Liquidity, Payment Systems, and the Real Economy: Implications from India’s Demonetization Xiaodan Gao∗

Na Zhang

National University of Singapore

Fudan University

Abstract With a cashless society gaining popularity across the world, the smooth functioning of payment systems becomes essential to facilitating transactions. In this paper, we examine the role of payment systems in supporting economic activity and price stability through the lens of India’s demonetization. We find sharp drops in real economic activity and aggregate price during the first two months following a demonetization shock, and show that the impact arises from underdeveloped payment systems. We develop a model to illustrate the mechanism through which payment systems affect real economy and use the model to estimate the effects of payment-system disruptions of different scales.

JEL Classification: C21, E21, E31, E52, G29 Keywords: Demonetization; Cash; Deposits; Payment systems; Real economy.

∗

Corresponding author. Department of Strategy and Policy, NUS Business School, 15 Kent Ridge Drive, Singapore. Email address: [email protected]. Zhang is sponsored by the Shanghai Pujiang Program. We would like to thank Yuri Tserlukevich, Ting Xu and Chu Zhang for insightful comments and suggestions.

1

1

Introduction

In recent years, numerous financial innovations have emerged to facilitate electronic and mobile payments. The development of electronic payment systems gradually turns the economy into a cashless one, featured as the gradual phasing out of currency from circulation and its replacement with convertible deposits. Sweden leads the trend toward a cashless society; cash transactions are expected to drop to 0.5% by 2020 (Kireyev, 2017). A similar de-cashing trend is observed in other Nordic countries and is gaining popularity across the world. To develop and support a cashless economy, pervasive and well-functioning payment systems are essential. Any wide-scale disruptions of payment systems, possibly due to cyber-hacking or technology failures, can lead to a systemic crisis. Understanding the resulting economic consequences will be instructive for future policy responses; nevertheless, such economy-wide payment-system malfunctions in a cashless economy have been rare, if not absent. This paper provides novel insights into the problem by drawing implications from India’s demonetization—86% of the currency in circulation is invalidated when payment systems are underdeveloped— and uses the event to understand the role of payment systems in supporting economic activity in a cashless society. On November 8, 2016, the Prime Minister of India in an unscheduled live televised address announced the demonetization of all 500- and 1000-rupee banknotes and the issuance of new 500- and 2000-rupee banknotes in exchange for the old ones, with the main objective to tackle black money and convert it into the banked and taxable parts of the economy. The demonetization came as a shock, and led to chaos in India. Using monthly data from the CEIC India Premium Database, we find that real economic activity fell by 5.4% within four months, while aggregate price dropped by 2.96% in the first two months and then increased

2

by 4.3%. In this paper, we empirically and theoretically demonstrate that the adverse impact of demonetization arises from underdeveloped electronic payment systems, and use our calibrated model to structurally estimate the real effects of payment-system disruptions of different sizes. Specifically, we first provide empirical evidence to support our conjecture regarding the importance of payment-system underdevelopment in India in explaining the contractionary effects of demonetization. We find that a 1% increase in the number of point-of-sale (POS) machines generates a 0.267% increase in economic activity. This suggests that without the expansion of payment systems— the 47% increase in POS machines—during demonetization, the economy would experience an additional 12.5% drop in economic activity, with the total decline amounting to 17.9%. We further exploit cross-state variation and find that states with a lower pre-demonetization POS diffusion rate experience more severe contractions, which confirms the importance of electronic payment systems in absorbing the negative impacts of demonetization. We then build a model to understand the mechanism. We modify the cashin-advance (CIA) model by allowing for electronic payments and incorporating a constraint on the supply of electronic-payment services, called a paymenttechnology constraint. This constraint limits the use of electronic money in transactions and determines the effective liquidity supply in the economy. In the model, households solve a standard consumption-saving problem. They value consumption, leisure, and real cash balances and face persistent shocks to money-supply growth. Cash balances and demand deposits are required to purchase consumption goods. We assume that cash earns zero interest but yields direct utility. The latter feature is used to capture all of the motives for holding cash—such as payment habits and tax avoidance—other than the limited access to electronic-payment services. Deposits are assumed to earn a return that is

3

proportional to risk-free nominal interest rates. This feature is used to reflect the interest earned on checking and saving accounts and any discounts on products bought with debit and credit cards. Deposits can be used to purchase consumption goods through electronic payments, but are subject to a constraint. To facilitate consumption purchases, households optimize their money holdings by managing cash balances and demand deposits. When demonetization pushes money to flow from cash into bank deposits, the payment-technology constraint starts to play a role: It reduces the effective aggregate liquidity that can be used for consumption purchases in the economy, which explains the large drop in economic activities. Moreover, the significant decline in effective liquidity drives down the anticipated inflation and leads to a lower aggregate price. As new banknotes are gradually injected back into the economy, households expect an increase in effective liquidity in the future, which pushes up prices and causes a sharp rebound in inflation. We use the demonetization shock to identify the constraints on electronicpayment services in India. Our calibrated model is able to reproduce the contractions and price dynamics following a demonetization shock. In particular, it predicts a 18% drop in aggregate consumption, very close to the estimate obtained in the empirical section. We further use our model to quantify the adverse effects of payment-system disruptions in a cashless economy. We find that when the scale of payment-system failures increases from 25% to 75%, the initial output drop and initial employment drop rise from 0.03% and 0.05% to 0.38% and 0.6%, respectively. Finally, we examine the effectiveness of monetary policy in a cashless economy. We find that traditional monetary policy remains effective, and the holding cost of deposits relative to risk-free bond can be another useful monetary policy instrument to stabilize the economy, similar to the policy tool proposed by Barrdear and Kumhof (2016).

4

Our paper contributes to the literature in three ways. First, it examines the effect of financial-technology (fintech) innovation, which has recently attracted significant attention from academic researchers, policy makers, and financial practitioners. Most of the literature focuses on its effects on information asymmetry (Cong, He, and Zheng, 2017; Liberti, Sturgess, and Sutherland, 2017); information generation (Franks, Serrano-Velarde, and Sussman, 2016); financial market efficiency (Farboodi and Veldkamp, 2017); the availability and cost of funding (Jagtiani and Lemieux, 2017; Parlour, Rajan, and Walden, 2016); and the performance of traditional financial intermediaries (Buchak, Matvos, Piskorski, and Seru, 2017; Wolfe and Yoo, 2017). This paper focuses on payment systems and examines their importance in facilitating transactions and stabilizing the real economy by relying on India’s demonetization to infer the costs of constrained payment systems in a cashless society. Second, it complements the literature on the impact of fintech innovation on households’ cash demand. Attanasio, Guiso, and Jappelli (2002) and Alvarez and Lippi (2009) study the role of the diffusion of automated teller machines (ATMs). Attanasio, Guiso, and Jappelli (2002) find that a wider availability of ATMs facilitates cash withdrawals and lowers average cash holdings, while Alvarez and Lippi (2009) argue that ATMs tend to increase households’ cash demand by generating a precautionary motive. Our paper differs from these studies by focusing on electronic payments that are facilitated by payment systems. We suggest that without a strong preference for holding cash, the development of electronic payment systems relaxes transaction technology constraints and reduces the demand for cash. Third, our paper sheds light on the degree of price flexibility in response to demand shocks. The speed of price adjustment is a central question in macroeconomics, which helps to determine the role of sticky prices for the monetary

5

transmission mechanism. However, no clear consensus on price rigidity has been reached in the literature (Bils and Klenow, 2004; Klenow and Kryvtsov, 2008; Nakamura, 2008; Kehoe and Midrigan, 2015). Our empirical result suggests a sizeable and prompt adjustment in aggregate price following demonetization. This challenges the importance of sluggish price adjustment as a leading explanation for the large effects of monetary policy on the real economy. The paper is organized as follows. Section 2 presents the estimated impacts of demonetization on Indian economy and illustrates the importance of payment systems in absorbing the negative effects. Section 3 lays out the model and characterizes the optimal money demand in the presence of payment system constraints. Section 4 derives the steady-state behavior of the aggregate economy and demonstrates the role of payment systems’ development in determining optimal liquidity demand. Section 5 performs quantitative analyses, shows the model’s ability to generate the contractionary effects of demonetization, and estimates the costs of payment-system disruptions. Section 6 concludes.

2

Empirical Facts

In this section, we empirically estimate the impact of India’s demonetization on domestic economy and demonstrate the importance of payment systems in supporting economic activity and price stability. We begin with a brief introduction of the event.

