A pandemic and economic slowdown: the case of India

ABSTRACT

In this article, we use a novel dataset on new investment projects undertaken, available for states in India in quarterly frequency, to forecast economic activities in the wake of the ongoing Covid-19 pandemic. Panel data estimates for the country, employing fixed effect and Generalized Method of Moments, show a drop of approximately 50% in investment count during the financial years 2019–20 to 2021–22, compared to the counterfactual projection of an increase of 38% over the same period. State-level analysis reveals similar patterns, even though with considerable heterogeneity – relatively economically well-developed states may witness higher reduction in investment compared to the less developed ones. Given that incidence of Covid cases is also relatively higher in the more developed states, our findings can help in formulating policies appropriate for each state. This study can be generalized in the following three ways: it can be extended over a longer period of time, it can be applied in future crisis periods, and it can be made more precise given more granular data is available.

KEYWORDS:

• Investment
• dynamic panel
• system GMM
• forecast
• Covid-19

JEL CLASSIFICATION:

• C23
• E17
• E27

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ The contraction due to government restriction has been documented by Goolsbee and Syverson (2021).

² Goodell (2020) points out this deficit and suggests possible avenues of research on the economic and financial impact of Covid-19.

³ We provide more evidence on similarities between investment and GDP, later in the paper.

⁴ The Official GDP data, available at the website of Ministry of Statistics and Programme Implementation, validates this.

⁵ Majority of this drop, of the order of 28%, is projected to happen in 2021–22 FY.

⁶ Ashraf (2020b) uses high frequency Total Daily Stock Market Returns data as a proxy for economic activity and use this to measure impact of Covid-19 on the economy for a set of 64 countries.

⁷ The data that support the findings of this study are available from CMIE. The data is available [from the authors] with the permission of CMIE. We use the data as available on 31 March 2021.

⁸ We cannot calculate rank correlation (a) at quarterly frequency, since NSDP is not available at that frequency, and (b) for 2020–21, since NSDP for that FY is not available for most states in India.

⁹ We also consider data up to 2019–20 FY to construct the counterfactual scenario, in which the pandemic does not happen.

¹⁰ In India, the lockdowns started from the last week of March, which is very close to the start of 2020–21 FY.

¹¹ Results from state-specific seasonal autoregressive moving average (SARMA) supports this feature. Results are available on request.

¹² Our regression results support the choice of lags. We find all four lags are significant (Table 5).

¹³ Repo rate does not have cross-sectional variation, since one repo rate is announced for the whole country.

¹⁴ We must mention that in our counterfactual model, this dummy variable is absent.

¹⁵ Figure 4 confirms the spatial agglomeration of investment by region.

¹⁶ We see an increase in annual investment count for the state of Gujarat. However, there was substantial drop in the last two quarters of 2020–21.

¹⁷ Figure 4 shows a spatial pattern; while majority of eastern states exhibit lower number of new investment projects undertaken, better performing states are mostly concentrated in western and southern regions. The cross-sectional independence test (based on Pesaran 2004) also confirms this.

¹⁸ We emphasize that our forecast based on these models can easily be extended for further periods. But we restrict ourselves for the current FY due to the ongoing threat of arrival of different mutants of SARS-COV2 Virus.:

¹⁹ We choose $k=4$, for our quarterly data.

²⁰ We perform BP test, since in our case, T > N.

²¹ In fact, the predictions based on GMM are closer to the actuals as compared to FE estimates.

²² Dynamic forecast works in the following way: it forecasts the values for the period (t + 1) in the same way as static forecast. Subsequently, it uses the forecasted values from (t + 1) period to predict for period (t + 2), and so on. However, the model parameters remain same throughout.

²³ The Hausman test rejects the RE model below 1% level.

²⁴ We obtain our results using xtabond2 command of STATA following Roodman (2009).

²⁵ We use forecasted $reporate$, using ARIMA model.

²⁶ For FY 2020–21, the forecasted growth rates using GMM (−34%) are closer to actuals (−29%).

²⁷ Some of the states severely affected by the pandemic have already resorted to phase-wise lockdown.