What do high-frequency indicators tell us about economic activity in South Asia?

|

This page in:

Image
Earth Night At South and Central Asia From Space. Jayjune69/Shutterstock.com
Earth Night At South and Central Asia From Space. Jayjune69/Shutterstock.com

Last year, South Asia imposed some of the world's strictest lockdowns to combat the spread of COVID-19. As a result, the region's economy shrank by 5.4 percent in 2020, its largest-ever decline.

As noted in our recent South Asia Economic Focus, recovery is underway but uneven as some countries, especially those dependent on tourism, are still reeling from the pandemic impacts. The report uses a number of high-frequency indicators including electricity consumption, nighttime lights intensity and nitrogen dioxide emissions (at the urban level) to understand the nature of the recovery. The blog discusses the two main indicators used to understand changes to GDP.

Measuring the full extent of the crisis and recovery is no easy task as GDP numbers only provide a partial estimate in four of the eight South Asian countries—national accounts figures in Bangladesh, Bhutan, Nepal, and Pakistan become available after mid-2021.  Unlike national accounts, high-frequency data and non-traditional economic indicators, which are available sooner, paint a more detailed picture of the economic impact of the pandemic and breadth of the recovery.

Given the lag in publication of GDP numbers, nowcasting quarterly economic indicators for countries with quarterly national accounts also provides more up-to-date information. Nowcasting is the prediction of the present, the very near future, and the very recent past state of an economic indicator (the term is a contraction of the words "now" and "forecasting"). The type of high-frequency indicators we would use to nowcast GDP is usually at an even higher frequency than quarterly. For this reason, we built a quarterly activity indicator by combining information from different high-frequency variables for four countries with quarterly GDP data: India, Sri Lanka, Maldives, and Nepal.

We use the Least Absolute Shrinkage and Selection Operator or LASSO (Tibshirani, 1996 and Meinshausen, 2007) to both select the most relevant economic activity variables and enhance the prediction accuracy of the model for each of these four countries. We cover as long a period as possible in each country (dependent on when quarterly GDP series became available and the availability of high-frequency indicator series). The method produces a lead indicator of GDP for the immediate period before it is officially published.

The results suggest these four countries in the region are at different stages of the recovery.

Figure 1 shows how the LASSO indicator keeps track of the steep drop in output and year-on-year contraction in the second quarter of 2020 in all countries. The bars show actual growth estimates consistent with the annual data presented above. The decline was the largest in Maldives where tourism came to an abrupt standstill.

In India, the contraction of the LASSO Indicator was 23.4 percent, nearly identical to the fall in officially reported GDP, though it over-predicts growth in the last quarters of calendar year 2020.

In Sri Lanka, the LASSO Quarterly Economic Indicator fell by 15.6 percent in Q2, then recovered to reach 2.6 percent year-on-year growth in the fourth quarter.

In Nepal, with data available for Q3 2020, the LASSO procedure confirms that the economy was still in negative territory in the third quarter of CY2020. Overall, the procedure performs relatively well, which gives us confidence in the accuracy of the nowcast for the fourth quarter of 2020, with India growing, Sri Lanka consolidating the recovery and Maldives still in negative territory.

Figure 1.  Quarterly Indicators suggest all countries started recovery around the third quarter of 2020

Nowcasting economic Indicators based on LASSO regressions, percent YoY growth (CY)

Nowcasting economic Indicators based on LASSO regressions, percent YoY growth (CY)

Note: Data in calendar years. The line denotes the model prediction and bars the actual values. The nowcasting index uses the set of variables that provide the most accurate prediction through a k-fold cross validation procedure. The actual data used for Nepal is preliminary, from the Central Bureau of Statistics "Rebasing of National Accounts Statistics".

Source: Authors using CEIC and Nepal's Central Bureau of Statistics.

For countries where quarterly GDP are not available (Afghanistan, Bangladesh, Pakistan, Bhutan) it is not possible to use the LASSO procedure to nowcast GDP. We then  follow a different method to produce monthly indicators of activity. We use principal component analysis (PCA) to develop a composite indicator that can trace the turning points and trends in activity indicators relative to pre-COVID levels—though it is more difficult than with LASSO to make a precise estimate of how large or small the actual change will be.

The PCA is a commonly used dimensionality-reduction technique to extract the maximum variance (largest eigenvalue) across different dimensions of the data set. In this context, we construct monthly economic indicators from different groups of high-frequency variables (Table 1). Our selection on high-frequency variables is based on four criteria:

  • The variables should fit into three categories of information: a) mobility indicators correlated with economic activities; b) domestic activities represent key sectors of the economy; c) trade activities related to GDP in the past.
  • The variables should be released in a timely manner and without significant publication lag.
  • The variables should have higher or equal frequency in relation to monthly frequency (daily, weekly, monthly).
  • The variables should be seasonally adjusted and measured in levels.

After selecting the high-frequency variables, we first standardize each variable, by subtracting the mean and dividing by the standard deviation. Then we apply PCA to generate the loadings of principle components. The monthly economic indicators are essentially the linear combinations of selected variables, weighted by loadings. To compare to pre-COVID level, we index economic activity indicators with January 2020 equal to 100.

Figure 2 shows the results for Bangladesh and Pakistan. The indicator suggests a clear improvement in economic activity starting in the third quarter. By the beginning of the fourth quarter, economic activity in both countries seems to have returned to pre-COVID levels, reflecting the pickup in mobility and the recovery of domestic production as well as trade activities, which takes a prominent weight in the PCA measure.

Figure 2. Economic activity indicators suggest Bangladesh and Pakistan also recovering to pre-COVID levels by Q4 2020

Table 1 Indicators selected by the principal component analysis for each country
Economic activity indicators suggest Bangladesh and Pakistan also recovering to pre-COVID levels by Q4 2020 Indicators selected by the principal component analysis for each country

Note: Economic Activity Indicator was constructed by using Principal Component Analysis. Variables included in Economic Activity Indicator for South Asian countries was summarized in Table 1.2.

Source: Authors using Google mobility report, Oxford COVID-19 Government Response Tracker, United Nations, World Bank.

 

The choice of indicators that are inputs into both the LASSO and PCA method can theoretically change every period. Some underlying indicators such as the Google mobility indicators—which measure the visits to workplace, retail, etc. compared to pre-pandemic levels—proved to be very important predictors in 2020.

As the economies of South Asia return to normal, other indicators such as tourist arrivals will continue to be important. Also, for smaller countries where economic activity is dominated by just a few sectors—such as hydropower production in Bhutan, or agricultural production in Afghanistan—understanding the activity of core sectors may be sufficient to make an assessment, as there may not be enough high-frequency indicators to use the PCA methodology.

Authors

Valerie Mercer-Blackman

World Bank Senior Economist, South Asia Office of the Chief Economist

Sebastian Franco Bedoya

Research Analyst, Office of the Chief Economist for South Asia

Join the Conversation