2.1

India’s Demonetization

On November 8, 2016, the Prime Minister of India announced the demonetization of all 500- (US$7.80) and 1000-rupee (US$16) banknotes in an unscheduled live televised address. In the announcement, he declared that the use of all Rs 500 and

6

Rs 1000 banknotes would be invalid after midnight and announced the issuance of new Rs 500 and Rs 2000 banknotes in exchange for the old ones. Rs 500 and Rs 1000 notes were commonly used in daily transactions and accounted for 86% of all banknotes in circulation. The Reserve Bank of India (RBI) set a window of 50 days—i.e., until December 30, 2016—to deposit the invalidated banknotes as credit in bank accounts, up to a daily limit that varied over time.1 The government and RBI also imposed limits on daily cash withdrawals from ATMs and bank branches to stabilize the flow of new currencies following demonetization. There are four main reasons for India’s demonetization. The first is to tackle black money and put it into the banked and taxable parts of the economy. The second is to reduce the cash circulated in the economy, which is directly related to domestic corruption. The third is to fight terrorism funded by counterfeit currency in India, and the last is to foster the use of bank accounts and digital payments to render Indian economy less cash dependent.

2.2

Demonetization and the Real Economy

We use regression analyses to determine the effect of demonetization on the real economy, using data collected from the CEIC India Premium Database. We use monthly data for estimation, with our sample covering the period from January 2011 to February 2017.2 Compared to quarterly data, monthly data provide more information about the dynamics of the economy during the post-demonetization period. We start by estimating the effect of demonetization on real economic activity, which is proxied by electricity demand in our analysis, as most economic activities 1

The limit was set at Rs 4,000 per person in the first week after the announcement, Rs 4500 per person in the second week, and Rs 2000 per person from November 18 on. All exchange of banknotes was abruptly stopped after November 25, 2016. 2 We use CPI indices with 2012 as the base year. The data series are from January 2011, which we use as the starting period of our sample.

7

nowadays require power. We use dummy variable Dt to represent the change in policy, which takes the value one for the post-demonetization period and zero otherwise. Subscript t denotes time. In addition, we control for weather— temperature and rainfall—to rule out the possibility that changes in electricity demand are driven by climate changes. We also include year dummies τy and month dummies φm .3 The former are used to capture the over-time changes in population and aggregate shocks that may affect power demand. The latter are added to account for the variation in electricity demand due to seasonal factors. Our baseline model is given as follows:

electricityt = α0 + α1 Dt + α2 weathert + τy + φm + t ,

(1)

where the dependent variable is in log form. Table 1 summarizes our sample’s descriptive statistics. Over the time period, the average log electricity demand is 11.36 thousand KGWH per month. The mean monthly maximum temperature in India is about 30.60 C and the mean monthly minimum temperature is about 19.56 C. The actual rainfall per month is 20.36 mm on average, with the mean rainfall deviation from average being -10.85 mm. [Table 1 about here.] Estimation results of regression model (1) are reported in Table 2. Column (1) presents results without other control variables except year and month dummies. Column (2) reports the estimates of our baseline specification, which uses maximum and minimum temperatures as weather controls. It suggests that on average, log electricity demand after demonetization fell by 0.056 from 11.47, 3

We have only two observations for 2017. To distinguish demonetization effect from time effect, we let 2016 and 2017 share the same year dummy.

8

which translates into a 5.4% drop in real economic activity. Column (3) uses an alternative measure of weather, and Column (4) includes the stock market index to capture households’ expectation for the future economy. Both confirm an adverse impact of demonetization. [Table 2 about here.] We next examine whether aggregate price responds to demonetization. This exercise helps to shed light on the degree of price rigidity following a demand shock. We use a specification analogous to regression model (1) and use logged consumer price index (CPI) as the dependent variable:

CPIt = α0 + α1 Dt + α2 weathert + τy + φm + t .

(2)

Table 3 summarizes the estimation results. As in Table 2, Column (1) reports results without weather controls, Columns (2)-(3) include different measures for weather, and Column (4) adds all covariates, including the stock market index. All coefficient estimates on demonetization are positive but not statistically significant, while all other control variables, with the exception of the stock market index, have no effect on aggregate price. [Table 3 about here.] At first glance, price appears to be sluggish and does not react to demonetization. However, this unresponsiveness may result from the situation in which price fluctuated over time and the positive and negative changes offset each other. To determine the reason for the near-zero overall effect, we take a closer look at the post-demonetization period by allowing the effects of the shock to vary over time. We include two dummies to the regression model. The first dummy equals 1 for the first two months after demonetization—November and December 9

2016—and zero otherwise. The second dummy equals 1 for January and February 2017 and zero otherwise. Results are reported in Column (5) of Table 3. Our estimates suggest that aggregate price dropped by 2.96% in the first two months, then increased by 4.3%. This prompt adjustment goes against the price-stickiness assumption often used in monetary theories. To rule out the possibility that the change in aggregate price was driven by a particular component of the CPI basket, we look into the changes in each main category—food and beverage, tobacco, clothing and footwear, housing, fuel and light, and miscellaneous. Results are reported in Table 4. [Table 4 about here.] Columns (1)-(6) report results for each component of CPI. The pattern by which that prices first fell and then rebounded is found across all categories, although of different magnitudes. Specifically, during the first two months, food and beverages experienced the largest drops by 4.8%, while tobacco and the miscellaneous category—which includes education, medical care, recreation, and so forth—were least affected, and fell less than 1%. During the subsequent two months, with more and more new banknotes circulating in the economy, prices started increasing; all categories other than food and beverage increased by more than 5%. We also examine the price responses of vegetables, which are of particular interest as a component of CPI, and report the results in Column (7). As expected, the perishable-goods industry suffered the most: On average, it experienced a 20% decline in price. Lastly, we provide state-level evidence regarding the large negative impacts of demonetization by using state-level dependent and independent variables. Slightly different from the baseline specification, we add state fixed effects and replace year fixed effects with state-specific year fixed effects. We use the former to control for the time-invariant differences in demographic and economic structures 10

across states, and the latter to account for the time-varying annual population in each state and state-specific shocks. [Table 5 about here.] Results are presented in Table 5 and confirm the findings derived above. On average, electricity demand dropped by 3.7% after demonetization, slightly less than the estimate obtained using aggregate data. Columns (3)-(5) present results for price. They suggest that prices dropped by 2.2% in the first month, recovered by 1.5% in the second, and continued to climb up by 12.2% afterwards.

2.3

The Role of Payment Systems

According to the RBI’s annual report on August 30, 2017, 99% of the demonetized notes were deposited or exchanged for new currency. This implies that most people have managed to preserve their fortunes. In addition, as shown in Figure 1, from October 28 to November 25, 2016, 77% of cash in circulation was reallocated into bank deposits, and time deposits were five times greater than pre-demonetization currency holdings. This suggests that households converted invalidated cash to bank deposits and had sufficient money in bank accounts for consumption. As such, neither a negative wealth shock caused by the elimination of black money and unreported income nor a reduction in money supply explains the observed economic contractions. A natural question arises: What is responsible for the negative impacts of demonetization? [Figure 1 about here.] We next show that the underdeveloped payment system in India is the main driver behind the significant contractions following demonetization. In particular, we exploit the over-time variation in the degree of payment-system development 11

after demonetization and show that the expansion of electronic payment systems during that period helps to absorb the adverse impacts of the shock. We use the diffusion of POS machines to proxy the development of electronic payment systems. Figure 2 plots the number of POS machines from April 2011 to February 2017. As shown, following demonetization, the supply of electronicpayment services surged dramatically: The number of POS machines increased by 47% within four months, from 1.51 million to 2.22 million. [Figure 2 about here.] To show the importance of electronic payment systems in muting the negative effects of demonetization, we control for the number of POS machines. We expect different effects of payment systems on the real economy before and after the shock and thus add two interaction terms to the regression models (1) and (2). One is the interaction between the pre-shock dummy and the log of lagged POS machine number, and the other is between the post-shock dummy and the log of lagged POS machine number. Tables 6 and 7 report the estimation results for economic activity and aggregate price, respectively. [Table 6 about here.] As shown in Table 6, the availability of POS machines has significant positive impacts on economic activity during demonetization. According to Column (4), a 1% increase in the number of POS machines generates a 0.267% increase in electricity demand. This implies that the 47% increase in the number of POS machines following demonetization prevented a further 12.5% drop in economic activity. In other words, without the expansion of POS during demonetization, the decline in real economic activity would amount to 17.9%. In Column (5), we control for the liquidity services provided by cash and find similar results.

