research papers on effect of demonetization on banking

Impact of the Indian “demonetization” policy on its export performance

• Published: 14 September 2021
• Volume 62 , pages 2799–2825, ( 2022 )

Cite this article

• Bidisha Lahiri   ORCID: orcid.org/0000-0003-1884-0537 1 &
• Anurag Deb 1  

598 Accesses

Explore all metrics

We examine the export impact of the Indian government’s surprise move to invalidate 86% of existing currencies overnight. This ‘demonetization' was undertaken in an attempt to eliminate illegal and untaxed transactions, and to make the financial system more reliant on electronic transactions. This policy was implemented within a production structure that significantly depended on cash and informal credit sources, and a banking infrastructure that did not reach all corners of the economy. Due to the unexpectedness of this policy, the impact is predicted to have been stronger in the short run than the long run as the economy adjusted to the new system. The synthetic control method (SCM) achieves this, and with its ability to accommodate scenarios where the policy impact may vary across post-treatment periods, provides a suitable tool for our analysis. Additionally the unexpectedness of the policy circumvents anticipation issues making the scenario an ideal candidate for SCM analysis. No long-term improvements though significant short-term losses in exports values due to this policy are uncovered that are robust across alternate specifications and estimation methods.

Similar content being viewed by others

Back on track a macro–micro narrative of italian exports.

The weak rupiah: catching the tailwinds and avoiding the shoals

Export failure and its consequences: evidence from Colombian exporters

Avoid common mistakes on your manuscript.

1 Introduction

On the 8th of November 2016, the government of India made the surprise announcement that the largest denomination of currency Rs.500 and Rs.1000 currency notes no longer remained legal tender. This accounted for 86% (14.180 billion INR) of the total amount of currency notes in circulation and represents a major natural experiment scenario in economics. The move was not announced prior to implementation due to the objective of stifling illegal, unaccounted or untaxed transactions collectively known as the “black market,” to eliminate counterfeit currencies and to move the economy toward one that uses more streamlined and transparent electronic-transactions. Public reaction to this policy has been strong, ranging from it being hailed as a harbinger of the modern efficient India, to its being compared to a wrench being thrown in the erstwhile growth-rate of India.

The implications of this unforeseen and massive “de-monetization” policy for the exports by India are analyzed in this paper. Given that 98% of transactions (68% of value) in India took place in cash Footnote 1 (Pricewaterhouse Coopers 2015 ), the demonetization policy can be expected to have had an impact on India’s trade flows. Since the banking sector in India is not very well developed and access to the internet is relatively limited, a significant (and legal) proportion of the economy was more cash-dependent and felt the unintended consequences of the demonetization policy. However, the long-term impact on Indian exports could be positive, negative or insignificant, respectively, depending on whether the policy intervention resulted in a financial transition that helped, crippled or did not significantly affect India’s international trade transactions, which needs to be empirically uncovered. The insights from our research are relevant for any developing country looking to implement a policy change within the constraints of a banking and formal credit sector that does not yet reach all corners of the economy.

It is widely recognized that the demonetization policy was implemented during a time of economic stability (Lahiri 2020 ). Given that it was an unusual policy implemented during usual times, endogeneity issues like pre-existing factors leading to the adoption of the policy are not a driving concern in the Indian demonetization scenario. Since we need to track the short-term variations in export, we consider monthly data available at the national level. Also, analysis of the intertemporal changes in the Indian export flow by itself is not sufficient to capture the effect of demonetization since these changes may be driven in part by global patterns. This necessitates that the intertemporal export trajectory of India be compared to the trajectory of other similar countries. The synthetic control method (SCM) achieves this, and with its ability to accommodate scenarios where the policy impact may vary across post-treatment periods, provides a suitable tool for our analysis, while the unexpectedness of the policy circumvents anticipation issues making it an ideal candidate for SCM analysis. We further implement a difference-in-difference type estimator after the synthetic control to quantify the policy impact.

We find that the demonetization policy, in its attempt to move the economy toward a financial system that relies on more digitized transactions, did not bring any long-term benefits for exports but did entail significant short run sacrifices to the extent of 7.7 billion dollars that we capture in this paper.

While the SCM remains a well-recognized technique, recent advances in econometric methods allude to potential biases of this method and attempt to improve it in various ways. Augmented synthetic control by Ben-Michael et al. ( 2018 ), Imperfect synthetic control by Powell ( 2018 ), Robust synthetic control by Amjad et al. ( 2018 ), Nonparametric synthetic control by Cerulli ( 2019 ), Synthetic difference-in-difference by Arkhangelsky et al. ( 2019 ), Matching and Synthetic control by Kellogg et al. ( 2020 ) each address specific constraints of the synthetic control method. We explore several of these recent methods as robustness assessments of our SCM analysis.

The rest of the paper is organized as follows. Section 2 provides a summary of relevant literature.

Section 3 describes the SCM method. Section 4 describes the data and model specification. Section 5 presents the main results while Sect. 6 explores robustness analyses. Finally, Sect. 7 discusses the potential causes behind the observed results and Sect. 8 summarizes the conclusions.

2 Background literature

2.1 literature on indian demonetization policy.

Singh and Singh ( 2016 ) couch the examination of the Indian demonetization experiment with similar efforts undertaken by different economies in the past, where the outcome ranged from political instability (Nigeria 1984), economic instability (Myanmar 1987), surge in inflation (Zaire 1993, North Korea 2010), loss of confidence on the government (Soviet Union 1991) as well of instances of the process working smoothly (Russia 1998). This makes a compelling case to examine the impact of demonetization for the Indian economy.

Since India’s demonetization policy is recent, there exists relatively less research about the effects of this policy, though it is an expanding research area. Chodorow-Reich et al. ( 2020 ) and Bhavnani and Copelovitch ( 2018 ) employ district level cross-sectional and difference-in-difference (DID) specifications, respectively, to ascertain the regional effect of demonetization on the economic activity, banking sector, credit markets and political outcomes. Zhu et al. ( 2018 ) look at the impact of the policy on rural poor and use a dummy for the post intervention periods. Beyer et al. ( 2018 ) look at the district level night-light intensity following demonetization to capture the effects on economic activity using a DID approach. Karmakar and Narayanan ( 2019 ) find that households that had bank accounts had significantly higher consumption levels compared to households that did not have bank accounts in the month immediately following the demonetization shock.

