Business Problem

The ministry of Finance possesses a lot of data with respect to multiple financial indicators of all the states. A forecast of these indicators for the next 5 years, will guide the ministry in planning the fiscal budget for various states.

The dataset in usage has the data of all 28 states, however, we have chosen to construct forecasting models for the state of Andhra Pradesh. However, it should be noted that, due to the separation of Telangana from Andhra Pradesh in 2014, the forecasts will be an aggregation of both the states.

Client: Office of Finance Minister of Government of India

How the forecasts will be used: This forecasting model will help the government predict various financial indicators and accordingly perform fiscal planning for the state of Andhra Pradesh and Telangana for the time 2016-2020 (5 years).

Forecasting Problem and data

o forecast the fiscal indicators, we have 6 datasets each one related to one indicator. The data is available from 1980-81 to 2015-16. Below is the list of fiscal indicators we are trying to forecast:

1. Nominal Gross state domestic product (GSDP)
2. Gross Financial deficit
3. Capital Expenditure
4. Social sector expenditure
5. Revenue deficit
6. Own tax revenues

Forecasting Model

Double Exponential Smoothing and Linear Regression models performed optimally for the data-sets in the project. The data sets are aggregated on an annual level and do not show substantial seasonality, only increasing trend.

Recommendations:

• Only time is the independent variable. Need other external variables for better forecasting the indicators
• No seasonality in the time series of the indicators. Quarterly data would have helped to see whether seasonality is present.
• It could be helpful to use some of the relevant data sets as predictor variable in building the forecasting model for other indicators. Many of the indicators are interdependent of each-other, so we need to know how the factors interact with each other in the regression model too.
• During application of this model, separate prospective CAGR of Population density, general allocation funds, financing surplus budget, revenue, and other macro-economic indicators of Telangana and AP should be factored in the model. Also, a roll forward approach for training data set should be used every period so that the model keeps improving with real-time data.