my journal 2

[Yamato]

Holy cow, 40 systems created in three weeks at the most, total at 114, and most of them are excellent.

Ooh, it makes me wonder... ooh, it makes me wonder...

http://www.youtube.com/watch?v=7YZb8s7Kxa4

I could create about 10 more (at the most), but right now I am a bit tired and I will resume tomorrow.

Also, I am getting to a point where I forget the systems that I've created and I create them all over again. It happened with 3 weekly systems on GC, SI, HG. But guess what? It was a good thing. The 3 systems I re-created were better than the original ones I had created (fresh systems, which had not been automated yet), I used them to replace the previous ones.

 

[Yamato]

Nope. You know what? Unexpectedly, I have a widespread sense of fatigue that makes me want to stop creating any more systems. 40 new systems is plenty. Let alone the fact that I feel there's a high probability that I won't find any more systems to create - now I am only waiting for ideas from some friend on new systems on HG and SI.

Recapitulating, unless this guy gives me ideas for HG-SI systems, I will end here my back-testing/creation phase, that is I will stop testing previous systems on copper, and therefore stop creating new ones, because:

1) I am dead tired
2) I am satisfied with 40 systems
3) I probably wouldn't find any more systems (>50% chance)
4) at this point I risk losing count, messing things up (e.g.: putting a system in the wrong folder, classifying it wrongly), etcetera. Better to do less stuff neatly and properly, than to do more stuff disorderly.

You know, this Sleepwalking movie is not that great: unoriginal story maybe, or not told very well, or too many things are being told and it's not coherent/focused enough... it's hard to follow. However, having said this, I also have to say that this scene is excellent:

Actually, I got to the part with Dennis Hopper, who's playing the dad/grand-dad. The movies has got some good points. You see miserable children and grand-children, and then you find out that they are like that because the father/grandfather is a total asshole. I can relate to that a lot. On the other hand the story is not very well told.

Holy cow, I can relate to this part (and the part leading to it):
http://www.subzin.com/quotes/Sleepwalking/33

01:26:41 You piece of ****.
01:26:42 You...
01:26:46 You ruined everything.
01:26:53 You ruin everything!

Still, a pretty bad movie, even though with some good parts. Anyway, always better than any movie with tom cruise or similar.

Crappy movie... rottentomatoes fails it, and rightly so.

I guess woody harrelson is not infallible after all: I just watched two crappy movies in a row with him in them, The Grand, and Sleepwalking. For the rest, he usually acts in excellent movies. Oh, and also "Surfer, Dude" was pretty bad.

 

[Yamato]

note to self: systems can but shouldn't run unattended for a week

The weekly automation (no need to do anything for a week) can be done but it is not safe. It is best to check every day, in case of errors or even power outage problems, or TWS problems (which increase the longer you keep it running - a daily restart is healthy, like sleeping). However, I have managed to achieve it like this:

First of all, you install TWS_Start and run it once and select the option that sets the autologoff time to 99:00 (which cannot be done manually). That way TWS never gets turned off (except on weekends maybe). Once you do this, you can stop using TWS_Start, because that was the only thing it's useful for.

Modify e.xls as follows (see file in folder):

15.45 to 15.55 "copy_print" macro on the module "mav" is stopped, by reading "false" on a cell on "s" sheet, that tells it to stop the loop at that time

17.40 on ThisWorkBook.Workbook_Open the macro ThisWorkBook.Workbook_Open is scheduled for 17.40

17.40 and 5 times thereafter at 30 minutes intervals, the request history macro on the "common" module checks (on a cell of the history request sheet) if history was all downloaded correctly (all cells must say "finished"). There's a problem on that macro that can be easily fixed, that says it can't run the self-looping macro line.

I could fix the last problem and automate the whole thing, but I won't do it, at least for now, because I am afraid that if I do, I will be tempted to stop checking every day, and this could cause stray trades to produce a lot of damage. Stray trades only happen once a year, but if you keep it running for an entire week, this could happen more often, as rebooting is a healthy thing. Running and checking the systems every day only takes 15 minutes and it's precisely like brushing one's teeth. You can't avoid it and after a while you get used to it.

Once I get home I need to copy this text on a file named "why_I_am_not_using_this_file.txt" and put it in the same folder as the .xls file.

