I'm drying running a system and would welcome people's view's on whether they thinking it is returning a big enough profit and any comments regarding the sample size.
I have data for 250 instances with very specific criteria and over 2000 instances with looser criteria.
On the 250 instances it is returning profit after comm of on average £5 per instance having laid for £100 at a price no more than 1.99 (but more often about 1.6).
The market can probably take £300 lays the the prices without any major effect on it. So it does not have major scalability.
Strategy Returns?
I think what you need to consider first is whether your backtesting is in-sample or out-of-sample. In other words did you tweak your strategy until you found 250 working instances, or are those 250 instances found after you had fully developed your strategy? If it's the former then there's a possibility you've just overfitted your strategy to the historical data. If you're unsure, I'd advise paper trading until you're more confident and have properly tested your strategy out-of-sample.
I think you should also consider what the potential drawdown could be, as that will help to determine how much capital to invest and how much to keep in the bank. It's common that consistent strategies aren't massively scalable though, because it's unlikely that the market is quite so inefficient in any particular way.
I didn't quite understand what you meant about 2000 loser instances though. If you know they will be losing can't you just flip the trade (eg. back instead of lay, and lay instead of back?)? Or are they only losing due to the commission you have to pay? Or do you mean that you don't know which will lose and which will win, but that the 250 wins outweigh the 2000 losses?
I think you should also consider what the potential drawdown could be, as that will help to determine how much capital to invest and how much to keep in the bank. It's common that consistent strategies aren't massively scalable though, because it's unlikely that the market is quite so inefficient in any particular way.
I didn't quite understand what you meant about 2000 loser instances though. If you know they will be losing can't you just flip the trade (eg. back instead of lay, and lay instead of back?)? Or are they only losing due to the commission you have to pay? Or do you mean that you don't know which will lose and which will win, but that the 250 wins outweigh the 2000 losses?
Haha, I can't be bothered to pick people on grammar.
We have many different nationalities on this forum, so it's amazing they correspond in English!
However, I've noticed on all forms of social networking that your example has become the most common mistake from everyone, including English speakers
We have many different nationalities on this forum, so it's amazing they correspond in English!
However, I've noticed on all forms of social networking that your example has become the most common mistake from everyone, including English speakers
Xitian thanks for your reply.
The 250 is the more "targeted" sample taken from the "looser" less targeted)2500.
I have got a spreadsheet which i can adjust to give me the p/l for any amount of the samples. So i do have a handle on exactly what returns would have been made and also what the largest amount of losses i would be running.
Does anybody have any comments regarding the figures i have posted on whether they think that these returns are adequate?
The 250 is the more "targeted" sample taken from the "looser" less targeted)2500.
I have got a spreadsheet which i can adjust to give me the p/l for any amount of the samples. So i do have a handle on exactly what returns would have been made and also what the largest amount of losses i would be running.
Does anybody have any comments regarding the figures i have posted on whether they think that these returns are adequate?
There are too many variables - imagine the difference between a system that lays 1/10000 shots and shows a profit of £3000 over 3000 observed instances... does not mean it is a winning system. or a system which backs at around evens but is in profit 10% over 3000 instances - it may well be profitable.
However if I understand correctly you have 2500 instances which is not profitable, you have then added further conditions to make it profitable if you only placed bets on 250 of the original 2500?
If this is the case then this is over-fitting to historical results. You need to take what you believe is profitable and then apply it to the 250 future results you collect.
Have a look around google for backtesting strategies; paper trading and statistical significance to give you a good idea of the kind of issues you need to address.
Good Luck on finding a strategy.
However if I understand correctly you have 2500 instances which is not profitable, you have then added further conditions to make it profitable if you only placed bets on 250 of the original 2500?
If this is the case then this is over-fitting to historical results. You need to take what you believe is profitable and then apply it to the 250 future results you collect.
Have a look around google for backtesting strategies; paper trading and statistical significance to give you a good idea of the kind of issues you need to address.
Good Luck on finding a strategy.
I have a sample set up 2500 and when i apply certain criteria the sample narrows down to 250. If i were to have placed £100 lay bets (at price no higher than 1.99) in these events i would be showing a profit which would be £1250 after commission.
Hence where i get the £5 per instance profit.
Hence where i get the £5 per instance profit.
Hi Buyshirts,
I've attached a spreadsheet that does strategy comparisons. You can play around with the numbers in the cells highlighted in Grey to see what effect commissions and strike rate will have on different strategies.
Regards,
Switesh
I've attached a spreadsheet that does strategy comparisons. You can play around with the numbers in the cells highlighted in Grey to see what effect commissions and strike rate will have on different strategies.
Regards,
Switesh
You do not have the required permissions to view the files attached to this post.