than setting them in an in-sample and then testing for out-of-sample performance. This makes sense to me and I tend to think that out-of-sample testing handicaps the good systems although it is of help for screening out bad systems.

What are your opinions/experiences on this one?

I dont like any optimisation. I've always found that optimised parameters are too unstable, and they tend to shift around way too much over time. I'd rather diversify across a broad range of parameters, and then use position sizing to allocate capital as appropriate.

I'd possibly go so far as to say that you cant trade TA based systems mechanically without a great deal of diversification.

I might be talking through my **** on this one, as I use totally random entries, and the effects of random chance might possibly have a greater effect on my results than someone who is more mechanical, but from the results I've seen from others, I think I'm right.

 

[avano]

I think you are talking absolutely crap. Forward test whatever system you have, provide a few live calls if you wish, and stop posting twaddle.

Another one with a bad attitude. Probably some loser big time.

Listen dude, when you use a portion of your data to test out of sample you deprive your system in the design phase from conditions that may not be present in the in sample. I guess you have never encountered this because you just osund like a parrot of material you read in forums. But serious trading system developers like me know this is a problem. Now please leave this thread. You have nothing to contribute. You do not have the knowledge to contribute anything here like the other dude. Go play some video game please.

 

[alphadude]

Another one with a bad attitude. Probably some loser big time.

Listen dude, when you use a portion of your data to test out of sample you deprive your system in the design phase from conditions that may not be present in the in sample. I guess you have never encountered this because you just osund like a parrot of material you read in forums. But serious trading system developers like me know this is a problem. Now please leave this thread. You have nothing to contribute. You do not have the knowledge to contribute anything here like the other dude. Go play some video game please.

Son; you may be foremost expert in system dev; so cut to the chase and backup your claims with statements or at least backtest results.

Walk the walk. we have many twadles around here.

 

[Liquid validity]

when you use a portion of your data to test out of sample you deprive your system in the design phase from conditions that may not be present in the in sample.

I would agree that you have a point, no OOS means more data to base your algo design on.
If the OOS is disregarded, I personally think its only safe to do so if the trade
frequency is high enough to allow a decent live forward test without spending years
in forward testing.

Thats generally the idea with OOS, to replicate forward testing anyway.
In short, I would prefer a live forward test of at least 300 trades to an OOS test anyday.
That is simply for the reason that backtesting is never spot on compared to a live forwards test anyway.
Backtesting is only meant to test execution and give a rough idea of performance.
In summary, either an OOS test or +300 sample size forward test is obligatory.
Doing away with both is asking for trouble.

For me personally that is the juggling act that must be dealt with.
Higher trade frequency = decent live forward test
Downside is higher commisions, and slippage will also be more of a factor due to shorter
trade duration.
Having said that, with a random or semi random entry, positive slippage can reduce or even balance
out negative slippage, reason being random (or semi random) is not always chasing price.

 

[Liquid validity]

I dont like any optimisation. I've always found that optimised parameters are too unstable, and they tend to shift around way too much over time. I'd rather diversify across a broad range of parameters, and then use position sizing to allocate capital as appropriate.

Completely agree with your points on optimisation, its just automated curve fitting.

 

[rsh01]

Go play some video game please.

Good idea thanks, I will!
Your sentence...
"I have found out that some systems work best in real trading if one uses the full price history to set their parameters"
is the biggest no sh*t statement I have heard, but then you go and top it with..
"when you use a portion of your data to test out of sample you deprive your system in the design phase from conditions that may not be present in the in sample."

Good skills! And I gave you advice which others have also...fwd test it. But go ahead if by stating the obvious helps you then please continue with your 'I have found' when you really haven't found anything statements. I am sorry I bothered to open the thread.

 

[alphadude]

I would agree that you have a point, no OOS means more data to base your algo design on.
If the OOS is disregarded, I personally think its only safe to do so if the trade
frequency is high enough to allow a decent live forward test without spending years
in forward testing.

Thats generally the idea with OOS, to replicate forward testing anyway.
In short, I would prefer a live forward test of at least 300 trades to an OOS test anyday.
That is simply for the reason that backtesting is never spot on compared to a live forwards test anyway.
Backtesting is only meant to test execution and give a rough idea of performance.
In summary, either an OOS test or +300 sample size forward test is obligatory.
Doing away with both is asking for trouble.

For me personally that is the juggling act that must be dealt with.
Higher trade frequency = decent live forward test
Downside is higher commisions, and slippage will also be more of a factor due to shorter
trade duration.
Having said that, with a random or semi random entry, positive slippage can reduce or even balance
out negative slippage, reason being random (or semi random) is not always chasing price.

spot on.

to say the least; OOS in its pure theory; does work.

however; in reality OOS is not OOS. it is IS. why? because of hindsight.
once you have tested your strategy in IS; then in OOS; and it failed in the latter. You went ahead and changed the rules; then tested again in IS then OOS until the OOS got good results.

once you have tested the strategy in OOS one time; that's it. You have hindsight about it; and hence you cannot claim later that your strategy worked in OOS.

as long as you only use OOS once; and only once for a particular strategy; then you have to go through forward testing before putting any money in it.

experience suggest that forward testing of 2 months; or ~300 trades; is a must; in order to address the the biases (beside the curve fitting bias).

 

[avano]

spot on.

to say the least; OOS in its pure theory; does work.

however; in reality OOS is not OOS. it is IS. why? because of hindsight.
once you have tested your strategy in IS; then in OOS; and it failed in the latter. You went ahead and changed the rules; then tested again in IS then OOS until the OOS got good results.

once you have tested the strategy in OOS one time; that's it. You have hindsight about it; and hence you cannot claim later that your strategy worked in OOS.

as long as you only use OOS once; and only once for a particular strategy; then you have to go through forward testing before putting any money in it.

experience suggest that forward testing of 2 months; or ~300 trades; is a must; in order to address the the biases (beside the curve fitting bias).

Why do you talk about things you do not understand with such certainty? Can you at least qualify them with "In my opinion..."?

OOS testing makes sense only if both IS and OOS samples were drawn from the same distribution. Otherwise it is nothing of the sort you claim exhibiting your ignorance of the subject. If the distribution is non-existent or unknown, as it is the case with many financial series, oos tell you nothing. Learn not to talk about things you do not understand. Is this why you hide in forums?