Market facts | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

GBP Patterns, Facts, Trivia 133 replies

The facts 4 replies

Market & Trading Facts 60 replies

Pairs objective facts 0 replies

Facts about technical indicators 10 replies

Attachments: Market facts
Exit Attachments

Market facts

  • #1
  • First Post: Edited 4:18pm Dec 7, 2011 3:29pm | Edited 4:18pm
I thought this is the kind of thead I would want to read so let's see if this will work.

So, please post only proven (scientific or otherwise proven) facts about the market. The more insightful the better.

Feel free to quote posts or other sites but please, no opinions of any kind, just facts.
  • #2
  • Edited 3:45pm Dec 7, 2011 3:34pm | Edited 3:45pm
I'll start..

Some fx pairs are more correlated than others. By trading two non-correlated pairs (or any markets) you will diversify your profit curve is usually smoother (drawdowns don't cluster as they do in closely correlated instruments)

Correlations for fx pairs can be found for example here

http://www.forexticket.co.uk/en/tools/01-01-correlation
  • #3
  • Edited 3:54pm Dec 7, 2011 3:44pm | Edited 3:54pm
Here is another:

Market is not normally distributed. If someone else would like to provide some good links (including data about kurtosis) feel free to do so

edit: Here is one link related to this Stock Market Prices Do Not Follow Random Walks: Evidence from a Simple Specification Test
  • #4
  • Dec 7, 2011 3:51pm Dec 7, 2011 3:51pm
continuing from previous post..

So as market is not normally distributed, how it is distributed and how does this affect our trading?

Here is one link with some distribution curves and some more information about how ít affects actual trading results http://www.automated-trading-system....-distribution/

Here is another related to the first post, this also talks about skew, kurtosis and more..

http://www.automated-trading-system....ing/#more-1116
  • #5
  • Dec 7, 2011 4:04pm Dec 7, 2011 4:04pm
Different currenct pair turnovers vary quite a lot. Turnovers and a lot of more data can be found from "Triennal Central Bank study" by bank for international settlements.

http://www.bis.org/publ/rpfxf10t.pdf
  • #6
  • Dec 7, 2011 4:10pm Dec 7, 2011 4:10pm
One more, then I'll see if someone else would like to contribute..

CFTC publishes a weekly report called "Commitments of traders". You can use this report to see how much different types of players market have invested to different markets (for example currency futures).

http://www.cftc.gov/MarketReports/Co...ders/index.htm

Report looks like this
http://www.cftc.gov/dea/futures/financial_lf.htm

This site publishes charts with report data below the chart
http://timingcharts.com/
  • #7
  • Dec 7, 2011 5:35pm Dec 7, 2011 5:35pm
Hedge Funds along with other leveraged spec funds determine where the market will go. All they have to do is trade and wait. Trading profitably truly is an information game. Retail traders on average are uninformed and lose their money. The reason retail traders continue to trade is mostly just wishful thinking. Retail trading now accounts for ~ 10% of fx daily turnover...

Source: Carol Osler's most recent research paper. She even mentions Oanda's "profitable" traders. Good read.
Attached File(s)
File Type: pdf Norges_Bank_Working_Paper_2011_10.pdf   844 KB | 592 downloads
  • #8
  • Dec 7, 2011 5:55pm Dec 7, 2011 5:55pm
market closes above or below open most of the time

so then P(A)= Ʃ/n
AVT INVENIAM VIAM AVT FACIAM
  • #9
  • Edited 7:24pm Dec 7, 2011 7:13pm | Edited 7:24pm
True Range is not symmetrical distributed but Log-Normal distributed.

Source: http://www.forexfactory.com/showthre...56#post2719756


btw Here is the real data test to prove that forex market does not follow normal distribution.
  • #10
  • Dec 8, 2011 12:42am Dec 8, 2011 12:42am
Volatility is clustering.

It's also very interesting that autocorrelation is "insignificant".
(autocorrelation means correlation of asset price compared to previous price)

One source:
http://www.proba.jussieu.fr/pagepers...clustering.pdf

So basically volatility follows volatility but up or downmoves don't follow each other.
  • #11
  • Dec 8, 2011 12:47am Dec 8, 2011 12:47am
This is also interesting:

Web search queries predict stock market trading volume

http://www.technologyreview.com/blog/arxiv/27281/

(paper)
http://arxiv.org/abs/1110.4784

This is even more interesting when combined with the other fact that volume and volatility are correlated (see previous post linked paper). Therefore web search queries predict both volume and volatility.
  • #12
  • Dec 8, 2011 1:08pm Dec 8, 2011 1:08pm
Time-varying volatility explains a significant part of the excess kurtosis of price distributions. Inversely, when accounting for conditional volatility as opposed to unconditional volatility, the returns distribution becomes more "normal".

Also, as a general rule of thumb: the greater the timeframe, the more normal the distribution (e.g. 5 minute returns show higher kurtosis than monthly returns.)
The nail that sticks out gets hammered back in
  • #13
  • Dec 8, 2011 1:14pm Dec 8, 2011 1:14pm
Quoting Noloqy
Time-varying volatility explains a significant part of the excess kurtosis of price distributions. Inversely, when accounting for conditional volatility as opposed to unconditional volatility, the returns distribution becomes more "normal".

