Wish List

Quote: Originally posted by uhrbi

What do you think of a feature to only enter a new trade when there is no position? (Marketposition = 0) Regards, uhrbi

GPU

Quote: Originally posted by dpetaz

Hello Peter, since GSB needs a lot of CPU power, did you think to increase the computational speed using also the GPU? According to some documents I read, this could improve performance 10x to 60x, for high end video card.

Quote: Originally posted by rws

Here is another aplication that was able to improve speed 100-1000X doing this and his open code in Lazarus was already very fast. This was only price patterns but it could work for any type of scripts especially because GA is a parallel calculation. http://mechanicalforex.com/kantu-system-generator Quote: Originally posted by dpetaz   Hello Peter, since GSB needs a lot of CPU power, did you think to increase the computational speed using also the GPU? According to some documents I read, this could improve performance 10x to 60x, for high end video card.

Quote: Originally posted by rws

I understand you have other priorities and rewriting software for supporting a GPU is a lot of work. Comparing OHLC data is even less math and it was 100-1000X faster. Especially in parallel processing which could be a great portion of Wf in GSB and GA a GPU shines. http://www.nvidia.com/object/what-is-gpu-computing.html Nvidia stock is up so much because their GPU chips are so efficient in machinelearning and parallel processing. Even VMware uses GPUs now and that is running a complete operating system. Google tensorflow runs very well on GPU. https://www.tensorflow.org/programmers_guide/using_gpu

Quote: Originally posted by Carl

GSB already has the possibililty to switch off/on the "secondary filter". I would like to have the possibility to switch off the "primary filter", to see what kind of secondary filter works best on a particular ticker and session.

Quote: Originally posted by Carl

GSB already has the possibililty to switch off/on the "secondary filter". I would like to have the possibility to switch off the "primary filter", to see what kind of secondary filter works best on a particular ticker and session.

Quote: Originally posted by cyrus68

The Sharpe ratio is already calculated in GSB and is reported in the performance stats. It would be extremely useful to have it available in the panel, alongside the other performance metrics, so that we could rank systems according to their Sharpe-ratio score. The Sharpe ratio calculates return, per unit of risk - where risk is defined as the standard deviation of returns. It is the best overall measure of the quality of a system. It would also be useful to have the Sharpe ratio reported in the WF results, for the OOS and Current curves.

Quote: Originally posted by cyrus68

The Sharpe ratio is already calculated in GSB and is reported in the performance stats. It would be extremely useful to have it available in the panel, alongside the other performance metrics, so that we could rank systems according to their Sharpe-ratio score. The Sharpe ratio calculates return, per unit of risk - where risk is defined as the standard deviation of returns. It is the best overall measure of the quality of a system. It would also be useful to have the Sharpe ratio reported in the WF results, for the OOS and Current curves.

Quote: Originally posted by rws

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2745220 Quote: Originally posted by cyrus68   The Sharpe ratio is already calculated in GSB and is reported in the performance stats. It would be extremely useful to have it available in the panel, alongside the other performance metrics, so that we could rank systems according to their Sharpe-ratio score. The Sharpe ratio calculates return, per unit of risk - where risk is defined as the standard deviation of returns. It is the best overall measure of the quality of a system. It would also be useful to have the Sharpe ratio reported in the WF results, for the OOS and Current curves.

Quote: Originally posted by zdenekt

It would be nice to have a "Win/Loss ratio" as fitness.

Quote: Originally posted by cyrus68

Peter: I like the idea of a user definable field. To be able to switch between % profitable and the Sharpe ratio would be ideal. Also, if it would be possible to have the Sharpe ratio displayed, after WF, for the OOS and Current curves. At the bottom of the parameters table perhaps. rws: I can't comment on the ssrn paper until I've read it.

