Monetary policy and transaction costs: Empirical analysis of consumption function for the United States of America (1988-2014)

Gomes, Carlos Eduardo; Dias, Maria Helena Ambrosio Article Monetary policy and transaction costs: Empirical analysis of consumption function for the United States of America (1988-2014) EconomiA Provided in Cooperation with: The Brazilian Association of Postgraduate Programs in Economics (ANPEC), Rio de Janeiro Suggested Citation: Gomes, Carlos Eduardo; Dias, Maria Helena Ambrosio (2020) : Monetary policy and transaction costs: Empirical analysis of consumption function for the United States of America (1988-2014), EconomiA, ISSN 1517-7580, Elsevier, Amsterdam, Vol. 21, Iss. 1, pp. 91-103, https://doi.org/10.1016/j.econ.2020.02.001 This Version is available at: https://hdl.handle.net/10419/266951 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/ Available online at www.sciencedirect.com ScienceDirect HOSTED BY EconomiA 21 (2020) 91–103 Monetary policy and transaction costs: Empirical analysis of consumption function for the United States of America (1988–2014) Carlos Eduardo Gomesa,∗, Maria Helena Ambrosio Diasb aDepartment of Economics, Federal University of Roraima (UFRR), Brazil bGraduate Program in Economics, State University of Maringá (UEM), Brazil Received 11 January 2018; received in revised form 16 July 2018; accepted 6 February 2020 Available online 14 May 2020 Abstract In macroeconomic theory, different approaches discuss the ability of monetary policy to affect real variables in the long run. This research proposes the empirical application of a theoretical model that includes nominal rigidities arising from transaction costs and real rigidities arising from the firms’ competition structure. Thus, based on data from the US economy we propose to test the theoretical results combining the existence of shocks arising from both real factors and factors affecting the money market, establishing tests for shocks of supply and aggregate demand. The concern is to investigate if the economic policy is capable of affecting the long run (average) values of the real private consumption. Time series analysis methodology of Structural Vector Autoregressive (SVAR) models was used to test long run relationship for US economy. The ability to establish the long run relationships between these aggregates with some degree of confidence, allow simulating shocks by impulse response functions and variance decomposition. The theoretical results of the model indicate that productivity, transaction costs (number of transactions with credit/debit cards) and currency money tend to positively influence consumption, whereas delinquency have a negative effect on consumption. We tested two empirical specifications and both confirmed the theoretical results for the US economy and, therefore, the research can help policymakers to measure the long run consequences of their decisions, that is, the results converge on the non-neutrality of money in the long run in the US economy. Keywords: Monetary policy; Policymakers; Transaction costs; SVAR; US economy JEL classification: C32; E21; E52 © 2020 The Authors. Production and hosting by Elsevier B.V. on behalf of National Association of Postgraduate Centers in Economics, ANPEC. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/ licenses/by-nc-nd/4.0/). 1. Introduction Throughout the evolution of macroeconomic theory, different approaches have been discussing the ability of monetary policy to affect real variables in the long run. In general, monetary policy can affect indicators of production, employment and consumption in the short run through shocks. If the economic system has rigidities or imperfections, the difficulty of automatic adjustment could insert relevant action for the policymakers in the short run. ∗Corresponding author. E-mail addresses: [email protected] (C.E. Gomes), [email protected] (M.H.A. Dias). https://doi.org/10.1016/j.econ.2020.02.001 1517-7580 © 2020 The Authors. Production and hosting by Elsevier B.V. on behalf of National Association of Postgraduate Centers in Economics, ANPEC. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 92 C.E. Gomes, M.H.A. Dias / EconomiA 21 (2020) 91–103 This research proposes the empirical application of the macroeconomic model of Dias (2002),1which includes nominal rigidities arising from transaction costs and real rigidities arising from the firms’ competition structure. As a theoretical result, the currency would affect the consumption and purchasing power of agents in both the short and long run. Transaction costs play a prominent role in this analysis and may alter the effects of monetary policy on economic aggregates. Still on the theoretical model, it should be emphasized in its derivation that the representation of the agent, which is homogeneous, follows the presentation of Blanchard and Kiyotaki (1987), in which the agent produces the goods and also purchase them. Based on data from the United States economy from 1988 to 2014, the purpose of this study is to test the theoretical results of Dias (2002), which combine shocks from both real factors (productivity, technology, real incomes and market structure) as well as factors related to the money market (currency and transaction costs), establishing tests for shocks of supply and aggregate demand. Thus, the research intends to contribute to the academic debate of macroeconomic approaches and to the analysis of shocks in the North American economy stemming from economic policy or structural factors of the system. In particular, this research proposes to identify the effect of long run monetary policy on consumption, considering transaction costs, competition structure (market power), productivity and prices. Therefore, indicators for transaction costs (such as the number of transactions with credit/debit cards and delinquency rates) and productivity were used to achieve the proposed objectives. In this context, the research is based on the new macroeconomic theories of contract rigidities and imperfect competition to verify the theoretical model and to find the economic relations on the effects of the shocks of demand and supply to be tested by the empirical methodology. In order to reach the goals, the empirical analysis uses the time series methodology of Vector Autoregressive (VAR) models and Structural Vector Autoregressive (SVAR) models. As it is intended mainly to test the long run relationships of variables, SVAR models are appropriate because they have the capacity to establish the long run relationships between these aggregates, allowing the analysis of shock simulation through impulse response functions, in addition to analysis of variance decomposition.2 Several studies on monetary policy were performed using SVAR such as of Carlino and DeFina (1999); Di Giacinto (2003) for the United States; De Lucio and Izquierdo (1999) for Spain; Beenstock and Felsenstein (2005) who used the VAR for Israel; among others. Barro (1977, 1978) conducted empirical tests and concluded that the party does not expect the currency affects the product. Mishkin (1982) repeated the Barro test and found similar results and also concluded