 

[Yamato]

stairway to hell

****, I got home, late because the ACE team was busting our balls all day long and slowing us down. Damn idiots.

So I got home and found this dude, the CL_ID_3, losing 1000 dollars. Why does it have to be so hard for my systems to perform consistently? I am being persecuted by my systems. Unless something incredibly good happens in the next three days, we're going to have another bad week of drawdown. After reaching 14300 on April 11th, we've been in a drawdown that still is far from being over. Now we're below 11000. Yeah, because they lost 1000 yesterday and 1000 today. Damn, do they lose money fast...

Another few days and one whole month of drawdown will be a reality. Can we recover before the 11th of May? Ooh, it makes me wonder...

http://www.youtube.com/watch?v=7YZb8s7Kxa4

I was realistically expecting 5000 of profit every month, whereas last month they made nothing, and this month they lost 2000 already. They now owe me 5k of profits for April, and 2k for what they've lost so far, and 5k for May. Can they catch up and deliver 12k by the end of May?

 

[Yamato]

Damn. Not going well at all: they lost a lot from CL_ID_3. Losing from GBP_ID_5 as well. Kind of screwed, because they tend to lose and win all at the same time. Damn, damn, damn.

Equity is now at 10500, the lowest valley we've had lately.



The biggest fall so far has been 5k, at the end of last year. Considering the leverage is higher now, with many more contracts traded, I would be ok if we went down a whole 5k before starting to rise again. If we went down below 9500, I would get kind of depressed, but it still would not have exceeded the worst drawdown, which, for these systems traded, is 13k (which would mean going back to 1000 of profit).

Anyway, the longest drawdown so far has lasted 3 months, from june to september. This time it's only lasted 3 weeks. I can still take it.

 

[Yamato]

Tonight it's pretty early and I am already in bed. I might be able to sleep.

I skipped some bull**** chat with my dad, who usually shows no interest in what I say and think. Sometimes I am in the mood to take some non-existent conversation with him (he's either in lecturing mode or interrogation mode). And sometimes I am too depressed to take more abuse by others. This is the story of my life: I tell my mom I got beaten up, and I am told "oh, poor people, they must have had a tought childhood...". I tell my dad I got beaten up and I get told I am an idiot. Never any sympathy or support or encouragement by my parents.

So, ok, I am pretty depressed and so I might sleep tonight. Or then again I might not.

I am done with the 114 systems, and i will start their automation next weekend or the one after.

The sore throat is gone. The hot water is still missing. Now I'll watch a movie or just start it and finish it tomorrow:
http://www.putlocker.com/file/3LYRRMAKTQECO4C0

I am in a moment of weakness, because of the drawdown the systems are experiencing. I cannot take any more fighting right now. I need to stay by myself, peacefully, and not fight against any outside enemies. It's going to be this way for this whole month probably. Not because the equity will keep dropping, but we have first to recover from 4k of losses and also break to the upside and exceed the previous peak, before I can say the drawdown is over and I am satisfied with my systems. Right now I am pretty depressed. I will feel ok once we reach 16k.

 

[Yamato]

Ok, that movie was too depressing and I stopped it.

I will watch this one now:
http://www.putlocker.com/file/F8YAMC4AS88OOSG#

Probably will manage to fall asleep only in one hour.

I need the weekend to happen soon. These people at work are busting my balls.

Plus I have some general frustration related to other self-censored matters.

 

[Yamato]

work ahead

After coming back home today and in the future, I will have to do this thing, which I woke up thinking of: improve systems that failed.

It is reasonable to keep systems that failed. But, once they blatantly failed, it is even more useful to slightly change them - if there's anything good in them. Or even reverse them if they failed too much.

I will change systems that have more than doubled their maximum historical drawdown and that have been unprofitable in forward-testing (since i started forward-testing them in July 2009).

Such systems are about 10. If I won't be able to improve them without changing their nature, then I will leave them as they are and keep forward-testing them.

The best example of these failed systems is CL_ID, which for the past had a max drawdown of 9k and since July 2009 has had a max drawdown of 35k. It is down 22k since July 2009. Another example is EUR_ID_3, which is down 13k, has a drawdown of 22k vs the max historical drawdown of 6k.