12

[Table 7 about here.] Similarly, results in Table 7 support the idea that the growth in electronicpayment services largely absorbs the effects of demonetization on prices. A 1% increase in POS machines leads prices to rise by approximately 0.2%. This suggests that prices would have dropped further—by an additional 9.4%—after demonetization if the payment system had not been developed during the shock period. Note that using the lagged number of POS machines and including the demonetization dummy help mitigate endogeneity concerns by rendering our analysis less likely to suffer from reverse causality and problems with omitted variables. Moreover, even if reverse causality exists, it would imply a negative relation between number of POS machines and dependent variables. That is, low aggregate demand due to cash shortage would prompt retail stores to install POS machines and accept electronic payments, which is opposite to the effect we find above. We further demonstrate the robustness of our results by exploiting crosssectional variation—that is, the different diffusion rates of POS machines across states before demonetization. We construct state-level number of POS machines as follows. We find data on the number of POS machines offered by each bank and the number of each bank’s branches in each state. The sum of each bank’s POS number weighted by the branch share of each state in each bank is our measure of the POS number for each state, given by

P OSs,t =

X n

bank POSn,t ×

number of branchess,n,t , total number of branchesn,t

where s, n, and t denote state, bank, and time, respectively. We use the average growth rate of POS machines per 1000 people before demonetization in each state to proxy their corresponding degree of payment-

13

system development. We classify states in the top quartile of the distribution as the ones with high POS diffusion rates and those in the bottom quartile as the ones with low POS diffusion rates. The dummy variable, diffusion rate, is equal to 1 for the former group and zero for the latter. We then add the growth rate of POS machines and the interaction term between the demonetization dummy and diffusion rate dummy to the state-level analysis. To ease the interpretation, we focus on the period 2011M1-2016M11 to get rid of the effects of the POS expansion following demonetization in each state. The identification of the effects of payment-service availability comes from the cross-state variation in the degree of payment-system development before the shock. Results are summarized in Table 8. [Table 8 about here.] Our estimation results support the important role of payment systems in absorbing the negative impacts of demonetization. States with a low diffusion rate of POS machines experienced severe contractions—a 4.7% drop in economic activity and a 2.1% drop in price. By contrast, demonetization has a much weaker impact on states that have wider access to electronic-payment services. In those states, the economic activity did not decline and the drop in price was approximately 50% smaller. One potential concern for the interpretation of the effects of payment systems in Table 8 is that the POS diffusion rate may reflect the size of black money and unreported income in each state. That is, states with more black money and unreported income tend to rely more heavily on cash transactions and use electronic payments less frequently, which leads to a lower POS diffusion rate. Demonetization may then act like a large negative wealth shock, hit those states harder, and generate more severe contractions. However, as mentioned earlier, 99% of the invalidated notes came back to the central bank. This implies that 14

demonetization fails to achieve its main objectives and barely hurts black-money holders and households with unreported income. Overall, results in this subsection provide strong evidence in support of the importance of payment-system underdevelopment in explaining the contractionary effects of demonetization.

2.4

Discussion

In summary, we find significant contractions following demonetization. Real economic activity proxied by electricity demand shrank by 5.4%, while aggregate prices dropped by 2.96% in the first two months, then rebounded quickly and rose by 4.3%. We also show that payment systems play a key role in supporting the real economy. Without the development of payment systems during demonetization, economic activity and aggregate price would further decrease by 12.5% and 9.4%, respectively. We next present a parsimonious model featuring electronic-payment constraints and perfectly flexible prices to explain the phenomena found in this section.

3

Model

We present a modified version of the basic CIA model. In the model, the economy is populated by a unit measure of identical infinitely lived households and firms. Households value consumption, leisure, and real cash balances. They face persistent shocks to money-supply growth, a CIA constraint, and a paymenttechnology constraint. Firms hire labor and rent capital in a perfectly competitive market to produce a single good used for both consumption and investment, and face uncertainty from productivity.

15

3.1

Preferences, Technology, and Productivity Shocks

The representative household has a periodic utility function given by 1−Φ

u(ct , mct , 1

[ac1−b + (1 − a)mc1−b ] 1−b (1 − nt )1−η t t − nt ) = +Ψ , 1−Φ 1−η

(3)

where ct , mct and nt denote, respectively, the household’s consumption, real cash holdings, and labor supplied to market activities. The inclusion of real cash balances in the utility function is to capture the motives for holding cash in addition to transactional purposes, such as payment habits and tax evasion. Parameter a ∈ (0, 1] is the share assigned to consumption in utility, b controls the elasticity of substitution between consumption and real cash balances, Ψ is positive and captures the household’s preference for leisure, and Φ and η capture the elasticity of intertemporal substitution in consumption, cash balances and leisure. The household’s time preference is characterized by a constant discount factor β. Firms combine capital stock k and labor n to produce a good y. The economy’s production function has constant returns to scale and is given by

α yt = ezt kt−1 n1−α , t

(4)

where zt represents the total factor productivity and the parameter α ∈ (0, 1) is the share of capital in production. Firms rent capital and hire labor at respective market rates rt and wt . The exogenous productivity shock z is assumed to follow an AR(1) process,

zt+1 = ρz zt + εz,t+1 ,

16

εz ∼ N (0, σz2 ).

(5)

The parameter ρz governs the persistence of z, and the innovation εz is normally distributed with variance σz2 .

3.2

Government Sector and Money-Supply Shock

C D ¯ t−1 consists of cash Mt−1 The money supply M and demand deposits Mt−1 . The

former is assumed to earn zero interest, while the latter earns a positive interest rate (1 − χ)it−1 , where it−1 is the nominal interest rate on risk-free bonds and χ ∈ (0, 1) captures the cost of holding deposits relative to bonds. Interest payments on demand deposits are financed by lump-sum taxes T1,t , that is, D (1 − χ)it−1 Mt−1 = T1,t .

(6)

Money enters the economy through lump-sum transfers T2,t made to households by the government, which take the following form:

¯t = M ¯ t−1 + T2,t = (1 + θt )M ¯ t−1 , M

(7)

where the growth rate θt is random with mean θss . The deviation ut = θt − θss follows an AR(1) process given by

ut+1 = ρu ut + εu,t+1 ,

εu ∼ N (0, σu2 ),

(8)

where εu is the innovation to the money growth rate and ρu is the shock persistence.

17

3.3 3.3.1

The Household’s Problem Timing

The timing of events within a period is as follows. Households enter the period C D with cash Mt−1 , demand deposits Mt−1 , and one-period risk-free nominal bonds

Bt−1 . They learn the current-period shocks zt and θt and receive transfers Tt = T2,t − T1,t = Pt τ1,t − Pt τ2,t = Pt τt from the government, where Pt is the aggregate price at period t. Households supply labor and rent capital to firms. The goods market opens first. Households purchase investment goods I with credit, which will be settled at the end of the period. Purchases of investment goods augment the household’s capital stock according to the accumulation equation: It = kt − (1 − δ)kt−1 ,

(9)

where the parameter δ ∈ (0, 1) is the capital depreciation rate. The purchase of consumption goods c is subject to a CIA constraint. That is, households purchase consumption goods with cash, demand deposits, and transfers. In addition, households face a payment-technology constraint: Deposits can be used to purchase consumption via electronic payments, but cannot exceed current electronic-payment capacity.4 The constraint in real terms takes the following form:

ct ≤

 mct−1 md + τ2,t + min [1 + it−1 (1 − χ)] t−1 − τ1,t , γ m ¯ ss , 1 + πt 1 + πt

4

(10)

Note that our model assumes away the possibility that households can finance consumption by making cash withdrawals from deposit accounts. This is a reasonable assumption due to the tax-avoidance motive. More specifically, cash can be used to protect income from taxation, while deposits cause income to be monitored and cannot act as a tax shelter. As such, if households eventually need to withdraw money from bank accounts and purchase consumption goods with cash, they would have no incentive to deposit money in their bank accounts in the first place. Alternatively, the constraint on payment technology can be extended to a constraint on the availability of ATMs and electronic-payment services. The resulting optimal choice between deposits and cash will be unchanged.