Quote: Originally posted by jptann

Peter: I am using the GSB a lot for the ES markets and trading them with good results. One recommendation that I have to fix a limitation of the program is to add more sophistication in Stops and Profit targets. Just having a fixed number is no longer good enough. Maybe a suggestion would be to handle custom stop and profit target code like you handle custom indicator codes. This would allow me to add some sophisticated stops and profit targets (actually, never use profit targets) but would love to iterate Break even ratchet type stops, or others. There are a number of standard ones and you could look on the TS Forums for examples. Look for LeBeau_Stops and also scale in and scale out logics. If you allow for more flexibility here for stops, scaling in and out, profit targets, etc. it would take the package to the next level. A methodology similar to the "custom indicator" section would be excellent here. Look at _StopsFlexible function in TS by MarkSanDiego. It may give you some ideas. Thanks in advance

Quote: Originally posted by rws

Here a nice description of Lebeau http://stockcharts.com/school/doku.php?id=chart_school:techn... I have tested it on pricepatterns and sofar it is the best exit I found. Still I am trying to find better exits like other pricepatterns.

DStat exit

Quote: Originally posted by Petzy

DStat is a stop that is mainly used for daily swing trade strategies. It tends to give smooth equity curves together with high hit-rate strategies. (For example RSI2) I am not sure if it will be good for GSB and intraday strategies but I think it might be a good idea to try it. The strategy comes from E2 Trading in Sweden. (e2trading.com) The exit works like this. 1. Exit on first close that is profitable. (Current Close > Entry price) 2. Always Exit on close at bar 5. Even if the trade is not profitable /Peter

Quote: Originally posted by Petzy

DStat is a stop that is mainly used for daily swing trade strategies. It tends to give smooth equity curves together with high hit-rate strategies. (For example RSI2) I am not sure if it will be good for GSB and intraday strategies but I think it might be a good idea to try it. The strategy comes from E2 Trading in Sweden. (e2trading.com) The exit works like this. 1. Exit on first close that is profitable. (Current Close > Entry price) 2. Always Exit on close at bar 5. Even if the trade is not profitable /Peter

Quote: Originally posted by Petzy

Quote: Originally posted by Petzy   DStat is a stop that is mainly used for daily swing trade strategies. It tends to give smooth equity curves together with high hit-rate strategies. (For example RSI2) I am not sure if it will be good for GSB and intraday strategies but I think it might be a good idea to try it. The strategy comes from E2 Trading in Sweden. (e2trading.com) The exit works like this. 1. Exit on first close that is profitable. (Current Close > Entry price) 2. Always Exit on close at bar 5. Even if the trade is not profitable /Peter I realized that I can get a somewhat simular effect with a profit target and a time exit. I think the original DStat exit will be better though. I tested on @FDAX 30 minutes. Exchange time 1400 - 1700. After a very quick test it looks like it is a good kind of exit.

Quote: Originally posted by cyrus68

It would be extremely useful to enable the exporting of the WF trade report as a csv file. Of course, it is possible to get the report via TS. But it's quicker and more convenient to do it directly from GSB, when you want to test many systems.

Quote: Originally posted by cyrus68

Yes, I realise you can export to PA and then Excel. However, this is a 2-step procedure. And you are left with a lot of redundant files in PA that need to be cleaned up. The files you would want in PA are the few that have passed all tests and have the best potential for portfolio inclusion. I need to do many tests on multiple WF results before I get to that point. If there isn't a lot of programming involved, exporting directly from GSB to csv would be helpful.

Quote: Originally posted by emsjoflo

I would like to have Nth profitable exits added to GSB -- similar to the Dstat exit Petzy mentioned. In Mechanical Trading Systems, Beann has several systems that exit after 2 or 3 profitable bars (or days) or after 5 unprofitable. That has the advantage of scaling the stops to the volatility of the market.

Quote: Originally posted by emsjoflo

I really want it for nth profitable days but if bars are easier, I could make that work too. I haven't tried it in TS yet

Quote: Originally posted by boosted

I've been using Standalone up til now. With Petzy's help I got Manager and Workers up and running. I currently max Workers (95ish %) at 4 with i7 7700 4.2Ghz 65GB RAM Take my wish with a grain of salt since I just started using Workers, but it would be nice if there was either a background color (ie green) that highlights Unique Systems row so at a glance you know when all Workers are done tranferring their systems to Manager OR do something like this: Unique Systems: 4,500 of 6423 (so you can get an idea how many were created and how many left to be finished) or maybe 4500/6423 or Percentage done (ie 67%) Something simple but from my first use of Workers it would of been nice to know how much longer I had til I could do further work in Manager.

