Workflow plan

19 replies

munchie

Customer, bbp_participant, community, 63 replies.

Visit profile

7 years ago #115415

Hi guys,

I’ve been using SQ for a little while now and would like to sound out advice from other sq users as to how you plan your workflow?

Seems that the generation part of sq is pretty straight forward but the testing, retesting, WFM are a bit more fiddly? How I presently do it is to generate strategies, send to retest on M1, delete those that do not achieve my minimum filtering values, save to folder, do next test, save to next folder etc etc. I have read the recent ebook which had helped bit does anyone do things differently to the above? As testing takes find and can sometimes take few days, it becomes difficult to track/remember where you are at when yo come back the next day etc etc. Would welcome a discussion as to what others do to organise the workflow and it’s many steps.

Sent from my iPhone using Tapatalk

0

Karish

Customer, bbp_participant, community, sq-ultimate, 444 replies.

Visit profile

7 years ago #138626

Right now i quit use SQ3, waiting for SQ4,

 

my workflow was:

#1 Random generation on M1/Tick (depends if searching for scalping or swing strategies),

[DATES IF M1: (IS)=Asirikuy From the beginning Till 2003 + OOS From 2003~Today]

[DATES IF TICK: (IS)=Ducascopy Tick From 2003 Till 2010 + OOS From 2010/2012~Today]

#2 Retest again on M1 + MonteCarlo, [SAME IS&OOS]

#3 Filter,

#4 Retest again on Tick + MonteCarlo, [SAME IS&OOS]

#5 Filter,

#6 personally i dont use WFM, i just run more then 2000 MonteCarlo with huge randomization of the data theoretically gives me the same thing (thats just my opinion i trust the MonteCarlo more),

     and i filter those strategies that are do not perform well after those MonteCarlo tests,

#7 Then i try the strategies that have passed the tests until now inside the Improver with M1 with different pair this time and try to get different results with slightly different parameters that i let it to randomize and i lock some parameters if needed, and then i will retest again on M1 + MonteCarlo,

#8 then retest again + MonteCarlo on Tick, [SAME IS&OOS]

#9 if some strategy passed the same testes on 2~3 pairs you got a robust strategy,

#10 forward test it immediately on demo account, after sometime restest it again with new data on SQ with the same settings, check if it fits the demo forward test,

 

[IF YES=Great open small live account and keep on going]

[IF NOT=Might wanna check your broker OR strategy’s settings might be the time that is differs or daily open/close etc.. Try to find the problem.]

 

i will recommend as always to watch Threshold’s videos about how to NOT over optimize the strategies and etc..

i will recommend as always the M1 data from Asirikuy also

 

*

I DO NOT use RealTicks testing, i prefer FIXED spread.

I DO NOT use WFA/WFM, takes time.., i rather prefer using 1000 MonteCarlo couple of times to randomize the results of it, 2 times minimum i can reach up to 10 times (10,000 MonteCarlo).

I DO NOT use OPTIMIZATION tab EVER!, you will break the strategy!, i rather use IMPROVER tab.

i DO NOT use OpenPrices when testing, i use SelectedTF.

i DO NOT use Time depended indicators (Pivots/Daily OHLC/Hour/Minute/Day) better that way, do not use any time indicators/parameters.

ALWAYS ALWAYS ALWAYS retest your strategy from SQ3 in MT4 backtester with the same settings/parameters/spread/data, BEFORE going realtime DEMO/LIVE, SQ3 can give different results then MT4, and after all you will be trading thru MT4 so you MUST retest and see if it gives you the same results!, MT4 backtester results are the results you must pay attention to and not SQ3 in case of different results in between them! (hopefully the new SQ4 backtester will fix some or all of the bugs in this part).

0

Heilpraktiker

Customer, bbp_participant, community, 46 replies.

Visit profile

7 years ago #138637

Hi guys,

I’ve been using SQ for a little while now and would like to sound out advice from other sq users as to how you plan your workflow?

