Did India’s Demonetisation of 8 Nov 2016 boost uptake of digital payments systems

An analysis of data from the Reserve Bank of India

Abstract

On Nov 8 2016 at about 8 pm, the Prime Minister of India announced that effective midnight that day the existing stck of Rupee 500 and 1000 notes would cease to be legal tender. This policy came to be known as India’s Demonetisation. These two denominations of currency notes made up 86% of the currency notes in circulation.

Though not mentioned as an objective of the policy initially, one of the benefits claimed for it at a later date was that it would lead to a significant shift away from payments using paper money and towards digital payments.

Two years on from Nov 2016 it is germane to ask, ‘Did digital payents really take off?’

In this analysis I use a time series analysis forecasting approach applied to monthly data from RBI on the volume, and the total value, of financial transactions by each of several different digital payments systems.

Taking the data upto October 2016 (the month before demonetisation) and using Autoregressive Integrated Moving Average (ARIMA) modelling to forecast future growth, I show that the actual growth observed in the months after Nov 2016 was mostly within the expected range of the forecast levels.

My analysis shows that any growth in the volume of of digital payments post-demonetisation can be explained almost completely on the basis of trend growth. Digital payments were growing before Nov 2016 and they continued to grow post-demonetisation at about the same rate. There is little direct evidence of any causal effect of demonetisation that was either long term or sustained.

Source and description of data

The data for this analysis come from publicly available data in the Database of Indian Economy published on its website by the Reserve Bank of India. Specifically, the time series data on Payment Systems Indicators records for each month the total volume (i.e. number of transactions in millions), and the total value of the transactions (in billions of rupees) for each of the non cash payments systems in use.

A screen shot of the data is attached in Appendix 1.

The payment systems indicators that I downloaded are:

##  [1] "Month"                 "RTGS_vol"             
##  [3] "RTGS_val"              "CCIL_vol"             
##  [5] "CCIL_val"              "Paper_vol"            
##  [7] "Paper_val"             "Retail_vol"           
##  [9] "Retail_val"            "IMPS_vol"             
## [11] "IMPS_val"              "NACH_vol"             
## [13] "NACH_val"              "AllCards_vol"         
## [15] "AllCards_val"          "CreditCards_vol"      
## [17] "CreditCards_val"       "CreditCardsATM_vol"   
## [19] "CreditCardsATM_val"    "CreditCardsPOS_vol"   
## [21] "CreditCardsPOS_val"    "DebitCards_vol"       
## [23] "DebitCards_val"        "DebitCardsATM_vol"    
## [25] "DebitCardsATM_val"     "DebitCardsPOS_vol"    
## [27] "DebitCardsPOS_val"     "PPI_vol"              
## [29] "PPI_val"               "mBanking_vol"         
## [31] "mBanking_val"          "CardsTotalNumber"     
## [33] "CreditCardsNumber"     "DebitCardsNumber"     
## [35] "ATMNumber"             "POSNumber"            
## [37] "DigitalGrantTotal_vol" "DigitalGrandTotal_val"

The other data in the RBI database are subsets of these; for instance RTGS data is broken down into Customer transactions, Interbankl transactions and others. I have restricted my analysis to the total for selected modalities of payment.

Analytical methods

Each column of the data set can be seen as a time series at regular monthly intervals. If a time-series is stationary or can be transformed into a stationary time-series by suitable transformation, it can be modelled using statistical regression. Such regression breaks down the perioic variation or noise into a smooth trend component, a seasonal component and what is left is a residual unexplained variation. The parameters of the regression equation allows the time series to be forecast into the future for any given number of periods, to generate an expected value at each time period and, because of the residual component, an 80% or 95% upper and lower confidence interval.

This is standard time series analysis, and the statistical programme R provides some powerful tools for the purpose. Keen readers should see the monograph by Rob J Hyndman

I used the R package ‘forecast’ to carry out the ARIMA analysis and the R package ‘ggplot2’ to generate the charts.

