Menu
Sister Sites

Options

Bookmark Thread

First Page First Unread Last Page Last Post

Print Thread

Similar Threads

Brokers with low Stop Levels (CFD or other acc types) 1 reply

Two types of trends and one follows after the other 0 replies

Needed Median, Highest, and Lowest Price Range Indicator 1 reply

Median of a Price Series 8 replies

Need an indicator that calculates the median price 3 replies

Attachments: Close Price VS Median Price (or other price types)
Exit Attachments

Close Price VS Median Price (or other price types)

There is one consideration that I have not seen so far in this thread:
Highs and lows are always seen at top and bottom of any candle
However Close and open will reverse when compared to its opposite:
Close is immediately below High in bullish candle, whereas
Close is immediately over Low in bearish candle.
What difference does it make?
Close is always close to Highs in Bullish candle and always close to Lows in Bearish ones, which in my opinion gives a closer picture to reality, if you want an average closer to the extremes, as opposed to a real down the middle average in which case Median is your choice.

Dan
Quoting Proximus
so there is no volatility difference between X second of a candlestick from Y second of a candle stick (representing M1 candles by using real HQ tick data to calculate it),on average atleast
I think it was you how said that what looks random to some trader looks orderly to another with more advanced methods. This is the perfect example. Your methodology for doing this analysis was very poor, and thus your result looks random.

Here is the same test done the proper way with no manual intervention, with two years of tick data. I have 7 years, but I don't want to wait 5 minutes until they are loaded. I used EUR/USD ticks from FXCM. I also double-checked the result with OANDA/Dukascopy ticks (no difference). After resampling to 1 sec interval the number of ticks increases, since ticks are added for seconds which don't have any, that's why 41M ticks were loaded, but 51M were counted on the S1 chart.

Top: total move grouped by 1 second / 10 second intervals
Bottom: total move grouped by 1 minute / 10 minute intervals

Notice the beautiful spike at second 0 from each M1 candle / minute 0 from each H1 candle. Also notice the spikes at second 30 and minute 30. On the minute chart you can also see periodic spikes at the start of M15 and M5 candles (minute 0, 5, 10, ...). These are created by M5/M15 traders (retail and institutional)

https://dl.dropboxusercontent.com/u/...nge_bucket.png
Thanks Adal! I was about to do this after reading this thread. You saved 30 min of my day.
For what I can contribute to this thread I agree that the difference between close and median price based MA is pretty small. I think therefore that if a strategy makes very different results using one or the other carries a high suspicion of overfitting.
No greed. No fear. Just maths.
Quoting Adal
{quote} I think it was you how said that what looks random to some trader looks orderly to another with more advanced methods. This is the perfect example. Your methodology for doing this analysis was very poor, and thus your result looks random. Here is the same test done the proper way with no manual intervention, with two years of tick data. I have 7 years, but I don't want to wait 5 minutes until they are loaded. I used EUR/USD ticks from FXCM. I also double-checked the result with OANDA/Dukascopy ticks (no difference). After resampling to 1...

  1. you measured the (BID+ASK) /2 , while i measured the ABS(BID[i]-BID[I+1])
  2. you only counted the nr of ticks, while i measured the difference between the BID prices (i could have done it on ASK too, but it would have been the same) ,which is the volatility between ticks
  3. THIS DOESNT PROVE that volatility is higher at the first sec/min/hour of a bar, it only shows that there are more ticks there.

My test shows that the distance between ticks is random, so probably, combined with your method it may show that volatility is too, because if its random and there are more ticks at first second/min, then indeed the volatility is different there.



You should have also measured the ABS(BID[i]-BID[I+1]) and correlate the distances between one another, especially the first/last seconds with the rest of the seconds.If the correlation is very high like i observed, then indeed the CLOSE PRICE is better.

I will also check your claims, and see if the tick count is different ,and you should check my claims too.

