Can you tell with naked eyes? | Page 7 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

If you could go back - one Forex thing you'd tell youself? 64 replies

Trade with Naked Eyes 1,757 replies

Looking on a naked chart, can you pick the best strategy? 32 replies

Can you tell how the fundamentals are from the monthly chart? 3 replies

Trading as naked as you can get 0 replies

Attachments: Can you tell with naked eyes?
Exit Attachments

Can you tell with naked eyes?

  • #121
  • Apr 15, 2010 9:49pm Apr 15, 2010 9:49pm
Great. Thanks for your comment. I find that you are the champion over there. As my understanding of your comment, their fake charts there look really fake. And the fake charts here is indistinguishable for you.

Quoting Trad3r
That MIT study is a crap, the smoothed charts in "mandrill", "bulls", "bears" and other are the real, in "reindeer" is the opposite and in other is the chart with the fast price movement... I can't distinguish if the charts posted here are fake or real and I'm first in one contest and second in other two, you can see it here: http://arora.ccs.neu.edu/v4/tool/menu.jsp judge for yourself.
  • #122
  • Apr 15, 2010 10:00pm Apr 15, 2010 10:00pm
To create "freshness" in the experiment, they may get a random start point on the price chart to present. It is highly possible that some chart-portion are repeated, and this repetition manifest the memory factor. This problem has not been addressed in their paper.

But I just realised from trad3r, he is one of the champion there. The MIT fake charts look really fake and can't compare to the quality of fakeness on this thread. So this is another critical weakness of the study.

Quoting FXEZ
A variable taken from a gaussian distribution would be called gaussian random. A variable taken from a completely distribution would be called completely random.




I realize that there may be a language barrier in play but read the quote I posted again from the paper - fresh means new, as in data that the subject hasn't seen before. Personally I'm not employed by MIT to handle such complaints about their scholarly papers, and if I were I would claim my salary as being far too low for the aggrivation. If you have a problem with the research...
  • #123
  • Apr 15, 2010 10:45pm Apr 15, 2010 10:45pm
I just played the spot the fake game at ARORA website and it's fun.

Now let's do some math here after the fun.

I use the game named Reindeer for example. There are 40 trials in this game. Each success guess will add approximately 1,000 points while the wrong guess will minus about 1,000 points. The top 3 participant scored above 40,000 points and the initial point is 10,000 points, therefore we can say that they have approximately 35 success guess with 5 wrong guess.

So we plug the number into binomial distribution to check the probability to have at least 35 success guess for those who has no skills. In excel the formula will be 1 - Binomdist(34, 40, 0.5, true) = 0.00007% (Probability of success in single trial is 50% for those who has no skill)

So the results speak for itself. The chances that the top 3 scorer are the result of luck (No skill) is just 0.00007%, which means you need to have conduct 10,000,000 trials so that 7 trials will have hit above 40,000points.

What about if someone can hit that 1 in 100 trials? Can we measure how far his skill skew away from luck? So in this time the chances is 1% and we do some manipulation in excel with the above formula, 1 - binomdist(34, 40, 0.704, true) = 1%. In order words, we can say that the chances that his probability of success in single trial is 70.4%, which is 20.4% above luck!

Since I dont think Trad3r has played 100 times in this particular game, therefore i concluded he skill give him at least 20% advantage above random (luck) to pick the real chart.

However please do not confuse that the ability to spot the real chart has something to do with profitability. You can spot it right 70% or even 90% but it does not mean your winning rate is 70% or 90%. Your winning rate is depends on the probability that the price hit your target before your stop loss, not anything else.
  • #124
  • Apr 15, 2010 10:59pm Apr 15, 2010 10:59pm
Quoting kk007
I don't have a direct practical application initially. It is my belief that if I dig an issue interesting to me deep enough, some benefits will eventually come. A direct practical application is not what I am looking for to begin with this thread. Nevertheless, this one below I just come up with may be a practical application?

False Positive (Type I problem): My trading system trigger a trade signal based on a historical price movement, but the price does not move as predicted.

False Negative (Type II problem): My trading system...
Unless your randomized chart has similar probability distribution with real market (High kurtosis and fat tails), else I dont think you can use a randomized chart to check the validity of your system.

Like I mentioned previously it is the price distribution that matter, not randomness.
  • #125
  • Apr 15, 2010 11:09pm Apr 15, 2010 11:09pm
Quoting ForexQuant
Unless your randomized chart has similar probability distribution with real market (High kurtosis and fat tails), else I dont think you can use a randomized chart to check the validity of your system.

Like I mentioned previously it is the price distribution that matter, not randomness.
Precisely, because the randomized charts cannot simulated exactly the real price movement, the trading system should not be triggered.

If the simulated price movement derivated from a real market price and the trading system still trigger a signal, the system is bad.
  • #126
  • Apr 15, 2010 11:20pm Apr 15, 2010 11:20pm
Quoting kk007
Precisely, because the randomized charts cannot simulated exactly the real price movement, the trading system should not be triggered.

