Balance Forecasting



Hi fellows

The problem statement is to do balance forecasting for lacs of accounts in banking sector. Forecasting needs to be done separately at daily, weekly, monthly and quarterly level, for which different models will be created. Obviously this needs to be done in an automated way as we are dealing with lacs of accounts.

As multiple algorithms are available for dong forecasting and any algorithm can beat others depending on the data, what would be the best way to handle this.

Looking for your valuable inputs. Thanks !!


According to me ARIMA or SARIMA models are the best models as they can take care of all the three components of time series. Just do trails to get the best set of AR,I,MA values.

Correct me if am deviated.:slightly_smiling_face:


Arima models are good, but how do we justify that Arima model will give best accuracy for all the accounts. For some accounts, may be some other algorithm will work better. It just depends on the data.


You can do an overall forecast with ARIMA or group your different accounts which will require some EDA work. You will have to do some maintenance either monthly for the daily models and quarterly for the monthly models.