Analysis and Results

The analytical steps

I illustrate the steps in the analysis using the mobile banking payment systems indicator as a template before repeating the same analysis for the following indicators:

  1. Digital Grand Total by value
  2. Digital Grand Total by volume
  3. All cards by value
  4. All cards by volume
  5. Retail sales by value
  6. retail sales by volume
  7. RTGS payments by value
  8. RTGS payments by volume

Mobile payments by value

Chart 1 shows the raw data in a standard time series chart. I have partitioned the data into a pre and post Nov 2016 series to show the effect of DeMonetisation. The indicator was rising steeply from 2016 onward, post demonetisation there was an acute spurt upward that was not sustained. By mid 2017 it had fallen back to levels seen a year earlier. Since then there has been a steady rise with the slope about the same as 2 years earlier.

The key question is this: using just the data we had pre Nov 2016 what would be have predicted for, say April to Aug 2018? First we need to establish that this time series - i.e the data from April 2014 to Oct 2016 - is capable of being made stationary, i.e we need to understand the trend, seasonal and residual components of the data series. This partitioning of the variability is shown in Chart 2

Now we can use the forecast package in R to use this data to forecast the future. Note that for the purposes of this ‘thought experiment’ we act as if we are in Nov 2016; the only data we have are upto Oct 2016 and we ask ourselves, ‘What is the best prediction we can make how mobile banking payments will grow over the next 22 months - to Aug 2018. We then overlay the data that have infact observed and see if what we have observed was ’predictable’ back in Nov 2016.

Chart 3 shows the result of this analysis. The white line is the mean of the predicted values of the indicator, the shaded areas on either side are the 80% and 95% confidence bands, and the red line is the actual observed values of the indicator.

As can be seen the values observed are well within, indeed more recently they are slightly lower than waht we might have predicted in Nov 2016. I believe this is strong evidence of the lack of any appreciable effect of DeMonetisation on the value of mobile payments. They are higher than they were in late 2016, but then they have been growing steadily since 2014 and any growth post DeMonetisation is merely part of trend growth

Analysis and results of other payment system indicators

For the subsequent indicators, I present just the final plot of the forecast together with the actual data observed to draw a conclusion whether there was a change that was outside the range of statistically predictable values.

Volume of Mobile banking payments

Chart 4 shows a dramatic rise in the volume of mBanking transactions. It was clearly in excess of anything that could have been predicted in Nov 2016 using the data available up that point.

Chart 4 (mBanking volume) is in sharp contrast to Chart 3 (mBanking value). It would appear that whereas the total value of transactions done each month using mBanking has gone up in keeping with the pre Nov 2016 trend, the number of such transactions has gone up well beyond trend growth. This would suggest that the average value per transaction has fallen back sharply at some point after Demonetisation.

Average spend per transaction in mBanking payments

Chart 5 shows precisely this trend. I have not done any predictive modelling here, the chart tells its own story. Clearly, the average size of mobile Banking transaction rose briefly above 15,000 Rupees but then fell back rapidly over 2017 to levels of just over 5,000 Rupees that were last seen in early 2015. Possible reasons for this could be a backlash against charges for remittances and payments using mobile based payment services.

Total digital transactions by value

Total Digital Payments is perhaps the most pertinent indicator to track. This indicator is the sum of the following indicators:

  • 1.1 Customer transactions through Real-Time Gross Settlements (RTGS)
  • 1.2 Inter-bank RTGS transactions
  • 2 CCIL operated payments systems
  • 3 Paper Clearing
  • 4 Retail Electronic Clearing
  • 5 Card Payments (debit and credit cards)
  • 6 Pre-paid Payments Instruments (includes m-Wallet, PPI cards and Paper Vouchers)

Chart 6 shows that total digital transactions have been steadily growing throughtout the period 2004 to 2018. This growth has continued after DeMonetisation in Nov 2016 but this has been trend growth. The post Non 2016 growth is within the range that could have been predicted before DeMonetisation by statistical modelling using the data available at that point.

Total digital transactions by volume

Chart 7 shows that the volume of transactions included in the Total digital payments indicator rose sharply immediately after Demonetisation and after some moderation continued over the next 22 months to be just above the 95% upper confidence limit of the range that might have been predicted in late 2016 based on the data then available.

Plastic card payments by value

Chart 8 shows that the value of transactions made using Plastic cards after DeMonetisations is above the range that could have been predicted in Nov 2016 based on the trend data then available. But this finding needs to be seen in the context of the known pattern of debit card usage in India. Predominantly it is used to draw cash at ATMs.

