Deadline June

[Adamus]

Thank you Cedric the Frog. It seems not all Frenchmen deserve to be decapitated and have their heads impaled on spikes in the Tower of London.

It took me a while to surmount my internal revulsion at the idea, but after his comments, I had no choice but to investigate.

It does seem to be a good idea to seperate long and short trading systems. Here's the optimisation and walk-forward for the Aussie Dollar again, this time with long and short trading optimised seperately (but the results combined):

Not only did I get profit in the walk-forward, but it wasn't massively distorted by the huge surge in autumn 2008 as before, i.e. it would still be profitable without that.

I hope this is all worth it. It takes ages. One of the first things I'm going buy if this all starts making money is Trading Blox.

 

[Adamus]

oops no more photos allowed.

ok so I'll have to attach them & it'll only show thumbnails.

BP: couldn't find a profitable optimisation, so ended up with this and the walk-forward is still profitable. If I start from 2009, after the massive autumn 2008 credit crunchorama, I get something a little more likely.

 

[Adamus]

Corn optimisation also rubbish

Couldn't get a decent optimisation for Corn. The best I could find doesn't make enough to cover transaction costs.

It took a long time to do this. Looking at the other markets in the basket that I still have to optimise, it might take a few days. Labour intensive is the word.

Anyway, from the optimisation stats and the walk-forward stats, it's difficult to see that they're the same system. Don't know whether that's indicative of curve-fitting (regardless of whether it made a profit - could be due to the huge surge in 2008 that I'll just never see a loss) or whether this is something I can ignore.

What I'd like to do is have an optimisation statistical report, and see similar values for stuff like max loss, max drawdown, trade frequency, win/loss ratio, profit per trade etc - but within what size range, I don't know. Maybe experience will be the only way to find out.

Here are the equity curves. Ignoring 2008 and the massive surge, I guess the equity curves at least look familiar. Loss making! Well, if it increases diversification and reduces correlation, it might be worth including. I'll find out that at the end.

The equity curves for 1996-2008 and 2008-2010 obviously have different x axis scales, so I included the whole equity curve in one.

 

[Adamus]

Lost in Fangorn Forest of optimisation and walkforwards

Here's Crude.

The optimisation gave a great result for 1996-2008, but the walk-forward from 2008-present was dire.

The max drawdowns in the walkforward was $33K, and the max drawdown in the optimisation was only $13K.

So has the walk-forward blown out the trading system? Is it curve-fit? Or is this just down to the massive volatility in Crude?

This is only the 4th in the basket and I've got 21 more to do, so I'll leave the jury out on this one, but more bad walk-forwards and I'll write off this system.

 

[Yamato]

I don't know exactly what you mean by "walk-forward", but if you mean what I meant years ago, when I tried using it, by periodically re-optimizing parameters (every 6 months), then I would suggest replacing it with self-adapting parameters based on volatility, closes, ranges... so that you just have one single system, which is always the same and works for the past 10 years.

I kept my post short, just in case I am off-topic. Feel free to ignore my quick and somewhat superficial comment.

 

[Adamus]

no, I mean a walk-forward as in a test of the trading system with optimised parameters on a piece of historical data it wasn't optimised on. For this whole basket, I optimised on 01/1996-06/2008 and I walked forward 07/2008 to present. That worked but the stats showed drawdowns for serious accounts only.

I'm trying to improve on that by re-optimising, market by market, on the same windows. I'm not worried that the initial optimisation was curve-fit because the data set is so big - 12 years of 25 markets = 75,000 daily bars. But the data sets for the individual market optimisations are only 3000 bars, so I'm worried that I'm curve-fitting.

Crude certainly looked like it, but that market was wild for 2008 / 2009, wilder than ever before.

By the way, what would a 'self-adapting parameter' look like? I only use measures which are relative, never absolute. Is that what you mean?

 

[Adamus]

I know what you're going to say Travis and I just realised myself:

I could optimise on hourly bars back to 1996 and I would have 50,000 bars.

3000 bars say their owners like to curve-fit.

But 47,000 more will make a robust systems :jester:

Yes I rambling but I just got excited - don't you remember those dumb cat food ads?

