I have a low temporal frequency irregular dataset with a value available every 40 to 48 days. I have another set of time-series data over the same period at 12 day frequency. The pattern of the two datasets are related (but not the values), and I'd like to use the pattern of the 12-day dataset to interpolate the 40-48 day data to a higher temporal resolution (e.g. daily or 10 days). I'll be implementing the analysis in Python (or R would be fine too), but I'm not sure of which method I should be looking to use. Any advice would be much appreciated.
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
4
you can use temporal disaggregation methods please see this article: https://journal.r-project.org/archive/2013-2/sax-steiner.pdf
-
$\begingroup$ Nice answer. Could you provide more info from the link in your answer directly? this would help if the link expires $\endgroup$ Commented Jan 5 at 11:55
-
$\begingroup$ we have two kins of data high frequency(hf) and low frequency(lf).there are two mainly approach for increasing lf to hf. 1-according to another hf data 2-without care about another hf data. $\endgroup$– far hadCommented Jan 11 at 13:10
-
1$\begingroup$ We have two data types: high frequency (HF) and low frequency (LF). There are two main approaches to increase LF to HF: 1) regressing LF on HF (LF ~ HF) and 2) regressing LF solely on time (LF ~ t). The first method requires a close relationship with another HF dataset, while the second is independent of specific HF data. To illustrate, if GDP data is annual and you aim to make it high frequency, the ideal HF indicator for a close relationship would be the Consumer Price Index (CPI). $\endgroup$– far hadCommented Jan 11 at 13:17
-
$\begingroup$ Better to add it to the answer than as a comment $\endgroup$ Commented Jan 11 at 13:44
$\begingroup$
$\endgroup$
We deal with two types of data: high frequency (HF) and low frequency (LF). When we want to increase the frequency of LF data to match HF data, we have two main strategies:
1.Regression of LF on HF (LF ~ HF): This approach involves using the relationship between LF and HF data to estimate LF values at higher frequencies. For example, if we have monthly HF data and yearly LF data, we can use regression techniques to estimate LF values for each month based on the corresponding HF data.
2.Regression of LF solely on time (LF ~ t): In this method, we only consider the passage of time to estimate LF values at higher frequencies. It doesn't rely on specific HF data but rather on the assumption that LF data changes gradually over time.
To clarify, let's take an example: Suppose we have yearly GDP data (LF) and we want to make it high frequency. If we choose the first method, we would need another dataset with high frequency, like the Consumer Price Index (CPI), which correlates closely with GDP. However, with the second method, we could estimate GDP values for each quarter, month, or even day solely based on the passage of time without relying on specific HF data.