The international trade aspect has not been explored in the above literature that has examined the various other effects of the demonetization policy. While the existing regional analyses used variations in regional characteristics to control for the heterogenous effect of the demonetization policy, export flows at the monthly frequency are available only at the national level necessitating the use of other countries as controls. Our contribution is in using the data-driven synthetic control method (SCM) for examining the export effects of the demonetization policy, since India is a significant player in the world economy.

2.2 Literature on SCM applications

Though the SCM is a relatively new econometric technique, there has been substantial application of this method due to its versatility. Below we discuss three areas of SCM application relevant to our study.

Given that our research question focuses on the trade dimension of India, we discuss the applications of SCM to trade outcomes. Hosny ( 2012 ) is one of the earliest applications of SCM in which he evaluates the trade loss arising from Algeria’s delayed participation in the Greater Arab Free Trade Area agreement where SCM helped to create the counterfactual of Algeria having joined earlier. Hannan ( 2017 ) examines the impact of bilateral trade agreements on bilateral trade flows. Barlow et al. ( 2017 ) look at the impact of NAFTA on Canadian imports of corn-syrup.

The policy change under consideration for India is similar to a monetary policy; hence, we next turn our attention to this literature, though in the Indian demonetization case there was not an explicit target to expand or contract the money supply. Chamon et al. ( 2017 ) evaluate the impact of the effectiveness of the exchange-rate sterilization policy undertaken by the Brazilian central bank. Due to the unannounced sudden nature of the policy, they were able to use the SCM analysis by creating a synthetically created doppelganger of Brazil. Aregger and Leutert ( 2017 ) apply the SCM to examine the impact of an unconventional combination of monetary policies on the Swiss Franc during 2009–2011. Karaman and Yıldırım-Karaman ( 2017 ) look at a change in inflation targeting policies in Turkey on inflation and growth outcomes.

Third, since we are interested in the macroeconomic trade outcomes of a national policy, we turn our attention to studies looking at national level data. Billmeier and Nannicini ( 2011 , 2013 ) look at several economies in transition and examine whether economies that have experienced economic liberalization grow faster than those that have not. Born et al. ( 2019 ) use the SCM to evaluate the national economic impacts of the Brexit vote. Using the placebo tests, they infer that Britain is doing significantly worse than the synthetic control group. Gharehgozli ( 2017 ) evaluates the economic costs to Iran of recent sanctions. Adhikari et al. ( 2018 ) look at the impact on per-capita GDP of different countries from the various reform policies adopted by the respective countries.

The synthetic control method (SCM) introduced by Abadie and Gardeazabal ( 2003 ) and further developed in Abadie et al. ( 2010 ) and Abadie et al. ( 2015 ) matches the treatment unit with a combination of untreated units and is suitable to our research question for three reasons: First, it is difficult to find a single country that can be considered a control unit for India. Even if we used several countries as controls, the difference-in-difference (DID) method would apply equal weights to all these countries in the estimation. The SCM gets around this issue by using data-driven methods to assign weights to the control units to create a doppelganger whose pre-intervention characteristics and trends match that of the treatment unit India. Second, the SCM allows the policy effect to vary across post-treatment periods, making it ideal to demarcate the short-run versus long-run effects. Third, if the policy is expected to have an impact on several variables, as in our research focus on the demonetization policy, then a DID estimation equation will suffer from endogeneity problems. The SCM, on the other hand, by matching countries on their pre-intervention characteristics minimizes this problem. Abadie et al. ( 2010 ) prove that if the synthetic control matches the pre-treatment trajectory of the outcome variable of the treated country, then the bias caused by time varying changes in confounders in the post-treatment periods goes towards zero as the pre-intervention period increases.

The SCM method is especially suited for scenarios that are unexpected, so that expectations about the policy do not affect the pre-treatment behavior in the treated unit, and hence avoids bias during the creation of the synthetically matched control.

Suppose the data comprises of \({\varvec{J}}+1\) countries over a timespan of T periods with \({{\varvec{Y}}}_{{\varvec{j}},{\varvec{t}}}\) representing the outcome variable for \({\varvec{j}}=\mathrm{0,1},2,\dots .{\varvec{J}}\) and \({\varvec{t}}=\mathrm{1,2},\dots .{\varvec{T}}\) . If country \({\varvec{j}}=0\) , henceforth called the treated unit, experiences a policy intervention at period \({\varvec{t}}={{\varvec{T}}}_{0}\) , then periods \({\varvec{t}}<{{\varvec{T}}}_{0}\) are pre-intervention periods, while periods \({\varvec{t}}\ge {{\varvec{T}}}_{0}\) are post-intervention periods, and the \({\varvec{J}}\) countries ( \({\varvec{j}}\ne 0\) ) are considered the donor pool.

Let \({{\varvec{X}}}_{0}\) denote a ( \({\varvec{k}}\boldsymbol{ }\times 1)\) vector of pre-treatment characteristics of the treated unit. It comprises of values of the outcome variables and may additionally include a vector \({{\varvec{Z}}}_{0}\) of pre-intervention observed explanatory variables. Footnote 2 Let \({{\varvec{X}}}_{1}\) be the ( \({\varvec{k}}\boldsymbol{ }\times {\varvec{J}})\) matrix of similar values for all the donor countries. The SCM method assigns weights \({{\varvec{w}}}_{{\varvec{j}}}\) (s.t. \({{\varvec{w}}}_{{\varvec{j}}}\boldsymbol{ }\ge 0\) and \({\sum }_{{\varvec{j}}=1}^{{\varvec{J}}}{{\varvec{w}}}_{{\varvec{j}}}=1)\) to countries in the donor pool such that the optimal choice of the weights \({{\varvec{W}}}^{\boldsymbol{*}}=({{\varvec{w}}}_{1}^{\boldsymbol{*}},\boldsymbol{ }\dots ..,\boldsymbol{ }{{\varvec{w}}}_{{\varvec{J}}}^{\boldsymbol{*}})\) minimizes the distance || \({{\varvec{X}}}_{0}-{{\varvec{X}}}_{1}{\varvec{W}}\) ||.