Plastic cards transaction volumes

Chart 9 shows that the growth in number of transactions using plastic cards (debit and credit cards) has been steady but unspecatcular. The growth in the period since DeMonetisation is in the middle of the range of values that could have been predicted in Nov 2016 based on statistical modelling of the data available at the time.

Charts 8 and 9 need to be interpreted in light of known patterns of of card use in India. Debit cards are widely held. As of August 2018, Indians held 980 million debit cards but only 41 million credit cards. Debit cards are used in the main to draw cash out of ATMs. In August 2018, debit cards were used 805 million times in an ATM (with an average transaction amount of INR 3,400); and 357 million times at a Point of Sale terminal (with an average transaction amount of INR 1370).

Credit cards are used relatively more frequently at POS terminals (144 million transaction in August 2018, vs 0.8 million times at an ATM).

Immediate Payment Service (IMPS) payments

Chart 10 shows the trend growth in the total value of payments made using IMPS - the Immediate Payments Service. The growth in the period after DeMonetisation has been a continuation of the trend prior to that event. The actual growth observed is just above the predicted line but well within the 80% confidence range.

IMPS Payments transaction volumes

Chart 11 shows the trend growth in the volume of transactions settled using IMPS - the Immediate Payments Service. The growth in the period after DeMonetisation has been a continuation of the trend prior to that event. The actual growth observed is just above the prediction line but well within the 80% confidence range.

RTGS payments by value

Real Time Gross settlements are used for payments over INR 2 Lacs.

Chart 12 shows the predicted trend in growth using ARIMA modelling and the observed growth. After Demonetisation growth in RTGS payment volumes has continued largely on predictable trend.It fluctuates quite a bit (for example see the spike in 2012) but the observed numbers post-DeMonetsiation has stayed mostly with in the 80% confidence band, occasionally going above it but still within the 95% confidence band.

RTGS payments by volume

Chart 13 shows that the volume of transactions using the RTGS payments system has continued to grow. The growth after Demonetisation has been within the 95% range that could have been predicted in Nov 2016 using ARIMA modelling of the data available upto that date.

Currency in circulation

Currency in circulation of course dipped dramatically in the immediate aftermath of DeMonetisation. The RBI was ill-prepared to replenish the old notes with new ones fast enough.

Chart 14 shows how the currency in circulation fell and then gradually recovered. But has it recovered fully?

Some commentators have argued that thanks to the DeMonetisation of 8 Nov 2016, India now has a lower stock of currency in circulation than it might have had. The putative benefits of this are of course linked to the notion that increasing use of cashless or digital transactions means a more open and formalised economy, leading to reduced levels of tax evasion. This theory itself is open to question: witness how tax evasion and compliance are big issues facing Governments in rich countries too where the level of digital payments is far higher. Large business entities with teams of tax lawyers will always try inegnious ways to avoid paying tax.

Nevertheles, it is worht examining whether the growth in currency stock following Demonetisation has indeed been lower than expected. In order to answer this question I looked at the weekly data on RBI’s liability arising from the currency issued and in circulation. This data set goes back to the week ending 02 July 2004 and ends on week ending 26 Oct 2018.

Chart 15 shows the ARIMA model projection from Nov 2016 onwards. Clearly though cash has returned and the quantum of currency in circulation is in excess of the amount that prevailed before Demonetisation, it has not kept up with the projected growth in cash supply. The short fall amounts to about 2 trillion Rupees in much of 2018. It is important to note that this so-called shortfall is a statictical concept and not a real objectively defined shortfall. Exactly how much currency should be in circulation to enable ordinary commerical activity is for the RBI to decide. What my analysis shows is that assuming the growth in cash supply observed 2004 to late 2016 continued at about the same rate for the next 2 years then we would have expected some 2 lakh crores more in circulation than was in fact the case.

Chart 16 shows this gap in the expected amount of currency and the actual for the period following DeMonetisation. There are many ways to look at this gap. For one, it could be seen as a severe contraction of the currency supply through out calendar year 2017. Conditions eased somewhat towards the end of that year. and for much of 2018, the currency stock while still above pre-DeMonetisation levels, fellshort of the predicted growth by about 2 lakh crores.

