Have you tried Build Alpha

[srinir]

Anyone can torture data long enough and come up with examples that beat metrics, any metrics, anytime. This is statistics 101.

If you are trading based on past performances and patterns, you are torturing the data. You may be better of buying and holding.

Even that based on past performances, but atleast it is over very long period. So it may be stable.

 

[srinir]

Mark Brown, you have very strong words for someone who doesnt have a license. I'll leave it at that as I imagine you'll be very confrontational and probably enjoy bashing other people's hard work and efforts for no reason. Not sure what I did to you but hey the internet you can do what you want, I guess.

No one takes him seriously. After 20+ years he is still deciding whether tradestation or multicharts is better

 

[sle]

If you are trading based on past performances and patterns, you are torturing the data.

In fairness, there is nothing wrong with a bit of data dredging if you are correcting your statistical significance properly (Bonferroni correction is the easiest one, albeit rather conservative).

 

[srinir]

In fairness, there is nothing wrong with a bit of data dredging if you are correcting your statistical significance properly (Bonferroni correction is the easiest one, albeit rather conservative).

That was exactly my point.

He selling software for backtest, not a blackbox or some snake oil service (Hello Kim). It is upto user to use parameters that seems to fit for their purposes.

In fairness his software has lot of means to test for robustness of the system. He also uses variance testing and other means and also tests against random noise etc. You can also do out of sample testing at the end or beginning of the data, roll the testing period etc. One can bring their own data, so he is not selling some fitted system.

poo pooing as statistics 101, without knowing what exactly his software does is little too far.

 

[sle]

That was exactly my point.

We are in a violent agreement then This said, I have no opinion on the actual software

 

[bogitrade]

In fairness his software has lot of means to test for robustness of the system. He also uses variance testing and other means and also tests against random noise etc. You can also do out of sample testing at the end or beginning of the data, roll the testing period etc. One can bring their own data, so he is not selling some fitted system..

You are not up to speed with data mining bias. Those are tests valid for unique models. Using those tests in multiple testing makes them useless. Software has many hard wired rules and tries to find the best combination for given time-series. This is data mining and under condition of repetition it generates spurious correlations that can pass every conceivable test. I looked at the site and the examples appear post selections to justify the validation metrics. But the main issue here is demo availability. Except if you suggest buying this first and then figuring out if it is good for your purposes. Is that what you are suggesting?

 

[srinir]

But the main issue here is demo availability. Except if you suggest buying this first and then figuring out if it is good for your purposes. Is that what you are suggesting?

You need to get upto speed on your reading comprehension. Where did i suggest anything or made comment about demo availability?

Providing demo is vendor discretion. He has not come here to push his product. If customer does not want buy without trying, then they can walk.

Regarding data mining, Trader who trades on 200 dma is based on data mining, Investor who invests on 60:40 portfolio is data mining. If you are uncomfortable with those parameters then there is always Treasuries.

Regarding spurious correlations, Software already tests and plots graph on 3 separated data sets for the same signal. You can always export the results and further testing. Even the software provides Python coding to do further test on different time series. You can apply same signals on multiple symbols in the software itself for robustness check.

 

[sle]

You are not up to speed with data mining bias. Those are tests valid for unique models. Using those tests in multiple testing makes them useless.

As I said above, data mining is not bad per se as long as you apply the right corrections for family-wise error.

Regarding data mining, Trader who trades on 200 dma is based on data mining, Investor who invests on 60:40 portfolio is data mining.

To be fair, these were probably created from unique hypotheses and optimized for specific parameters after. That's not really data mining in the form that requires any forms of correction.

 

[Sergio77]

StrategtQuant goes for $990 and the developer in Czech Republic offers a demo and all extra help.

Price Action Lab (DLPAL) offer a free trial except code as I remember. UI is little outdated but results are usually good.

Same with Adaptrade the developer offers demo identical to full version. Price is $995.

These are some of well-known and tested programs from experienced developers that have the time to provide demos and deal with all support issues. They are very responsive and helpful all of them. I have all three softwares above and each is useful in its own way. I could use Buildalpha but only if I could test for at least 14 days before buying on my own with the help of my expert programmer.

 

[srinir]

Here is another one. It costs only $60k

http://www.tradingsystemlab.com/howtoorder.aspx

No free trial