A Different Approach to CIC Forecasting
Forecasting cash in circulation (CIC) is a key activity of central banks as it is an important variable in monetary policymaking, affecting overall liquidity in the economy and guiding their currency issuance operations.
The Bangko Sentral ng Pilipinas (BSP) has issued a working paper looking at a balance sheet approach to forecasting CIC 1.
Cash cycle
The paper starts by explaining the cash cycle and the need to estimate currency requirements by denomination based on currency management indicators and the prevailing macroeconomic conditions. It identifies four motives for demanding cash:
Transactions – Positively related to economic activity.
Opportunity costs – High inflation and/or interest rates reduce demand for cash.
Precautionary motive – Uncertainty and reduced appetites for risk reduce the demand for cash.
Other – For example, older people use more cash.
CIC in the Philippines
In 2020 economic activity fell and there were fewer person-to-person payments, which depressed cash transactions, but increased demand to hold cash as a precaution. Since then, economic conditions have improved and the long run shift of people working in family operated farms and businesses to sectors with formal wage structures resumed, both of which increased transactional demand.
CIC as a proportion of GDP grew at a rate of 6% a year between 2001 and 2013.
The mean was 6.3% and the mode 6.2%. These figures include 6.8% growth in 2008 and 6.9% growth in 2009 during the global financial crisis. From 2014 to 2019 the CIC growth rate accelerated to an average of 7.7%.
From 2016 to 2019 GDP and CIC both increased in step with each other, but in 2020 CIC rose to 11.4%, predominantly as a precaution. This is reflected in a steep increase in the volume of the highest value banknote withdrawn from the BSP, the P1,000, which rose to 33% of notes withdrawn, up from 28% in 2019.
From 2001 onwards a number of other factors drove this level of growth.
In 2004 the ratio of foreign currency deposits to peso deposits was 46.3%. By 2020 it was 17.1%. This drop had largely happened by 2013, when it was 18%.
Household consumption relative to GDP has also been strong. In 2020 it reached 75% but the lowest it reached was in 2009, when it fell to 70%.
The Philippines has enjoyed a low interest rate environment throughout.
People have a wide range of financial instruments available for use in transactions and as a store of value. This has reduced cash demand.
Cash forecasting
CIC is forecast based on macroeconomic variables that reflect or proxy the motives to hold cash. Real GDP is used to capture transactional demand and inflation opportunity cost. The flow equivalence of the aggregate CIC forecasts and unfit currency retirement are incorporated in models used to determine the currency order. The fitness of the banknotes, and their lifespan, are affected by economic conditions. The velocity of circulation determines the level of wear.
Generally central banks use statistical models such as Auto Regressive Integrated Moving Averages (ARIMA) and Co- integrating Regression Analysis (CRA) to forecast. Based on these, combined with the existing inventory and the buffer stock requirements, the final currency order by denomination can be set.
BSP is currently conducting studies to include digital transactions in its CIC forecasting.
Demand-side forecasting of CIC today
Literature is based on univariate models that depend on the series’ own history or time series models. These consider macroeconomic variables that affect the demand for cash. Seasonal ARIMA models are commonly deployed, which rely on previous observations of CIC to forecast its future value. Poland, the Maldives and Qatar are said to use this approach. The European Central Bank has employed both exponential smoothing and ARIMA techniques.
Macroeconomic variable ARIMA with exogenous variables (ARIMAX) and Vector Auto Regressive (VAR) models are also used. The Bank of England uses an error correction model to estimate a long run relationship between CIC and macroeconomic and currency management variables.
Indonesia uses a hybrid ARIMAX model and artificial neural networks to handle linear and non-linear correlations. This approach is necessary because of Eid al-Fitr.
The Banca d’Italia has used both ARIMA and breakpoint regression, plus ARIMA and VAR. It found that ARIMA outperforms more complex models in terms of forecast accuracy. Including macroeconomic variables did not translate to better predictive performance.
Academics research has found that ARIMA is better than pure expert knowledge, but a combination of both is good particularly when there are periods of significant and unexpected change in CIC.
BSP’s alternative approach
The BSP’s CIC approach described here is not intended to replace existing demand- side CIC models grounded on the motive to hold cash. The proposed models are a viable alternative in generating CIC forecasts, allowing better matching of currency supply and demand.
This approach explores using the balance sheet of the central bank as the basis for modelling. The assets and liabilities of BSP are employed to forecast levels of CIC using univariate tie series models. Central bank assets, especially foreign exchange purchases, can be converted into CIC.
This balance sheet approach is useful in generating monthly forecasts since it does not rely on GDP figures, which are generated quarterly.
Whenever BSP purchases foreign exchange (FX), it effectively sells the local currency which, left ‘unsterilised’, could end up as cash. Demand for local currency emanating from foreign exchange inflows may, therefore, be reflected in the balance sheet.
