Standard deviation for irregularly spaced timeseries

K

kroxobor

Hi fellows,

Suppose I have a collection of irregularly spaced time series (tick data with time_obs(i+1) - time_obs(i) ranging from 2ms to 20ms). I want to calculate 1-minute standard deviation of returns.

The question is: should the number of observations be the same in the every sample of 1 minute series? i.e. for example, the first minute contains 50 observations and the second minute 100 has observations. After calculating standard deviations for both is it legal to make claims like volatility of price in the first minute is higher than in the second since std1 > std2?

Thanks in advance.
 
kroxobor said:
Hi fellows,

Suppose I have a collection of irregularly spaced time series (tick data with time_obs(i+1) - time_obs(i) ranging from 2ms to 20ms). I want to calculate 1-minute standard deviation of returns.

The question is: should the number of observations be the same in the every sample of 1 minute series? i.e. for example, the first minute contains 50 observations and the second minute 100 has observations. After calculating standard deviations for both is it legal to make claims like volatility of price in the first minute is higher than in the second since std1 > std2?

Thanks in advance.
More...

https://quant.stackexchange.com/que...of-a-sample-when-points-are-irregularly-space

GAT
 
use fill data

use the exact same last bar until new data replaces it. the purpose is to fill every slot for data, make it up. but you don't want it to influence, so make it the same. logical?
 
kroxobor said:
Hi fellows,

Suppose I have a collection of irregularly spaced time series (tick data with time_obs(i+1) - time_obs(i) ranging from 2ms to 20ms). I want to calculate 1-minute standard deviation of returns.

The question is: should the number of observations be the same in the every sample of 1 minute series? i.e. for example, the first minute contains 50 observations and the second minute 100 has observations. After calculating standard deviations for both is it legal to make claims like volatility of price in the first minute is higher than in the second since std1 > std2?

Thanks in advance.
More...
you should not be using time series methods at all. use the point process paradigm , where points are described by their inter-event durations
 
MarkBrown said:
use fill data

use the exact same last bar until new data replaces it. the purpose is to fill every slot for data, make it up. but you don't want it to influence, so make it the same. logical?
More...
you might be a good trader Mark but that's not good advice , the term for this is censoring in statistics
 
stochastix said:
you might be a good trader Mark but that's not good advice , the term for this is censoring in statistics
More...

let's say you have a series that updates every minute and a series that updates every five minutes to avoid the gap there is nothing "impacting" on the price to censor it. using the synthetic data fill method i described.
 
You must log in or register to reply here.
Back
Top