This is achieved through minimizing the mean square error \(({{\varvec{X}}}_{0}-{{\varvec{X}}}_{1}{\varvec{W}})\boldsymbol{^{\prime}}{\varvec{V}}(({{\varvec{X}}}_{0}-{{\varvec{X}}}_{1}{\varvec{W}})\) where \({\varvec{W}}\equiv ({{\varvec{w}}}_{1},{{\varvec{w}}}_{2},\boldsymbol{ }\dots .{{\varvec{w}}}_{{\varvec{J}}})\) is the ( \({\varvec{J}}\boldsymbol{ }\times 1)\) vector of the country weights and \({\varvec{V}}\) is a ( \({\varvec{k}}\boldsymbol{ }\times {\varvec{k}})\) symmetric positive semidefinite matrix. Following Abadie and Gardeazabal ( 2003 ), we choose \({\varvec{V}}\) such that the resulting synthetic control unit approximates the trajectory of the outcome variable of the affected region in the pre-intervention periods by minimizing the root-mean-square prediction error (RMSPE) of the outcome variable for the pre-intervention periods.

Once the weights \({{\varvec{W}}}^{\boldsymbol{*}}\) are optimally chosen using the above method, the pre-treatment values of the outcome variable for the constructed synthetic unit closely mimics that of the treated unit. Any discrepancy in the behavior of the outcome variable between the treated-unit and the synthetically created control unit after the policy intervention is thus attributed to the policy itself. The causal effects are estimated by \(\widehat{{\boldsymbol{\alpha }}_{{\varvec{t}}}}={{\varvec{Y}}}_{0,{\varvec{t}}}-\boldsymbol{ }{\sum }_{{\varvec{j}}=1}^{{\varvec{J}}}{{\varvec{w}}}_{{\varvec{j}}}^{\boldsymbol{*}}{{\varvec{Y}}}_{{\varvec{j}},{\varvec{t}}}\) for each post-intervention period.

This approach is a generalization of the difference-in-differences (fixed-effects) approach since the regional heterogeneity is allowed to vary with time. In difference-in-difference models, the unobserved heterogeneity is assumed to be constant across time and can therefore be eliminated by taking differences, which is not the case under SCM.

Following Abadie et al. ( 2010 , 2015 ) and Doudchenko and Imbens ( 2016 ) inferences are made in SCM based on the idea of randomly assigning the treatment-unit and randomly assigning the treatment-period, also known as “in-space placebo” effect and “in-time placebo” effect, respectively. Iterative application of the synthetic control method to every region in the donor pool creates a distribution of placebo effects for the untreated units that can be compared to the effect observed for the unit that actually received the treatment. A low probability of observing an effect larger in a non-treated unit than the effect observed in the treated unit would indicate that the treatment effect was significant for the treated unit. This is known as the “in-space” placebo tests which creates a distribution of placebo effects for the untreated units. Thus, the p value has the interpretation of being the proportion of control units that have an estimated effect at least as large as that of the treated unit and is a good indicator of the predictive power. Similarly, “in-time” placebo tests create a distribution of estimated effects for non-treatment periods.

Diff-in-Diff after synthetic control

To assess the treatment effect, Bohn et al. ( 2014 ) and Munasib and Rickman ( 2015 ) construct a difference-in-difference-type estimator after applying the synthetic control method. Let \({\Delta }_{{\varvec{T}}{\varvec{R}}}\) represent the treatment effect as a difference between the treatment unit and the synthetic unit, also differenced over the pre-treatment and post-treatment periods as expressed as follows:

The estimator adjusts the post-treatment effect with the pre-treatment quality of fit. This ensures that any significant post-treatment effects \(\left|{\overline{{\varvec{Y}}} }_{{\varvec{T}}{\varvec{R}},\boldsymbol{ }{\varvec{a}}{\varvec{c}}{\varvec{t}}{\varvec{u}}{\varvec{a}}{\varvec{l}}}^{{\varvec{p}}{\varvec{o}}{\varvec{s}}{\varvec{t}}}-{\overline{{\varvec{Y}}} }_{{\varvec{T}}{\varvec{R}},{\varvec{s}}{\varvec{y}}{\varvec{n}}{\varvec{t}}{\varvec{h}}{\varvec{e}}{\varvec{t}}{\varvec{i}}{\varvec{c}}}^{{\varvec{p}}{\varvec{o}}{\varvec{s}}{\varvec{t}}}\right|\) are not a result of the pre-treatment fit quality \(\left|{\overline{{\varvec{Y}}} }_{{\varvec{T}}{\varvec{R}},\boldsymbol{ }{\varvec{a}}{\varvec{c}}{\varvec{t}}{\varvec{u}}{\varvec{a}}{\varvec{l}}}^{{\varvec{p}}{\varvec{r}}{\varvec{e}}}-{\overline{{\varvec{Y}}} }_{{\varvec{T}}{\varvec{R}},{\varvec{s}}{\varvec{y}}{\varvec{n}}{\varvec{t}}{\varvec{h}}{\varvec{e}}{\varvec{t}}{\varvec{i}}{\varvec{c}}}^{{\varvec{p}}{\varvec{r}}{\varvec{e}}}\right|\) .

The permutation or randomization test, used to create the in-space placebo effects, is once again useful in testing the statistical significance of this estimate. If the cumulative density function of the complete set of Δ estimates is denoted by F (Δ), the p value from a one-tailed test of the hypothesis that Δ TR  > 0 is given by F (Δ TR ) (Bohn et al. 2014 ). The p value is the proportion of control units that have an estimated effect at least as large as that of the treated unit, once we have adjusted for the pre-treatment quality of fit (Bertrand et al. 2004 ; Buchmueller et al. 2011 ; Abadie et al. 2010 ).