Nice that you are up and running.
The way I usually setup GSB Manager is that I have restarts=1000. That will make GSB run "forever". After a day or two I Pause the manager and go through the systems that GSB found. If I find enough systems that are good I transfer them to TradeStation to double check that they work out of sample. If I think I need more systems then I resume GSB Manager again for a day or two.
So in this scenario I don't expect the worker to ever "be done". I just run until I have enough good systems.

Just a side note. The way I verify if my systems are good is this:
1. I export data from Tradestation for as long time back as I can get. The end date is 2018-01-01. (Sometimes 2018-03-31).

2. I run GSB Manager with Dates 2015-12-31 - 2100-01-01. And then I chose Dates Mode to Not Tradable. (See picture)

3. When I verify I Pause manager and then I change all systems to "Tradable". Alla systems that are ok Out Of Sample I put in Favorite.

4. I change all systems that I put in Favorite to Dates Mode=All and then I run Walk Forward Optimazation on them

5. All systems that are stabgle in WFO I transfer to Tradestation so I can get the final OOS from Tradestation.

6. All systems that I might want to trade I do WFO in tradestation to find out good values and Stop Loss.

7. Then I hit Resume on GSB Manager and let GSB find more systems. Or if I am happy with the systems I change to another market to find other systems in other markets

Quote: Originally posted by bill897

I would think drawdown percentage instead of profit targets and stops would be more beneficial. You have to protect your bottom line. If I've made a certain profit and hits a drawdown, I would think that it would lock in your profit. Profit targets are hypothetical. Let your profits run, cut your losses. How do you do that? The EA is suppose to handle that based on your indicators. But, if I'm sitting at my Tradestation, I can see if I'm green 10 and all I have to do is hit red 10. Is there drawdown percentage in GSB?

Quote: Originally posted by Petzy

Quote: Originally posted by boosted   I've been using Standalone up til now. With Petzy's help I got Manager and Workers up and running. I currently max Workers (95ish %) at 4 with i7 7700 4.2Ghz 65GB RAM Take my wish with a grain of salt since I just started using Workers, but it would be nice if there was either a background color (ie green) that highlights Unique Systems row so at a glance you know when all Workers are done tranferring their systems to Manager OR do something like this: Unique Systems: 4,500 of 6423 (so you can get an idea how many were created and how many left to be finished) or maybe 4500/6423 or Percentage done (ie 67%) Something simple but from my first use of Workers it would of been nice to know how much longer I had til I could do further work in Manager. Nice that you are up and running. The way I usually setup GSB Manager is that I have restarts=1000. That will make GSB run "forever". After a day or two I Pause the manager and go through the systems that GSB found. If I find enough systems that are good I transfer them to TradeStation to double check that they work out of sample. If I think I need more systems then I resume GSB Manager again for a day or two. So in this scenario I don't expect the worker to ever "be done". I just run until I have enough good systems. Just a side note. The way I verify if my systems are good is this: 1. I export data from Tradestation for as long time back as I can get. The end date is 2018-01-01. (Sometimes 2018-03-31). 2. I run GSB Manager with Dates 2015-12-31 - 2100-01-01. And then I chose Dates Mode to Not Tradable. (See picture) 3. When I verify I Pause manager and then I change all systems to "Tradable". Alla systems that are ok Out Of Sample I put in Favorite. 4. I change all systems that I put in Favorite to Dates Mode=All and then I run Walk Forward Optimazation on them 5. All systems that are stabgle in WFO I transfer to Tradestation so I can get the final OOS from Tradestation. 6. All systems that I might want to trade I do WFO in tradestation to find out good values and Stop Loss. 7. Then I hit Resume on GSB Manager and let GSB find more systems. Or if I am happy with the systems I change to another market to find other systems in other markets

