Time Series Analytic Models for Forecasting Vehicular Registration Volume in the Indian Context

In this paper, time series analytical models developed to capture the trend of vehicle registration in India has been elaborated. For this, twenty two years of motor vehicle registration data (2001–2022) have been used. In all, nine optimal models namely, linear, quadratic polynomial, cubic polynomial, exponential, power, logarithmic, grey (1, 1), and fuzzy time-series models have been developed. The models so developed were rigorously evaluated with mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE), R2, p-value, and scattering index metrics. Theil’s U test has also been administered on the models to check their veracity. Among the models, polynomial, exponential, linear, grey, and fuzzy models stood out to be the best with low values of error metrics, i.e., MAE, MAPE, RMSE and scattering index 10–12.5, 5/8%-14.5%, 11–12.5, and 1.8%-4.8% respectively. These models have also shown high values of coefficient of determination (R2) in the range of 0.93–0.96 indicating their forecasting accuracy. Thiel’s-U statistic is also found to be very low for these models and was found to be in the range 0.04–0.97 proving their authenticity as best forecasting models. The projection of vehicular population for the next five years (2023–27) has also been done using the best of the models. The estimated figures are in the range of 340–400 million. Further, the relationship between GNP and the vehicle registration numbers is also explored and interestingly, the exponential relationship with a high R2 value of 0.98 has been the best fit.