The problem of Optimization, and subsequent failure... - page 2
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Ok, here's a question from a NOOB. That would be me of course. If taking the Advise of most veteran traders, and not rely on back-tests what am I suppose to justify my strategy based on? These same traders in my Opinion are NOT going to open up their Account History and say Hey here's the results. And if their systems are based off a (top of head),(how i feel) whats the word (Subjective) feeling, how can they teach it to anyone? Or yet program it?
They all say Change with the market but most have No definition of what constitute a market change. I would gladly Adopt the system of my dear Advanced trader if they could show Proof it works and I'm not talking about over 1 year or less. If putting my money on Back-test Profitable over 10 years is Noob-ish, I'm afraid the Pros aren't providing much alternative either.
Don't give me that, if China joins the Eur fundamental (as if they can) then my system is doomed. Everyone's system would be doomed lol... A system trader would be dumb to ignore such Change.
Getting back on point - to the guy who wrote this, I'm in the same Shoes as you. 95% of back-tests Fails it seems. 4% works within short time like a year and then fails. What I decided is to accept results that at worse break-even within a year. Then I bunched all these together as one system. My profit factor Never exceed 1.5 which was what i was hoping for when I started. Well it did, but with those systems they probably placed less than 1 trade per week or month.
On closing - I don't expect anything of the market except to Lose like the 95% of people like the books say. So I have NO pie-in-the-sky view of Automated EA's, its just a place to Start. I'm currently Live-Demo-Testing my Combined Strategy hopefully it can get the 1.3 Profit Factor like the back-tests. Wish me luck!!
It should be understood that the lowest bar interpolation in the Strategy Tester (ST) is M1. In demo and Live trading, bar interpolation can be in seconds. It only take 1 pip to make or miss an EA controlled trade action. I find to make ST/Optimization more reliable, EA should only be active on M1 increments. This way, EA will ignore less than M1 interpolation. This is not a fix all problems with relying on Demo/Live results to be more compatible with ST results, but for me, it helps.
Another factor is the fixed spreads and stoplevels in ST and the latency times. ST has instant execution, while demo and live have this latency time issue to overcome.
you can significantly reduce the time impact by creating a ram-drive and mounting the partition as your ...\tester\files\ folder. [...]
This might be a well known approach but I figured I'd mention it on the off-chance it hadn't been openly discussed here before.
I've not heard this suggested before, but I've tried it myself... with limited success. I didn't see that much of a speed gain, and I was running the whole of MT4 from RAM, not just the \tester\files directory.
Yeah I started out warehousing the entire MT4 installation on a ramdisk but eventually migrated to a model that just stored the files that were being written/read/written/read repeatedly as that turned out to be my rate-limiting step.
Obviously the speed-up value you can expect from doing this will be entirely limited to the amount of attempted read/writes per second your EA/strategy-tester is performing.
If you aren't exceeding the capability of spindle-disks (on the order of 250-500 IOPs for small-file read/writes) then boosting that limitation to 10,000 IOPs isn't likely to net you a noticable improvement in computation speed. For me it was on the order of a 2x improvement (for the particular EA's I was backtesting, naturally) and became all the more value-added when I started loading my quad-core rigs with 4 simultaneous instances of MT4 running backtests in parallel (which invites 4x the opportunity for disk thrashing).
(...) I find to make ST/Optimization more reliable, EA should only be active on M1 increments.
Forcing the EA to be active in M1 increments just so as the Tester's results will be more reliable makes no sense to me (IMHO)... We should find a way to make the Tester fit what we want to do with the EA and not the opposite... Or alternatively, we need a measure of the accuracy of the Tester's results. I usually require that SL/TP be at least a decent multiple of the average movement (high-low) in M1 bars OR a decent multiple of the spread (discussed here -> https://www.mql5.com/en/forum/124026/page2#275527) in order for the Tester's Results to be considered 'reliable'.
If taking the Advise of most veteran traders, and not rely on back-tests what am I suppose to justify my strategy based on?
The hard uneasy answer to your question is that the justification should first come from a real theoretical basis as a foundation for the strategy. What inefficiencies is the Strategy trying to exploit? What is the theoretical basis for it's forecasting? Why should it work in theory?
If you can answer these questions before writing the first line of code (and way way before doing any kind of testing) then u are on the right path. When u get to the point of Testing, if u get a positive result, then at least u know that it would have worked in the past (within the limited accuracy inherent in the Tester). This still does not mean it will work in the future, but it does give u a first positive indication that u r on to something. U then need to take a much deeper look into the result and use all the tools mentioned in phillip's post to analyze the result (and as mentioned - the default report template is not sufficient).
Optimization, if done at all, should be done with extreme care and only with a theoretical justification. This part is extremely hard to achieve and extremely easy to screw up... Hence, it should be avoided or done with extreme care (and I recommend that beginners avoid it altogether).
p.s. I am sure some of what I said might be controversial... It's all IMHO.
Would THIS method be likely to produce acceptable results?:
-Adjust/optimize on a small slice of history, say 1 month or thereabout, getting eg 50 trades or some other statistically signifficant number...
-Shift his time window back and forth in history and test the result for each location, if not good readjust and re-test.
By evening out the adjustments so they will work reasonably well on 'all' months many years back they are more likely to work equally well in the future...
Prettty obvious that the profitability for any month will be just a small fraction of what a profit peak optimization would indicate, but maybe it could still be profitable.
A lot of work involved, dunno if I have the stomach for it... What do you think ?
Well if you are looking for the trade strategy that produces good results for the middle month, the previous month, and the subsequent month then why not just run a backtest over the entire 3-month period and assess the results?