BSP assets equal its net foreign assets (NFA) and its domestic claim (DC) in relation to transactions with residents. Its NFA is dominated by its gross international reserves. In 2019 its NFA represented 90% of its assets but, due to BSP granting loans to the government during the pandemic, this had fallen to 87% in 2020.
The Philippines has strong FX inflows because it has strong macroeconomic fundamentals, remittances from overseas and inflows from exports. It has also benefited from the unconventional monetary policies of advanced economies which has moved FX liquidity to emerging economies. The result is an accumulation of BSP’s international reserves.
What makes up LOTRM
Liabilities other than reserve money (LOTRM) consists of CIC and BSP’s reverse repurchase facility, overnight deposit facility, term deposit facility and other equity and treasury International Monetary Fund accounts.
LOTRM as a proportion of BSP’s liabilities has risen from 32% in 2019 to 44% in 2020 due to the expansionary monetary policy pursued to stimulate the economy.
Assets and liabilities
At the end of 2020 BSP’s capitalisation was P500 billion and its accumulated reserves were P120.9 billion, equivalent to 2.4% of its assets.
Increasing BSP’s assets through reserve accumulation leads to an equal increase in liabilities. Increasing foreign reserve assets could be offset by increasing its liabilities through reserve money, although this is potentially inflationary if left unsterilised.
To be consistent with the inflation target, the increase in reserve money may be partially counteracted and mopped up by selling market instruments through open market operations, where BSP sell securities (buying local currency) with an agreement to buy (sell local currency) them back in the future, or by accessing BSP’s deposit facility.
Sterilisation can take place through the reserve purchase facility, overnight deposit facility or term deposit facility. These increase BSP’s liabilities under LOTRM, effectively offsetting the accumulation of assets while moderating inflationary risk. Local currency converted through FX purchases that are not sterilised could be kept as deposits in the banks or withdrawn as cash, thereby increasing CIC.
Given that reserve accumulation entails BSP’s sale of domestic currency to fund such FX purchases (and sterilising the domestic liability injected through the BSPs open market operations and deposit facility), it is not surprising that the correlation between GIR on the asset side and CIC and LOTRM on the liability side is significantly higher at 92.1%, as they generally move in tandem.
Framework in Forecasting CIC: Balance Sheet Approach
Assets equal liabilities and equity, but in this case equities are not included because they are negligible. This can be expressed as:
NFA + DC = RM + LOTRM
Reserve Money consists of CIC, Liabilities to Other Depository Corporations (LODC) and Liabilities to Other Sectors (LOS). To isolate CIC in the formula:
CIC = (NFA + DC) - (LODC + LOC + LOTRM)
Or, to express it more simply:
CIC = (Assets) – (Liabilities other than currency issued (LOTCI)
CIC is positively correlated to total assets, so if assets increase, CIC increases holding LOTCI constant. CIC is negatively related to LOTCI, so if assets are constant and LOTCI increases, CIC falls.
Methodology
Three time series models were explored: (i) an Autoregressive Distributed Lag (ARDL) Model, (ii) an Autoregressive Integrated Moving Average with Explanatory Variables (ARIMAX) Model, and (iii) an ARIMAX- Generalized Autoregressive Conditional Heteroskedasticity (GARCH) Model.
The work found that for deseasonalised CIC, lagged values of assets and lagged values of LOTCI are statistically significant as explanatory variables. The results were consistent with assets being positively associated with CIC.
The rise in BSP assets has been driven by reserve accumulation over the years. Some of the local currency converted through purchases of foreign reserve assets – which were not sterilised – end up as cash resulting in an increase in CIC.
As expected, LOTCI was negatively related to CIC. Holding assets fixed, an increase in LOTCI should lead to a decrease in CIC to maintain parity between assets and liabilities. If assets are held constant and LOTCI are increased – converting some of the circulating cash into deposits or other forms of liabilities – then CIC is expected to decline.
Finally, CIC levels in prior periods were shown to be related to CIC in the present period. This was expected as the historical behaviour of CIC should theoretically provide information on its future values.
In terms of in-sample fit, the goodness-of-fit statistics points to the ARIMAX and GARCH models as the superior models.
Conclusion
BSP’s work in this area took place during the pandemic. The paper points out that as the pandemic fades away, the precautionary motive may gradually decline. Equally, the transaction motive may gain ground as economic activity starts to recover.
As a result, it remains to be seen how the models will perform when CIC levels start reverting to their pre-pandemic behaviour, if that happens at all. The authors felt confident that the models would still produce fairly accurate forecasts, given the computed in-sample and out-of-sample performance statistics.
‘Demand for cash’ models are ultimately more interpretable than this balance sheet approach as they are anchored on inflation, economic output, dummy variables for economic shocks. Even if the balance sheet approach offers more frequent and potentially more accurate forecasts, the demand-based model still offers rich economic information.
Subscriber content
Read the full article
Full access to Cash & Payment News articles, newsletters and archives.