 

[Yamato]

small complaining break

The bitch from ACE came to have a meeting with us (the 7 people in my office) and she was late as usual. The very person that is supposed to make us more efficient is always late.

Not just me but also my colleagues complained about this stupid process, that will be going on for a total of 6 months, trying to make us more efficient, and instead it is just making us worse and wasting our time.

We are all quite pissed off.

 

[meanreversion]

The number of systems you create is fairly astonishing. I recall you have several broad categories for each system, but in my experience whilst hunting for systems I felt I could live with, good systems are hard to find without resorting to curve fitting.

If you have a system and run it on two different markets, is that two systems, or one? Do your systems generally work well on different markets, or mainly just on a chosen few?

 

[Yamato]

Yes, good questions.

I have about 10 categories of strategies, and almost all of them work on all the 16 futures I am trading, with small variations such as:

1) daily range requirement - inevitable customization because the different markets don't have the same daily range

2) hour of entries and exits - inevitable customization because we can't expect the currencies, the german bund, the stock index futures to have exactly the same time cycles.

Oh, and to answer the other question, if the same system applied on two markets is one or two, I would say two, because I get different results, the trades are made at different times, I measure performance differently, I may trade one and not the other one... for all purposes I have to regard them as two different systems. Also because in some cases there might be the mentioned differences (in which case they're undebatably two different systems). In any case it's much more practical to consider them 2 systems. But if you mean to say that instead of having created 114 systems, I have only created 10 main strategies (and variations of those), then I can agree with that as well. Except that you must not think that I simply created 10 algorithms and applied them to 16 markets, just like that, because that's usually not the case (there's a lot more work required). And the equity curves, the sharpe ratios, the returns, everything is multiplied by 114. There's 114 different behaviours, so that is why I always refer to them as 114 systems.

Regarding the risk of curve fitting, it is avoided by these methods:

1) I keep my parameters limited to two or three per system.
2) I use the out-of-sample. Optimization is done on the in-sample and then what works in the optimized period has to work also on the out-of-sample.

I have now reached a point where if you give me a good new symbol (liquid enough and not correlated to the ones already traded), I will be able, almost for sure, to create 10 profitable systems on it and add them to my portfolio. At this point, though, I am also running out of good symbols to add, for various reasons, all mentioned on this journal.

However, if it is true that you give me a new symbol, and I create 10 systems on it, then it means that one way or another, including all variations and flavors, I might have not 10 but about 20 systems that sum up all my existing strategies. Yeah, because in some cases I mix one strategy with another, and so that creates a different system.

Then, before concluding this post, I must also add that the majority of my systems are not worth trading, and that only about one third are really good. One third seems to be unprofitable, and one third breaks even. And I am talking about the forward-testing, because of course in back-testing they've all been excellent, otherwise I would not have implemented them. Those 33% of break-even systems might be dormient and waiting to make money. Instead, regarding the 33% that are unprofitable I don't have any hope left. That is why I am planning on changing them, and fixing them. And backing up the previous and abandoned systems before removing their trades from my forward-tested records and replacing everything with the new ones.

 

[Yamato]

some venting out about ACE idiots

I need to relieve some stress from work.

Today I came home at 8 pm instead of 3 pm.

That sucks. Why? Usual team of ACE morons trying to make us more efficient. Ever since they've been there "helping" us, I have been having trouble getting home at 3 pm. That is terrible.

During all these years, I've developed the most efficient ways to order things, and do my work. Four years of optimizing what I am doing, with intelligence and hard work. They came, these two ACE idiots, out of the blue, without knowing anything about my job, and start telling me and the others to do things in a different way. It could even be a good method, if it were implemented and coached by intelligent people, and best if it were done on stupid lazy people. The problem is that the ones supposed to make us more efficient are two idiots, and they're using their method on me, a hard working intelligent person. I feel like killing them.

It's almost hilarious - it would be if it didn't last six months - as I feel like I am in one of those Fantozzi movies, where they tell the story of a typical employee, who gets screwed by the bosses and all the office characteristics. Or in Office Space for that matter. But Fantozzi renders the idea better.

A bunch of people kissing up to the boss and saying how great ACE is (or being quiet at best) and no one daring to say the truth that it's all bull**** and useless and it's hurting our office. Today we had yet another meeting and one of my colleagues said something against it, so I joined in, and after a few minutes all hell broke lose and we were all saying **** against this ACE project. Finally I got some satisfaction. Finally some people were on my side to fight for the self-evident truth.