Many thanks to gordon, 1005phillip, JJC and other Advanced Programmers/Traders on this site. I've had allot of my questions answered by you guys before I joined the forum yesterday. Please keep helping us Newbies.
Please allow my background and theory. I'm from the Blackjack world. We run Billions of simulations to get system expectation/standard dev/risk-of-ruin etc. We tend to hold those result as Static or as Phillip put it (Stationarity Notion) within standard error and deviations. Now I'm Not at applied Science/Mathematics level so I couldn't give you the equation for calculating three standard deviations on any data. What I did become good at was Comparing apples-apples which sd/system was better.
In Trading, I was warned ahead of time from my BJ forums that Card-Counters unlike Poker players have a harder time transitioning to trading because of that Stationarity Notion. Please forgive the Gambling Analogies. I had so much hope for Trading thinking it would be better than gambling because of how our Society treats these two occupations but that's looking more and more like a false assumption. The biggest Advantage I can see with trading is the Table-Limit is Theoretically unlimited and the House is not gonna kick you out for winning Too-Much money. However, there seem to be an invincible house here.
When Phillip wrote earlier about newbie mistakes and conceptions, it was like he was talking directly to me. I was like "wow - how could he know - the man don't even know me - lol". I was kinda shocked because it sounded like he didn't believe in back-tests. But now I understand that it's more of the quality of the back-tests.
A few weeks ago I was on a popular Forex chat-room and there was Not a single EA trader who passed through there that whole week. Almost everyone there did not Believe in EA nor the results of back-tests. But when I asked them about their system Expectation/Profit Factor/Maximum Draw-down, no one had a clue. How can anyone execute proper Money-Management without these numbers. And alot of these guys were self-proclaimed seasoned traders with years under their belts. Most of them dodged the ? of if they were profitable or not.
This and everything else just reinforces my notion that either No-One know what they're doing or the People who know are not sharing. Again, its like a Poker game where one guy (Millionaire traders and Banks) takes the other guys (Everyone else) money right. So why should anyone give me truthful information when I'm their feeding ground.
I don't have any set Theory on the markets right now. It seems Efficient 90% of the time and Trends/News/could be exploited 10% of the time. However, most trader miss 5% of that move waiting for the setup to develop and for end signals. If I had to select a Theory upon which to test First it would be trading Fundamentals - good news = buy || bad news = sell lol. However I don't have the technical know-how to program and back-test this type of strategy in order to get Optimum stop-loss and take-profit levels.
So I turn toward Technical trading which In my mind equals Psychological trading trying to exploit whats happening in that 90% of the time movements. I use the most popular TA-Indicators in their native forms or rules. I keep it simple meaning no combination of indicator. Then Optimize it to Death over 10 years, if it stays profitable than I'll add it to my Combo System. I try to add systems which Do-Not Correlate so i have every mix of the pot as you can imagine. If they all worked for the last 10 years and then decide to Fail now, Oh well.
I'm trying Not to be a Full-Cup here therefore I'm gonna go through all the Article recommended by Phillip and Others. Truth is I've already glanced over those articles long before I seen it in this thread. It's very well written but was too technical for my level at the time. The only thing I toke from reading it in the past was how to create External Variables to Optimize on. This Process is turning into more that I would have bargained for in the past. I promised my self, I was not gonna become a programmer to count-cards. Doesn't appear like that's gonna fly here. Looks like I'll need to become a full Mathematician/Programmer/Economist/Psychologist to become a Successful Trader.
I don't mind the learning process or challenge. However, the question here is where does it end. Don't even bother mentioning Manual trading as I was losing sleep trying to Manually trade on a demo account with fake money. Having an alert wake me up at 3am for a trade-setup when it could just place the trade is a NO-NO. I'd much rather spend that stress time at the Blackjack tables where the Cocktail waitress and Dealers are hot. Not to mention all those fly Honeys steaming up on the slot machines. The system expectations seem to be the same for BJ to Forex but the compound per independent event (bet) is faster at blackjack. 1% for traditional card-counter vs maybe 2% for traditional trading. 5% for Over/Under-13 blackjack and is there anyone with a 5% ROI per Trade in Forex or Anything. Those wondering why I'm not rich by Over/Under-13 blackjack, here's the math $25 table Max - frequency of 4 bets per hour - do the math.
You are on the right path ubzen. Master the art of walking before attempting to run. And yes this cottage industry has its share of charlatans who got burned trying to take shortcuts in their efforts to skip straight to running.
Everyone has a unique trading psychology (just as in professional gambling), and as such what is best for me is not likely to be best for you when it comes to trading styles, etc. But I will share with you what worked for me...learn to successfully trade manually first, then worry about automating trade strategies. That isn't to say you need to automate your manual strategy, automation enables an entire class of strategies that aren't practical for manual approaches.
But what comes with mastering manual trading is a deeper understanding and appreciation for what a successful automated strategy ought to look like, smell like, act like, etc. There are things you know you don't know and then there are things you don't even know that you don't know. Manual trading provides the opportunity for you to uncover what you don't know about trading and forces you to master them (at your own pace and learning curve) or perish.
Automated trading reduces the likelihood that you will be confronted with the experience of uncovering something you were ignorant about, and those flaws usually become fatal to your account in time.
You want to be in the position where you can trade your automated strategy with confidence, that requires you to know what exactly it has been designed to deftly handle versus the things you know you didn't bother to make it capable of mitigating. If you approach it like a blackbox that is going to churn out the dollars then you are far better of hitting the slots and enjoying the comforts and company that go with it :)