Seems that the generation part of sq is pretty straight forward but the testing, retesting, WFM are a bit more fiddly? How I presently do it is to generate strategies, send to retest on M1, delete those that do not achieve my minimum filtering values, save to folder, do next test, save to next folder etc etc. I have read the recent ebook which had helped bit does anyone do things differently to the above? As testing takes find and can sometimes take few days, it becomes difficult to track/remember where you are at when yo come back the next day etc etc. Would welcome a discussion as to what others do to organise the workflow and it’s many steps.

Sent from my iPhone using Tapatalk

Workflow:

Generating H1 strategies of EURUSD, GBPUSD, USDJPY with full asirikuy M1 dates over 30 years

Ranking for databank: >2000 trades, Net Profit IS per 0.01 >1000, Net Profit OOS > 300

After a few days, if databank has reached 2000 trades i filter out and delete all strategies, which have:

Stability <90%

Stagnation >1000 days

Return/Drawdown-Ratio <30

Complexity >21

these criterias i make harder and harder until 2-3 strategies rest. I am looking to the code, whether strategies are sounding and logical for trading.

Then i improve first entry rules, then order types, then exit rules with all the 30 years data. Values should get better, but complexity should not get higher, or i dont take the improvement.

After the improvement i do the simple optimization. I am looking for a mix of good stability (>93%), stagnation (30) .

Then i put the optimized strategy in the WalkForwardMatrix and let it run. It needs a few hours with all the data, RAM is going over 15GB.

If the strategy passed the WFM, i do the WalkForwardOptimization with the time period generated of WFM.

If the strategy passed the WFO, i incubate in real account with microlots and do the MonteCarloSimulations (200) for risk estimation.

What do you think about it? 🙂

0

_Cujo

Customer, bbp_participant, community, 102 replies.

Visit profile

7 years ago #138649

Here’s my current workflow. It looks janky here, but it makes more sense in the spreadsheet. I had to start writing them down, as the dates I use for certain things, and what I’m testing for (separate spreadsheet), got impossible to remember.

 

5WZFXRM.png

 

It’s my 7th WF (hence name, far left “WF7”, which is also one part of the strategy name string when saving, etc.. so I know the time frames it was generated, tested, etc..etc..).

 

The dates are all calculated off formula, so it’s live dates. The red dates are the “anchor” dates, the others calculate off them, but I’m typically working in blocks of months.

 

Each stage has criteria it has to pass. If it passes, I just bump it along to the right until it’s incubating on the platform, then live. Criteria stored elsewhere.

 

I plan dates ahead (for example here goes to November 2016), as I split the dates of back testing and forward testing, as I’m not patient enough to wait for it to live testing for months, or whatever.

 

When it’s live on the platform, I assess it it every 10 trades (sometimes that takes days or weeks, sometimes it’s longer, in some cases, haven’t gotten there yet…). If it’s profitable after 10 trades, I keep it, check at 20, if it’s profitable at 20 keep it, check it at 30, at which time I put it live, or ditch it.

 

You see I have very simple criteria (“am I happy if it’s running now”). In reality, I have strictly quantified metrics I’m looking at each stage, which even vary inside the color coding below, for example, I’m looking for slightly different things inside retest 1,2 and 3, but they have a common theme. But THAT spreadsheet is a total snarling mess. 🙂

 

Anyway, yeah, took 7 workflows to get here. 🙂

 

If you’re surprised at the short time frame to generate, keep in mind, I’m predominantly trading futures, and eminis specifically, so things like “30 years of Asikury” data are not available, since eminis started in the late 90s. I test as far back as possible, but keep some data, “just in case”. Also, the eminis started at different times, and I don’t want a work flow per symbol, so I take a WF across the asset class.

 

Oh, yeah, I keep commentary for weeks and 1/2 weeks since SQ only lets you specify # of trades, not for example 1 trade per week (*hint hint*).

0

munchie

Customer, bbp_participant, community, 63 replies.

Visit profile

7 years ago #138651

my last post got deleted! thanks for the posts and details on your workflows, how are you both finding it? 

i am on my third workflow now, and it looks like this

 

60396dca84bce4e7a47ee5f5089036c9.png

 

fc93d8721564d7d51c5066e9b2c7b75c.png

 

based on the ebook approach and the first one is a filter SS taking the initial generated strategies through the filter hoops until it is at the end stage (far right) ready for incubation.