Also, as a general rule of thumb: the greater the timeframe, the more normal the distribution (e.g. 5 minute returns show higher kurtosis than monthly returns.)
Although OP said no opinion here but I strongly disagree with your statement that the greater the timeframe, the more normal the distribution.
  • #14
  • Dec 8, 2011 2:03pm Dec 8, 2011 2:03pm
Quoting ForexQuant
Although OP said no opinion here but I strongly disagree with your statement that the greater the timeframe, the more normal the distribution.
I don't mind discussing about facts if the argument can be decided by facts, that's how science is made

It would be interesting to see some data about this as I also don't think the above is true (the higher the timeframe, more normal the distribution), in fact I think it's opposite - the lower the timeframe, the more normal the distribution but I don't have any data to back this up.. if anyone else has then please post it.

I could whip up some data later if I have time..
  • #15
  • Dec 8, 2011 3:23pm Dec 8, 2011 3:23pm
Quoting ForexQuant
Although OP said no opinion here but I strongly disagree with your statement that the greater the timeframe, the more normal the distribution.
Maybe you're right - there's little room for opinion, so lets examine the facts. I'll look it up on scholar.google.com, and if I can't find a source I'll analyze the data myself. I encourage you to do the same

By the way, Mikkom, great idea setting up this thread.
The nail that sticks out gets hammered back in
  • #16
  • Dec 8, 2011 3:56pm Dec 8, 2011 3:56pm
I just did some quick tests.

Monthly returns go back to 1990; 5 minute returns go back to Jan 2011. Of course, one could object to that, though I believe the results hold in general.

Considering the log returns in more detail:

5 minute:
11471 obs; .3945 skew; 13.68 kurt
Month:
261 obs; -.3549 skew; 3.97 kurt

Jarque Bera test would reject normality for both.


R = (S_{t} - S_{t-1}) / S_{t-1}
r = ln(S_{t}/S_{t-1})
Attached Image(s) (click to enlarge)
Click to Enlarge Name: eurusd5min.png Size: 21 KB
Click to Enlarge Name: log eurusd5min.png Size: 20 KB
Click to Enlarge Name: eurusdmonth.png Size: 24 KB
Click to Enlarge Name: log eurusdmonth.png Size: 23 KB
The nail that sticks out gets hammered back in
  • #17
  • Dec 9, 2011 2:30am Dec 9, 2011 2:30am
Quoting Noloqy
I just did some quick tests.

Monthly returns go back to 1990; 5 minute returns go back to Jan 2011. Of course, one could object to that, though I believe the results hold in general.

Considering the log returns in more detail:

5 minute:
11471 obs; .3945 skew; 13.68 kurt
Month:
261 obs; -.3549 skew; 3.97 kurt

Jarque Bera test would reject normality for both.


R = (S_{t} - S_{t-1}) / S_{t-1}
r = ln(S_{t}/S_{t-1})
I think you are right. I have done a normality analysis on 5min, 15min and 60min returns. I also divided the data into 3 sessions in order to reduce the seasonality effect. The results speak for itself, 3 sets of data indicated that the kurtosis reduces on higher timeframe.

Once again this proves that what we think may not always right without support of numbers. I must apologize for what I have said earlier.
Attached Image(s) (click to enlarge)
Click to Enlarge Name: EU_5Min.PNG Size: 61 KB
Click to Enlarge Name: EU_15Min.PNG Size: 60 KB
Click to Enlarge Name: EU_60Min.PNG Size: 59 KB
  • #18
  • Dec 9, 2011 2:59am Dec 9, 2011 2:59am
Great stuff :-) I was thinking more about fat tails than kurtosis in general.. The distribution above seems to be such that I think kurtosis represents peakedness but not that much fatness of tails.. Any idea what would be a good measurement for fatness of tail in general?

Thanks a lot for the data!

edit: here is some stuff from higher timeframes
http://www.andreassteiner.net/perfor...p%3B_Fat_Tails

Quote
Results from analyzing stock returns: Daily, weekly and monthly returns have significant leptokurtosis according to the conventional test based on kurtosis. Leptokurtosis test based on L is rather significant for certain daily and weekly returns, but less so for monthly returns. Further, tests based on peakedness indicate a decline in peakedness according to the increasing length of the time interval. As weekly and monthly returns are sums of daily returns, they become more and more normal according to the Central Limit Theorem (CLT implies that the distribution of an average will tend to be Normal as the sample size increases, regardless of the distribution from which the average is taken except when the moments of the parent distribution do not exist.). That’s the reason, weekly and monthly returns tend to have less leptokurtosis than daily returns.
  • #19
  • Dec 9, 2011 3:09am Dec 9, 2011 3:09am
Quoting mikkom
Great stuff :-) I was thinking more about fat tails than kurtosis. The distribution above seems to be such that I think kurtosis represents peakedness but not that much fatness of tails.. Any idea what would be a good measurement for fatness of tail in general?

Thanks a lot for the data!

edit: here is some stuff from higher timeframes
http://www.andreassteiner.net/perfor...p%3B_Fat_Tails
Since the kurtosis reduces on higher timeframe, is that also means the returns will eventually normal distributed on extreme high time frame? Anyone else got data to prove it?
  • #20
  • Dec 9, 2011 4:51am Dec 9, 2011 4:51am
As Mikkom noted, volatility follows volatility.

I already mentioned that a primary driver of the kurtosis we just observed is time-varying volatility.

This could lead us to hypothesize that the variance of 5 minute volatility is higher than the variance of monthly volatility. This hypothesis makes intuitive sense, as on NFP friday, volatility is much greater than during the Japan session. The longer the timeframe, the less the influence of such seasonal effects, and the more closely the volatility will float around the unconditional volatility.

Anyone interested in researching this in more detail, could run a GARCH(1,1) on returns of different timeframes (e.g. 5 minute versus daily,) and measure the variance around the unconditional mean.

Does anyone know if it is possible to insert parts of latex text in this forum?
The nail that sticks out gets hammered back in
0 traders viewing now