18

where πt =

Pt Pt−1

− 1 is the inflation rate, mdt−1 is the real demand deposits

carried from the preceding period, parameter γ ∈ [0, 1] captures the degree of payment-technology development, and m ¯ ss is the real money supply at steady state. Inequality (10) suggests that electronic-payment development γ affects the economy’s total effective liquidity for consumption. Following the closure of the goods market, households receive their labor earnings and capital rents from firms, given the wage rate and real interest rates. They enter the asset market, in which they manage their asset portfolios by allocating resources in cash, demand deposits, and nominal risk-free bonds. The budget constraint that households face in real terms is given by

ct + It + bt + mct + mdt ≤ yt + (1 + it−1 )

mc md bt−1 + t−1 + [1 + it−1 (1 − χ)] t−1 + τt . 1 + π t 1 + πt 1 + πt (11)

Note that because technology is subject to constant returns to scale and markets are perfectly competitive, households’ labor and capital income in period t is equal to real output yt . 3.3.2

Set-up

The households’ objective is to maximize their lifetime expected utility by choosing consumption, investment, working hours, real bond holdings, real cash balances, real demand deposits, given the wage rate, real interest rates, and inflation rates. The optimization problem can be characterized by the following Bellman equation:

 V (zt , kt−1 , mct−1 , mdt−1 , bt−1 ) = max u(ct , mct , 1 − nt ) + βEV (zt+1 , kt , mct , mdt , bt ) , (12)

19

subject to constraints

ct ≤

 mct−1 md + τ2,t + min [1 + it−1 (1 − χ)] t−1 − τ1,t , γ m ¯ ss , 1 + πt 1 + πt

ct + It + bt + mct + mdt

mct−1 mdt−1 bt−1 ≤ yt + (1 + it−1 ) + + [1 + it−1 (1 − χ)] + τt , 1 + π t 1 + πt 1 + πt mct ≥ 0, mdt ≥ 0.

3.4

Market Clearing

In the economy, we have four markets to clear—the goods market, bond market, money market, and labor market:

yt = ct + kt − (1 − δ)kt−1 ,

(13)

Bt = 0,

(14)

¯ t = M C + M D, M t t

(15)

and the labor market clears by Walras’ law.

3.5

Optimal Money Demand

In this subsection, we characterize households’ optimal money demand—cash holdings and demand deposits.

20

3.5.1

Optimal Cash Holdings

The condition for optimal real cash balances is given as follows:

λt = umc (ct , mct , 1 − nt ) + βE(λt+1 + µt+1 )

1 , 1 + πt+1

(16)

where λ and µ denote the Lagrangian multiplier associated with the budget constraint and the CIA constraint, respectively, and umc is the first derivative of the utility with respect to real cash balances. The left-hand side of equation (16) gives the marginal cost of carrying one additional unit of real cash balances into the subsequent period, which is the foregone marginal utility of resources in the current period. The right-hand side represents the marginal benefit and has three components. The first is the marginal utility directly yielded by real cash balances, umc (ct , mct , 1 − nt ), which reflects the household’s strong preference for holding cash. The second is the expected present value of cash added to the resources (λt+1 ) in the subsequent period. The last is the liquidity services (µt+1 > 0) provided by cash to purchase consumption goods in the subsequent period.

3.5.2

Optimal Demand Deposit

Similarly, the optimal demand deposit is given by

λt = βE[λt+1

1 + it (1 − χ) 1 + it (1 − χ) + µt+1 1 mdt [1+it (1−χ)] ], −τ1,t <γmss 1 + πt+1 1 + πt+1 1+πt+1

where 1 mdt [1+it (1−χ)] 1+πt+1

−τ1,t <γmss

(17)

is the indicator function and equals 1 if the specified

condition in the subscript is satisfied. Unlike cash, the demand deposit yields zero direct utility. However, it earns a return proportional to nominal interest rates, and thus adds more to the

21

resources in period t + 1 and provides more liquidity services in period t + 1 if the payment-technology constraint is not binding. We next illustrate how the payment constraint affects households’ choices between cash and demand deposits in steady state.

4

Steady-state Equilibrium

In this section, we derive the steady-state behavior of the aggregate economy and characterize the role of limited access to electronic payment services in affecting households’ portfolio choices.

4.1

Monetary Neutrality and Nonsuperneutrality

Our model departs from the basic CIA framework in three dimensions. First, we decompose money into cash and demand deposits. Second, we allow households to earn a non-zero interest rate on demand deposits; however, this yields no direct utility. Third, we impose an electronic-payment technology constraint. Including these features does not change the key properties of the basic CIA model. Money remains neutral—an increase in the nominal money supply has no impacts on the economy’s real equilibrium—as a result of perfect flexible prices in the model. However, the model exhibits monetary nonsuperneutrality, that is, monetary policy affects the real economy through inflation. This is because households’ utility depends on real cash balances and leisure. A change in inflation affects the costs of holding cash, and thus leads to a portfolio adjustment between cash and demand deposits. The change in real cash balances, in turn, affects labor supply, output, capital investment, and consumption.

22

4.2

Steady-state Cash and Demand Deposits

However, the three new features impact the choice between cash and demand deposits. Specifically, the first feature makes no changes in the costs and benefits of holding cash and demand deposits; adding it to the basic CIA model does not change model solutions. The second feature differentiates their holding costs. The interest earned on demand deposits is proportional to the nominal interest rate, which is determined by inflation. Because of this, the inflation cost of holding deposits is smaller than that of holding cash, and changes in inflation will cause resource reallocation between them. The last feature, which is the emphasis of our paper, imposes a friction on holding deposits. The constraint on payment technology results in the illiquidity of excess deposits and makes cash more desirable. We next characterize steady-state cash and deposit decisions in the following proposition and denote steady-state values using the superscript ss. Proposition 1 Assume A =

1+θss , (1−χ)(1+θss −β)+1+θss

then in the steady state, optimal

cash and demand deposits have the following properties: (1) if 0 < a < A, then mcss = m ¯ ss , mdss = 0 and µss = 0; (2) if a = A, then mcss = m ¯ ss , mdss = 0 and µss > 0; (3) if A < a <

A , (1−γ)b

1

1

then mcss = A b m ¯ ss , mdss = (1 − A b )m ¯ ss < γ m ¯ ss , and

µss > 0; (4) if

A (1−γ)b

≤ a ≤ 1, mcss = (1 − γ)m ¯ ss , mdss = γ m ¯ ss , and µss > 0.

Proof. From the Euler equations for cash holdings and consumption, we derive the following condition: umc (css , mcss , 1 − nss ) 1 − a css b (1 − χ)iss (1 − χ)(1 + θss − β) = ( c ) = = , c uc (css , mss , 1 − nss ) a mss 1 + iss 1 + θss (18) 23

where the last equality follows that πss = θss and 1+iss = (1+πss )(1+rss ) =

1+θss . β

1

(1−a)(1+θss ) ] b css . Rewriting equation (18) gives that mcss = [ a(1−χ)(1+θ ss −β) 1

(1−a)(1+θss ) When [ a(1−χ)(1+θ ] b > 1, the CIA constraint does not bind, and therefore ss −β)

risk-free bonds are preferred to demand deposits because of higher interest earnings. In this case, households allocate all money to cash. 1

(1−a)(1+θss ) When [ a(1−χ)(1+θ ] b = 1, mcss = css = m ¯ ss . The latter equality can be ss −β)

proved by contradiction. Suppose mcss = css < m ¯ ss . Complementary slackness implies that µss = 0 and thus mdss = 0, which contradicts with mcss < m ¯ ss . 1

(1−a)(1+θss ) When γ < [ a(1−χ)(1+θ ] b < 1, mcss < css = m ¯ ss . The proof of the latter ss −β)

equality is similar to the one above. It is straightforward to show that if it does not hold, µss = 0 and thus mdss = 0. In other words, households hold demand deposits only if they provide liquidity services. 1

(1−a)(1+θss ) When [ a(1−χ)(1+θ ] b ≤ γ, the proof is by assuming the contrary. Suppose ss −β)

mdss > γ m ¯ ss . Due to the electronic-payment constraint, the marginal benefit of holding an additional unit of demand deposits is lower than that of bond savings, which indicates that households would have incentives to lower their deposit holdings. This adjustment will continue until the marginal unit of demand deposits can be used to purchase consumption goods via electronic payments, that is, mdss = γ m ¯ ss and µss > 0.