Its peer review, if we want a real scientific method here.

Quoting PipMeUp
Thanks Adal! I was about to do this after reading this thread. You saved 30 min of my day. For what I can contribute to this thread I agree that the difference between close and median price based MA is pretty small. I think therefore that if a strategy makes very different results using one or the other carries a high suspicion of overfitting.
Precision is important, no matter how small is the difference, we need to choose the best one always.
"There's a sucker born every minute" - P.T. Barnum
Quoting Proximus
{quote} you measured the (BID+ASK) /2 , while i measured the ABS(BID-BID[I+1]) you only counted the nr of ticks, while i measured the difference between the BID prices (i could have done it on ASK too, but it would have been the same) ,which is the volatility between ticks THIS DOESNT PROVE that volatility is higher at the first sec/min/hour of a bar, it only shows that there are more ticks there. My test shows that the distance between ticks is random, so probably, combined with your method it may show that volatility is too, because if its...
I know exactly what you measured.

I measured ABS( MID[i+1] - MID[i] ), or more exactly ABS ( (BID[i+1] + ASK[i+1]) / 2 - (BID[i] + ASK[i]) / 2 )

The ABS difference between MID[i+1] and MID[i] that you missed is done by the abs(s_df.diff()) call - http://pandas.pydata.org/pandas-docs...rame.diff.html

diff(x) where x is a vector returns a new vector { x[1] - x[0], x[2] - x[1], ... x[n] - x[n - 1] }

So esentially the same thing as you did, but I used the mid price instead of just the bid.

And I didn't count the number of ticks. I added together all the ABS (...) from above for each second (0, 1, ...), so the histogram shows the TOTAL MOVEMENT done during each second/minute, not the number of ticks as you assumed.

So I measured what you intended, but I didn't use correlation, I used total abs movement since it's more graphical and easier to see.
Quoting Proximus
Precision is important, no matter how small is the difference, we need to choose the best one always.
Best with respect to what? Mid-price will lead to a smoother result since it is already an average. But an additional averaging increases the lag a little bit.
Low/High are not evenly spaced in time. Is this a problem for your application or not?
No greed. No fear. Just maths.
  • Post #27
  • Quote
  • Edited 1:35pm Sep 4, 2014 1:23pm | Edited 1:35pm
Quoting Adal
{quote} I know exactly what you measured. I measured ABS( MID[i+1] - MID ), or more exactly ABS ( (BID[i+1] + ASK[i+1]) / 2 - (BID + ASK) / 2 ) The ABS difference between MID[i+1] and MID that you missed is done by the abs(s_df.diff()) call - http://pandas.pydata.org/pandas-docs...rame.diff.html diff(x) where x is a vector returns a new vector { x[1] - x[0], x[2] - x[1], ... x[n] - x[n - 1] } So esentially the same thing as you did, but I used the mid price instead of just the bid. And I didn't count...
Ok i see, i`m not really familiar with that programming language, so i must have been read it wrong.

I will add them together aswell just to check it again, as the correlation is a linear tool as i understand, and my data was non-linearly picked.

Quoting PipMeUp
{quote} Best with respect to what? Mid-price will lead to a smoother result since it is already an average. But an additional averaging increases the lag a little bit. Low/High are not evenly spaced in time. Is this a problem for your application or not?
Best in respect to reality, that price which represents its time period with the most accuracy.One tick is represented best by the (BID+ASK)/2, One M1 candle is represented best by ....?
Yes i know that the mid price is the best one, but MT4 unfortunately doesnt have tick chart, so the best i have is HQ M1 data.

Looks like i just have to separately analyze the data on a tick chart for best performance, as it seems that any other method will have some errors in it.And even then it will depend heavily on the quality of the tick data.
"There's a sucker born every minute" - P.T. Barnum
Quote
Best in respect to reality, that price which represents its time period with the most accuracy
Then it is the close price because it is the most recent one!