If the simulated price movement derivated from a real market price and the trading system still trigger a signal, the system is bad.
Nope. If i can figure out the parameter of price distribution of the market, then i can use a random number generator to simulate the real price movement. Mandelbrot did it by using his multifractal trading time technique but just too bad i hardly understand it.
  • #127
  • Apr 15, 2010 11:26pm Apr 15, 2010 11:26pm
Quoting ForexQuant
Nope. If i can figure out the parameter of price distribution of the market, then i can use a random number generator to simulate the real price movement. Mandelbrot did it by using his multifractal trading time technique but just too bad i hardly understand it.
Your method make distribution assumptions on the market data, but market data cannot be exactly modelled by these assumptions.

But this is not my main point. My main point is the faker is the chart, the more the trading system should not send out trading signal.
  • #128
  • Apr 15, 2010 11:37pm Apr 15, 2010 11:37pm
Quoting kk007
Your method make distribution assumptions on the market data, but market data cannot be exactly modelled by these assumptions.
It can be done. I bet 100 bucks for it.

The main reason that LTCM failed not because they used randomized data to test their model but they use wrong distribution assumption to model the market.
  • #129
  • Apr 15, 2010 11:55pm Apr 15, 2010 11:55pm
Quoting ForexQuant
It can be done. I bet 100 bucks for it.

The main reason that LTCM failed not because they used randomized data to test their model but they use wrong distribution assumption to model the market.
Sorry, it is not something I am so interested that I will pay 100 bucks for. I don't know how can you verify that your simulation is exactly the same as the real.
  • #130
  • Apr 16, 2010 1:06am Apr 16, 2010 1:06am
Quoting kk007
Sorry, it is not something I am so interested that I will pay 100 bucks for. I don't know how can you verify that your simulation is exactly the same as the real.
How you verify that your randomized chart is not same as real chart?
How you verify that a fair coin simulation is as good as real fair coin?
How you verify that a fair dice simulation is not same as real loaded dice?

If you know the answers then you know my answer to your question. No offence but i find that no point to keep argue with you on this issue anymore since you are not open your mind enough to accept other people's opinion. While you asking proof from other people, why not you show proof to support yourself 1st?

You opened a few threads for this topic and I as well as others were just trying to help and shared our idea, but looks like you are looking for something that support your thought rather than looking for an alternative idea. I think I should stay on sideline and let some others to continue the discussion.
  • #131
  • Edited 2:43am Apr 16, 2010 1:58am | Edited 2:43am
KK007,

I think I understand where you're coming from, and I hope I have an answer that will satisfy you.

Real and randomly generated ('fake') charts are going to appear much the same, for at least two reasons. Firstly, fake charts are going to exhibit apparent trends (e.g. higher highs) and patterns (e.g. pinbars, wedges, head and shoulders, multiple bounces at certain levels, etc etc) just like real charts do, because trends and patterns - at least as the naked eye perceives them - will always be generated, from any stream of data, purely as a matter of course. Hence that in itself proves nothing, except possibly that how easy it is for the naked eye (to quote Taleb) to be 'fooled by randomness'.

The key lies in the detail. On a real price chart, trends are going to occur fractionally more often, and last, on average, fractionally longer; certain patterns and levels are going to provide subsequent directional bias fractionally better than 50/50. Couple that with all of the other 'contextual' data that a trader has access to, e.g. time-of-day volatility measures, behavior before and after news announcements, average daily ranges, multiple time frame analysis, correlation, fundamentals, using strength to pick which pairs to trade, ...... and no doubt dozens of other 'discretionary' variables, and possibly some trade management skills ....... put all of that together, and there's your edge.

The fractional differences (between real and fake charts) that I mentioned are, both qualitatively, and in terms of their frequency, too small for the naked eye to perceive. They need to be counted and analyzed, objectively, over significantly large amounts of data, especially if one is seeking to profit from a predominantly 'mechanical' approach. That's my second point: a mere 500 data point 'chart' generated by Excel is incapable of proving anything.

Hope that helps.
  • #132
  • Apr 16, 2010 2:31am Apr 16, 2010 2:31am
Hi Hanover, could you elaborate to me your "time of day volatility measure" thing ? What do you mean by that ? What is that when looking at the chart ?

Thanks, please PM me if you feel you might bring the topic off track.

Thank you very much.

mark
  • #133
  • Apr 16, 2010 2:52am Apr 16, 2010 2:52am
Quoting aiyahmarklah
Hi Hanover, could you elaborate to me your "time of day volatility measure" thing ? What do you mean by that ? What is that when looking at the chart ?

Thanks, please PM me if you feel you might bring the topic off track.

Thank you very much.

mark
All I mean is that certain pairs might tend to move greater distances during different sessions, and that by averaging this over a large number of days, it could provide some kind of statistical edge. For example, I think you'll find that USDCAD moves further, on average during the American session than it does during the Asian session. Hence, everything else being equal, it would make more sense to trade it during the American session.