Quote: Originally posted by boosted

Quote: Originally posted by Petzy   Quote: Originally posted by boosted   I've been using Standalone up til now. With Petzy's help I got Manager and Workers up and running. I currently max Workers (95ish %) at 4 with i7 7700 4.2Ghz 65GB RAM Take my wish with a grain of salt since I just started using Workers, but it would be nice if there was either a background color (ie green) that highlights Unique Systems row so at a glance you know when all Workers are done tranferring their systems to Manager OR do something like this: Unique Systems: 4,500 of 6423 (so you can get an idea how many were created and how many left to be finished) or maybe 4500/6423 or Percentage done (ie 67%) Something simple but from my first use of Workers it would of been nice to know how much longer I had til I could do further work in Manager. Nice that you are up and running. The way I usually setup GSB Manager is that I have restarts=1000. That will make GSB run "forever". After a day or two I Pause the manager and go through the systems that GSB found. If I find enough systems that are good I transfer them to TradeStation to double check that they work out of sample. If I think I need more systems then I resume GSB Manager again for a day or two. So in this scenario I don't expect the worker to ever "be done". I just run until I have enough good systems. Just a side note. The way I verify if my systems are good is this: 1. I export data from Tradestation for as long time back as I can get. The end date is 2018-01-01. (Sometimes 2018-03-31). 2. I run GSB Manager with Dates 2015-12-31 - 2100-01-01. And then I chose Dates Mode to Not Tradable. (See picture) 3. When I verify I Pause manager and then I change all systems to "Tradable". Alla systems that are ok Out Of Sample I put in Favorite. 4. I change all systems that I put in Favorite to Dates Mode=All and then I run Walk Forward Optimazation on them 5. All systems that are stabgle in WFO I transfer to Tradestation so I can get the final OOS from Tradestation. 6. All systems that I might want to trade I do WFO in tradestation to find out good values and Stop Loss. 7. Then I hit Resume on GSB Manager and let GSB find more systems. Or if I am happy with the systems I change to another market to find other systems in other markets Thanks for sharing your process. I did have a few questions though. You mention "After a day or two I Pause the manager and go through the systems that GSB found. If I find enough systems that are good I transfer them to TradeStation to double check that they work out of sample." Are you doing this after your steps 1-4 or prior? I would of figured you would of had GSB do the OOS which would mean no need to put into TS for OOS check, unless you wanted to confirm the GSB OOS results are truly what they should be and in fact work in TS too, as they should without error. Regarding your step #2 it seems you would have GSB take care of that automatically by setting your validation period to closely match your dates noted therefore not needing to take the extra step(s) of placing your Favorites list of strategies into TS, since the results can be seen immediately in the GSB OOS portion. Maybe I am not understanding the reasoning behind that step. Please elaborate. What is the purpose of doing a final WFO in TS if GSB has already completed a WFA on your list of stable strategies that passed GSB WF process? That would seem to be a very long time consuming process unless you instead use EWFO. I like your thorough process but trying to understand the reasoning behind your steps since some of them seem to be something GSB takes care of internally and quicker with EWFO vs TS WFO. Again, thanks for sharing your process like to understand some of the thought process behind it. Thanks.

I do step 1 and 2 first. That is the starting point you might say.
after step 2 I let gsb run for a couple of days and then I go to step 3.

The reason I do yet another OOS in step 5 in TS is because it makes me feel even more comfortable with the strategies. If you noticed I didn't export data all the way up until todays date in step 1. That way I have a true OOS test in TS with completly new data.
I have never had a problem that GSB don't show the same result in GSB as in TS. That have never been an issue for me.

Regarding the question about using Validation to check OOS automatically instead of using the Tradable/Not tradable function. -There was an issue some time ago that GSB might overfit strategies if using the Validation function. (That is probably true about all programs that automatically finds strategies). I think it was most obvious in Crude Oil strategies(?). Using the method in step 3 is better.
(If you haven't yet. Read this thread: http://trademaid.info/forum/viewthread.php?tid=117

The reason I do a final WFO in TS is because I use EWFO. The WFO in TS produce the data you later import into EWFO. I don't do many WFO in TS. Just the systems I am considering trading so it is a limited amount.
WFO in both GSB and TS have a fitness goal of NetProfit. In EWFO I can experiment with different settings to get a more straight equity curve and less drawdown which I like. GSB WFO is very good so you don't really need to do this step. But since I can, I do. And it's fun

Quote: Originally posted by Petzy

One thing to add to the wish list Some markets works better with Short trades only and some with Long only. And some with both long and short. When validating a market I would like to have the Performance tab show long, short and both in three different columns. It would be a fast way to see if the market is very bias to one sort of trade or not.