“DID-rank test” is a second test based on the difference-in-difference estimator after synthetic control (Abadie et al. 2010 ; Munasib and Rickman 2015 ). This ranks the absolute value of the magnitude of the difference-in-difference of the treatment state against all the placebo difference-in-difference magnitudes. If DID-rank is 1 then the estimated impact of the intervention in the treatment state is greater than any of the estimated placebo impacts.

4 Model specification and data description

Our baseline model uses the lagged values of the dependent variable as the pre-treatment covariates for creating the synthetic control group. The incorporation of additional macroeconomic predictors assesses the robustness of the results. We consider countries’ GDP, tariff rates, change in exchange rates, capital formation and FDI flows that are traditionally used as the determinants of a country’s trade flows (Booth et al. 2015 ; Minford 2016 ; Ebell et al. 2016 ; Born et al. 2019 ) as additional pre-intervention period observed covariates for calculating weights to generate the synthetic doppelganger for India in this robustness specification. The results with the added covariates remain similar to that of the baseline model.

The outcome variable: merchandise exports values (in million US$) at the monthly level comes from WTO, Footnote 3 the monthly exchange rates comes from IMF, and the macroeconomic indicators are combined from IMF International Financial Statistics, World Bank’s WDI and the Datastream international databases Footnote 4 . The donor pool covers 40 countries that have non-missing data for the relevant variables over this time period. Country level monthly data are available starting in April 2009 with the intervention period being Nov 2016. The dataset for the post-intervention periods was cut off at May 2017 to prevent contamination from subsequent Indian tax-change policies that were anticipated in June 2017 and implemented in July 2017. Table 1 presents the characteristics of the sample countries in the pre-intervention periods.

The two-month average (September–October 2016) for Indian exports preceding the demonetization policy was $22.73 Billion while following the policy shock the two-month average (November–December 2016) for Indian exports was $21.93 Billion. However, this change in mean might also be partially caused by other global factors that affected all countries and is hence not necessarily a true indicator of the policy effect, which we uncover using SCM.

The results section is broken into two parts: a section that explains the construction of the synthetic controls, and a section that details the inferences drawn from the constructed counterfactuals.

5.1 Construction of the synthetic control for India

India is one of the larger exporters in the world market; hence, one of the restrictions was that countries that get positive weights in creating the synthetic unit for India are countries that are relatively significant players regarding export values. This was achieved by considering countries whose average export values during the pre-intervention periods was at least one half of that of the Indian economy, thus eliminating Argentina, Chile, Colombia, Costa Rica, Ecuador, Finland, Greece, Hungary, Israel, Morocco, New Zealand, Paraguay, Peru, Philippines, Portugal, South Africa and Tunisia. The purpose was to make the set of countries homogenous to some extent, so that they can be considered to be members of a high-powered international players “club.” Otherwise the constructed synthetic control might result in a good fit but not make much sense economically. We will call this the baseline specification.

Abadie et al. ( 2010 , 2015 ) suggest also restricting the donor pool if there exist countries in the donor pool whose trajectory in the pre-intervention periods was significantly different from the trajectory of India. The inclusion of such donors would make it difficult to disentangle whether a difference in post-intervention match quality is attributable to the lack of fit or to the demonetization policy. The placebo effects may be quite large if those units where not matched well in the pre-treatment period causing the p values to be too conservative. To circumvent this problem, China and Denmark that have more than 5 times the pre-demonetization root-mean-square prediction error (RMSPE) of India are eliminated from the donor pool during robustness checks.

Additionally, countries that get a significant weight in the initial estimation are removed from the donor pool one at a time to ensure that the results are not driven by the characteristics of the specific donor in the leave-one-out test.

The baseline model of analysis uses the lagged values of the dependent variable- merchandise exports values to create the synthetic counterpart for India, while the incorporation additional explanatory variables explained in Sect. 4 is one of the ways to assess the robustness of our results. Among the available covariates that have economic meaning, only those that are available for all units, and that improve the quality of pre-treatment fit, i.e., reduce pre-RMSPE, are used.

The weights of the countries in the synthetic control created for the various specifications are presented in Table 2 . Since Switzerland is seen to get a high weight in the baseline model, we also evaluate the results excluding Switzerland from the donor pool.

5.2 Inference

We use the in-space placebo test, the in-time placebo test and the diff-in-diff after synthetic control to make inferences regarding the policy impacts.

5.2.1 In-space placebo test

Figure  1 presents the effects graph for the months leading up to the policy and in the aftermath of the policy. Following the policy intervention, no significant change in trend is apparent graphically. This indicates that the demonetization policy, in its attempt to move India towards a financial system more reliant on electronic transactions, appears not to have significantly improved or crippled Indian exports. This is consistent with the results presented in Table 3 and is examined more closely.

Effects-graph for model with high pre-RMSPE donors removed for export values (in million $)

Using the “in-space placebo” test suggested by Abadie et al. ( 2010 , 2015 ) and Doudchenko and Imbens ( 2016 ), we make inferences regarding the effects of the policy shock for every period, as explained in Sect. 3. The p value has the interpretation of being the proportion of control units that have an estimated effect at least as large as that of the treated unit.

Table 3 presents the numerical values both of estimated coefficients and p values for exports. Column (i) provides estimates that creates the synthetic control based on the lags of the outcome variables, column (ii) provides the estimates when countries with pre-intervention RMSPE greater than five times that of the RMSPE of India prior to demonetization have been removed, column (iii) additionally removes Switzerland, and column (iv) creates the synthetic control by matching on the additional macroeconomic indicators (see Table 1 ) that might affect a country’s export volumes.