Did you read the papers I posted here? http://www.forexfactory.com/showthre...54#post7713954
They may make you change your mind about "tick is more acurate". The good news is that the filter proposed by Corsi is easy to implement.
No greed. No fear. Just maths.
Quoting PipMeUp
{quote} Did you read the papers I posted here? http://www.forexfactory.com/showthre...54#post7713954 They may make you change your mind about "tick is more acurate".
Funny you would say that, because both of those papers start with tick data (not with candles). In fact, I've skimmed though hundreds of academic papers related to finance and 99% of them worked with tick data. And not even a single one of them had candles in them.
Just for the sake of it, I've also plotted the TOTAL NUMBER OF TICKS (red, right axis) on top of the TOTAL (ABS) MOVEMENT (blue, left axis). Same data as before. Almost two years of EUR/USD ticks from FXCM.

As expected, there is an almost perfect correlation between the two. No surprise here, many papers also found this result - volatility (total movement) is directly related to the volume, and the tick count is a very good proxy for volume on markets where volume information is hard to get (like forex).

https://dl.dropboxusercontent.com/u/...ick_bucket.png
Quoting Adal
{quote} Funny you would say that, because both of those papers start with tick data (not with candles). In fact, I've skimmed though hundreds of academic papers related to finance and 99% of them worked with tick data. And not even a single one of them had candles in them.
Really? I read so many papers too... that's the first one where ticks are used. Usually that's daily or weekly or even longer term. We may not search the same kind of info.
No greed. No fear. Just maths.
I watch quantitative papers on arxiv/ssrn.

Yes, some use daily or hourly close data. But not even one I've seen used OHLC. So if they don't use ticks, they use only C.
Quoting PipMeUp
{quote} Then it is the close price because it is the most recent one! Did you read the papers I posted here? http://www.forexfactory.com/showthre...54#post7713954 They may make you change your mind about "tick is more acurate". The good news is that the filter proposed by Corsi is easy to implement.
Tick data is the base data, if the tick data is not accurate then how can the candles be accurate ?

It's like if you build a house and the foundation is made by jelly instead of concrete

In both cases its the median, or (MEDIAN_BID +MEDIAN_ASK)/2 if its based on candlesticks, but in MT4 only MEDIAN_BID because there is no historic ask price there.
It cant be the close because the close price misses out 99% of the data, as if there is a volatile event at second 34 of a M1 candle, if you were to look at the close price it doesnt averages in the volatility.

In all cases its the median, i know this, but if Adal's claim is true, then we have a problem because there is an inconsistency at the CLOSE price, some EA's really use candles instead of ticks.

Quoting Adal
{quote} Funny you would say that, because both of those papers start with tick data (not with candles). In fact, I've skimmed though hundreds of academic papers related to finance and 99% of them worked with tick data. And not even a single one of them had candles in them.
Quoting PipMeUp
{quote} Really? I read so many papers too... that's the first one where ticks are used. Usually that's daily or weekly or even longer term. We may not search the same kind of info.
Depends, if Adal is trading on arbitrage then logically tick data is better for him, but for econometric studies, they usually used M1 data do daily data.
I think it depends on the methodology.

However i am looking at microstructure so for me the tick data is better now
"There's a sucker born every minute" - P.T. Barnum
  • Post #34
  • Quote
  • Last Post: Sep 4, 2014 5:20pm Sep 4, 2014 5:20pm
Quoting Proximus
{quote} = I will add them together aswell just to check it again,=
So far the 2 sums of it are: 2.40554999999937000000 (side seconds) > 2.37033999999944000000 (mid seconds)

Which in pipettes are, rounded : 240555 pipettes and 237034 pipettes, and the difference is only 3521 pipettes which is too small, but i think its still random, however i will segment it into 10 second groups, when i`ll have a little time and check it more deeply.
"There's a sucker born every minute" - P.T. Barnum
0 traders viewing now