You could run the same kind of analyses for different days of the week, to see if some days were better 'trading days' than others. And analyze behavior around certain high impact news announcements. And so on.

These are all examples of phenomena that might not be obvious by simply comparing real and fake charts, which was a point that I was attempting to make in my previous post.
  • #134
  • Apr 16, 2010 3:24am Apr 16, 2010 3:24am
Quoting hanover
All I mean is that certain pairs might tend to move greater distances during different sessions, and that by averaging this over a large number of days, it could provide some kind of statistical edge. For example, I think you'll find that USDCAD moves further, on average during the American session than it does during the Asian session. Hence, everything else being equal, it would make more sense to trade it during the American session.

You could run the same kind of analyses for different days of the week, to see if some days were better 'trading...
Ah I get it, thank you.
  • #135
  • Apr 16, 2010 4:38am Apr 16, 2010 4:38am
Thanks hanover,

Thanks for your understanding of my quest here. As, H.Rearden said "real good advice in here is very very rare", I think your this post is the most valuable one on this thread.

First, you know exactly what I want. Second, your experience is rich. More importantly, your language and effort in sharing your view is really good.

I believe many of the earlier posters have good intention to share, but a lot of them trying to convince me not to dig into the issue, saying that it is pointless. Nevertheless, I think as far as it is a keen concern for me, there is a point to continue the quest.

Your response hit my key concern. Few earlier posts have ever touch the key issue. I totally agree that the key lies in the details. And a number of "discretionary" factors does play a role. The edge you discribed is valid, and people using this edge is not playing a random game, as if trading on a fake chart.

I feel I am continue to benefit from digging the issue. The issue together with your post make me reflect the key elements in discretionary trading decision from a different angle. Your post signals the right direction to go, from my point of view. I hope the discussion on this thread onwards can give insight to some other people too.

Finally, as H.Rearden said, "the discussions drift into personal issues between members and their ego's", I think it is a journey require the tackling of many personal problems. The random charts may not be disturbing for a person, but it could be disturbuing for another.

kk007

Quoting hanover
KK007,

I think I understand where you're coming from, and I hope I have an answer that will satisfy you.

Real and randomly generated ('fake') charts are going to appear much the same, for at least two reasons. Firstly, fake charts are going to exhibit apparent trends (e.g. higher highs) and patterns (e.g. pinbars, wedges, head and shoulders, multiple bounces at certain levels, etc etc) just like real charts do, because trends and patterns - at least as the naked eye perceives them - will always be generated, from any stream of...
  • #136
  • Apr 16, 2010 4:41am Apr 16, 2010 4:41am
No offense, but your questions are very easy for me to answer.

1. How you verify that your randomized chart is not same as real chart?
i. Generate a randomized chart and put it on the left of the screen.
ii. Open the real chart and put it on the right of the screen.
iii. Compare the left and the right

2. How you verify that a fair coin simulation is as good as real fair coin?
3. How you verify that a fair dice simulation is not same as real loaded dice?
i. The concept behind throwing a coin or a dice is to produce 2 or 6 outcomes with equal probability.
ii. With a proper random number generator, you can simulate it easily. In Excel for the coin case for example, an equation =ROUND(RAND(),0) can do the trick. I think you know how to do it.
iii. For verification, generate a big number of occasions with the random number generator. And throw the same big number of coins.
iv. Compare the two outcomes using statistical tests for any significant differences.

For real market data, the variables behind the price movement cannot be estimated in terms of quality and quantivity, I don't think anyone can exhaustively get all the variables behind. Unlimited statistical analysis have been done on this, I think a statistical model capture 100% of the price movement do not exist. Perhaps, I am wrong; you may be an exception.

Best Wishes

Quoting ForexQuant
How you verify that your randomized chart is not same as real chart?
How you verify that a fair coin simulation is as good as real fair coin?
How you verify that a fair dice simulation is not same as real loaded dice?

If you know the answers then you know my answer to your question. No offence but i find that no point to keep argue with you on this issue anymore since you are not open your mind enough to accept other people's opinion. While you asking proof from other people, why not you show proof to support yourself 1st?

You opened a few...
  • #137
  • Apr 16, 2010 6:56am Apr 16, 2010 6:56am
If you know it's going to move, it's 'tradeable'.
  • #138
  • Apr 16, 2010 6:58am Apr 16, 2010 6:58am
How you verify that a fair coin simulation is as good as real fair coin?
  • #139
  • Apr 17, 2010 10:35pm Apr 17, 2010 10:35pm
The MIT aurora test is useless for me. I had never ever traded with line chart. They should have provided candlestick charts as well.
  • #140
  • Apr 17, 2010 10:58pm Apr 17, 2010 10:58pm
As I've said before....an inexperienced trader will see randomness whereas an experienced trader will see structure.
0 traders viewing now