Some have seen this as one of the objectives of the exercise - to make the Indian economy less dependent on cash and more reliant on digital or non-cash means of settlement of commercial transaction obligations. But that was always a post-hoc justification for DeMonetisation; it was not one of the principal or even originally stated aims. Besides, it is a moot point whether having a lower quantum of currency in circulation is a good or bad thing. There is no economic theory that holds that other things being equal, a formalised bank- and digital transactions-based economy provides greater citizen welfare.

Both means of using money as a medium of exchange - currency cash and digital money - come with costs.

With currency the costs of printing, securely storing securely, distributing to banking outlets, handling and counting, and the risk of theft and counterfeiting are borne by the State, the public sector. Only the cost of loss is borne by the private individual.

With digital money, the costs of maintaining the technological infrastructure, owning and maintaining Point of Sale machines, access to the internet, owning and maintaining the means with which to connect to the internet (computer or mobile), all fall on the individual citizen. These costs are what justify the commission charged by payments systems companies.

For many better off citizens, the ease and convenience of digital money transactions will justify the costs.

For many other poorer citizens, digital money systems are unaffordable and currency money offers them the only alternative to be involved in economic activity.

Summary of results

ARIMA modelling of the time series data upto October 2016 - the month before DeMonetisation - shows that for most of the digital payments systems indicators, the growth in the 22 months following (November 2016 to August 2018, the latest month for which RBI has published data) has been ‘on trend’.

Discussion

Newspaper reports as well as a research paper by Maiti SS of RBI suggest that there has indeed been a shift from cash and paper based systems to digital payments. These reports were based on analysis comparing the value of digital transactions a few months after Demonetisation with corresponding values before that event.

This approach is almost certaionly flawed because it does not take into account the fact that as India’s economy grows and as financial technology and the necessary digital infrastructure spreads the uptake of digital payment systems grows organically. Any increase over a period of time cannot confidently be causally attributed to a specific one-off policy intervention with any degree of certainty.

Ascribing causality is always fraught. The danger is that those who espouse a particular policy or defend a policy move by ‘their hero’ will have a natural inclination to pluck selectively at data before and after the policy to show that a change has occurred. But we need contantly to be aware of the well known logical trap known as the ‘post hoc ergo propter hoc’ fallacy. Translated from Latin it means ‘After this therefore because of this’. Just because event B occurs at some point in time after A, it does not follow that A caused B. To be able to show that A caused B, one would have to show convincingly that B would not have occured had it not been for A.

In the case of economic data like payments systems indicators there is steady change (usually growth, but sometimes also decay) over time, and there is fluctuation from one period to the next. By selectively choosing time points one could make whatever argument one wishes to. Therefore, with time series data the best and most rational approach is to consider the whole data set an look for patterns.

Time series analysis is a well developed technique. A time series can be modelled using standard statistical regression techniques and, just as with any other regression model, the model parameters can be used to ‘predict’ how it will pan out into the future. I use the word ‘predict’ within quotes to make the obvious point that this is not prediction in the astrological or tea-leaf reading sense but rather statistical prediction. As with all statistical prediction, there is a mean expectation and surrounding it one can set up a range within which the observed actual values can be expected to land with a given degree of confidence.

My approach in this analysis has been in the nature of a thought experiment. If we were back in time in early Nov 2016, when the Demonetisation that happened on 8 Nov was not on anyone’s horizon, and we looked at the RBI data set on payments system indicators, ans asked ourselves this question: “Given this data where do we think the numbers will be 15 months to 22 months later?”.

My analysis shows that what really happened - the actual observed data - was for the most part within the range that we might have predicted.

Demonetisation can be credited for at most a marginal effect on one or two digital payment systems indicators - namely, a) the number of mobile payments, though not on the total value of them,
b) the number of total digital transactions, but again not on the total value of them,
c) the value of platic card transactions but not on their number, but note here the predominant use of debit cards in India is to draw cash at ATMs.

But these account for relatively small amounts of money. The big ticket items are RTGS, and IMPS and here both the numbers of transactions and the total value are within the range that could have been predicted back in Oct 2016 with the data then available.

Conclusion

Demonetisation cannot be claimed to have led to a step change in the use of digital means of payments as opposed to cash.

Appendices

Appendix 1

Screenshot of RBI’s webpage showing the top few rows and columns of the payment systems data set.