 

what do you guys think? would be great to hear others approach to this topic as its well confusing!

0

_Cujo

Customer, bbp_participant, community, 102 replies.

Visit profile

7 years ago #138653

my last post got deleted! thanks for the posts and details on your workflows, how are you both finding it? 

i am on my third workflow now, and it looks like this

 

60396dca84bce4e7a47ee5f5089036c9.png

 

fc93d8721564d7d51c5066e9b2c7b75c.png

 

based on the ebook approach and the first one is a filter SS taking the initial generated strategies through the filter hoops until it is at the end stage (far right) ready for incubation.

 

what do you guys think? would be great to hear others approach to this topic as its well confusing!

 

Looks very workable, 2 pieces of advice, more process things, than specific:

1) Put your data  dates in there. Great that this is your 3rd WF, but make sure yo comment your dates, so you know in future and for past what date you tested over, etc. I made my dates formulas, so it’s always live ( I anchor on things like end of months, etc. not today), that way, I can know from any strategy, what data it was generated on, what date range  it’s robustness tests where over, etc.

 

2) Make sure you systematise what you’re looking for, so write down the metrics to pass each step (you probably have this already, I keep it in a different file just for neatness). For example, if you delete any strategies that have ret/dd less than 2.00, or whatever whatever….that way, when you tweak things (say this workflow you look for profit factor, next work flow you look for profit, or again whatever, just example), you can trace back when you find strategies that are working well in real time, and identify attributes (be it specific metrics you looked for, or whatever) that worked in the past, then you can data mine heavily around that.

 

But, I like your workflow layout, mine is hard to show on a forum, but works ok, when scrolling across one of my monitors.

0

Gui

Customer, bbp_participant, community, 59 replies.

Visit profile

7 years ago #138655

Hi guys,

I’ve been using SQ for a little while now and would like to sound out advice from other sq users as to how you plan your workflow?

Seems that the generation part of sq is pretty straight forward but the testing, retesting, WFM are a bit more fiddly? How I presently do it is to generate strategies, send to retest on M1, delete those that do not achieve my minimum filtering values, save to folder, do next test, save to next folder etc etc. I have read the recent ebook which had helped bit does anyone do things differently to the above? As testing takes find and can sometimes take few days, it becomes difficult to track/remember where you are at when yo come back the next day etc etc. Would welcome a discussion as to what others do to organise the workflow and it’s many steps.

Sent from my iPhone using Tapatalk

 

Hi, here is my way of doing. Still at the beginning of the journey though!

 

I haven’t looked into strategies below 1H yet. I believe robustness lies in reasonable expectations and proper portfolio management.

My target per year on MC results: 20 to 40%, max consecutive loss, positive expectancy, max drawdown 10%

 

1/ I use Random Generation on selected time frame only on Dukascopy data (13 years), 5 years IS and 5 years OOS, last 3 years left for future OOS

2/ I delete any strategy not matching my requirements (RDD/ yearly return/expectancy/profits)

3/ I retest everything on 1 minute TF with first 10 years IS and last 3 years OOS (add other TF and currencies)

=> for tested main currency: I delete all strategies unbalanced between the IS/OOS in term of ratios. I want things to be coherent and of course matching my requirements.

=> for other TF/currencies: I delete any strategy collapsing or not linear, or not ending up profitable (>0) at end of the test.

4/ What is left (around 20/50 strategies at that stage) will go for MC at 200 tests.

=> I delete all strategies with negative profit (left with 5 to 10 at that stage)

5/ with the remaining strategies I run a WFM, those who passed are in the final contender folder (30 to 60% pass).

6/ I run each final contender in retest in one minute TF:

= run in other time frame

= run in other currencies

= run MC 2000

7/ I collect the CSV files of each test, and it goes into a consolidated file, for analysis and benchmark against existing/future strategies. Each pseudo code, is copy pasted to 1 tab of the excel doc.