 

[Yamato]

You're absolutely right: I had written the post below before reading your last message above, but you predicted what I would write very accurately. As you read what's below this dotted line, bear in mind that I wrote before reading your last post above (it happened because I started writing, got interrupted, and it took me one hour to finish my post):

------------------------------------------------------------------

That's correct. That is what I meant by self-adapting parameter, but I only mentioned it because I thought you were doing walk-forward optimisation, which would mean you re-optimize parameters every six months or so. But I had understood it all wrong. What you are doing makes perfect sense, and that's how i did it, too.

One more things, which I already told you. You do realize that if you switch your timeframe (and consequently your strategies) from daily candles to 15-minute candles, it's like multiplying your data set by... for futures... 15 times 4 times 24 = 100 times? This would guarantee you against curve-fitting, and it would yes reduce your strategies gains but conversely drawdowns, too. I already told you before. That's why we spoke of disktrading.com data, and so on. I think you're missing out on important intraday action with your daily timeframe approach.

Usually everyone says intraday is bad because they're referring to emotional intraday trading, which causes many of us to lose control and blow out our accounts (particularly me). But, as far as automated trading, "intraday" is not bad at all, not a bad concept, not a thing to stay away from, because, obviously, automated systems are not emotional and can take any number of trades per day, and, the more they take, the more you can diversify and reduce your drawdown. If you instead go for the big trades, that will necessarily mean bigger drawdown, and that is always bad.

 

[Adamus]

another failed optimisation / walk-forward, this time Cocoa.

Another rubbish optimisation result.

So another question on the north face of the optimisation / walk-forward learning curve: is this because this system just doesn't work on cocoa for the walk-forward window, or did I curve fit it? The drawdown in the walk-forward is well within the drawdown range seen in the optimisation window. Lots of stuff is the same - the trade frequency, % winners, etc.

 

[Adamus]

next up......

If I look at my basket again (here's the original table again), I've optimised the first 6 rows and not got much better - although I think I may have reduced the losses.

I could have left the Aussie $, the BP and Crude alone. The walk-forwards with the original were stellar results, but they're caused by the Credit Crunch surge of autumn 2008 and the rest of their walk-forward periods were loss-making.

What I shall try to do is optimise the losing markets - CD, EJ, GB, GC, KC, SF, SP & X and only the rest if I have time now.

 

[Adamus]

One more things, which I already told you. You do realize that if you switch your timeframe (and consequently your strategies) from daily candles to 15-minute candles, it's like multiplying your data set by... for futures... 15 times 4 times 24 = 100 times? This would guarantee you against curve-fitting, and it would yes reduce your strategies gains but conversely drawdowns, too. I already told you before. That's why we spoke of disktrading.com data, and so on. I think you're missing out on important intraday action with your daily timeframe approach.

I'm playing around with that GBP 15 min data. I see it's not Globex, and some of the gaps between the days are frightening compared to the EOD stuff I'm used to.

I'm having problems getting TradeStation to display the data, it keeps going haywire when I try to load the whole 1996-2008 range.

I emailed disktrading.com with some questions, asking if they can construct the continuous contracts to the specifications I use. I guess I might not get a reply until Tuesday.

 

[Adamus]

Just found out from disktrading.com that their tick data continuous contracts for futures are stitched together without any back-adjustment.

So the choices are:

(1) do the back-adjustment myself - disktrading provides a spreadsheet with the list of roll-over dates - although I bet I would have to find out what the gap on the close between the contracts was.

(2) stick with forex - no back-adjustments necessary - but how correlated are they too each other - isn't just like one big market?

(3) purchase data from somewhere else - tickdata.com??? $$$$ gulp!

(4) ignore roll-overs - hmm, that would probably lead my optimisations to trade at the roll-overs a lot!

(5) stick with EOD for now

any other options?

 

[Adamus]

Forex it is then. I'll get the tick data from disktrading and run my trading system against it.

I doubt I could trade 25 different forex pairs. I don't want to trade the minor pairs where the liquidity and slippage are issues.

USD:EUR
USD:GBP
USD:CHF
USD:JPY
USD:AUS
USD:CAD
EUR:GBP
EUR:JPY

maybe:
EUR:CHF
GBP:JPY

That would make it ten. Of course you can't see volume data. I'll have to search for a liquidity data table on the net.

 

[Yamato]

All very logical and reasonable. Thanks for letting me know about disktrading.com data's problems. They didn't stop me from building profitable systems. You're right about your objections, that you might develop an edge precisely at those dates. But then again it might not happen, depending on the systems you're using.

