Holy Grail?

[episteme]

There is no way from what you say to work out what could be doing this. I do know however that anything I have ever produced that gave results like this from back testing was soon followed by a DOH!

If you can't see the problem now then try and find a completely different way of testing it. 20 mil is probably worth the effort.

 

[N Rothschild]

the problem with systems and back testing is this.

Max. Consecutive 18
Largest loss -644048.34


you said start with 50,000? if you had those 18 concecutive loses right out the gate, your done, blownup

 

[skanker]

Any suggestions on a completely different way to test it? Sorry for being such a newbie.

 

[skanker]

Rothschild,

point taken. maybe I need to limit my exposure to be able to absorb 18-20 consecutive losses.

 

[N Rothschild]

the problem in any system is whether you start in a down period or an up period for the system, this can have a massive effect on the outcome

 

[astoria]

Can someone analyze this?

Avg. Profit/Loss 4435.91

Winners 4313 (60.87 %)
Avg. Profit 12209.15
Avg. Profit % 6.48 %
Largest win 923883.76
Max. Consecutive 34
Largest win 923883.76

Losers 2773 (39.13 %)
Avg. Loss -7654.25
Avg. Loss % -4.21 %
Max. Consecutive 18
Largest loss -644048.34

Max. trade drawdown -644048.34
Recovery Factor 20.07
Profit Factor 2.48

are you running this off NT?

 

[Mr Flibble]

Did you hide any data from a certain date? If not its sounds like you have shaped a system around your data. I can do plenty of these and as soon as I add the missing data its quite shocking the difference. Get happy when your system does equally well with the data date filter off and when it live tests nicely.

 

[skanker]

Ok not quite the holy grail as I have found a future reference in it. I guess I didn't win the lottery. I am going to create a new thread to place the code and see if people can help me refine it.

 

[Charlton]

There is no way from what you say to work out what could be doing this. I do know however that anything I have ever produced that gave results like this from back testing was soon followed by a DOH!

If you can't see the problem now then try and find a completely different way of testing it. 20 mil is probably worth the effort.

Time to apply your 3 letter word to post 28

You too had a future reference.

Charlton

 

[episteme]

Thought so - as I say surprisingly easy to do.

You have to be careful testing - for example with metatrade if you look up a timeframe then you are seeing the whole of that timeperiod - so a lot of the time it contains future information.

 

[episteme]

I remember another strange one. Was using some EOD data and had an awesome system. Problem was the low and high were the theoretical low and high of the day (i.e what the index high/ low would have been if all the individual stock high/ lows had happened at the same time). I still don't understand to this day what use that data was to anyone.

 

[skanker]

If there is a system that works well on the S&P 500 but not on the NASDAQ, do I dismiss it because it doesn't work well across different markets?

 

[Adamus]

no, but you need to be able to see why. You need to be able to tell that you haven't curve-fitted your system by over-optimising it.

 

[simcom]

I remember another strange one. Was using some EOD data and had an awesome system. Problem was the low and high were the theoretical low and high of the day (i.e what the index high/ low would have been if all the individual stock high/ lows had happened at the same time). I still don't understand to this day what use that data was to anyone.

I had the same problem - cost me weeks of wasted time and dashed my hopes of billions!

I think the data I had the issue with was Yahoo EOD Dow 30. Only when somebody on here told me about it did I realise why the results seemed so amazing.

 

[meanreversion]

I use Amibroker and would advise you to go back and check precisely what trades you are generating, as the software can sometimes deal at invalid rates. For example, if your stop rate to sell is 25, and one day the market closes at a recent low of 26, then the following day it opens (gap lower) at 20 but then later makes a high of 22 before going lower, the machine will stop you at 22 (it's the rate closest to 25 that it can see).

Obviously you would not sell at 22 in the real world, because you don't know it's the high!

Amibroker is an excellent piece of software, but please use with caution. Metastock has similar issues, it will allow you to deal at the low or high of the day.

 

[DionysusToast]

A few questions will sort this out.

1 - what time on average is your morning trade placed ? Are you assuming a fill at 9:30 EST ?
2 - what time on average is your evening trade placed ? Are you assuming a fill at 16:00 EST ?

How many cents profit/loss are your average trades ?