If your using EFS, enter the highest return you've had...

[gspiker]

Spike,

But if you trade on indicators you always have to wait until the bar has closed. Otherwise you are never sure that the condition to trade will be met.

We were talking systems, not indicators. But even switching the terms, one would have to define indictor. Anything I can code is an indicator to me...

If you backtest a system and rely on the results you have to remember that the backtest only knows the value of an indicator on the close of that bar. A backtest will never take all the different values inside of that bar.

This is only true if:

1) You are using an indictor who's math you don't understand, and therefore can't reproduce the interim values. It is not hard to do this for most standard indicators

or

2) The indicator itself doesn't allow reproduction of the interim values for some reason. This isn't true with most standard indicators.

But that said, I will agree that with most std. indicators you are correct, it is best to wait for end of bar. But it isn't even close to an absolute rule for all indicators, and I have seen many who's performance will suffer greatly by waiting for the end of bar. Which was my entire point. There are very few absolute rules in trading.

Garth

[BakedWafer]

I am currently adding some different rules to my system which could allow me to see a serious increase in gains, far above the 80% + in backtest I was seeing. I will keep you all posted.

-J

[spike500]

Gspiker,

I agree, i talk about indicators. If we are talking systems than the situation is completely different.
I talk about the traditional indicators (macd, rsi, cci ….) where you calculate the value on the close of a bar by default.
Example: RSI has to be higher than 80 to go short. Before the close of a bar you get a quote that is high enough go generate a sell signal . But If you are unlucky and the close of the bar is much lower, the signal to go short can disappear again. In a backtesting situation you would never have had that short signal because the only value that you have to calculate with is the close of that bar. You can only anticipate if you would take the high instead of the close, because that value can be known before the close of the bar. But that signal will, in backtesting, only be generated at the close of the bar because backtesting goes from closed bar to closed bar.

I think that the difference in point of vue between us comes from a different approach of building a system. We think probably in a different way and have therefore probably a completely different situation.
That’s why it is sometimes interesting to know what others do, because it can give you new ideas, it can broaden your insight in the market.


PS: I hope my English is not too bad. But I must say that I don’t see to many grammatical errors in my postings.




By the way it is at this moment 8.30 in the morning over here, so don't think that i can't sleep.

[gspiker]

Spike,

PS: I hope my English is not too bad. But I must say that I don’t see to many grammatical errors in my postings.

Your english is great! I hope you keep posting, I enjoy hearing what you have to say.

Garth

[spike500]

Just some thoughts, maybe usefull.

I do not believe that prices are “at random”. To me there is a logic in the movements of the quotes. The behaviour of all investors together generate the moves in stock markets. So if you can quantify the behaviour of the mass you can predict more or less what will happen. The accuracy of the prediction diminishes exponentially in relation to the factor “time”.

I can visualize on my screens more or less what the mass probably will do. In this way I can anticipate and prepare myself in advance to take action. This is a terrible advantage to the other investors. I can take position and once the herd of investors starts to move they build up the momentum that I need to make money.
It's also important to eliminate the "noise" in your calculations. But the problem is: how can you do this, and what's noise and what isn't?
The answer is simple: if you see that, after filtering, your indicator gave good results then this means that you did a good job.
The only problem is that it is a hell of a job to get the noise out. Especially if you don't know where to start.
Investing is not easy.

My hobby:

[TraderJO]

spike500,

I also wanted to say that your English & grammer are excellent. In fact, better than a lot of Americans these days.

[soylent]

Spike --

That's one sexy car!

This is all I've worked up to...

-c

[spike500]

BakedWafer,

could you do the ultimate test for your system?

Take data from another period than the one you backtested and apply the system WITHOUT MAKING ANY CHANGES to the parameters or formulas you use. Check than if the results are still as good. If they are good it will mean that it might be a good system. If the results are not good this will proof that your system was optimized for the testing data and therefore it will only work on the backtested data. This would mean that the system is worthless.

You can also test it on another commodity.

I think this is really important. It is useless to have a system that gives good results in backtesting but gives totally different returns in real trading.
I am critical because i want to avoid that other people losse money on the same mistakes that i made years ago. I have the impression that you develop a system and trade it without profoundly testing it. The fact that within 1 or 2 days you have a new and better version confirms to me these thoughts.
You are motivated and enthusiastic about trading, but be carefull.
In daytrading you can loose lots of money in a very short time.