You know, in a sense it doesn't matter how properly you develop profitable systems, and if you do things in an unorthodox way. The most important thing is to start making money as fast as possible (as long as it's also reliable of course), and then, with that money, you can definitely buy tickdata.com data, hire programmers (in my case), pay my cousin to give me ideas for systems (he's a physics major), and so on. If I can make money with my ideas, so much the better. Why should I waste time and resources when I can make money on my own, with poor data, and poor programming skills? The point is to make money and not to do things academically, with a lot of footnotes and stuff like that.

The same applies to you and your limited resources (money and time, since you're in a rush to get done by June).

But among all the academic things you can discard, don't discard intraday trading and testing because it's not academic/pedantic/superfluous and because most of all the opportunities will come from there. I already wrote about it in several posts on this journal. If you want to discard something, discard looking for perfect data. Better to have bad data on an intraday timeframe, than to have good data but only on a daily timeframe.

Also, regarding which currency futures to trade, I would try to develop several systems on just these three currency futures: EUR, GBP, JPY. They're the ones with the most volume. I opened an account with IB and selected all the futures I'd trade and back-test on, simply based on their volume on my TWS platform.

 

[Adamus]

Judging from what I know of your systems Travis, you won't be affected much by gaps. But I'm running moving averages and they will go haywire.

I'm certainly going to trade those 3 - I understand that even CAD and AUS are deep and liquid enough for decent trading. Funnily there isn't even one currency in the top 20 contracts by volume:

[FONT=Courier New]
[SIZE=3]
Kospi 200 Index Options              KRX    2127.98
Eurodollar Futures                   CME    348.2
Euro-Bund Futures                    Eurex  254.1
TIIE 28-Day Interbank Rate Futures   Mexder 85.3
Eurodollar Options                   CME    160.52
E-mini S&P 500 Index Futures         CME    174.5
10-Year T-Note Futures               CBOT   181.27
DJ Euro Stoxx 50 Index Futures       Eurex  118.1
Euribor Futures                      liffe  137.79
Euro-Bobl Futures                    Eurex  131.01
Euro-Schatz Futures                  Eurex  116.15
One-Day Interbank Deposit Futures    BM&F   94.4
DJ Euro Stoxx 50 Index Options       Eurex  76.89
5-Year T-Note Futures                CBOT   102.89
S&P 500 Index Options                CBOE   58.45
Taiex Index Options                  Taifex 62.92
30-Year T-Bond Futures               CBOT   73.72
Sterling Futures                     liffe  56.7
E-mini Nasdaq 100 Index Futures      CME    62.16
TA-25 Index Options                  TASE   50.14[/SIZE]
[/FONT]

 

[Yamato]

You're quite right about the CAD and AUD and even the CHF, but definitely forget the EUR:GBP. I've checked it out a few months ago, because I wanted to trade it discretionary, and there's like 100 contracts traded every day.

This the EUR:
http://futuresource.quote.com/chart...&o=&a=D&z=800x550&d=LOW&b=CANDLE&st=VOI(1,1);

This is the EUR/GBP CME future (I don't remember the symbol):
http://www.cmegroup.com/CmeWeb/html.wrap/daily_bulletin/gbulletin_current.html

On the link above for some reason the EUR is listed as 50k contracts rather than 300k, which is more the case, but anyway the EUR/GBP is listed as having 82 contracts traded. Even if you multiplied that by 6, you'd still have a pretty poor volume.

There! I found a good measurement of volume on the CME chart, which lists it exactly at a few hundreds contracts per day, as I expected:
http://www.cmegroup.com/popup/mdq2...._2010_EUR_GBP&type=p#study=VOI;bartype=CANDLE


REGRET AND PLAN

I've got a big problem with not having developed systems on those, big regret. Because I do have the data, and if I traded them, I wouldn't have have just 40 systems but 60, and my diversification would be awesome. Also the DX should be traded. Actually you gave me an idea, and I am right now considering whether I should indeed build another 20 systems. If my eyesight assists me, I will. Another problem is that excel is the most tiring program for your eye-sight. You see, I created a structure for just 10 systems, and somehow I managed to fit 40 systems into it, into one page. If I expand the systems to 60 or more, I will not be able to look at systems at once any more.