But usually I am like the fool on the hill, and I "know that they're the fools":

All my life I have been going against the current, and speaking the uncomfortable truth. Not out of sacrifice, but because it's more comfortable for me to say how things are than to lie. So I've been speaking the comfortable truth.

The bitch slammed her door again. If I meet her, it will be hard, but I must force myself to not say anything. Because it's also a natural thing to say hi to people, but in this case she must know at least that I don't consider her a friendly neighbour. She's a bitch and I wish she died. I hope heels get stuck in some machine or hole and she dies as a consequence of it. Her fault is mere stupidity - she doesn't think that the door could bother people, she doesn't think that it can be closed without being slammed. She is simply stupid. Nothing against me. Stupid people are worse than selfish intelligent people. They do unnecessary damage, to others and to themselves. I hope i can make enough money to move away from stupid people, at a safety distance. I am highly intolerant of stupid people.

Ok, time to try to go to sleep.

 

[meanreversion]

The other issues I've encountered with running multiple systems are

a) capital
b) error avoidance

I rarely have more than about 5-10% of my account exposed to the market, but in keeping my risk down, I limit the number of systems I could conceivably run concurrently.

I also keep a thorough journal/list of ALL trades, such that I can compare the trades I've done with the backtest (this helps to avoid coding errors) and also that I know precisely my returns/commissions/equity swings etc.

I try to keep my procedure as tight as possible with the intention of keep errors to a minimum.. I've also coded algorithms to assist with order management, but all the same, I'd feel hard pushed to run more than 3-4 systems.

Do you have any issues with keeping on top of your systems, or are you reasonably happy to let the algos run in the background?

 

[Yamato]

Replying as I read.

Capital is a problem, yes, especially considering that I have none. But for that I have resorted to other people, who also work with me in assessing the validity of my systems, and make it impossible for me to interfere with my systems, because this arrangement would stop immediately if I started gambling with their capital.

I don't understand the concept of "error avoidance".

Regarding the systems running concurrently, I would be ok with all systems running at once and all capital being used at once (as margin), but the fact is that it doesn't happen, ever. The more systems I add (16 systems being traded now) and the less likely it will happen. The average number of systems trading concurrently is 2, the maximum number has been probably 5, and every once in a while there are no systems running. At least for a few hours a day is the norm. It's rare to see days where no trades are placed, but it happens.

I keep exactly the same registries of trades as you do (back-tested, forward-tested, live traded), and have checked the coeherence of back-testing with forward-testing and with live trading, to a reasonable extent (considering the amount of work required).

Regarding the coding and the running of systems, I am a self-taught and rudimentary/beginner vba programmer, and yet being a compulsive perfectionist (double-checking things all the time), I am now running 74 systems on one excel workbook and haven't seen any bugs in years. Have not missed one trading day in years, and the only problems were caused by two power outages and the UPS running out of power. I am confident that I'll be able to bring the systems to 114 and still run them without any freezing or bugs. But probably I could not bring their number to 200. By that time however, I would be focusing on eliminating the systems that aren't working at all and replacing them with new ones. I don't think I can keep more than 150 systems all clear and ordered in my mind.

Regarding "keeping on top of my systems", I don't need to check them more than once a day, at about 11 pm CET. That is all the work they would require. However, I enjoy looking at them and praying occasionally for profit to increase and markets to go my way.

I have even found out how to make them run for the entire week unattended but I don't feel it's necessary and I feel that it would unnecessarily increase the risk of heavy damage by a stray trade, which only happens once a year (after I make some changes usually), but since running the systems takes me the time it takes me to brush my teeth, I should keep doing it, because it's a healthy habit.

The other big risk I fear is this: I have disabled excel's auto-save feature to avoid any inteferences (it can potentially cause more problems than it solves) and what happens if I go on vacation for a week and when I come back I found out that there was a power outage on Friday and I've lost a whole week of trades? I would only enable such a feature if I had to stay at the hospital. But then it would be best to simply not run the systems for a week.

 

[Yamato]

uplifting day

http://dictionary.reference.com/browse/uplifting

–adjective
inspirational; offering or providing hope, encouragement, salvation, etc.

