# How do I treat zero inflated univariate time series data?

The data I am handling is a rainfall data. The only columns are "Date" and "Rainfall".

The day that is not raining will be accounted to zero, therefore it is a zero inflated dataset.

I am looking to use "pscl" library from R and all the example I saw is either using multivariate or bivariate data.

Is it not possible to use this method to handle zero inflated data? If no, should I ignore the violation of normality and proceed the forecasting?

• Why did you decide on pscl? – kjetil b halvorsen Mar 30 at 12:32
• Why don't you just use ARIMA and/or exponential smoothing - it looks like all you are trying to do is predict a future results? I think in practice normality and other assumptions get ignored in time series and will in any case be primary important for the confidence interval.If you just want to predict a point estimate, I am not sure that even matters. I admit I have never heard of zero inflated data in the concept of time series. – user54285 Mar 30 at 21:26
• @user54285 I am researching on univariate rainfall data, zero indicates non rainy day. – Seow Apr 2 at 4:28
• I would look at arima in the forecast r product and exponential smoothing models in the smooth product. Both are designed to deal with univariate time series. I have never read anything in the literature about them that says it matters if the value is zero or not. Of course any statistic that does not vary is going to have problems. I know nothing about your topic but if you are predicting rain don't rain logistic regression would seem logical (or linear probability models if you accept their logic). I am not sure where these are in R, I use SAS for them. – user54285 Apr 2 at 16:49
• @user54285 no matter what it is regression based, i have to uphold the assumption of normality, having many zeros making it impossible to be normal – Seow Apr 3 at 18:19