We are currently performing analysis on data similar to the title.
We cannot explain the contents of the data in detail because of the rules, but the data was collected using the following methods.where y is the variable to be explained and x is the variable to be explained.Also, let t be the time.
At time t, y(t) is observed.A most recent explanatory variable x (T|t-100<T<t), (record x of the most recent 100 events) is collected.Also, the explanatory variable x is known to have long-term memory (strong correlation between times), and the event time t does not take the continuous value as expected in a normal time series analysis.Specifically, it is a discontinuous event that follows the Poisson point process.
Because the data is similar to the above, it does not take advantage of normal time series analysis like ARIMA model.Also, when you try to estimate OLS, there is a strong correlation between variables, so multiple collinearity is a problem and you cannot estimate the correct parameters.
The goal of this analysis is a parameterized estimate of the descriptive variable (not a prediction, but an identification problem).If you have any ideas, please let me know.
python r
I know all the basic time series analysis (models such as Arima, var, etc.Unlike normal continuous time series data, this data is discrete sampling.(meaning that the sense of observation time series is not equidistant.)In that respect, the analysis does not take advantage of the normal time series analysis.As an image, I think it's easy for you to imagine the very dirty data before the time-change theorem in the unsteady-point process is applied.
© 2024 OneMinuteCode. All rights reserved.