The results are found to be consistent across the various specifications: The estimates indicate that for exports there was a negative effect for three months following the implementation of the policy but it was statistically significant for only the first two months. For November 2016, the month in which the policy was implemented, the estimated effect averaged across the four specifications was a loss of about 4.2 billion dollars of exports and about 3.5 billion dollars in December 2016. This is approximately 19% and 16% of Indian monthly exports, relative to the export value of the month immediately preceding the policy shock.

The intuition of the results presented in Table 3 can be better understood by zooming in to the months immediately after the demonetization policy using the p value graph for India in Fig.  2 . In November and December of 2016, the negative effect on the exports by India is stronger compared to that for most countries in the donor pool during this time. The significance of the negative effect is stronger in the first month than in the second month for export values. In the subsequent months however, the divergence for India gets closer to that for other countries, resulting in higher p values that imply the absence of any significant long-term effect.

p value graph for exports by India with high RMSPE donors dropped

5.2.2 In-time placebo test

To be further convinced of the relevance of the significant negative effect immediately following the policy, it is useful to be assured that the treatment unit was not significantly different from the synthetic control immediately prior to the policy shock. To ensure this, we evaluate the “in-time” placebo tests whose results are reported in Table 4 . In the estimation models, we specify the hypothetical policy intervention dates to be October and September 2016, which are earlier than the true policy date of Nov 2016. The significant two-period negative effect is still picked up only following the true policy period indicating that it is an impact of the policy shock. Also, the p values are high for the pre-intervention periods prior to Nov 2016 indicating that the treatment unit India was not statistically different from its synthetic unit immediately prior to the policy shock.

5.2.3 Diff-in-diff after synthetic control

While change in the long-term trend was a possibility in the Indian demonetization experiment, the SCM analysis only indicates a significant deviation for the two periods following the policy shock. We evaluate the robustness of this finding using the DID test after synthetic control.

This test adjusts for pre-treatment fit while evaluating the two period post-treatment effect uncovered for India. This is especially useful in our case where there is not a significant change in trend, and so assures that any short-term deviation detected by SCM is not a result of the quality of pre-treatment fit.

The average deviation of the treatment unit India from its synthetic counterpart for the pre-treatment period (D1) is small in magnitude and less than 0.002 of the average pre-treatment value of the outcome variable in all three specifications, thus indicating a good fit.

Table 5 shows that for the two-periods immediately following the currency invalidation, the DID effect for India was between 3.9 to 4.3 Billion USD per period across the different specifications. The p values for the DID effect for India are 0.000 and the DID-rank for India is 1 in all specifications, indicating that in the two periods following the demonetization policy, India experienced the greatest negative effect compared to all donor countries, indicative of the effect being an outcome of the policy. However, no significant policy effect is detected beyond the initial two periods for all specifications.

The DID rank test results can be clearly understood from Fig.  3 , where India ranks highest for the 2 periods following the policy intervention, but does not rank significantly differently from the donor pool beyond those two periods.

DID rank test after synthetic control for different post-treatment horizons, in the model with high-RMSPE donors removed

5.2.4 Other relative measures

Results similar to above emerge when using the post-RMSPE/pre-RMSPE statistic suggested by Abadie et al. ( 2010 ) and used in other SCM applications. This relative-measure based statistic normalizes with respect to the pre-treatment fit in evaluating the post-treatment effect to ensure that pre-treatment fit quality is not driving post-treatment effects. As before, for the two periods immediately following the policy intervention, India has a high rank that implies a significant policy effect during this time. Beyond these two periods, the exports of India do not show any effect significantly different from the donor pool. The results are shown in Fig.  4 a, b.

a Post-RMSPSE/pre-RMSPE for the two periods immediately following currency invalidation. b Post-RMSPSE/pre-RMSPE beyond the two periods following currency invalidation

6 Robustness checks for the short run impacts

To ensure that the two-period effect is truly an outcome of the policy, we deploy further robustness checks.

6.1 Leave-out-one test based on Synthetic Control method

We run the analyses leaving out each of the donors that received a weight greater than 10% in the initial specifications, to examine if the two-period significant negative effect still emerges, and we find that it does, as presented in Table 6 .

6.1.1 Other restrictions on the donor pool

As is apparent from Figs.  3 and 4 a, India stands out from the donor pool for measures that evaluate post-treatment effects after adjusting for pre-treatment deviations for the short run impact. Still, the figures show that several other countries are also characterized by average post-treatment deviations which are larger than their average pre-treatment fit (that is, the post-RMSPE/pre-RMSPE relative measure greater than 1). We therefore restrict the donor pool of countries as a robustness check in the “Appendix A1,” but the results are very similar to those which we obtain for our baseline specification.

6.2 Robustness checks based on other estimation techniques

Although SCM remains a popular method for evaluation of policy effects, we undertake some recent extensions of this method that attempt to correct for potential biases in the SCM method. Of the various methods available, we detail three methods that we explore.

Augmented Synthetic Control method (ASCM): this method developed by Ben-Michael et al. ( 2018 ) estimates the bias in the SCM estimate due to covariate imbalance and then de-biases the original SCM estimate, similar to bias correction for inexact matching. Negative weights are possible since ASCM extrapolates outside the convex hull of the control units. This ensures much closer balance, reducing bias, but rests more heavily on modeling assumptions, such as linearity.

Nonparametric synthetic control method (NPSYNTH): In this method advanced by Cerulli ( 2019 ), a kernel function gives a higher weight to countries that are closer Footnote 5 to the treatment unit, and penalizes (i.e., gives lower weight to) countries that are farther from the treatment unit. As the distances differ every year, the weights also vary by year, unlike the SCM that relies on a single vector of weights for all pre-treatment periods. For the post-treatment periods, each donor unit is given the average of its pre-treatment period weights to construct the synthetic control.

Matching and Synthetic Control method (MASC): Kellogg et al. ( 2020 ) show that since the SCM uses a convex weighted average of the untreated units to create a synthetic untreated unit, it is susceptible to interpolation bias. On the other hand, other commonly used matching estimators, such as nearest-neighbor matching, suffer from potentially extrapolating too much when suitable untreated units are unavailable. The MASC estimator uses a rolling-origin cross-validation procedure to trade-off between interpolation and extrapolation bias to combine the matching and synthetic control estimators through model averaging.

