EViews.com

Dear forum members,

I am facing problems in order to interpolate data in a dataset about danish wheat prices. Right now I have weekly data ranging from 2005 to 2013. My problem is that some values are missing and I dont know how to handle that without loosing too much information of the data set.

The problem is that not certain values are missing - otherwise I would just use the build-in methods of eviews to estimate them with normal interpolation methods - but rather 3 months in each year. My fear is that if I use the interpolation methods on this huge lack of data I simultaneously loose quite a bit of information. My thought was to estimate a VAR-model and then just plugging in the weeks i need data for and get a value for the dependent variable. However, due to I am not an expert in time series modelling I wanted to ask if someone has an idea how to ideally handle this?

I am using Eviews 8 and I am thankful for every hint...

\Tysken

1) What is the purpose of your study? (structural estimation or forecast)
2) What is the reason behind those missing values? (no transaction or access problem)
3) Do you really have to work with weekly frequency? Why not monthly or quarterly?

Hi trubador

the purpose is the evaluation of price transmission in different swedish grain markets. I try to evaluate how the prices in Sweden, Danmark and France are moving together in a long-run relationship via VECM/Engle and Granger 1987. Furthermore, I want to detect a structural break in the end of 2008 that may have occured after an important local demander in Sweden has massively increased its demand for wheat.

The reason of missing values is a simple access problem. Because I deal with spotprices, values are very hard to obtain. However, the source of the existing data is the danish agency for agriculture that is simply not observing the prices in the time of summer, because the crop of winter-wheat is basically sold and the demand is covered from storage. Because of time pressure for the study I hoped for a method that may be appropriate to estimate the data.

I am using a weekly frequency because I hade safe access to other time series I am evaluating. Because the prices are quite volatile I hope for a better accuracy. I am evaluating the data of only 8 years, quarterly and monthly data seemed to imply too less observations that are not reflecting the changes within a short time (what I considered to be crucial for the analysis of price transmission).

Do you have any ideas or suggestions? Otherwise I was thinking of not including the price series for danish wheat and shrink the analysis to the countries Sweden and France. I was thinking it is pointless to evalute 52 values a year if 30% of observations are missing...

Thanks for your help
Best
Tysken

First of all, I suggest you to try Gregory-Hansen approach if your sole purpose is to detect a structural break in a cointegrating relationship. You can find EViews’ code here: http://forums.eviews.com/viewtopic.php?f=15&t=976#p3427
It seems the prices change in increments of 5, which may be due to rounding. I do not have any expertise on this market, but you can think of considering zero changes (i.e. flat prices) for missing periods. This would be similar to contracting the sample with removing NAs.
If you have any prior information as to how supply-demand dynamics affect/determine prices, then you can estimate a structural model to approximate missing values. I am afraid any other approach could lead to data snooping.
If you convert to monthly frequency, you’ll deal with less missing values. In this case you can consider using interpolation methods.
You are already aware of the trade-off here, so there is nothing much to add. You should decide between dropping an endogenous variable from the model and losing information due to missing values (or to frequency conversion).

Hi trubador,

thanks for you reply. I appreciate that you help me and people a lot in this forum and that you shared a part of your experience. I read about the Gregory Hansen approach, thanks for providing the scheme. I am sure I will have some more questions in that regard :)

Again, many thanks, it helped me a lot

Best from Sweden
Tysken