4.3

The Role of Payment Constraints

To show the effects of payment constraints on the choice between cash and demand deposits, we need to understand the different incentives for carrying them in different scenarios. The results established in Proposition (1) allow us to do so. Preferences for cash and pure liquidity services are captured by 1 − a and µ, respectively. In case (1), the preference for cash holdings is so strong (0 < a < A) that all money is allocated to cash. In such a scenario, the marginal unit of cash 24

does not provide liquidity services (µ = 0), and deposits have zero value and are strictly dominated by cash and risk-free bonds. In case (2) when a increases and equals A, the benefit of holding cash remains higher than that of holding demand deposits. But in this case, the marginal unit of cash is held to serve two purposes: satisfy preferences and meet liquidity needs for consumption purchases. In case (3) when a continues to increase but is lower than

A , (1−γ)b

households

start holding demand deposits. Cash accounts for a certain fraction of money demand, which is between 1 − γ and 1. In this case, the incentives for holding cash are the same as case (2), and demand deposits are positive and provide liquidity services. A In the last case ( (1−γ) b ≤ a ≤ 1), payment constraints start to play a role and

cash is held for an additional reason compared to case (3). Due to households’ limited access to electronic-payment services, an additional unit of demand deposits cannot be translated into liquidity used to facilitate transactions. To increase the effective liquidity for consumption purchases in future, households choose to hold more cash and keep demand deposits at the level of electronicpayment capacity γ m ¯ ss . The discussion above suggests that when the preference for cash is relatively weak, the presence of electronic-payment constraints shapes households’ demand for cash and deposits and determines the effective aggregate liquidity in the economy.

5

Calibration and Quantitative Analysis

In this section, we apply the model described in Section 3 to data. We start by calibrating model parameters and show that our model is able to reproduce

25

the contractionary effects of India’s demonetization. We then use the calibrated model to estimate the real effects of payment-system disruptions in a cashless economy.

5.1

Calibration

The time period t in the model corresponds to one month. We group model parameters into two sets. The first set consists of parameters whose values are standard in the literature. The second is determined such that the implied steadystate features of our model are aligned with aggregate Indian data from 2004 to 2016. We set the household time discount factor β = 0.998 to match an average annual real interest rate of 2.3% during the sample period.5 The production parameter α is set to 0.36, following Bosworth, Collins, and Virmani (2006). The capital depreciation rate δ is set to 0.25% per month, which indicates an annual depreciation of 3% to match the average undepreciated capital ratio of 97% in the sample period. This is close to the annual estimate 4.26% provided by Erumban and Das (2014), who use a sample period from 2000-2013. The average monthly growth rate of M 1 in India during 2004M1-2016M10 is 1.1%.6 We therefore set θss = 0.011. Parameters that govern household utility include Ψ, η, Φ, a, and b. The value of Ψ is chosen to be 0.88, so that the household spends one-third of its time in market work. Estimates of the Frisch elasticity of labor supply for India are rarely available and vary widely (Skoufias, 1996; Gine, Martinez-Bravo, and Vidal-Fernandez, 2016). Given that India has low per capita income, we choose ηnss −1 the elasticity ( 1−n ) = 0.5, which is the lower bound of the estimates [0.5,1] ss

for developed countries. This implies that η = 4. Kapoor and Ravi (2017) exploit 5 6

The relatively low real interest rate is caused by the high inflation rates during the global financial crisis. Here we exclude the post-demonetization period to eliminate its effects on the estimate.

26

a change in the Indian banking legislation in 2011 to estimate the intertemporal elasticity of substitution in consumption

1 Φ

= 2.2. We follow their estimate and

set Φ to be 0.45. The elasticity of cash demand with respect to nominal interest rates is given by 1b , and we choose b = 4, the midpoint of the range [1.39, 8] suggested by the literature (Mammen, 1999; Padhan, 2011). The calibration of parameter a will be discussed later. We have two new parameters in the model—the costs of holding demand deposits relative to bond savings χ and the tightness of electronic-payment constraints γ. We set χ = 0.4 to capture the interest rates earned on demand deposit accounts and any purchase discounts provided by the linked debit and credit cards. We identify the underdeveloped payment systems γ together with cash-holding preferences 1 − a. As illustrated in Proposition 1, it is not easy to disentangle the motives behind the marginal unit of cash holdings, which could be due to either strong preference or limited access to electronic-payment services. However, demonetization allows us to tell them apart: It forces households to exchange cash for deposits, regardless of holding motives, and prompts them to purchase consumption goods by electronic payments due to cash shortage. As such, the average electronic payment-to-M 1 ratio in November 2016 gives us the estimate of the tightness of payment constraint γ in India, which is 0.448. Considering the possible endogenous development of electronic payment facilities induced by demonetization within that month, we proxy the expansion of the electronicpayment system using the difference between the growth rate of POS terminals in November 2016 and the average growth rate of POS terminals from October 2015 to October 2016, which is 3.6%. We therefore set γ = 0.412. Given γ, we choose a to match the average cash-to-M1 ratio 0.6 during the pre-demonetization period, which gives a = 0.999. Our estimates of payment-constraint and cash-preference

27

parameters suggest that before demonetization, payment constraints in India were not binding and households held cash due to strong preference. Table 9 summarizes the parameter values. [Table 9 about here.]

5.2

India’s Demonetization

We next use our calibrated model to study India’s demonetization and show that the payment-technology constraint is the crucial element in reproducing the key patterns of the event. To model the Indian government’s demonetization policy, which invalidates a large fraction of currency notes, we now add to the baseline model a demonetization shock. We assume that the banknote supply is sufficient to meet households’ cash demand in normal times and is equal to the real money supply κmss with κ = 1. Demonetization is modeled as an unanticipated decrease in κ. We set ∆κ=-0.6 to match the drop in cash-to-M 1 ratio to 40% during the post-demonetization period and allow κ to gradually recover over time. All other parameters are fixed at their values in Table 9. Figure 3 illustrates the responses of key economic variables to a demonetization shock. The blue solid line with circles plots the dynamics for the model without a payment-technology constraint, while the black solid line with stars shows the patterns for our model. [Figure 3 about here.] As shown, without a payment constraint, demonetization has little effects on the real economy. In response to a significant drop in the currency supply, households simply reallocate money from cash to deposits and purchase goods via electronic payments, which forces electronic payments to go up by 50%. 28

Total effective liquidity in the economy, which is the sum of cash and electronic payments, barely changes; inflation, consumption, investment, employment, and output also remain almost unchanged. By contrast, in an economy with an electronic-payment constraint, a -60% cash-supply shock generates a remarkable reduction in effective aggregate liquidity. Demonetization shifts money from cash to bank deposits—which cannot be fully used for consumption, because of limited access to electronic-payment services. Consumption therefore drops significantly by 18.8%. To compensate for the loss in utility, households substitute toward leisure and cut labor supply, which ultimately leads to a decline in output. The small drop in output arises from the low value of leisure and thus a minor decline in employment level. Goods market clearing implies a significant increase in capital investment. In addition, inflation initially falls following the shock, then rebounds quickly and moves toward steady state gradually, which is in line with the empirical findings in Section 2. This interesting pattern is generated by the expectation of effective liquidity in the economy. Following demonetization, electronic-payment constraints limit the use of electronic money. This reduces the effective liquidity in the economy and drives down prices. However, the situation reverses itself in the subsequent period. As the government injects more and more cash back into the system, households expect a continued increase in effective liquidity in the future, which in turn pushes up prices and leads to a sharp increase in inflation.

5.3

The Value of Payment Systems

Furthermore, we quantify the value of payment systems by performing the following two exercises. First, we estimate the welfare cost of India’s demonetization resulting from payment-system underdevelopment and report results in Panel A of Table 10. To 29

do this, we compute the present utility during the post-demonetization period for the model with different degrees of payment-constraint tightness: λ = 0.412, λ = 0.5 and λ = 1. The ratio of the utility value for the model with constraints to that without constraints gives the welfare loss. With our baseline calibration λ = 0.412, we find that the welfare loss is 1.43% during the first 3 years following demonetization. If the constraint λ is relaxed to 0.5, the welfare loss drops to 0.64%. This suggests that a 20% increase in the supply of payment services would reduce the welfare loss of demonetization by 45%, which is economically significant. [Table 10 about here.] Second, we use our model to estimate the real effects of payment-system disruptions in a cashless economy. Specifically, we consider a demonetization shock which invalidates all the currency circulated in the economy (∆κ = −1) and examine the short-run responses of output and employment and long-run welfare losses in the presence of payment constraints. We vary the degree of constraint tightness λ and use it to proxy different sizes of payment-system disruptions. Results are reported in Panel B of Table 10. We find that when the scale of payment-system failures increases from 25% to 75%, the initial output drop rises from 0.03% to 0.38%, the initial reduction in employment goes up from 0.05% to 0.6%, and the 3-year welfare loss increases from 4.94% to 10.7% as a result of significant drops in aggregate consumption.