Yeah, today was an uplifting day.

Not at the office, where the usual was going on:

It was a financially uplifting day. It was a day providing hope, encouragement, and promising future salvation from the office.

My mood goes up as the equity curve goes up, and today it went up.

Other than this, I got home and now I am about start playing risk with some dude I met on the internet. But we need more players, so if anyone is interested, let me know.

 

[Yamato]

Ok, the risk dude kind of ditched me, unless he wants to start playing at midnight. I've been waiting for four hours already and no replies from him.

I won't be able, mentally speaking, to improve any failed systems today. But I will take the first step: selecting the failed systems.

All right, so we have 74 systems (of the previous ones).

Out of those systems, I will not test those recently created in the summer of 2007 or later, because they are too new.

I will use a pivot table to rule those out.

I am ruling out all those systems that have been created from wave 9 through 13, regardless of their performance.

Then I am ruling out all the systems have had a positive return in forward-testing. There's no question of fixing those. This alone rules out 36 systems out of 74, which is about half (yeah, pretty bad to see that only half of my systems have been profitable).

I am now left with just 18 systems that were both created over a year ago and that are still not profitable.

Now I have to see how many of those 18 have exceeded their maximum historical drawdown.

And now I am left with just 14 systems.

Now, since the future is always worse than the past, I want their drawdown to be exceeded by at least 1.5 times, or else I can't say they have failed yet.

This leaves me with 9 systems to be fixed.

But now I want to make sure those 9 systems aren't paired with similar systems that have instead proven profitable.



Ok, this is just perfect, because these nine systems above have all doubled or more the previous max historical drawdown, and have so far lost many thousands of dollars in forward-testing. They are total failures.

And besides, it's not like I will eliminate them. I will fix them with the addition of an extra filter, and using the out-of-sample approach and using what I've learned since I've created them. By the way let us also check when I've created them.

CL_ID	(200802)
CL_ON_4	(200910)
EUR_ID_3	(200802)
EUR_ID_6	(200911)
GC_ON_2	(200910)
JPY_ID	(200802)
JPY_ID_2	(200902)
JPY_ON	(200801)
ZN_ID_2	(200910)

More than half of them were created over two years ago. The other four were created a year and a half ago. Well, I definitely know much more now. In the next few days I will get busy fixing them, so at the end it will be like adding 49 systems rather than just 40.

 

[Adamus]

Hi Travis, good luck fixing those systems. A question: what backtesting period and out-of-sample period are you going to test them with?

The more I think about the periods you use, the more I think it's more robust than the longer backtesting period and shorter out-of-sample period that I use.

 

[Yamato]

Replying as I read.

Thanks for wishing me good luck. The backtesting period is from 2000 to now, more or less (depending on the symbol). The in-sample is from 2000 to 2005 included, the out-of-sample is from 2006 included to now.

Maybe I understand what you are saying if you are saying that the out-of-sample has to be shorter than the in-sample, but for some reason I prefer to do half and a half, so I have a guarantee that all types of markets are included in both samples. And, since a drawdown for my systms could last as long as two years, I cannot afford to have an out-of-sample periods of less than 5 years.

Then some flexibility in this methodology, after having created hundreds of systems, can be allowed for those systems that already work on several other symbols, so you know you're not being delusional in choosing parameters and coefficients. For example, if one algorithm already works on gold and silver, you could allow yourself to optimize the daily range coefficient throughout 10 years on copper, if you know that the volatility has changed in the last 5 years (as it has for gold and silver, too), knowing that your system would not be properly optimized as a consequence. That is the flexibility I allowed myself.

 

[Adamus]

One day when I have time I will do a study to see what the probablilty is that systems which work in forward testing then go on to work over a second forward test.

It could be the basis for a PhD or at least an article.

I think I would need more data than I have though. If I had enough data for 3 five year periods, that would be ideal. One 5 year period to build the system, one 5 year period to do the first forward test, and the last 5 year period for the second forward test. Maybe using 4 year periods would be easier, I'd need 12 years only - but then instruments like EUR has only been around for 10.

It would obviously be a lot of work coming up with many systems but they could theoretically all be recycled for use in real life and that is anyway what I'm doing, long term.

 

[JosephCass]

good luck