Forecasting India's GDP over 2015-2019 to assess the performance of Modi Government

Project Details




Shashank Suryae, Amol Kamath, Soumya Joshi, Mithilesh Parvataneni, Gary Rohan Singh, Rashi Choudhari





Problem Description
The project aims to forecast India’s GDP through 3 methods – employment, GDP trend and
the sum of 4 expenditure components (household, government expenditure, investment and
net exports) and to use the same to assess if the Modi Government has been able to change
India’s GDP growth trajectory over its 5 years in power (2015-2019). The client is the
Department of Expenditure, Ministry of Finance. Using the project, they can do a high-level
assessment of the impact of government policies and can also use it for pitching the success of the incumbent party, in case of favourable results. A favourable result for the client will show outperformance of actual GDP to the forecasted GDP, showing a change in trajectory, which could be attributed to the government policies. In terms of the implication of forecasting errors, a model that under-forecasts the GDP implies that the Government was successful in making India better as compared to the usual growth pattern and paints the Government in a favorable light while a model that over-forecasts does the exact opposite.

Brief description of data
The data from Penn World Tables (original source) was only available until 2014 and the fact that the employment data had missing links due to ILO not reporting retrospective data was a problem. The current calculations have been carried out basis the data from World Bank for India from 1960 to 2017. The data has been cleansed for any erratic values, missing values, reporting error, and converted to US$ Bn. (2011 ER equivalent) to account for any
irregular reporting or exchange rate effect in the entire data set.

Forecasting Methods
The project forecasts 6 series and uses the methods of Naive, Naive Seasonal, Moving
Average, Exponential, Double Exponential, Holt-Winters, Linear Regression additive, Linear
Regression Multiplicative and Linear Regression Quadratic. The data was partitioned to
create a validation period was of 5 years from 2010 – 2014 as that is the length of the
forecasting horizon. The balance observations were taken as the training period. The forecast period is 2015~2019. The metrics to judge the same include MAPE and RMSE.
Further, residual plots were used to check if trend and seasonality were captured. Across models, we see improvement from the benchmark of Naïve model. For every dataset, we compared the metrics and chose the model with lowest errors and appropriate residual plot. Using this model the forecasts were made.

Conclusions and Recommendations
As per the data, the GDP has outperformed the mean prediction in all 3 cases. It is worth noting that for the first 2 years, the actual GDP is between the UCL and LCL of the forecast, however, the GDP exceeds the UCL in the last 3 years. Further, the GDP forecast from
employment is significantly lower than the actual GDP. This leads to below conclusions:

  • The Modi government has been able to change India’s growth trajectory through its policies. Even its government expenditure has been higher than forecast.
  • GDP forecasted through employment is significantly lower - indicating an increase in productivity amongst the employed workforce.
  • It is likely that it takes time (~2 years) for government policies to show impact in terms of metrics, explaining the trend seen in the first 2 years

The client could use the work to showcase the achievements of the current govt. and the ruling party could use it for upcoming elections. This could also provide learnings for future policies.

Application Area: