RaboResearch - Economic Research

The race to predict India’s economic data

Economic Report

Share:

Also published on Bloomberg/Quint, July 18, 2019

  • There are two good reasons to use so-called nowcasting modelling techniques for the Indian economy. First, we have been able to improve our forecasting accuracy using these models. For example, we have accurately predicted India’s recent slump in economic activity.
  • Second, nowcasting can be used to validate recent claims that India’s GDP figures are overestimated. We do not find ground for these claims: our models do not show a significant break in series in the post-2011 period and they arrive at an estimated average annual growth rate of 7.0% in the period 2011-2019.
  • Going forward, we expect an ongoing debate about the quality of India’s statistics. Since solid and reliable statistics and forecasts are the cornerstone of global investment decisions, it is even more important that the new Modi government takes action to remove any lingering concerns.

On your marks…

My eldest son took up athletics a year ago. All the kids in his group are between six and nine years old, so the training is pretty playful – most of the time. But when they have to race against each other, the competition heats up. Not just between the kids, but among the passionate parents urging them on from the side-line.

Forecasting economic indicators often resembles a race as well. Each quarter, between 45 and 50 economists who participate in the Bloomberg survey lace up their running shoes to see who comes first, second and third. In other words: who is the most accurate? Every economist uses their own forecasting techniques. Some just take a guess at GDP growth by using a crystal ball, while others talk to clients or use economic models.

At RaboResearch we use two model approaches.[1] First, for medium-term purposes we use a model based on seasonally-trended extrapolation techniques which we combine with effects of policy measures announced by the government. In this approach the external sector is viewed within a closed accounting setting using a global econometric trade model.

Our second modelling approach is new. This year we have been developing nowcasting models which we use to override the short-term forecast. The basic principle of nowcasting (a contraction of the words now and forecasting) is the exploitation of data at higher frequencies (see Table 1) to get an early estimate of the target variable before the actual print is released.

Table 1: High-frequency input for Rabobank’s nowcasting models
Table 1: High-frequency input for Rabobank’s nowcasting modelsSource: Macrobond
Figure 1: Rabobank’s nowcasting methodology predicted the slump in economic activity
Figure 1: Rabobank’s nowcasting methodology predicted the slump in economic activitySource: CSO, Bloomberg, Macrobond, Rabobank

There are two very good reasons for taking a closer look at the potential of nowcasting in India. First, our nowcasting models accurately predicted India’s recent slump in economic activity (see Figure 1). Second, the models could make a useful contribution to the discussion about the accuracy of India’s GDP statistics. Recently, the former economic top advisor to the government Arvind Subramanian raised doubts about the reliability of India’s national accounts data, stating that annual GDP growth between 2011 and 2017 was only 4.5% on average, compared to official numbers of 7%. We can use our nowcasting models to see if this claim can be validated.

Get set…

To return to our running theme, let’s introduce our nowcasting models to you as three athletes in competition.

The goal of the three athletes (i.e. our three models) is to finish first (i.e. to accurately predict GDP growth). To test their fitness, we use ‘out-of-sample testing’. This means that instead of estimating the models for the entire data sample, we exclude the last GDP realisations from the sample to test their prediction accuracy. Put more simply: we blindfold our three contenders for the final part of the track and see how close they come to the finish line. Now let’s meet our three competitors.

Mr. Traditional

Mr. Traditional has a long track record of winning races. He does not believe in modern approaches to running, preferring to rely on proven concepts that helped him win in the past. For instance a traditional training scheme (i.e. vehicle sales), paying his sport club contribution (i.e. tax revenues), a good night’s sleep (i.e. electricity production), an oatmeal breakfast (i.e. oil consumption) and the running shoes he bought in the 1980s (i.e. inflation).

Mrs. Modern

Mrs. Modern has not been running competitively as long as Mr. Traditional has, so she lacks some experience. Mrs. Modern uses not only the traditional proven concepts, but also the more advanced tools and services now available in the sport to crank up her performance. For instance, high altitude training camps overseas (i.e. vehicle exports), data analytics to monitor her blood pressure and stamina, and advice from a sports nutritionist (i.e. services sector purchasing managers’ index).