5.4

Monetary Policy in A Cashless Economy

Lastly, we examine our model’s policy implications. The popularity of de-cashing has triggered heated discussion of the conduct of monetary policy in a cashless economy (Fung and Halaburda, 2016; Barrdear and Kumhof, 2016). In this 30

section, we discuss aspects of this topic. In particular, we study the effectiveness of traditional monetary policy when transaction settlement takes place with regular recourse to central bank reserves. To do so, we extend our baseline model by relaxing the frictions imposed on deposits and allow the following: (i) deposits generate direct utility the same as cash, with the new utility function u(ct , mt , 1 − nt ), and (ii) electronic payment services are widespread (γ = 1). Under these assumptions, cash is dominated by demand deposits, and all transactions are conducted by electronic payments. As shown by the blue solid line with circles in Figure 4, monetary policy remains effective. In response to a 2.5% money-supply shock with persistence ρu = 0.5, inflation increases by 3.65%, while real money balance and consumption fall by 1.14%. Due to the low value of leisure, employment and output barely change. [Figure 4 about here.] Moreover, the holding cost for deposits relative to risk-free bonds appears to be a feasible policy instrument in addition to open market operations, and is similar to the monetary policy tool proposed by Barrdear and Kumhof (2016). The black solid line with stars in Figure 4 illustrates the idea. A higher holding cost (χ) pushes up expected inflation. In the case of χ = 0.9, inflation rises by 4.14%, which is 13% higher than that without managing deposit-holding costs (χ = 0.4). Higher inflation generates a larger drop in real money balances and consumption. As a result, households value leisure more and further cut employment which leads to a larger drop in output. On the other hand, more current resources are allocated to capital investment.

31

6

Conclusion

In this paper, we examine the effects of payment systems on the real economy in a cash-free society by gaining insights from India’s demonetization. That is, we infer the role of payment systems in supporting economic activity in a cashless economy by analyzing the effects of underdeveloped payment systems on economic activity when a majority of currency in circulation is invalidated. We empirically find significant drops in economic activity and aggregate price following a demonetization shock, and provide evidence to support the importance of underdeveloped payment systems in India in explaining the economic contractions. To illustrate the mechanism, we propose an extension of the classic CIA model. In particular, we decompose money into cash and deposits and study households’ choices between these two margins in the presence of payment constraints. We show that payment system underdevelopment limits the effective aggregate liquidity in the economy and is the key to generating the negative impacts of demonetization. The calibrated model successfully reproduces the effects of the event and is used to estimate the value of payment systems. We also use our model to examine the effectiveness of monetary policy in a cashless economy. We find that traditional monetary policy remains feasible, and the holding cost for deposits relative to risk-free bonds can be another useful policy instrument to stabilize price and output.

32

References Alvarez, F. and F. Lippi (2009). Financial Innovation and the Transactions Demand for Cash. Econometrica 77 (2), 363–402. Attanasio, O., L. Guiso, and T. Jappelli (2002). The Demand for Money, Financial Innovation, and the Welfare Cost of Inflation: An Analysis with Household Data. Journal of Political Economy 110 (2), 317–351. Barrdear, J. and M. Kumhof (2016). The Macroeconomics of Central Bank Issued Digital Currencies. Bank of England Staff Working Paper No. 605 , 1–90. Bils, M. and P. Klenow (2004). Some Evidence on the Importance of Sticky Prices. Journal of Political Economy 112 (5), 947–985. Bosworth, B., S. M. Collins, and A. Virmani (2006). Sources of Growth in the Indian Economy. Brookings Institution 3 (1), 1–69. Buchak, G., G. Matvos, T. Piskorski, and A. Seru (2017). Fintech, Regulatory Arbitrage, and the Rise of Shadow Banks. NBER Working Paper Series, 1–76. Cong, L., Z. He, and J. Zheng (2017). Blockchain and Smart Contracts. Working Paper, University of Chicago, 1–43. Erumban, A. and D. Das (2014). Role of Capital in India’s Economic Growth: Capital Stock versus Capital Services. The IARIW 33rd General Conference (August), 1–46. Farboodi, M. and L. Veldkamp (2017). Long Run Growth of Financial Technology. Working Paper, New York University, 1–48.

33

Franks, J., N. Serrano-Velarde, and O. Sussman (2016). Marketplace Lending, Information Aggregation, and Liquidity. Working Paper, London Business School , 1–34. Fung, B. and H. Halaburda (2016). Central Bank Digital Currencies: A Framework for Assessing Why and How. Bank of Canada Staff Working Paper (November), 1–27. Gine, X., M. Martinez-Bravo, and M. Vidal-Fernandez (2016). Are Labor Supply Decisions Consistent with Neoclassical Preferences? Evidence from Indian Boat Owners. IZA Discussion Papers, 1–40. Jagtiani, J. and C. Lemieux (2017). Fintech Lending: Financial Inclusion, Risk Pricing, and Alternative Information. Working Paper, Federal Reserve Bank of Philadelphia, 1–46. Kapoor, M. and S. Ravi (2017). Elasticity of Intertemporal Substitution in Consumption in the Presence of Inertia: Empirical Evidence from a Natural Experiment. Management Science, forthcoming. Kehoe, P. and V. Midrigan (2015). Prices are Sticky after All. Journal of Monetary Economics 75 (October), 35–53. Kireyev, A. (2017). The Macroeconomics of De-Cashing. IMF Working Papers, 1–26. Klenow, P. J. and O. Kryvtsov (2008). State-dependent or Time-dependent Pricing: Does It Matter for Recent U.S. Inflation?

Quarterly Journal of

Economics 113 (3), 863–904. Liberti, J., J. Sturgess, and A. Sutherland (2017). Information Sharing and Lender

34

Specialization: Evidence from the U.S. Commercial Lending Market. Working Paper, MIT , 1–55. Mammen, T. (1999). In India’s Economic Prospects: A Macroeconomic and Econometric Analysis, pp. 1–380. Nakamura, E. (2008). Pass-Through in Retail and Wholesale. American Economic Review P&P 98 (2), 430–437. Padhan, P. (2011). Stability of Demand for Money in India: Evidence from Monetary and Liquidity Aggregates. International Journal of Economics and Finance 3 (October), 271–282. Parlour, C., U. Rajan, and J. Walden (2016). Making Money: Commercial Banks , Liquidity Transformation and the Payment System. Working Paper, University of Michigan, 1–33. Skoufias, E. (1996). Intertemporal Substitution in Labor Supply: Evidence from Micro Data. Journal of Development Economics 51, 217–237. Wolfe, B. and W. Yoo (2017). Crowding out Banks: Credit Substitution by Peer-to-Peer Lending. Working Paper, SUNY Buffalo (1-56).

35

100

18

80 90 time deposits

currency and demand deposits 10 12 14 16

70

8

time currency

deposit

time deposit

Figure 1: Money Allocation after Demonetization. This figure plots the dynamics of currency with the public, demand deposits by the public, and time deposits with banks. The vertical red line marks the start of demonetization. The sample is constructed from the database maintained by the Reserve Bank of India over the period 2013-2017. The units of variables are trillions.

36

2.5 number of POS (million) 1.5 2 1 .5

Figure 2: The Growth of Electronic Payment Services after Demonetization. This figure plots the total number of point of sale (POS) terminals in the economy over time. The vertical red line indicates November 2016. The sample is constructed from the CEIC India Premium Database over the period from April 2011 to February 2017.

37

Output (y)

Consumption (c)

0.2

5 0 percent

percent

0.1 0 −0.1 −0.2

without payment constraint with payment constraint 5

10

15

−5 −10 −15 −20

20

5

60

0.1

40

0 −0.1 −0.2

15

20

Electronic Payment

0.2

percent

percent

Employment (n)

10

20 0

5

10

15

−20

20

Inflation (π)

5

10

15

20

Demonetization Shock (κ)

40

0

percent

percent

20 0

−20

−40

−20 −40

5

10

15

−60

20

5

10

15

20

Figure 3: Effects of Demonetization. This figure plots the impulse responses of output (y), consumption (c), employment (n), electronic payments, and inflation rates (π) to a -60% cash-supply shock κ. We compare two models: (i) model without a payment-technology constraint (γ = 1) and (ii) model with a payment-technology constraint (γ = 0.412).

