Leading edge ensemble econometric time series methods yield superior results, both for short term as well as at the desired level of granularities if leveraged with right attributes.
By Soumya Kanti Ghosh & Bikramjit Chaudhuri
With the ongoing innovations in the financial sectors, forecasting the demand of currency has become increasingly difficult for the central banks, the world over. This is especially true for the developing economies and emerging markets, and the South and South-east Asian countries (India included) are no exceptions to this. Because of a relatively buoyant economy in the recent years, a significant increase in cash holdings was witnessed in many of these economies which have not been anticipated well by some of the central banks. Therefore, the cost of printing currency notes incurred by these banks exceeded budgetary estimates.
One needs to evaluate the most prevalent existing practices of currency demand forecasting against this backdrop. Most of the previous modelling work on money aggregates in these banks had focused on their potential use in monetary policy, either as policy targets or as indicators. The previous empirical work had concentrated more on testing the stability of money aggregate demand functions, with the objectives to identify structural relationships based on economic theory. However, these traditional approaches do not always yield a highly precise outcome, especially for short-term forecasting. Accuracies of these outputs also are not always of the desirable order for granular predictions, particularly at the denominations levels.
In the Indian context, in the recently published Monetary Policy Report (MPR), there is a specific mention on short-term and long-term factors that impact currency demand in India. While the nominal GDP growth rate, rate of deposits and even festive season impact demand for currency in the long run, specific idiosyncratic factors such as elections, demonetisation and even festive demand also impact currency demand, but more in the short run. Specifically, the results show that the income elasticity of currency demand is marginally above unity, implying an almost one-to-one relation between the nominal GDP growth and currency growth.
Herein lies a contradiction and a possible structural change that might be currently impacting currency demand in a completely different manner. We believe that the presence and increased circulation of smaller denomination notes in the recent times necessitates a fresh look at modelling currency demand, focusing mostly at denomination levels.
For example, the growth in currency in circulation (CIC) to growth in nominal GDP that was on an average just below 1 during FY11-15, has jumped to around 1.5 in FY16 and FY19 (excluding the demonetisation year FY17 and remonetisation year FY18 as outliers). Such a large elasticity reflects primarily increasing and forced transactions through smaller currency notes like Rs 200 as larger notes are not getting adequately circulated, vindicated by a plunge in income velocity of money in FY19. For example, preliminary estimates show that the decline in income velocity in FY19 on a year-on-year basis has been the maximum since FY01.
Therefore, even with a 17% growth in CIC in FY19, we are currently looking at a GDP growth slowdown indicating CIC as an inadequate harbinger of economic activity, as it used to be earlier.
As an illustration, let us look at the notes printed in FY19. Approximating based on the indents placed by the Reserve Bank of India (RBI), we forecast that the total amount of Rs 2,000 notes in the system is currently around 3.36 billion pieces, while the Rs 200 note count could be at 5 billion. Therefore, if a transaction of Rs 2,000 is to be substituted with Rs 200 only, it implies that Rs 200 notes must be printed 10 times more, which will amount to as many as 50 billion pieces.
(Interestingly, Rs 10 notes are forecasted to be maximum at 35 billion pieces by March 2019). Alternatively, this means that to sustain a transaction of the same amount now, the volume of smaller currency notes goes up and, by default, the value of CIC also goes up, but this might not indicate a direct causation with economic activity.
So, where do we go from here?
We believe that leading edge ensemble econometric time series methods yield superior results, both for short term as well at the desired level of granularities if leveraged with the right attributes. Parameters such as withdrawal quantum by the member banks, retired currencies or soiled notes, amount of currency’s buffer stock and currencies in circulation used in conjunction with specific dummies yield reliable forecasted estimates both for the short term as well as by currency denominations. All such factors when augmented with shock events yield even higher accuracy. The impact of elections, for example, in some of these economies, once factored in the model, provides impressive outcomes.
There could be a number of reasons why a suitably crafted econometric time series model leveraging appropriate factors might be more reliable than the traditional structural demand function. For example, it is highly likely that innovation in payments technologies will have a strong impact on the demand for currency in the long term. As a result, a money demand equation that does not allow satisfactorily for innovation will be inadequately specified. Additionally, the structural relationships in an error correction model will only have an impact on the economy in the long run. So, while a structural model might still be useful for longer-term currency forecasting in some of these countries, for the short term and for granular level predictions by denominations, the approach outlined above (bottom up rather than top down) will always yield superior outcomes.
This is all the more important when the CIC has expanded by a sharp Rs 3 lakh crore in FY19 (we leave out Rs 5 lakh crore expansion in FY18 as it was the year of remonetisation), against an average of Rs 1.5 lakh crore from FY11-FY16.
(Ghosh is group chief economic advisor, SBI; Chaudhuri is senior vice-president, Datamatics Global Services. Views are personal)