=> I am surprised to see sometime that I developed some EURUSD_H1 strategies which perform better when used in GBPUSD_M30 for example!

=> then I adapt my portfolio according to new strategies and results from Max drawdown and Profits resulting from MC, not the backtest.

8/ Demo account immediately

9/ no strategies from SQ yet live, will evaluate after X trades (ETA 10 weeks).

 

Cheers,

Think different

0

_Cujo

Customer, bbp_participant, community, 102 replies.

Visit profile

7 years ago #138656

I have a ton of comments on your WF, but this jumped out at me…and I’m to lazy to comment on it all:

 

=> I am surprised to see sometime that I developed some EURUSD_H1 strategies which perform better when used in GBPUSD_M30 for example!

 

What you’re doing there, depends on your point of view…

 

It’s either data mining bias. 

 

OR, it’s (sounds like inadvertent) testing for robustness.

 

1 is good, 1 is bad. There’s no specific right answer, but do things consciously. 🙂

 

Read up on David Aronson, especially his book – Evidence Based Technical Analysis (it’s very, very good), I also posted some interesting videos of him talking about it on the forum a week or so ago.

 

Enjoy the journey, it’s great software, and will really help you make money, provided you use it correctly. Have a written WF is a great start.

0

Gui

Customer, bbp_participant, community, 59 replies.

Visit profile

7 years ago #138657

I have a ton of comments on your WF, but this jumped out at me…and I’m to lazy to comment on it all:

 

 

What you’re doing there, depends on your point of view…

 

It’s either data mining bias. 

 

OR, it’s (sounds like inadvertent) testing for robustness.

 

1 is good, 1 is bad. There’s no specific right answer, but do things consciously. 🙂

 

Read up on David Aronson, especially his book – Evidence Based Technical Analysis (it’s very, very good), I also posted some interesting videos of him talking about it on the forum a week or so ago.

 

Enjoy the journey, it’s great software, and will really help you make money, provided you use it correctly. Have a written WF is a great start.

 

Thanks Cujo, will look into David Aronson. But honestly, so far I found only break out strategies. The same as everybody else I guess. 😉

Not much room for curve fitting then, since the rules are very basic.

Cheers,

Think different

0

_Cujo

Customer, bbp_participant, community, 102 replies.

Visit profile

7 years ago #138667

Thanks Cujo, will look into David Aronson. But honestly, so far I found only break out strategies. The same as everybody else I guess. 😉

Not much room for curve fitting then, since the rules are very basic.

Cheers,

 

Curve fitting and data mining are different. Data mining has to do with selecting lucky outcomes. If you’re going through millions and millions of strategies, eventually, you get ones that looks good based on pure luck (maybe you’ve heard the saying, or a version of – 1000 monkeys at a typewriters eventually end up with Shakespeare, that’s datamining). 

 

Curve fitting can be adding in enough conditions  – “I’ll get better results if I put in a 21 EMA, and then a 2 period RSI”, etc… or taking out things like “what if I didn’t trade on Tuesday, and didn’t trade Friday at 8:00AM”, etc.. to get results that look good and are curve fitted to a specific set of data, but fall apart in other OOS tests.

 

Regarding only finding break out strategies…not sure where you’re hearing “everyone” is only getting breakout strategies? I am going to guess you’re trading forex? Maybe it’s true for forex, and SQ only finds breakouts (I don’t trade forex), but for futures I find trend following pretty good. What little I know of forex, I THINK it’s meant to work well with trend follow as well, but it probably depends on the time frames you’re using. 

 

There’s whole bodies of literature underpinning trend following as one of the more reliable trading strategies (as reliable as anything in trading can be). Try exploring along the three main themes-

1) linear regression (using ADX and linear regression is a “classic” combination)

2) MA cross overs (basically where many people start anyway)

3) Looking for higher highs and lower lows, the whole Donchian channel thing, which can also be a form of breakouts, maybe this is what you mean for breakouts?