38

Output (y)

Consumption (c) 1 percent

χ=0.4 χ=0.9

0.01 0 −0.01

2

4

6

8

10

0 −1 −2

12

0

5

−2

percent

percent

10

0 −5

2

4

6

8

2

4

−3

Investment (I)

10

0

4

percent

percent

6

−1

2

4

6

8

10

12

2

4

6

8

10

12

10

12

10

12

2 0

12

2

4

Nominal Interest Rates (i)

6

8

Monetary Shock 3 percent

1

0.5

0

10

Inflation Rates (π)

1

−2

8

−4 −6

12

6

Employment (n)

x 10

Money Balances (m)/ Electronic Payment

percent

percent

0.02

2

4

6

8

10

2 1 0

12

2

4

6

8

Figure 4: Responses to Monetary Policy Shocks in a Cashless Economy. This figure plots the impulse responses of output (y), consumption (c), investment (I), employment (n), real money balances (m), inflation (π), and nominal interest rates (i) to a 2.5% money supply shock with persistence ρu = 0.5.

39

Table 1: Summary Statistics: Regression Analysis Table 1 presents descriptive statistics for the variables used in regression analyses. The sample is constructed from the CEIC India Premium Database for the period 2011M1-2017M2. Variables Mean Median Std. Dev. 25% 75% Obs. Aggregate-level: Electricity demand (log) 11.36 11.36 0.09 11.29 11.42 74 CPI index (log) 4.72 4.74 0.12 4.62 4.83 74 CPI index: food and beverages (log) 4.73 4.77 0.14 4.63 4.87 74 CPI index: vegetables (log) 4.82 4.84 0.21 4.64 4.99 74 CPI index: tobacco (log) 4.73 4.74 0.16 4.61 4.88 74 CPI index: fuel and light (log) 4.69 4.72 0.11 4.61 4.79 74 CPI index: housing (log) 4.70 4.72 0.12 4.60 4.79 74 CPI index: clothing and footwear (log) 4.72 4.75 0.14 4.61 4.84 74 CPI index: miscellaneous (log) 4.68 4.71 0.10 4.61 4.76 74 Max temperature 30.60 31.06 3.29 28.56 32.29 74 Min temperature 19.56 20.76 4.34 15.18 23.69 74 Actual rainfall 20.36 9.81 20.81 3.85 37.54 74 Rainfall deviation -10.85 -15.1 36.22 -31.2 6.75 74 Stock market index (log) 9.05 8.95 0.22 8.87 9.28 74 No. of POS terminals (log) 13.80 13.87 0.32 13.52 14.01 71 Currency-to-M1 ratio 0.596 0.609 0.042 0.598 0.614 74 State-level: Electricity demand (log) CPI index (log) Max temperature Min temperature Actual rainfall Rainfall deviation Growth rate of POS terminals

6.53 4.82 29.72 19.09 27.29 0.40 0.05

40

6.93 4.83 30.89 21.24 10.66 -0.30 0.03

2.08 0.09 6.05 6.66 38.25 20.97 0.176

4.96 4.76 26.36 14.44 1.33 -0.72 -0.024

8.37 4.89 33.41 24.63 41.20 0.14 0.092

2587 2553 2336 2336 2604 2605 2139

Table 2: Demonetization and Real Economic Activity Table 2 reports estimation results for regression model (1) on the demonetization dummy, maximum temperature, minimum temperature, actual rainfall, rainfall deviation from average, and stock market index. Year fixed effects and month fixed effects are included in the regressions. Heteroskedasticity-consistent standard errors are reported in parentheses. Significance levels are indicated by ∗ , ∗∗ , and ∗∗∗ for 10%, 5%, and 1%, respectively. The sample is constructed from the CIEC India Premium Database for the period 2011M1-2017M2. (1) (2) (3) (4) demonetization -0.026∗ -0.056∗∗∗ -0.028 -0.053∗∗∗ (0.015) (0.018) (0.017) (0.018) max temperature 0.047∗∗∗ 0.041∗∗∗ (0.006) (0.010) min temperature -0.024∗∗∗ -0.020∗∗ (0.007) (0.009) ∗∗∗ actual rainfall -0.003 -0.001 (0.001) (0.001) rainfall deviation -0.0002 0.0001 (0.0001) (0.0001) stock market index 0.008 (0.048) Year FE Yes Yes Yes Yes Month FE Yes Yes Yes Yes R-squared 0.8784 0.9440 0.9302 0.9455 No. of Obs. 74 74 74 74

41

Table 3: Demonetization and Prices Table 3 reports estimation results for regression model (2) on the demonetization dummy, maximum temperature, minimum temperature, actual rainfall, rainfall deviation from average, and stock market index. Year fixed effects and month fixed effects are included in the regressions. Heteroskedasticity-consistent standard errors are reported in parentheses. Significance levels are indicated by ∗ , ∗∗ , and ∗∗∗ for 10%, 5%, and 1%, respectively. The sample is constructed from the CIEC India Premium Database for the period 2011M1-2017M2. (1) (2) (3) (4) (5) demonetization 0.009 0.010 0.009 0.007 (0.017) (0.018) (0.017) (0.017) demonetizationmonth 1−2 -0.030∗∗∗ (0.006) demonetizationmonth 3−4 0.042∗∗∗ (0.006) max temperature -0.001 0.005 0.004 (0.003) (0.005) (0.004) min temperature 0.003 -0.0001 -0.005 (0.005) (0.006) (0.005) actual rainfall 0.0001 0.0004 0.0003 (0.0004) (0.0004) (0.0004) rainfall deviation 0.0000 0.0000 -0.0000 (0.0001) (0.0001) (0.0001) stock market index -0.004 -0.033∗∗ (0.021) (0.016) Year FE Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes R-squared 0.9928 0.9929 0.9929 0.9931 0.9966 No. of Obs. 74 74 74 74 74

42

Table 4: Demonetization and CPI: By Category Table 4 reports estimation results for regression model (2) for each main category of CPI. Covariates include demonetization dummies, maximum temperature, minimum temperature, actual rainfall, rainfall deviation from average, and stock market index. Year fixed effects and month fixed effects are included in the regressions. Heteroskedasticity-consistent standard errors are reported in parentheses. Significance levels are indicated by ∗ , ∗∗ , and ∗∗∗ for 10%, 5%, and 1%, respectively. The sample is constructed from the CIEC India Premium Database for the period 2011M1-2017M2. (1) (2) (3) (4) (5) (6) (7) food & tobacco fuel & housing clothing & others vegetables beverages light footwear demonetization1−2 -0.049∗∗∗ -0.009∗ -0.017∗∗ -0.021∗∗∗ -0.019∗∗∗ -0.008∗ -0.222∗∗∗ (0.010) (0.005) (0.007) (0.008) (0.007) (0.004) (0.064) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ demonetization3−4 0.019 0.086 0.062 0.067 0.064 0.056 -0.166∗ (0.013) (0.008) (0.008) (0.007) (0.007) (0.003) (0.088) max temperature 0.008 -0.005 0.002 -0.003 -0.001 0.001 0.044 (0.008) (0.004) (0.004) (0.005) (0.004) (0.003) (0.054) min temperature -0.010 0.010∗∗ 0.001 -0.004 0.004 -0.002 -0.050 (0.009) (0.005) (0.005) (0.006) (0.005) (0.003) (0.063) actual rainfall 0.0001 0.0001 0.0003 0.0003 0.0002 0.0000 0.002 (0.0006) (0.0002) (0.0003) (0.0005) (0.0003) (0.0002) (0.004) rainfall deviation 0.0000 -0.0001 -0.0000 -0.0000 -0.0000 0.0000 0.0002 (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0000) (0.0006) stock market index 0.001 -0.046∗∗∗ -0.086∗∗∗ -0.072∗∗∗ -0.054∗∗ -0.053∗∗∗ 0.101 (0.026) (0.013) (0.018) (0.022) (0.019) (0.013) (0.149) Year FE Yes Yes Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes Yes Yes R-squared 0.9921 0.9984 0.9939 0.9923 0.9971 0.9970 0.9681 No. of Obs. 74 74 74 74 74 74 74