[mooseman]

Spike
I'm thinking the size or length of time for your development period will affect your testing to see if you curve fitted.

I understand your point well if for instance you develop a system testing a week or so of day trading, or a month or so of swing trading. Then apply it to two or three years worth of data. Results will be different unless the longer period in sum reflects similar patterns in the shorter period.

Or if you develop a biased system using more bullish trends and test over more bearish trends, the results have to be different. My question is how much variation is to be expected? You seem to say if results over different time frames are different the system is curve fitted and it is bad. I think there will always be some variance. The point might be to define acceptable variance size and look for large variances then get a hande on why. Then change the system or bias to use it during similar periods like the development conditions - if you can identify such.
Dave

[Gavishti]

The markets do three things: go up, go down, go sideways. There are periods of high, low, and medium volatility. And then there are the outliers (price spikes or shocks). Your system must be able to at least survive all of the various combinations of these possibilities. And of course it must make a profit on at least one of the scenarios. What works on daily bars will most likely not work on intraday. Be sure that your tests are realistic. Many people develop systems that work great in the backtester but fail in realtime. Remember that realtime data is a stream, not completed bars.

[BakedWafer]

I already did this a long time ago. Since a 6-month period is obviously too short, I ran it back 5 years "spot-checking" too see if the system would of done what it was supposed to do, it did. The interesting thing is it only works on stocks that move around a lot. So in 2000 on QQQ, she was monster, taking huge profit gains.

Originally posted by spike500
BakedWafer,

could you do the ultimate test for your system?

Take data from another period than the one you backtested and apply the system WITHOUT MAKING ANY CHANGES to the parameters or formulas you use. Check than if the results are still as good. If they are good it will mean that it might be a good system. If the results are not good this will proof that your system was optimized for the testing data and therefore it will only work on the backtested data. This would mean that the system is worthless.

You can also test it on another commodity.

I think this is really important. It is useless to have a system that gives good results in backtesting but gives totally different returns in real trading.
I am critical because i want to avoid that other people losse money on the same mistakes that i made years ago. I have the impression that you develop a system and trade it without profoundly testing it. The fact that within 1 or 2 days you have a new and better version confirms to me these thoughts.
You are motivated and enthusiastic about trading, but be carefull.
In daytrading you can loose lots of money in a very short time.

[derekg]

Good discussion here. Lot of great points being brought up.

[spike500]

Originally posted by mooseman
Spike
I'm thinking the size or length of time for your development period will affect your testing to see if you curve fitted.

I understand your point well if for instance you develop a system testing a week or so of day trading, or a month or so of swing trading. Then apply it to two or three years worth of data. Results will be different unless the longer period in sum reflects similar patterns in the shorter period.

Or if you develop a biased system using more bullish trends and test over more bearish trends, the results have to be different. My question is how much variation is to be expected? You seem to say if results over different time frames are different the system is curve fitted and it is bad. I think there will always be some variance. The point might be to define acceptable variance size and look for large variances then get a hande on why. Then change the system or bias to use it during similar periods like the development conditions - if you can identify such.
Dave

Mooseman,

i agree for 100 % with you. There may be, and therewill be differences. But the differences must be acceptable. A difference of 50% or more is not acceptable.
Testing should be done over years of data to be sure that you hav had all kind of markets.

[spike500]

Originally posted by BakedWafer
I already did this a long time ago. Since a 6-month period is obviously too short, I ran it back 5 years "spot-checking" too see if the system would of done what it was supposed to do, it did. The interesting thing is it only works on stocks that move around a lot. So in 2000 on QQQ, she was monster, taking huge profit gains.

I'm glad for you that you did that. 5 years is long enough to have a mix of all kind of situations. My impression that you were going too fast was apparently wrong.

[soylent]

Curious:
When building day trading strategies (not scalping for a pt but looking for a the larger move on 20m or 60m bars -- assuming you look @ that) -- have you ever incorporated 'sentiment' into your backtesting?

Not necessarily bullish/bearish newsletters but Adv Decline, Volume up/down, TRIN or VIX?

Or, for that matter, have you tried using any of these when scalping?

-c