Sorry for the delay on this post
You can chose dates, and then choose weather to trade or not trade them.

then you can build systems. GSB will only look at the dates shown.
Then you can right click and change dates to non trade-able.
Then all the dates out of sample will be seen.
I'm lost on your spearman comment sorry. Spearman applies to walk forward only and is not applicable here. Spearman is only for comparison of WF data over the same dates, but with the parameters varying.

Quote: Originally posted by ProbTrader

Thanks for your response Peter. This is a good workaround, but it forces you to use the entire data set which you may not want to do. This also means that you don't have full control over start and end dates, which, for example, is relevant for seasonal commodities and scenario based trading. An example of this would be NG. Winter and summer contracts are different beasts, and should in most cases be treated as such (ie contract based and not as rollling nearbies - this is a different subject of course). So if you are trying to construct a "NG winter system", you'd want the IS and OS to cover a specific date range for each year, not the entire calendar year. The ability to specifically control IS and OS dates would allow this. (Never mind the Spearman comment. It was there to show that it is useful to see ONE acccumulated number for ALL periods.)

Quote: Originally posted by ProbTrader

WF: yep - understand how this works. OOS: In that case I must have misunderstood how the Trd/NoTrd logic works and how to control it. Let me take an example to show why I am confused. Lets say I want to accomplish the following (never mind the reason, this is just an example): - IS = 2001 and 2003 ONLY - OOS = 2008 and 2010 ONLY How would I accomplish this specifically? If I for example set the dates to 1/1/2001-12/31/2001 and 1/1/2003-12/31/2003 to "Trd", I will only see data for 2001 and 2003. So far so good. But what's next? If I change the dates to "NoTrd" ALL other dates will be included in the output/equity curve, ie ALL dates for which there is data, except 2001 and 2003. But I ONLY want to use 2008 and 2010 as OOS, and I want the stats to be the accumulated stats for these two years ONLY. I could obvioulsy save the systems and do a complete workaround where I re-run everything for a new set of dates with the systems that I choose. But my point is that I want to be able to RUN the IS and OOS on specific dates without having to run sims twice and manually change this. I think the NG example I mentioned above also illustrates this. But perhaps I have completely misunderstood something here... Thanks again.

Quote: Originally posted by ProbTrader

I am not so sure that testing systems for certain periods only (seasonality of contracts), and/or stress testing it for specific historical periods/events is that unusual (I happen to know and have worked with many traders that do just that). As I was alluding to above, I believe that this approach is preferred if you are developing systems on seasonal products, especially those where the contract specs de facto differ between different months of delivery, I think anyone will miss out by treating all contracts the same way, essentially as one continuous data series. Since they are different in their specs, have different fundamentals etc, they also have different price behaviour/distributions/volatility etc. I guess one analogy I would make is with gold/silver. You may find a system that works for both - great if you do as it is likely robust then - but realistically you are probably better off with two different systems. Another analogy would be to treat all hours of trading the same for something like ES, which you concluded is a bad idea in your session time video (lthanks for putting together that video btw - loots of good takeaways in it). Also, for stress testing any system I find this to be a very practical approach (eg if a long only system survives the last 20 big drops OOS, it may be a winner). I realize that not all system developers think in these terms, but please do keep it under consideration during the develpment of the optimization options.. Again, I find it superior to the current % settings for IS/OOS/Verification, and it is an addition I feel others would find useful as well. (Also, I do understand that you can hack a CSV file, but there are obviously lots of issues with this. Not only is this inconvenent, but you would have to transform the data as you end up with price gaps etc etc.) Many thanks.

Quote: Originally posted by ProbTrader

I am not so sure that testing systems for certain periods only (seasonality of contracts), and/or stress testing it for specific historical periods/events is that unusual (I happen to know and have worked with many traders that do just that). As I was alluding to above, I believe that this approach is preferred if you are developing systems on seasonal products, especially those where the contract specs de facto differ between different months of delivery, I think anyone will miss out by treating all contracts the same way, essentially as one continuous data series. Since they are different in their specs, have different fundamentals etc, they also have different price behaviour/distributions/volatility etc. I guess one analogy I would make is with gold/silver. You may find a system that works for both - great if you do as it is likely robust then - but realistically you are probably better off with two different systems. Another analogy would be to treat all hours of trading the same for something like ES, which you concluded is a bad idea in your session time video (lthanks for putting together that video btw - loots of good takeaways in it). Also, for stress testing any system I find this to be a very practical approach (eg if a long only system survives the last 20 big drops OOS, it may be a winner). I realize that not all system developers think in these terms, but please do keep it under consideration during the develpment of the optimization options.. Again, I find it superior to the current % settings for IS/OOS/Verification, and it is an addition I feel others would find useful as well. (Also, I do understand that you can hack a CSV file, but there are obviously lots of issues with this. Not only is this inconvenent, but you would have to transform the data as you end up with price gaps etc etc.) Many thanks.