Since most of them are in the development stage, some of the methods do not provide standard error estimates by period. Nonetheless, they are useful in examining if the synthetic controls constructed using alternate algorithms still uncover a negative effect for India relative to the constructed counterfactual. We include all countries in the sample set to constitute the donor pool so that these estimation algorithms have the maximum flexibility while choosing countries during construction of the synthetic unit.

The graph for the difference between the treated unit India and the synthetic control does not show any change in trend (Fig.  5 ). We look at the estimates by each period to capture the short run effects of the policy, presented in Table 7 .

Export difference between India and synthetic control, estimated by ASCM

Table 7 shows that these alternate econometric methods construct the synthetic control doppelgangers by giving weights to very different sets of countries from among the donors, assuring that the effects picked up are not a function of the donors chosen in the counterfactual. Despite this they all find a negative effect in the two periods following the policy shock. Column 1 reveals a statistically significant negative effect immediately following the policy shock in Nov 2016. The effect gets smaller but remains significant in the second period and becomes even smaller and not significant after two periods. The MASC and NPSYNTH estimators also indicate a similar negative deviation between the treated and the synthetically created control unit, though in the absence of standard error estimates no statements can be made regarding the statistical significance of these effects.

7 Underlying causes

Around the time when the demonetization policy was implemented, about 98% of transactions (68% of value) in India, including industrial production, was primarily cash based (Pricewaterhouse Coopers 2015 ) and hence the invalidation of existing cash froze much of the economic flows. Demonetization hindered the production process by disrupting the payments for inputs and sales of output for economic transactions that primarily used cash or informal credit channels (Beyer et al. 2018 ; Zhu et al. 2018 ; Chodorow-Reich et al. 2020 ), which in turn also led to the effect on exports. Cash-based exporters who could not meet their export related production obligations, as well as intermediate buyers of export goods who paid in cash were impacted by this effect. Footnote 6

The micro, small and medium enterprises (also known as the MSME sector) and the agricultural sector are specific examples of firms and industries that primarily used cash for their transactions like paying for supplies and paying for contract work. In 2015–2016, the MSME sector contributed about 40% of exports Footnote 7 and about 30% of the GDP for India. For the agricultural sector, the shares towards exports and GDP were about 15% and 16.5%, respectively. Footnote 8

For agriculture, using data on arrivals and prices from around 3000 regulated markets for 35 major crops, Aggarwal and Narayanan ( 2017 ) find that demonetization reduced trade value by 13% in the short run. 86% of land holdings in India were small and marginal farmers Footnote 9 who used cash for their purchases and accepted cash for their sales. When demonetization was implemented, the farmers were either selling their ‘Kharif’ (October–November) yield or sowing ‘Rabi ’ crops (sowed in mid-November). Footnote 10 Due to lack of cash, millions of farmers were unable to purchase seeds and fertilizers ahead of the ‘Rabi ’ season, and also their sales were held up due to the cash shortage of the buyers.

On the industrial production front, the small and medium enterprises (SMEs) are dominant players in some of India’s major export sectors namely textiles and garments, leather products, sports goods, gems and jewelry, and handicrafts among others. Typically, most of the SME firms hold between one to ten lakhs of INR cash (approximately $1,350–$13,500) in hand for their daily operations while the owners of tannery and gems businesses keep a couple of crores of INR Footnote 11 in hand because these businesses are entirely run on cash. Footnote 12 Kumar ( 2017 ) cites examples of the bicycle industry in Ludhiana, the brass industry in Moradabad, the diamond industry in Surat, which were all devastatingly impacted by demonetization. Many such businesses shut down because there was no cash to pay the workers and most of the workers in the micro-industries working in urban areas went back to their villages in the aftermath of demonetization since they could not be paid. About a thousand firms in Jamshedpur that operated as small-scale ancillary units to big industries were closed. These examples highlight the effect of the demonetization shock on the MSME sector that accounts for about 45% of industrial production in the Indian economy. According to the Reserve Bank of India (RBI), the MSME sector was adversely affected by this policy shock both because banking credit to this sector experienced a negative effect and also because they relied on informal credit channels and primarily used cash for their day to day operations.

The RBI report Footnote 13 also stated “Among various items of MSMEs exports, gems and jewellery, carpets, textile, leather, handlooms and handicrafts items are highly labour intensive and depend heavily on cash for working capital requirements and payment towards contractual labourers. Hence, export shipments of these sectors could have been impacted by demonetization”. This report also found that there was a “sharp decline in arrivals of animal hide in major leather clusters. In view of constraints on availability of raw material as well as transportation and labour bottlenecks, about 60 out of 100 respondents indicated that they were no longer taking export orders”.

As the above examples, research and official reports show, for firms and sectors that primarily used cash for their transactions like paying for supplies, paying for contract work and selling their output, the policy shock impacted both production and exports negatively. Table 8 presents the results where the dependent variable is the ratio of exports and GDP, Footnote 14 where both moved in the same direction during the policy shock. Column (i) presents the baseline model, while an alternate estimation to check the robustness of the results is presented in column (ii), where donors that get high weights in the baseline estimation are excluded. The effects emerge to be consistently negative indicating that the negative movement of exports was stronger than the negative impact on GDP, Footnote 15 most likely due to the strong impact of demonetization on the MSME sector that had a larger share in exports than in GDP. However, the effects are not statistically significant implying that we cannot reject the hypothesis that exports reacted with an intensity similar to that of GDP. Since both GDP and exports of India suffered a negative impact due to the demonetization policy, the effect on the ratio of these two variables is expected to be muted, as is reflected in the SCM results presented in Table 8 .