Other than these issues and my fatigue, there's no other factors that could stop me from doing this. Here's the steps I will follow:

1) find out for sure which currencies futures are most traded and focus on the first 4 or 5 which I am still not trading

2) gather the necessary data, and, if necessary, buy symbols.

3) test previous systems on the new data, and adapt them where necessary

4) automate the systems

This will take me months (one month for testing, one month for automating) but it is something I must do. I just told you how important diversification is, and I can't waste the opportunity to diversify more, especially after I re-read yesterday the part where Schwager was summarizing Larry Hite's trading (in red the relevant part):

There are two basic elements to Hite's trading philosophy. First, contrary to the opinion of many academics, Hite is firmly convinced that the markets are inefficient. This means that if you can develop a method that places the odds in your favor (and it doesn't have to be by very much), you can win. Second, an effective method is a necessary, but not sufficient, condition to win. In order to survive and thrive at trading, you also have to respect market risk. If you don't, sooner or later, it will get you. Hite controls risk rigorously by applying four basic principles:
1. His system never trades counter to the market trend. There are no exceptions, and he always follows the system.
2. The maximum risk on each trade is limited to 1 percent of total equity.
3. Mint carries diversification to an extreme. First, their system is really a combination of many different systems, selected not only for their individual performance, but also for their degree of lack of correlation with other selected systems. Second, Mint trades in an extraordinarily wide spectrum of markets (nearly sixty in all), encompassing exchanges in the U.S. and five foreign countries and diverse market groups including stock indexes, interest rates, currencies, raw industrial goods, and agricultural commodities.
[...]

Let's hypothesize that we only allow each system to risk 1% of our equity. If margin allowed us (since we're trading futures), we would want our capital and systems to be continually busy, producing 1% risk bets. If each system produced on average 1 bet per day and we had no margin concerns, which we do not and I'll later explain why, we would want 100 systems to be trading. The reason we don't have margin concerned is that more or less, if one CL contract is making me risk 2000 dollars, or hypothetically more (because the biggest loss, as you said, is always in the future), it so happens to be that its margin is also around that value. If the ZN is causing me a risk of 500 dollars, its margin required is not that small, but close, in the case of intraday margin. In fact, this happens because margin was defined based on risk. They require you to have more margin than you can lose in one day with a given future. On average, about twice as much. I am thinking out loud.

So that would mean that I can... these formulas are getting too complicated for my small brain. It's overheating.

What I do know is that, even now, my capital is not busy all the time, so more systems can only help, by increasing profit and reducing drawdown or at least keeping it more or less the same. Anyway, I can't see all the implications of this, but definitely more systems is a good thing. And more futures always means more systems, because you can usually apply the systems you're running on other futures, on the new futures, and they will work.

 

[Splitlink]

You're quite right about the CAD and AUD and even the CHF, but definitely forget the EUR:GBP. I've checked it out a few months ago, because I wanted to trade it discretionary, and there's like 100 contracts traded every day.

This the EUR:
http://futuresource.quote.com/chart...&o=&a=D&z=800x550&d=LOW&b=CANDLE&st=VOI(1,1);

This is the EUR/GBP CME future (I don't remember the symbol):
http://www.cmegroup.com/CmeWeb/html.wrap/daily_bulletin/gbulletin_current.html

On the link above for some reason the EUR is listed as 50k contracts rather than 300k, which is more the case, but anyway the EUR/GBP is listed as having 82 contracts traded. Even if you multiplied that by 6, you'd still have a pretty poor volume.

There! I found a good measurement of volume on the CME chart, which lists it exactly at a few hundreds contracts per day, as I expected:
http://www.cmegroup.com/popup/mdq2...._2010_EUR_GBP&type=p#study=VOI;bartype=CANDLE


REGRET AND PLAN

I've got a big problem with not having developed systems on those, big regret. Because I do have the data, and if I traded them, I wouldn't have have just 40 systems but 60, and my diversification would be awesome. Also the DX should be traded. Actually you gave me an idea, and I am right now considering whether I should indeed build another 20 systems. If my eyesight assists me, I will. Another problem is that excel is the most tiring program for your eye-sight. You see, I created a structure for just 10 systems, and somehow I managed to fit 40 systems into it, into one page. If I expand the systems to 60 or more, I will not be able to look at systems at once any more.

