I have 2 datasets. One is time series data of sale of homes by region by type:
Region Type SaleMonth Sales NE SF 201601 100 NE SF 201602 100 NE SF 201603 150 NE SF 201604 100 NE SF 201605 150
These data were manually entered by realtors and given to us by a data feed.
We also get another data set of registered homes from the county. However there is a 2-month lag for this data set. It is structured as follows:
Region Type SaleMonth Sales NE SF 201601 130 NE SF 201602 130 NE SF 201603 160 NE SF 201604 NE SF 201605
The weight is calculated as: Retail Transactions/Registered values.
So, assuming the weight is correct, I have to predict the Registered values for 201604 and 201605. I have tried regression, time series analysis, etc., but I wasn't able to get to the threshold weight.
Is this a candidate for neural net time series? Is there anything else I can do assuming the weight is correct.
Assuming it is incorrect, what is the best approach to determine what sliding window of values to use from the transactions to accurately predict the registered values?