12

9

Google Trends returns weekly data so I have to find a way to merge them with my daily/monthly data.

What I have done so far is to break each serie into daily data, for exemple:

from:

2013-03-03 - 2013-03-09 37

to:

2013-03-03 37 2013-03-04 37 2013-03-05 37 2013-03-06 37 2013-03-07 37 2013-03-08 37 2013-03-09 37

But this is adding a lot of complexity to my problem. I was trying to predict google searchs from the last 6 months values, or 6 values in monthly data. Daily data would imply a work on 180 past values. (I have 10 years of data so 120 points in monthly data / 500+ in weekly data/ 3500+ in daily data)

The other approach would be to "merge" daily data in weekly/monthly data. But some questions arise from this process. Some data can be averaged because their sum represent something. Rainfall for example, the amount of rain in a given week will be the sum of the amounts for each days composing the weeks.

In my case I am dealing with prices, financial rates and other things. For the prices it is common in my field to take volume exchanged into account, so the weekly data would be a weighted average. For financial rates it is a bit more complex a some formulas are involved to build weekly rates from daily rates. For the other things i don't know the underlying properties. I think those properties are important to avoid meaningless indicators (an average of fiancial rates would be a non-sense for example).

So three questions:

**For known and unknown properties, how should I proceed to go from daily to weekly/monthly data ?**

I feel like breaking weekly/monthly data into daily data like i've done is somewhat wrong because I am introducing quantities that have no sense in real life. So almost the same question:

**For known and unknown properties, how should I proceed to go from weekly/monthly to daily data ?**

Last but not least: **when given two time series with different time steps, what is better: Using the Lowest or the biggest time step ?** I think this is a compromise between the number of data and the complexity of the model but I can't see any strong argument to choose between those options.

Edit: if you know a tool (in R Python even Excel) to do it easily it would be very appreciated.

for python, the standard tool is pandas. It was specifically designed to deal with financial data timeseries. pandas timeseries

– seanv507 – 2015-01-03T23:08:17.697Care to expand a bit on what you mean by "unknown property"? – TheGrimmScientist – 2015-01-04T21:20:51.923