Other than these issues and my fatigue, there's no other factors that could stop me from doing this. Here's the steps I will follow:

1) find out for sure which currencies futures are most traded and focus on the first 4 or 5 which I am still not trading

2) gather the necessary data, and, if necessary, buy symbols.

3) test previous systems on the new data, and adapt them where necessary

4) automate the systems

This will take me months (one month for testing, one month for automating) but it is something I must do. I just told you how important diversification is, and I can't waste the opportunity to diversify more, especially after I re-read yesterday the part where Schwager was summarizing Larry Hite's trading (in red the relevant part):



Let's hypothesize that we only allow each system to risk 1% of our equity. If margin allowed us (since we're trading futures), we would want our capital and systems to be continually busy, producing 1% risk bets. If each system produced on average 1 bet per day and we had no margin concerns, which we do not and I'll later explain why, we would want 100 systems to be trading. The reason we don't have margin concerned is that more or less, if one CL contract is making me risk 2000 dollars, or hypothetically more (because the biggest loss, as you said, is always in the future), it so happens to be that its margin is also around that value. If the ZN is causing me a risk of 500 dollars, its margin required is not that small, but close, in the case of intraday margin. In fact, this happens because margin was defined based on risk. They require you to have more margin than you can lose in one day with a given future. On average, about twice as much. I am thinking out loud.

So that would mean that I can... these formulas are getting too complicated for my small brain. It's overheating.

What I do know is that, even now, my capital is not busy all the time, so more systems can only help, by increasing profit and reducing drawdown or at least keeping it more or less the same. Anyway, I can't see all the implications of this, but definitely more systems is a good thing. And more futures always means more systems, because you can usually apply the systems you're running on other futures, on the new futures, and they will work.

Hi Travis,

I don't believe half of your self-criticism. I think you are trying to make me feel good!

Good point about EURO/GBP.

 

[Yamato]

I am glad you appreciated my post, thanks. Trading is teaching me modesty. You see, usually I'd immediately stop doing something I am not good at (such as trading, formulas, math, excel), in order to preserve my illusion of being the best at everything. But, because I definitely want to make money, I was forced to keep at it, despite failing for years, and this - failing and losing money for 12 years - has finally forced me to admit that I am not the best at everything. That I suck, as a matter of fact. I am the last survivor of a generation of suckers. No one else has lost money for 12 straight years and is still trading.

 

[Splitlink]

If you can come up with research into statistics, the way you do, I do not understand how you can fail to make enough winning trades to overcome your losses.

I am impatient when it comes to reading posts that I consider to be too profound. I go straight to what I consider is information that I can use ie., your points on the nunber of trades per day that EURO/GBP does that. The same can be said of my reading of TRO's, Mr Charts, and other threads. They both say that their systems are so easy and then go into reverse by writing stuff that is incromprehensible (to me, at least)

Anyway, I'm rambling on.

 

[Yamato]

No, please keep writing because it's very interesting. Thanks for placing me next to some legends of this forum. What I can say is this: sometimes people are good at doing things but not at explaining them. Maybe that's the case with these guys. Other times people are good at explaining things, but not as good at doing them, which is very likely to be my case. I like explaining things clearly to people, and especially to myself. I think out loud as I write.

The reason I have been unprofitable for 12 years is that, until very recently, I never stopped my discretionary trading. And, even when my automated trading was making money, I managed to lose it with discretionary trading. But now I am done with that. I am done with trying to succeed at that. I failed, I am not good, because it's not easy and I don't have the right personality, and I will not try it ever again. It's wrong to avoid all the things you're not good at, but it's also wrong to want to succeed at everything you're not good at. Especially since my objective is not to ace discretionary trading, but to make money, and that is much faster by doing automated trading.

Yet the idea of not doing anything and letting the systems run unattended is unnatural, it is so unnatural, and uncomfortable, and unusual, that I kept fighting this concept for two years longer than necessary. If I hadn't, by now I would have been profitable for about two years. And what's worse is that I still don't know if I am done with the discretionary nightmare, or if I will want to get my ass kicked a few more years. There could be a big loss one day, and I'll feel the urge to make it all back by doubling up on the losing position... it happened recently. Sometimes it works, sometimes you blow out your account. In the long run this method (martingale approach) makes you blow out your account. The more systems I'll have, the smaller will the losses be, the less I'll be tempted to double up.