 

For breakouts themselves, yeah, I find SQ pretty good for them on both the breakout, but also the reversion to mean part. The problem for me is, I can’t get away from back testing the reversion with limit strategies, and they’re notorious for back testing ok, but failing in real life, due to not getting fills. So, I don’t do so much around breakout.

 

Opening gap trading strategies are also pretty well known. I find SQ pretty good here, but you have to be certain of your data (eg. the gaps are really there, not odd data spikes).

 

Once you’ve worked through all that, then mine around certain times, the various market opens, etc..etc….so, there are many, many, many options besides breakout strategies.

 

So, if you’re just looking for, or accepting and running, breakout strategies – you’re limiting your diversification. BUT, up to you, it’s your money. Just trying to be helpful, ignore it all if it’s not.  🙂

0

Gui

Customer, bbp_participant, community, 59 replies.

Visit profile

7 years ago #138671

Hi Cujo. Thanks for correcting me on my comment. I am indeed not a native English speaker. What I meant is that the breakout strategies I found (on forex indeed) are the one that I guess everybody found too, since they pop up in all random gens. Otherwise I am in line with the rest of your comments. I am new to SQ, but not new to trading (over 10 years of discretionary Forex trading). I have some of the strategies automated you mention already running, but not from SQ yet. One thing at a time, getting solid breakout strategies as a start is great, then I can look into further diversification markets and strategies, CFDs are on my to do list as well, but I need education and practice there before plunging!

Think different

0

_Cujo

Customer, bbp_participant, community, 102 replies.

Visit profile

7 years ago #138674

No worries mate, it’s not meant as a correction, just a conversation. 🙂

 

You’re clearly an experienced guy, so was/am interested to hear your thoughts. But, yeah, just takes time, SQ is good tool. I found I started in one place (wanting my specific strategies), but quickly got other places.

 

Hey, coincidentally, I was just reading this thread over on futures.io about different sets up. I was like, of funny, was just talking about different sets up over on SQ>

0

Heilpraktiker

Customer, bbp_participant, community, 46 replies.

Visit profile

7 years ago #138771

Für die Deutschen unter Euch nochmal unser “Workflow” mit ein paar Beispielen 🙂 Was haltet Ihr davon?

0

Patrick

Customer, bbp_participant, community, 424 replies.

Visit profile

7 years ago #138856

Curve fitting and data mining are different. Data mining has to do with selecting lucky outcomes. If you’re going through millions and millions of strategies, eventually, you get ones that looks good based on pure luck (maybe you’ve heard the saying, or a version of – 1000 monkeys at a typewriters eventually end up with Shakespeare, that’s datamining). 

 

Curve fitting can be adding in enough conditions  – “I’ll get better results if I put in a 21 EMA, and then a 2 period RSI”, etc… or taking out things like “what if I didn’t trade on Tuesday, and didn’t trade Friday at 8:00AM”, etc.. to get results that look good and are curve fitted to a specific set of data, but fall apart in other OOS tests.

 

Regarding only finding break out strategies…not sure where you’re hearing “everyone” is only getting breakout strategies? I am going to guess you’re trading forex? Maybe it’s true for forex, and SQ only finds breakouts (I don’t trade forex), but for futures I find trend following pretty good. What little I know of forex, I THINK it’s meant to work well with trend follow as well, but it probably depends on the time frames you’re using. 

 

There’s whole bodies of literature underpinning trend following as one of the more reliable trading strategies (as reliable as anything in trading can be). Try exploring along the three main themes-

1) linear regression (using ADX and linear regression is a “classic” combination)

2) MA cross overs (basically where many people start anyway)

3) Looking for higher highs and lower lows, the whole Donchian channel thing, which can also be a form of breakouts, maybe this is what you mean for breakouts?

 

For breakouts themselves, yeah, I find SQ pretty good for them on both the breakout, but also the reversion to mean part. The problem for me is, I can’t get away from back testing the reversion with limit strategies, and they’re notorious for back testing ok, but failing in real life, due to not getting fills. So, I don’t do so much around breakout.

 

Opening gap trading strategies are also pretty well known. I find SQ pretty good here, but you have to be certain of your data (eg. the gaps are really there, not odd data spikes).