Quote: Originally posted by saycem

When sorting results I would like to be able to shift select multiple column headings so for eg you could sort by avg trade and then # of trades etc. Similar to excel.

Quote: Originally posted by saycem

When sorting results I would like to be able to shift select multiple column headings so for eg you could sort by avg trade and then # of trades etc. Similar to excel.

Quote: Originally posted by edgetrader

I was studying Rob Carver and one of his themes is to separate some of the fitting/optimization from the PnL producing trades. In GSB one way to do this would be to target certain forecasting windows, e.g. 20 bars ahead or until the close of the day. Then the GSB output would be optimized with something like RMSE against the correct target values taken from "future" price data. After this type of fitting, higher GSB output values would correspond to higher predicted price moves over the next 20 bars. Another layer would then take over and translate the forecasts into entries, the simplest being if output > X go long. One benefit would be that most of the fitting can be done on all in-sample bars even if the resulting system only trades very sparsely in order to produce high avg trade.

Quote: Originally posted by admin

This is complex and I dont understand the finer detail. I also dont know what rmse is. GSB does have separate trades via the PNL using nth and date features. I think that is a great concept.

Here's an example. The blue line is standard TS moving average with length 20 displace 10. This means it peaks 10 bars into the "future" to draw the perfect line.

You'd give the 1bar momentum of that blue line to GSB as targets. If the line was 2808 a bar ago and now it's 2810, your momentum is +2. You'd want the GSB output on this bar to be +2. If it's 0.5 you'd have an error of 1.5. RMSE is just a fitness function, you could use others.
https://en.wikipedia.org/wiki/Root-mean-square_deviation

Then once you're done with this, another layer would take the GSB output and go long if it's above +3, flat if below +1 (for example).

Quote: Originally posted by edgetrader

When GSB can build its own bars from 1min source data, a great feature would be to support MC custom resolutions. For example, add custom range bars that calc the 20bar ATR on daily data and form a new bar if its high-low range, or open to close range as a variant, exceeds 5% of ATR. Force bar completion at user specified session end e.g. 1500. Always evaluate at full minute ends and don't cut minutes into half or add prices that weren't traded. TS also has advanced bar types that in my testing however couldn't be used for systems as they all added phantom prices or peeked into the 'future' to some extent. With MC you have control over the COM object's source code that's your custom bar plugin, so you could make advanced bars that are fully backtestable. https://www.multicharts.com/trading-software/index.php/Custo...

Advanced Bar Types

Quote: Originally posted by Gregorian

I've been working exclusively with Advanced Bar Types in MC and TS since just after GSB debuted. Here is a summary of what I've learned: In TS, Kase bars are the only type that do not create phantom bars. TS' implementation of range and Renko bars always generate phantom bars. Point-and-figure and Kagi bars are by definition completely unusable. Stick to Kase bars and you'll be OK. In MC, Point Original bars (look them up - it's a hidden bar type, only selectable via the Command Line), Flex Renko (when configured not to generate phantom bars), and the brand-new Kase bars in v12 are usable. I've had inconsistent results with Point bars, MC's version of range bars, therefore I do not recommend them. So far I am having the best results with MC's Kase bars, for which GSB generates more strategies than it does for TS' Kase bars. Clearly MC's implementation of Kase bars differs from TS', though I don't know how. I've never experimented with developing Custom Bar Types. That would be a whole new world. Note that with any of these, you must take into account gaps that can occur between bars. For example, there can be a 1 tick difference between the close of one bar and the open of the next, so you should add at least 1 tick of slippage in the Report Slippage to get a fair idea of what to expect in the real world. I'm very grateful that GSB uses "on close" orders in TS/MC, as that is the only order type that works reliably with Advanced Bar Types. Competitive products use "next bar at market", which makes some bar types unusable. The advantage of Advanced Bar Types is that you can focus on price action better and thereby generate more profitable strategies. The disadvantage is that there are so many more bars, that you cannot viably backtest for as long a duration. I believe the tradeoff is worth it. Others hold different opinions.