43

Table 5: Demonetization and the Real Economy: State-level Analysis Table 5 reports estimation results for regression models (1) and (2) at state level on the demonetization dummy, maximum temperature, minimum temperature, actual rainfall, rainfall deviation from average, and stock market index. Statespecific year fixed effects, month fixed effects, and state fixed effects are included in the regressions. Heteroskedasticity-consistent standard errors reported in parentheses are clustered by time. Significance levels are indicated by ∗ , ∗∗ , and ∗∗∗ for 10%, 5%, and 1%, respectively. The sample is constructed from the CIEC India Premium Database for the period 2011M1-2017M1. (1) (2) (3) (4) (5) electricity electricity CPI CPI CPI demonetization -0.037∗∗∗ -0.037∗∗∗ -0.002 -0.0003 (0.008) (0.007) (0.020) (0.021) demonetizationmonth 1 -0.022∗∗∗ (0.005) demonetizationmonth 2 0.015 (0.034) demonetizationmonth 3 0.122∗∗∗ (0.011) ∗∗∗ max temperature 0.005 0.0002 0.0002 (0.002) (0.0004) (0.0004) min temperature 0.010∗∗∗ 0.0000 -0.0001 (0.002) (0.0005) (0.0005) actual rainfall -0.001∗∗∗ -0.0000 -0.0000 (0.0001) (0.0000) (0.0000) rainfall deviation -0.0000 0.0000 0.0000 (0.0001) (0.0000) (0.0000) Year × State FE Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes State FE Yes Yes Yes Yes Yes R-squared 0.9965 0.9968 0.9125 0.9093 0.9112 No. of Obs. 2587 2295 2553 2261 2261

44

Table 6: Demonetization and Economic Activity: The Role of Electronic Payments Table 6 reports estimation results for regression model (1) on the demonetization dummy, interaction term between the pre-shock dummy and the log of lagged POS machine number, interaction term between the post-shock dummy and the log of lagged POS machine number, maximum temperature, minimum temperature, actual rainfall, rainfall deviation from average, stock market index, and currency-to-M1 ratio. Year fixed effects and month fixed effects are included in the regressions. Heteroskedasticity-consistent standard errors are reported in parentheses. Significance levels are indicated by ∗ , ∗∗ , and ∗∗∗ for 10%, 5%, and 1%, respectively. The sample is constructed from the CIEC India Premium Database for the period 2011M4-2017M2. (1) (2) (3) (4) (5) ∗∗∗ ∗ ∗∗∗ ∗ demonetization -11.41 -5.472 -9.229 -6.221 -8.340∗∗ (3.443) (3.100) (3.207) (3.714) (3.845) ∗∗∗ ∗∗ ∗∗∗ ∗ dummypost × no. of POS 0.393 0.232 0.388 0.267 0.422∗∗∗ (0.125) (0.112) (0.115) (0.144) (0.158) dummybefore × no. of POS -0.406∗∗∗ -0.148 -0.256∗ -0.165 -0.146 (0.149) (0.126) (0.135) (0.138) (0.142) currency-to-M1 ratio -0.849∗∗ (0.366) max temperature 0.045∗∗∗ 0.034∗∗∗ 0.034∗∗∗ (0.007) (0.011) (0.011) ∗∗∗ min temperature -0.023 -0.016 -0.018 (0.009) (0.011) (0.011) actual rainfall -0.003∗∗∗ -0.001 -0.001 (0.001) (0.001) (0.001) rainfall deviation -0.0002 0.0000 0.0000 (0.0002) (0.0001) (0.0001) stock market index -0.014 -0.013 (0.052) (0.050) Year FE Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes R-squared 0.8757 0.9376 0.9288 0.9400 0.9457 No. of Obs. 70 70 70 70 70

45

Table 7: Demonetization and Prices: The Role of Electronic Payments Table 7 reports estimation results for regression model (2) on the demonetization dummy, interaction term between the pre-shock dummy and the log of lagged POS machine number, interaction term between the post-shock dummy and the log of lagged POS machine number, maximum temperature, minimum temperature, actual rainfall, rainfall deviation from average, stock market index, and currency-to-M1 ratio. Year fixed effects and month fixed effects are included in the regressions. Heteroskedasticity-consistent standard errors are reported in parentheses. Significance levels are indicated by ∗ , ∗∗ , and ∗∗∗ for 10%, 5%, and 1%, respectively. The sample is constructed from the CIEC India Premium Database for the period 2011M4-2017M2. (1) (2) (3) (4) (5) demonetization -1.018 -0.826 -1.069 -1.146 -0.335 (0.928) (1.014) (0.996) (1.131) (1.249) dummypost × no. of POS 0.206∗∗∗ 0.201∗∗∗ 0.210∗∗∗ 0.220∗∗∗ 0.160∗∗∗ (0.039) (0.041) (0.040) (0.049) (0.060) dummybefore × no. of POS 0.136∗∗∗ 0.144∗∗∗ 0.137∗∗∗ 0.141∗∗∗ 0.134∗∗∗ (0.041) (0.041) (0.042) (0.044) (0.044) currency-to-M1 ratio 0.325∗∗∗ (0.112) max temperature 0.001 0.005 0.005 (0.002) (0.005) (0.004) min temperature -0.001 -0.004 -0.003 (0.004) (0.005) (0.005) actual rainfall 0.0002 0.0003 0.0003 (0.0003) (0.0003) (0.0003) rainfall deviation -0.0000 0.0000 0.0000 (0.0001) (0.0003) (0.0003) stock market index -0.030∗ -0.031∗ (0.016) (0.016) Year FE Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes R-squared 0.9957 0.9957 0.9957 0.9960 0.9964 No. of Obs. 70 70 70 70 70

46

Table 8: The Role of Electronic Payments: State-level Analysis Table 8 reports estimation results for regression models (1) and (2) at state level on the demonetization dummy, its interaction with the diffusion rate dummy, growth rate of POS machines, maximum temperature, minimum temperature, actual rainfall, rainfall deviation from average, and stock market index. State-specific year fixed effects, month fixed effects, and state fixed effects are included in the regressions. Heteroskedasticity-consistent standard errors reported in parentheses are clustered by time. Significance levels are indicated by ∗ , ∗∗ , and ∗∗∗ for 10%, 5%, and 1%, respectively. The sample is constructed from CIEC and RBI databases for the period 2011M1-2016M11. (1) (2) (3) (4) electricity electricity CPI CPI demonetizationmonth 1 -0.047∗∗∗ -0.047∗∗∗ -0.021∗∗∗ -0.020∗∗∗ (0.017) (0.017) (0.006) (0.006) demonetizationmonth 1 × diffusion rate 0.099∗∗∗ 0.095∗∗∗ 0.010∗∗∗ 0.011∗∗∗ (0.010) (0.008) (0.001) (0.001) POS growth rate -0.020 -0.016 -0.006 -0.006 (0.013) (0.013) (0.004) (0.004) max temperature 0.005∗∗ 0.0002 (0.025) (0.0004) min temperature 0.007∗∗ 0.0001 (0.003) (0.0004) actual rainfall -0.0002∗ 0.0000 (0.0001) (0.0004) rainfall deviation -0.0000 0.0000 (0.0001) (0.0000) Year × State FE Yes Yes Yes Yes Month FE Yes Yes Yes Yes State FE Yes Yes Yes Yes R-squared 0.9957 0.9947 0.9105 0.9126 No. of Obs. 1072 938 1072 938

47

Table 9: Model Parameterization Table 9 summarizes the parameter values used to solve the model at a monthly frequency. The sample period covers 2004 to 2016. Parameter Value discount factor (β) 0.998 capital share (α) 0.360 capital depreciation rate (δ) 0.0025 average money growth rate (θss ) 0.011 leisure preference (Ψ) 0.880 inverse of labor supply elasticity (η) 4.000 inverse of intertemporal substitution in consumption (Φ) 0.450 inverse of interest elasticity of cash demand (b) 4.000 cash-holding preferences (1 − a) 0.001 costs of holding demand deposits (χ) 0.400 electronic-payment technology constraint (γ) 0.412

48

Table 10: The Real Effects of Payment Systems Table 10 reports the short-run responses of output and employment and long-run welfare losses in the presence of payment constraints. Panel A shows the effects of India’s demonetization when the degree of payment-constraint tightness λ is 0.412 and 0.5, respectively. Panel B reports the effects of payment constraints of different sizes (λ=0.75, 0.5, and 0.25) when a demonetization shock invalidates all the currency circulated in the economy. ∆ output (%) ∆ employment (%) ∆ welfare (%) Panel A: ∆κ=−0.6 γ = 0.412 -0.03% -0.05% -1.43% γ = 0.5 -0.02% -0.02% -0.64% Panel B: ∆κ=−1 γ = 0.75 γ = 0.50 γ = 0.25

-0.03% -0.20% -0.38%

-0.05% -0.31% -0.60%

49

-4.94% -7.45% -10.7%