The negative effects were short lived primarily for two reasons: first, the arrival of new currency into the economy and second, the successive adjustments made by the government to ease the restrictions on the agricultural and MSME sectors. Chodorow-Reich et al. ( 2020 ) find strong, positive correlation between the arrival of new currency by December 2016 and ATM withdrawals. On the dimension of adjustments made by the government, announcement allowing farmers to use the old Rs.500 bills to buy seeds is an example Footnote 16 of such a reprieve directed towards the agricultural sector, while the RBI directive advising banks to use the facility of providing additional working capital limit to their MSME borrowers to overcome the cash flow mismatches Footnote 17 was an amnesty directed at the MSME sector.

8 Conclusions

Our analysis finds that following the demonetization policy that invalidated a large fraction of the Indian currency, there was a statistically significant negative impact on exports of the Indian economy for the first two months, reflecting the international economic opportunities lost due to this policy shock. By the synthetic control estimates, this amounted to a cumulative total of about 7.7 billion dollars and was equivalent to 19% and 16% loss of monthly exports for the two months following the policy intervention, though the impact fades after this period, indicating that it would be difficult to detect the effect by research that undertakes the analysis at the annual frequency.

The large segment of the Indian economy that used cash and informal credit for their daily transactions encountered disruptions in their economic transactions due to the demonetization shock. The penetration of the banking sector is insufficient across India, leading to hardships even for those firms that did have bank accounts. Firms and industries with adequate banking access also experienced impediments since the banks were overwhelmed by the sudden invalidation of existing currency. The banks themselves were constrained in exchanging the old currency for new, thus exacerbating the bottleneck even for those that could access the banks. These bottlenecks resolved after two months when the new currency replaced the old, leading to the dissipation of the effect. The lessons learned are applicable for any developing country looking to implement a major financial overhaul within the constraints of a banking sector that does not reach all corners of the economy.

The SCM analysis highlights the relevance of looking at the intertemporal trajectory of the treated unit relative to the experiences of the control units. When this is implemented, it emerges that the demonetization policy, contrary to popular belief, just as it did not bring about digitization driven significant long-term efficiency gains for India’s export performance, neither did it cause long-term damage nor should be held accountable for the slowdown of Indian export flows. The results from our current research show that an invalidation of existing currency without any accompanying infrastructural change is not likely to bring about any structural change in the international trade performance of an economy, but only cause significant short-term disruptions.

Data availability

Code availability.

By comparison, 55% of transactions in the USA take place in cash (14% of value).

We perform SCM for both specifications where the X matrix comprises only of the pre-intervention outcome variables, and also where the X matrix comprises of other pre-intervention observed covariates in addition to pre-intervention values of the outcome variables.

Source: International Trade Statistics by WTO.

The SCM is unbiased even if the values of the additional pre-intervention covariates are available for a single pre-treatment period (Abadie et al. 2010 ). Thus, incorporation of macro-variables available at the quarterly or annual frequency provide a suitable robustness check of the baseline results.

Where closeness is defined based on a pre-defined x-distance method.

The demonetization policy caused short-term devaluation of the Indian exchange rate: the average exchange rate was INR 66.8/$ in the 3 months preceding demonetization, INR 67.8/$ in the 3 months following demonetization and INR 66.09/$ in the subsequent 3 months. However, this is not the underlying cause of export deviation since currency devaluations are associated with improved export performance and should also affect import flows, neither of which were observed. Synthetic control results for import flows are presented in Appendix A2.

Government of India, “Unlocking the potential of MSME exports”, 2018: http://dcmsme.gov.in/Export_promotion_strategy(MSME).pdf

Government of India, “State of Agriculture in India 2015–16", 2016.

As per the Indian agriculture census of 2015–16.

Government of India, “Price Policy for Kharif Crops, Agricultural Marketing Year 2017–18", 2016.

One crore INR is approximately $135,000.

As reported by Anil Bhardwaj, Secretary General of the Federation of Indian Micro and Small and Medium Enterprises: https://scroll.in/article/823231/interview-demonetisation-has-hit-80-of-small-businesses-the-sector-is-staring-at-apocalypse

Reserve Bank of India, “How have MSME Sector Credit and Exports Fared?”, 2018: https://m.rbi.org.in/Scripts/MSM_Mintstreetmemos13.aspx

Since national GDP is available only at the quarterly frequency, interpolation was used for monthly GDP.

As the effects of the two months are in the fourth quarter of 2016, an analysis was also done using quarterly data, presented in Appendix A3. Results show negative effects that are not statistically significant, similar to Table 8 .

https://economictimes.indiatimes.com/news/economy/agriculture/government-allows-farmers-to-buy-seeds-with-old-rs-500-notes/articleshow/55539367.cms?from=mdr

https://www.rbi.org.in/commonman/Upload/English/Notification/PDFs/NOTI200AC07DFD09E8E4039B41DB4EEA0318300.PDF

Abadie A, Diamond A, Hainmueller J (2010) Synthetic control methods for comparative case studies: estimating the effect of California’s tobacco control program. J Am Stat Assoc 105(490):493–505

Article   Google Scholar  

Abadie A, Diamond A, Hainmueller J (2015) Comparative politics and the synthetic control method. Am J Polit Sci 59(2):495–510

Abadie A, Gardeazabal J (2003) The economic costs of conflict: a case study of the Basque Country. Am Econ Rev 93(1):113–132

Adhikari B, Duval R, Hu B, Loungani P (2018) Can reform waves turn the tide? Some case studies using the synthetic control method. Open Econ Rev 29(4):879–910

Aggarwal N, Narayanan S (2017) Impact of India's demonetization on domestic agricultural markets. Indira Gandhi Institute of Development Research, Mumbai, India

Amjad M, Shah D, Shen D (2018) Robust synthetic control. J Mach Learn Res 19(22):1–51

Google Scholar  

Aregger N, Leutert J (2017) Policy evaluation by the synthetic control approach: the case of the Swiss Franc. Working papers 17.02, Swiss National Bank, Study Center Gerzensee

Arkhangelsky D, Athey S, Hirshberg DA, Imbens GW, Wager S (2019) Synthetic difference in differences (No. w25532). National Bureau of Economic Research

Barlow P, McKee M, Basu S, Stuckler D (2017) Impact of the North American free trade agreement on high-fructose corn syrup supply in Canada: a natural experiment using synthetic control methods. Can Med Assoc J 189(26):E881–E887

Ben-Michael E, Feller A, Rothstein J (2018) The augmented synthetic control method. arXiv preprint. arXiv:1811.04170.