Quote: Originally posted by JasonT

Advanced bars are based on tick data? Is that right? I'm not sure about TS and MT but in Ninjatrader that is the case. Can you go back in time far enough to do suitable IS/OOS backtest and then anchored walk forward?

Quote: Originally posted by admin

I'm going to talk the bar types over with the programmer when we come to rebuild the data manager. I would like feedback if any other GSB users want the feature.

Quote: Originally posted by JasonT

Quote: Originally posted by admin   I'm going to talk the bar types over with the programmer when we come to rebuild the data manager. I would like feedback if any other GSB users want the feature. Depending on your data provider you get 360 days of tick data with Ninjatrader. You can buy up to about ten years of historical tick data for some instruments for ninja with some online providers. Despite the large number of bars that you can get in a 6 month or 360 day period, for me that is insufficient time (years) to capture enough different types of market conditions and prove system resiliency. I've produced (without GSB) a lot of different volume bar based strategies for ninjatrader, and have only been able to back test over 360 days. None of them passed by incubation testing. Perhaps the strategies were bad, or perhaps there was not enough data to test them adequately. Maybe both. Further I note that you need to be really careful about time stamping of each tick of the tick data - ie the data provider must use it and you've got to ensure that your historical tick data is time stamped in the same way that your live tick data is. Otherwise your back testing will be inaccurate. I've seen volume charts on two different computers (one on a VM in the US and one on a PC in Australia) and, for the same instrument, the volume bars build differently. This was due to the different sequence of arrival of the ticks due to IP routing and long trip times. Refreshing the charts fixed the problem, but this is something that needs to be considered for unsupervised auto trading. For those reasons my preference currently is to use minute based bars and many years worth of data for in sample and out of sample testing.

Thanks for the interest and comments on this idea. With Kase bars in TS, you can set them to a target range of 2 points and have them built from 1 minute data. The only issue I found in my testing is that TS peeks one bar into the future to construct the bars. For example, your current bar has a range of 1.75. The next minute bar that comes in has a range of 2.50. TS uses future knowledge to complete the current bar at 1.75 and build the next bar with 2.50.

The correct thing to do is leave the current bar open at 1.75 and wait for the next minute to complete, then make a bar with 4.25 range. The occasionally larger bars like this one with 4.25 range are what you have to accept when correctly building advanced bars out of 1min data. The big advantage will be that most GSB users have access to up to 20 years of 1min data and systems can be live traded with MC custom resolutions.

Quote: Originally posted by edgetrader

Thanks for the interest and comments on this idea. With Kase bars in TS, you can set them to a target range of 2 points and have them built from 1 minute data. The only issue I found in my testing is that TS peeks one bar into the future to construct the bars. For example, your current bar has a range of 1.75. The next minute bar that comes in has a range of 2.50. TS uses future knowledge to complete the current bar at 1.75 and build the next bar with 2.50. The correct thing to do is leave the current bar open at 1.75 and wait for the next minute to complete, then make a bar with 4.25 range. The occasionally larger bars like this one with 4.25 range are what you have to accept when correctly building advanced bars out of 1min data. The big advantage will be that most GSB users have access to up to 20 years of 1min data and systems can be live traded with MC custom resolutions.

Quote: Originally posted by edgetrader

Yes that's right, you can't trade it in TS unless you do a massive workaround that involves recreating the custom bars on user controlled arrays and recoding all indicators to work on these arrays. So the idea is to have it in GSB and let the users try it. If they find it worthwhile and want to trade custom bar strategies, they can do it on MC or another platform that allows to program the bar formation, such as NinjaTrader. I'm not thrilled about TS10 and believe MC .Net (for various reasons) is the future anyway. So what I suggest is that GSB provides MC custom resolution plugins to match GSB's custom bars. We can then live trade the strategies on MC .Net and EL edition.

Quote: Originally posted by tmlong18

It would be great to add matrix optimization for EWFO.

