Using yearly data for responsible forecasting the impact of COVID-19 on real GDP and employment in South Africa [part 1 of 2]

Michèlè Capazario (7 Min Read)


As an econometrics lecturer of mine was fond of saying:

“Macroeconomic forecasts are as inaccurate as astrology.”

I believe that there is an element of truth to this statement. When an individual, company, or country relies too heavily on forecasts to make their assessments about what the future will look like, they are naively setting themselves up to being misled. However, that does not mean that macroeconomic forecasts aren’t useful. When used in conjunction with insights from studies, survey and other data sources, and discussions with experts, macroeconomic forecasts can provide a window into the future- even if that window is half-cracked and half-dirty.

The quality of a macroeconomic forecast is directly related to the quality of the data at one’s disposal. A long quarterly or monthly time series of statistics is ideal for forecasting- this is because all forecasts, no matter how you spin them, are derived from historical trends in the data from which they come. Therefore, the more data at one’s disposal, the more variability (insight) one has to use as a means to forecast more accurately.

In many African countries, data is, however, extremely scarce. In fact, in most cases, macroeconomic data is only published for public use at an annual level, with most time series only going back 10-20 years. According to Hyndman & Kostenko (2007), this number of observations is above the minimum number of observations needed to conduct forecast analysis, but not by much.

Does this mean that accurate economic forecasting cannot be done in Africa?

In my opinion, the answer is a definitive NO.

While yearly data is not ideal when attempting to conduct sophisticated forecasts and econometric modelling, it is still possible to glean some interesting (and potentially accurate) viewpoints on what the future might hold at a macroeconomic level.

Moreover, most treasuries and central banks across Africa do have access to monthly or quarterly economic indicators, such as GDP and employment numbers, at a national and sectoral level, which informs their in-house forecasts. In such instances, the challenge is not that the data does not exist, but that it has not been made available to the public, especially at a sectoral level.

South Africa provides an ideal case study to illustrate that useful (and comparable) findings can come about from analysing yearly data. Specifically, we can consider the impact of COVID-19 on the South African economy. In normal circumstances, a macroeconomist would rely on an Autoregressive Integrated Moving Average (ARIMA) methodology to forecast certain economic indicators with small annual datasets[1]. These forecasts rely on historical data trends to extrapolate potential future trends with varying degrees of accuracy. However, without any real-time data for 2020, these historical trends would not adequately capture the impact of COVID-19 on particular macroeconomic indicators. To come up with a model that can estimate the impact of COVID on real GDP and employment numbers without 2020 data, a hybrid model (trained by assumptions found in literature) could be employed. One such hybrid methodology is summarised at a very high level below:

Instead of forecasting GDP growth or employment growth rates between 2020-2024, the model estimates potential scenarios for changes in these variables across a wide variety of possibilities. This scenario-based model is applied to national South African data below:


The upper lines of each forecast graph estimate the best-case scenario for South Africa (which is, unfortunately, highly unlikely) and suggest that GDP will only contract by 1.5% in 2020, with employment remaining stagnant.

The bottom of the fan charts represent the worst-case scenarios for the country. In these worst-cases, real GDP is expected to decline by as much as 8% in 2020. This resulting recession translates to employment declining by as much as 7% in the same year. In this case, the national economy could shed as many as 1.2 million jobs in 2020 alone.These “simple “yearly estimates are more or less in with official forecasts[2]:

  • The South African Treasury estimated that real GDP in the economy would decline by 7.2% in 2020,
  • The South African Reserve Bank estimated that real GDP would shrink in 2020 by 7%
  • According to BusinessTech and News24 (who quoted Business for South Africa), roughly 1.5 million South Africans would lose their jobs in 2020 due to the lockdown of economic activity.

Based on our yearly-data forecasts, the best-case scenario suggests that the South African economy can recover to pre-COVID growth levels by 2022. The more bearish forecast scenarios, however, suggest that recovery would only occur by 2023/2024 in terms of real GDP and employment growth alike.


It is clear that in this case, an annual model fares well against more sophisticated models which rely on quarterly and monthly data.The lack of frequently collected data in other African countries need not prevent economists from undertaking responsible and in this case, critically important, economic modelling work. How this approach is applied to yearly sectoral real GDP and employment data is the subject of the following blog.


[1] (The South African Treasury, 2020); (South African Reserve Bank, 2020)(News24, 2020); (BusinessTech, 2020)

[2] (Box, et al., 2015)



Box, G., Jenkins, G., Reinsel, G. & Ljung, G., 2015. Time Series Analysis: Forecasting and Control. s.l.:John Wiley & Sons.

BusinessTech, 2020. New Report Details South Africa’s Jobs Losses Before Lockdown. [Online]Available at:
[Accessed 11 August 2020].

Hyndman, R. & Kostenko, A., 2007. Minimum Sample Size Requirements for Seasonal Forecasting Models. Foresight, 6(Spring 2007), pp. 12-15.

News24, 2020. COVID-19 Peak Expected Late August, with SA Economy to Recover in 2 Years. [Online] Available at:
[Accessed 11 August 2020].

Okun, A., 1963. Potential GNP: Its Measurement and Significance, New Haven: Yale University.

South African Reserve Bank, 2020. Statement of the Monetary Policy Committee, Pretoria: South African Reserve Bank.

The South African Treasury, 2020. The COVID-19 Shock and the Revised Economic Outlook, Pretoria: The South African Treasury.