Is this auto trading system good enough for sale?

Like someone said before, testing for slippage and out of sample data are your next steps. It is not impossible to create a system that makes money but it isn't easy at all. Ive had a lot of systems that look great on backtest but they fail miserably when I test them live. Slippage is a bitch for shorter term trend following systems that need to get in the market fast without missing the trade. I have two systems that make money live and it was a lot of blood sweat and tears making them.

EDIT: I wouldn't sell them for under 1 million.
 
I suggest you offer it on Collective2. If you post three consecutive profitable months with similar stats, you'll get subscribers provided you don't charge more than 10% of the gross monthly profit. And you don't have to risk your capital to do it.

If the system blows up or fails to perform, you'll only be $78 out of pocket.
 
I'd agree, any system that reliable generates a profit (of any magnitude) and can scale is potentially worth millions.

The key is reliability - your backtest figures look good almost too good. Can you backtest on other data and get comparable results.

Then forward test on a live or demo account for at least 3 months. Live is far better because you'll get realistic transaction costs including slippage.

You can use Collective2 to provide an independant audit of results and if the system works well enough, you might even generate some income. Here's an example of the sort of stats that you'll get :- http://www.collective2.com/cgi-perl/system51583707

Good luck.
 
Quote from cwu8918:

ES: 07/20/2009 - 10/29/2010
Trades: 883
Net Profit: 53119
Max DD: 2068
Profit Factor: 2.03
Kelly: 27
PI: 7.72
More...

Can you explain what Kelly and PI stand for? What do they mean?
 
Quote from intradaybill:

Can you explain what Kelly and PI stand for? What do they mean?
More...

It looks like the OP used JBookTrader for backtesting, which is the open source software which I manage, so I'll answer this question.

In JBookTrader, "Kelly" stands for the Kelly Criterion and is calculated as follows:

Code: 
[b]kellyCriterion = 100 * (probabilityOfWin - (1 - probabilityOfWin) / winLossRatio)[/b]
where
winLossRatio = (grossProfit / profitableTrades) / (grossLoss / unprofitableTrades)
probabilityOfWin = profitableTrades / trades
trades is total number of trades

"PI" stands for the strategy Performance Index and is calculated as
Code: 
[b]PI = sqrt(trades) * averageProfitPerTrade / stdev[/b]
where
averageProfitPerTrade = netProfit / trades
stdev is the standard deviation of all trades
trades is total number of trades

I personally use PI almost exclusively, as I believe it's superior to all other performance metrics.
 
Quote from nonlinear5:

It looks like the OP used JBookTrader for backtesting, which is the open source software which I manage, so I'll answer this question.

In JBookTrader, "Kelly" stands for the Kelly Criterion and is calculated as follows:

Code: 
[b]kellyCriterion = 100 * (probabilityOfWin - (1 - probabilityOfWin) / winLossRatio)[/b]
where
winLossRatio = (grossProfit / profitableTrades) / (grossLoss / unprofitableTrades)
probabilityOfWin = profitableTrades / trades
trades is total number of trades

"PI" stands for the strategy Performance Index and is calculated as
Code: 
[b]PI = sqrt(trades) * averageProfitPerTrade / stdev[/b]
where
averageProfitPerTrade = netProfit / trades
stdev is the standard deviation of all trades
trades is total number of trades

I personally use PI almost exclusively, as I believe it's superior to all other performance metrics.
More...

Isn't PI too influenced by the number of trades? A system with 1000 trades will almost always have a higher PI than one with 100, even if it really isn't better.
 
Quote from logic_man:

Isn't PI too influenced by the number of trades? A system with 1000 trades will almost always have a higher PI than one with 100, even if it really isn't better.
More...

I'd like that.
 
Quote from logic_man:

Isn't PI too influenced by the number of trades? A system with 1000 trades will almost always have a higher PI than one with 100, even if it really isn't better.
More...

Yes, everything else being equal, the system with 1000 trades will have the PI about 3 times higher than the system with 100 trades: sqrt(1000) / sqrt(100) ~ 3

That's built in the PI to reflect that higher number of trades gives higher statistical significance to the system performance. I am not a statistician, so I can't say if that sqrt(trades) term is too dominant. It seems to replicate the "standard error", which is inversely proportional to the square root of the sample size: http://en.wikipedia.org/wiki/Standard_error_(statistics)

The PI formula closely resembles Van Tharp's System Quality Number (SQN): http://www.ninjatrader.com/support/forum/showthread.php?t=4320

I did see a reference somewhere about Van Tharp's recommendation to cap the sqrt(trades) term when the number of trades is too large.
 
Quote from nonlinear5:

"PI" stands for the strategy Performance Index and is calculated as
Code: 
[b]PI = sqrt(trades) * averageProfitPerTrade / stdev[/b]
where
averageProfitPerTrade = netProfit / trades
stdev is the standard deviation of all trades
trades is total number of trades

I personally use PI almost exclusively, as I believe it's superior to all other performance metrics.
More...

This PI is the inverse of the well-known coefficient of variation in Statistics, which is also called the Signal to Noise Ratio, and then multiplied by sqrt(trades).

This is a meaningless ratio for non-positive random variables, such as trade values.

http://en.wikipedia.org/wiki/Signal-to-noise_ratio#Alternative_definition

It is possibly some measure of the extent to which the distribution of trades overlaps zero. Really meaningless for trading, especially for systems that try to catch trends and make the most money, in which case this ratio will have a very small value.

Trash it...
 
Quote from intradaybill:

This PI is the inverse of the well-known coefficient of variation in Statistics, which is also called the Signal to Noise Ratio, and then multiplied by sqrt(trades).

More...

No, it's not the inverse of the SNR. The ratio of the mean to standard deviation is in both SNR and PI.
 
You must log in or register to reply here.
Back
Top