Quote: Originally posted by admin

Quote: Originally posted by tmlong18   It would be great to add matrix optimization for EWFO. Can you explain? You mean the table of results of runs vs periods etc than can out in the last build of TS wfo.exe years ago. I dont see the need for it as if you have hit parameter stability, there it will make no difference.

Quote: Originally posted by tmlong18

Quote: Originally posted by admin   Quote: Originally posted by tmlong18   It would be great to add matrix optimization for EWFO. Can you explain? You mean the table of results of runs vs periods etc than can out in the last build of TS wfo.exe years ago. I dont see the need for it as if you have hit parameter stability, there it will make no difference. I mean the matrix table of different fitness criteria weight. Sometimes it get a great result with NP(1)+ Avg Trade(1) but mess up with NP(1)+ Avg Trade(2) / NP(2)+ Avg Trade(1). A matrix table may easily evaluate the stability.

Have you seen this? Its fully programmable as to what fitness's you load.

Quote: Originally posted by engtraderfx

Couple of suggestions.. Take Profit / Stop optimising - following the guide to optimise using no stop loss / TP, would it be useful to run a chosen system after setting SL / TP & also rerun an optimisation & WF for SL / TP while in GSB ? Long / short chart - might be useful to give option to trend long vs short?

Quote: Originally posted by engtraderfx

You mean have stop / pt values genetically modified and adjust in WF? [At a basic level, just wondering if can enter TP / SL values in parameters after selecting a system & rerun in GSB, & then yes, run an optimisation or WF to determine appropriate values. Would be useful to use GSB than have to rereun in TS / MC]. You mean performance graphs have a optional line for long, short and both? [Yes just show a equity curve of short vs long as quick way to see comparison]

Quote: Originally posted by zdenekt

I wish the fitness function "Avg win / Avg loss ratio". Is it possible?

Quote: Originally posted by zdenekt

I wish the fitness function "Avg win / Avg loss ratio". Is it possible?

Quote: Originally posted by adcardoso01

Peter, I got a beginners question: As per Carl post, the development process is as follows: 1. run GSB 2. select good looking strat in GSB and perform WF in GSB 3. export code to TS 4. WF opt in TS 5. WF analysis in EWFO Does that mean that the EWFO has much more capabilities than the Built in WF in GSB? In summary, when buying GSB one should also consider buying EWFO in addition or when buying GSB you get all the functionalities of EWFO? Thx!

Quote: Originally posted by cyrus68

Would it be possible to enable GSB to enter trades at the open of the bar instead of the close? i.e. the user will be able to select whether trades are executed at the open or the close of the bar. This will enable swing trading daily bars for stocks, without facing problematic execution issues at the brokerage. Thanks

Quote: Originally posted by admin

Quote: Originally posted by cyrus68   Would it be possible to enable GSB to enter trades at the open of the bar instead of the close? i.e. the user will be able to select whether trades are executed at the open or the close of the bar. This will enable swing trading daily bars for stocks, without facing problematic execution issues at the brokerage. Thanks is this in the context of daily bars, or say 30 min bars you want to enter say at 830 instead of 900 on ES? I think the concept is ok, but will come after the additional secondary filters. (work starting fairly soon)

Quote: Originally posted by adcardoso01

Peter, regarding the WF within GSB, is there a way to do the following? 1) Choose to do Anchored or Rolling WF and select the Rolling period; 2) Optimize WF periods, ie, Optimize 1yr with OOS of 2mo, then Optimize 1.5 yrs with OOS of 2mo and choose which works best for example? Here I'm a little skeptical due to the excessive risk of curve fitting, but I'd like to hear your thoughts on it. What is your take on it? Thx!

Quote: Originally posted by adcardoso01

My point here is: What PERIODS (IS and OOS) result in best parameters when doing WF analysis. After that, the question that follow is: Are these parameters stable overtime? Im asking you because a thought that occurred to me was: Can we improve parameters stability somehow? Is it possible to find out what periods of WF result in more parameter stability? If we can run WF tests using steps of 1 month (for example: IS: 1yr / OOS 1mo, then IS: 0.9yr / OOS 1mo, then IS: 0.8yr / OOS 1mo, etc) and compare the parameter stability of those. That possibly could give insights of what is the optimum period (in terms of parameter stability) to re-optimize a Model... Just food for thought...