Mr. BVAR

Mr. Bayesian Vector Autoregressive (let’s call him Mr. BVAR for short) is the odd one out, as he is completely agnostic. He is only interested in winning and couldn’t care less how he achieves it. So it’s impossible to trace exactly what factors contribute to his success.

Go!

Figure 2: Mr. Traditional is all over the place after 2015
Figure 2: Mr. Traditional is all over the place after 2015Source: Rabobank, CSO, Macrobond

Now let’s see how our three contenders are performing, beginning with Mr. Traditional.[2] Figure 2 shows his race. After a strong start in the period before and during the Global Financial Crisis (GFC), his performance weakened significantly after 2015. This finding is more or less in line with the observations by Subramanian, who also registers a breakdown of his models when applied to more recent years. Especially on the last part of the track, where contenders have to wear a blindfold, Mr. Traditional runs all over the place, spinning in circles rather than heading straight for the finish line.

One reason for Mr. Traditional’s poor performance after 2015 might be that the economic structure of emerging markets, such as India, generally transforms at a rapid pace, and the services sector and financial markets have become much more important for the economy. This might also explain the breakdown of Subramanian’s models, since they are also based on relatively ‘traditional’ determinants of economic growth, such as credit, electricity and trade. Many economists have indeed criticized the methodology of Subramanian. The Economic Advisory Council to the Prime Minister (EAC-PM) even claimed that Subramanian’s analysis was based on ‘cherry-picking’ high-frequency indicators while ignoring data on services, agriculture, and taxes.

Mrs. Modern does use several services available to athletes, so let’s check if she is able to outperform Mr. Traditional (Figure 3).

Mrs. Modern performs much better than Mr. Traditional over the entire race. Even when she has to put on the blindfold, she manages to follow the running track, but in her enthusiasm she overshoots the finishing line and ends up in the long-jump sand pit.

In contrast to this slight overshoot, Mr. BVAR cuts corners during the race and ultimately grinds to a halt right before the finish line (see Figure 4). Given that Mrs. Modern overshoots and Mr. BVAR undershoots the finish line, we ultimately get the best results when both athletes join forces and run together.

Figure 3: Mrs. Modern keeps performing well
Figure 3: Mrs. Modern keeps performing wellSource: Rabobank, CSO, Macrobond
Figure 4: Mr. BVAR is cutting corners
Figure 4: Mr. BVAR is cutting cornersSource: Rabobank, CSO, Macrobond

Keeping the sport clean

Nowcasting modelling definitely has its upsides. We are able to use these techniques to accurately gauge economic activity in India, especially when we combine the two models that take proper account of India’s changing economic structure. From this perspective, we do not find indications that the official GDP data are overestimated: our models do not show a significant break in series in the post-2011 period and they arrive at an estimated average annual growth rate of 7.0% in the period 2011-2019. Of course, testing the reliability of India’s national accounts data requires other types of analyses, but we believe we have shown that nowcasting can be useful in this discussion as well.

Going forward, we expect an ongoing debate about the quality of India’s statistics, which will intensify now and again. Since solid and reliable statistics and forecasts are the cornerstone of global investment decisions, it is even more important that the new Modi government takes action to remove any lingering concerns. For instance, by installing an independent committee with statistical experts from all over the world to review the quality of the national accounts data. Just as the reputation of sports, such as athletics and cycling, can be damaged by doping allegations, the discussion about India’s statistical credibility could dent India’s popularity as an international investment destination.

Footnotes

[1] Of course, Rabobank has eyes and ears on the ground in Mumbai and elsewhere in India who keep in contact on a weekly basis. If there are important developments which are not properly accounted for by the models, we always take the liberty to override our model outcomes based on ‘expert opinion’.

[2] For a complete overview of our nowcasting models, the variables used per model, the estimation technique and results, and the out-of-sample forecast tests, we refer to our Economic Report Nowcasting the Indian economy.

Share:
Author(s)
Hugo Erken
RaboResearch Global Economics & Markets Rabobank KEO
+31 6 2223 1650

naar boven