how to wash out noise in high frequency financial time series

  • Thread starter Thread starter jemnbo
  • Start date Start date
J

jemnbo

Joined
7/23/08
Messages
5
Points
11
hi , everyone, i am jemnbo from China, i have a question needed to solve, anybody can help me?
the question is , we got the high frequency financial series, such as open , high, low , close , and volume , commonly in technical analysis , or any other systems we used, we only care about close prices , but , if we reconclude the other four series , we find that the series is not flat as we want, there is so many nosie or jumps in the series , how can we wash out the noise?

i made a step forward , for example , based on the herd effect, i ignored the trade volumes less that 50% quantile , then regenerate the five series, i found that the series become flatter , but not access to the confidence level .

what would you do to the queation?:-k:-k:-k:-k:-k
 
May I ask you are you a Quant in some company? Is this the a real-world problem happened in your job? Very interesting... Thanks.

Claude.
 
Hey Jemnbo,

I guess you need some papers by Jeannette Woerner.
One of her papers is in this book edited by Shiryaev: Stochastic Finance.("Power and Multipower Variation: inference for high frequency data" pages 343 to 364)

Good luck for the olympic games!
Bastian
 
In technical analysis, what you want to do ("remove the noise/fluctuation") is called detrend. There are many way to detrend a series (adding moving average, etc)
After you detrend, what you will see is a more stationary, flat line.
 
Like Andy said, in time series analysis, you need to remove the trend in the data set and what you want after detrending is a random noise component which needs to be as close to stationary and zero-mean as possible.

In general signal processing's point of view, take speech enhancement for example, you may use spectral subtraction to remove noise component in the frequency domain and recover the "clean" component.

Good to have my first post here:)
 
claudewang said:
May I ask you are you a Quant in some company? Is this the a real-world problem happened in your job? Very interesting... Thanks.

Claude.
Click to expand...

yes , i work in a hedge fund set up lastly in china, maybe it is feminine, but we have made a step bravely.
 
Shawn said:
Like Andy said, in time series analysis, you need to remove the trend in the data set and what you want after detrending is a random noise component which needs to be as close to stationary and zero-mean as possible.

In general signal processing's point of view, take speech enhancement for example, you may use spectral subtraction to remove noise component in the frequency domain and recover the "clean" component.

Good to have my first post here:)
Click to expand...

answer to Andy and Shawn

thanks for your advices , i catch your meaning and had tried the methods before , such as MA , EMA , and some econometrica methods such as ARMA GARCH etc, as you known, they are all backward looking methods and fixed constructions , if we choose the backword looking method , that means you had jumped into the time lag loops, the signals you cleaned limited by the lag parameters you choosed and can not switch properly according to the conditons now. so i wonder if we can made a properly region switching ??
PS,, as i known , the buy or sell signals sent by technical analysis methods is so tardy. we can not take a action in time.
 

Similar threads

W
Financial Time Series in R
Replies
6
Views
4K
statquant
statquant
Back
Top Bottom