 

Once you’ve worked through all that, then mine around certain times, the various market opens, etc..etc….so, there are many, many, many options besides breakout strategies.

 

So, if you’re just looking for, or accepting and running, breakout strategies – you’re limiting your diversification. BUT, up to you, it’s your money. Just trying to be helpful, ignore it all if it’s not.  🙂

 

1) linear regression (using ADX and linear regression is a “classic” combination)

 

Can you give as an example, please? I do know MA crossing strategies, but i have no idea how this str could look like… :huh: 

0

rafaeldelrey

Customer, bbp_participant, community, 18 replies.

Visit profile

7 years ago #139644

How do you use 2000 Monte Carlo?  I cannot use values above 1000.  You can type larger values, but whenever you change tabs and return, the value is changed back to <= 1000

 

Right now i quit use SQ3, waiting for SQ4,

 

my workflow was:

#1 Random generation on M1/Tick (depends if searching for scalping or swing strategies),

[DATES IF M1: (IS)=Asirikuy From the beginning Till 2003 + OOS From 2003~Today]

[DATES IF TICK: (IS)=Ducascopy Tick From 2003 Till 2010 + OOS From 2010/2012~Today]

#2 Retest again on M1 + MonteCarlo, [SAME IS&OOS]

#3 Filter,

#4 Retest again on Tick + MonteCarlo, [SAME IS&OOS]

#5 Filter,

#6 personally i dont use WFM, i just run more then 2000 MonteCarlo with huge randomization of the data theoretically gives me the same thing (thats just my opinion i trust the MonteCarlo more),

     and i filter those strategies that are do not perform well after those MonteCarlo tests,

#7 Then i try the strategies that have passed the tests until now inside the Improver with M1 with different pair this time and try to get different results with slightly different parameters that i let it to randomize and i lock some parameters if needed, and then i will retest again on M1 + MonteCarlo,

#8 then retest again + MonteCarlo on Tick, [SAME IS&OOS]

#9 if some strategy passed the same testes on 2~3 pairs you got a robust strategy,

#10 forward test it immediately on demo account, after sometime restest it again with new data on SQ with the same settings, check if it fits the demo forward test,

 

[IF YES=Great open small live account and keep on going]

[IF NOT=Might wanna check your broker OR strategy’s settings might be the time that is differs or daily open/close etc.. Try to find the problem.]

 

i will recommend as always to watch Threshold’s videos about how to NOT over optimize the strategies and etc..

i will recommend as always the M1 data from Asirikuy also

 

*

I DO NOT use RealTicks testing, i prefer FIXED spread.

I DO NOT use WFA/WFM, takes time.., i rather prefer using 1000 MonteCarlo couple of times to randomize the results of it, 2 times minimum i can reach up to 10 times (10,000 MonteCarlo).

I DO NOT use OPTIMIZATION tab EVER!, you will break the strategy!, i rather use IMPROVER tab.

i DO NOT use OpenPrices when testing, i use SelectedTF.

i DO NOT use Time depended indicators (Pivots/Daily OHLC/Hour/Minute/Day) better that way, do not use any time indicators/parameters.

ALWAYS ALWAYS ALWAYS retest your strategy from SQ3 in MT4 backtester with the same settings/parameters/spread/data, BEFORE going realtime DEMO/LIVE, SQ3 can give different results then MT4, and after all you will be trading thru MT4 so you MUST retest and see if it gives you the same results!, MT4 backtester results are the results you must pay attention to and not SQ3 in case of different results in between them! (hopefully the new SQ4 backtester will fix some or all of the bugs in this part).

0

Karish

Customer, bbp_participant, community, sq-ultimate, 444 replies.

Visit profile

7 years ago #139668

You just do 1000 or if you have a low performance machine 500 MC tests a couple of times over and over again,

hope SQ4 will have a 100K MC tests limit lol that will be huge if a strategy will pass a 100K MC test, ok ok i went too far :), 10K MC can prove great robustness,

but there should be a 100K MC limit though :)..

0

Viewing 15 replies - 1 through 15 (of 19 total)

1 2