Bertrand M, Duflo E, Mullainathan S (2004) How much should we trust differences-in-differences estimates? Q J Econ 119(1):249–275

Beyer RCM, Chhabra E, Galdo V, Rama M (2018) Measuring districts’ monthly economic activity from outer space. Policy Research Working Paper 8523, The World Bank, Washington, DC

Bhavnani RR, Copelovitch M (2018) The political impact of monetary shocks: evidence from India's 2016 Demonetization. Available at SSRN 3095228.

Billmeier A, Nannicini T (2011) Economies in transition: how important is trade openness for growth? Oxford Bull Econ Stat 73:287–314

Billmeier A, Nannicini T (2013) Assessing economic liberalization episodes: a synthetic control approach. Rev Econ Stat 95(3):983–1001

Bohn S, Lofstrom M, Raphael S (2014) Did the 2007 Legal Arizona workers act reduce the state’s unauthorized immigrant population? Rev Econ Stat 96(2):258–269

Booth S, Howarth C, Persson M, Ruparel R, Swidlicki P (2015) What if?? The consequences, challenges and opportunities facing Britain outside EU. Open Europe report 03/2015

Born B, Müller G, Schularick M, Sedlacek P (2019) The costs of economic nationalism: evidence from the Brexit experiment. Econom J 129(10):2722–2744

Buchmueller TC, DiNardo J, Valletta RG (2011) The effect of an employer health insurance mandate on health insurance coverage and the demand for labor: evidence from Hawaii. Am Econ J Econ Pol 3(4):25–51

Cerulli G (2019) A flexible synthetic control method for modeling policy evaluation. Econ Lett 182:40–44

Chamon M, Garcia M, Souza L (2017) FX interventions in Brazil: a synthetic control approach. J Int Econ 108:157–168

Chodorow-Reich G, Gopinath G, Mishra P, Naraynan A (2020) Cash and the economy: evidence from India’s demonetization. Quart J Econ 135(1):57–103

Doudchenko N, Imbens GW (2016) Balancing, regression, difference-in-differences and synthetic control methods: a synthesis (No. w22791). National Bureau of Economic Research

Ebell M, Hurst I, Warren J (2016) Modelling the long-run economic impact of leaving the European Union. Econ Model 59:196–209

Gharehgozli O (2017) An estimation of the economic cost of recent sanctions on Iran using the synthetic control method. Econ Lett 157:141–144

Hannan SA (2017) The impact of trade agreements in Latin America using the synthetic control method. IMF working paper, WP/17/45.

Hosny AS (2012) Algeria’s trade with GAFTA countries: a synthetic control approach. Trans Stud Rev 19(1):35–42

Karaman K, Yildirim-Karaman S (2017) The monetary policy change in Turkey in 2009 and its implications on inflation and growth. Bogazici J 31(1):1–21

Karmakar S, Narayanan A (2019) Do Households Care about Cash? Exploring the heterogeneous effects of India’s Demonetization. Research in Economics and Mathematics (REM) Working Paper 073-2019

Kellogg M, Mogstad M, Pouliot G, Torgovitsky A (2020) Combining matching and synthetic controls to trade off biases from extrapolation and interpolation (No. w26624). NBER

Kumar A (2017) Demonetization and black economy. Penguin Random House India Private Limited.

Lahiri A (2020) The great Indian demonetization. J Econ Perspect 34(1):55–74

Minford P (2016) Brexit and trade: what are the options? In: R Bootle et al (ed) The economy after Brexit. Economists for Brexit, pp 13–15

Munasib A, Rickman DS (2015) Regional economic impacts of the shale gas and tight oil boom: a synthetic control analysis. Reg Sci Urban Econ 50:1–17

Powell D (2018) Imperfect synthetic controls: did the Massachusetts health care reform save lives? Available at SSRN 3192710.

Pricewaterhouse Coopers (2015) Disrupting cash: accelerating electronic payments in India. https://www.pwc.in/publications/publications-2015.html

Singh P, Singh V (2016) Impact of demonetization on Indian economy. Int J Sci Technol Manag 5(12):625–635

Xu Y (2017) Generalized synthetic control method: Causal inference with interactive fixed effects models. Polit Anal 25(1):57–76

Zhu H, Gupta A, Majumder B, Steinbach S (2018) Short-term effects of India’s demonetization on the rural poor. Econ Lett 170:117–121

Download references

Author information

Authors and affiliations.

Department of Economics, Oklahoma State University, Stillwater, OK, 74078, USA

Bidisha Lahiri & Anurag Deb

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Bidisha Lahiri .

Ethics declarations

Conflict of interest.

The authors declare that they have no conflict of interest.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1.1 A1. More restrictive donor pool

As a robustness check, we exclude donor countries that display relative post-/pre-treatment deviations larger than 1. This comprises of Australia, Norway, Indonesia, Hong Kong, Czech Republic and Malaysia. This is done in an abundance of caution to ensure that no treatment or spillover effects happened in the donor pool countries. Table

9 shows the estimated effects using the restricted donor pool.

1.2 A2. Synthetic control treatment effect for imports

1.3 a3. synthetic control effect of export/gdp ratio at quarterly frequency, rights and permissions.

Reprints and permissions

About this article

Lahiri, B., Deb, A. Impact of the Indian “demonetization” policy on its export performance. Empir Econ 62 , 2799–2825 (2022). https://doi.org/10.1007/s00181-021-02126-z

Download citation

Received : 12 October 2020

Accepted : 26 August 2021

Published : 14 September 2021

Issue Date : June 2022

DOI : https://doi.org/10.1007/s00181-021-02126-z

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Currency invalidation
• Synthetic control
• Counterfactual

JEL Classification

• Find a journal
• Publish with us
• Track your research