that the expected part of the currency has a positive correlation with economic activity. Several New Keynesian authors assert that in the long run, money has the ability to change real variables. The treatment of time series has advanced considerably in the economic literature in the last decades. The empirical estimates in the literature follow the article of Blanchard and Quah (1989), which decompose a country’s specific shocks into permanent and transitory components. Transitory or short run shocks and permanent or long run shocks can be obtained using Structural Vector Autoregressive (SVAR) models.3These empirical models allow us to identify the constraint matrices of the coefficients of the variables, according to the relationships proposed in the theoretical model. Another advantage of SVAR models is in allowing the simulation of impulse response functions of contemporary exogenous variables over endogenous variables. Specifically, this study presents an evaluation of the permanent or long run shocks arising from variations in the quantity of money offered in the United States on the consumption and purchasing power of agents, considering the presence of real and nominal rigidities. The theoretical results of the model indicate that productivity, transaction costs (number of transactions with credit/debit cards) and real balance tend to positively influence consumption, whereas delinquency has a negative effect on consumption. Here, transaction costs have the role of reducing the time spent on transactions and, consequently, increasing the time spent in leisure, since the time used in transactions is discounted from leisure time. In relation to the currency, 1See the derivation of the theoretical model of Dias (2002) in Appendix II – page 23. 2Therefore, the presented results refer to the long run SVAR, that is, the restrictions imposed were long run. 3See Sims (1980, 1986). The use of SVAR models to test theoretical hypotheses was widely used in the RBC – Real Business Cycles literature. This literature allows the reader to understand the relationship between theory and empirical tests using the methodology, especially in the articles by Chari et al. (2005); Fernandez-Villaverde et al. (2007) and Dupaigne et al. (2007). C.E. Gomes, M.H.A. Dias / EconomiA 21 (2020) 91–103 93 ownership has the role of increasing the bargaining power, that is, the individual who has liquidity can bargain at the time of the transactions and, with this, the price paid for the goods/services tends to be lower. If the theoretical model results for the US economy are confirmed, the research can help policymakers measure the long run consequences of their decisions. This paper aims to contribute to the discussion of the determinants of consumption. As such, he dialogues with the main consumer theories, such as Modigliani and Friedman, for example, but he also says that we need to consider other variables in determining consumption, such as transaction costs and money. In addition, the currency is not considered neutral in the system. To fulfill the objective, the present work is structured in two sections, besides this introduction and the final considerations. In the first section, the methodology is presented, while the second section presents the empirical results and the discussions. 2. Empirical methodology The research proposal is to investigate the extent to which nominal shocks from monetary policy can affect long run aggregate consumption in the United States through the application of Structural Vector Autoregressive (SVAR) models and analysis of impulse response functions variance decomposition. The database for econometric applications to the US economy is based on the indicators provided by the World Bank, Federal Reserve Bank of St. Louis (FRED St. Louis), Organization for Economic Cooperation and Development (OECD) and Bank for International Settlements (BIS). The period of analysis is concentrated between 1988 and 2014.4 Time series are observations of the same variable over time, that is, they are formed by sequential observations over time, the product of a random process of data generation. The case of time series requires the stationarity5check of the series before performing any analysis, i.e., it is necessary to work with stationary series. According to Greene (2002), the stationarity or not of a series can imply significant characteristics in its behavior and its properties. The use of non-stationary series can lead to spurious regressions, that is, when analyzing some regression, one could find a relation between two variables that, in fact, does not exist, and could lead to the wrong conclusions. If the series takes some temporary shock, an estimation made with unitary roots will present the shock as permanent (explosive), even if it is a temporary effect. Thus, the first step in the analysis of time series is to identify whether or not the variables are stationary. The non-stationarity of a stochastic process is generated by the presence of unit root in the autoregressive process that generates the variable. Thus, tests on unit root play an important role, as they aid in the evaluation of non-stationarity. To know if a series is stationary or not, it is consensus in the literature on the subject to perform unit root tests.6 With all the stationary series, it is possible to estimate the VAR model.7Once the VAR estimates have been made, it is necessary to verify that all the roots are within the unit circle.8If this occurs, it is understood that the model does not have unit roots and that, therefore, it is suitable for the econometric analysis (to make predictions, for example). If any root of the polynomial is on the unit circle, the model cannot be used, since it has a unit root, unless a VEC is being tested. For Marques (1998), Granger’s Causality can be defined as: “X is said to cause Y in the Granger sense if the past values of X contribute to improve predictions of the current value of Y, all else constant”. The Vector Autoregressive (VAR), proposed by Sims in 1980 in the article entitled “Macroeconomics and Reality”, from which the analysis technique became widespread, becoming one of the most used techniques in the study of macroeconomics. In this paper, the author criticized the econometric models used until then, stating that these were not adequate to make predictions because they are static, because they do not adequately consider the autoregressive character of the series, besides being not precise with respect to the influence of the variables over the others. Thus, 4The periodicity of the data used will be monthly. 5Stationarity is an oscillation of the series around the mean and variance over time. 6For more information about unit root tests, see Applied econometric time series, of Enders (2010). 7We used the VAR methodology because we did not have cointegration – to verify the cointegration number, the following tests were performed: Eigenvalue and Trace Statistic. For more information about cointegration tests, see Applied econometric time series, of Enders (2010). 8In this case, we say that the inverse of the roots must be within the unit circle, since we are using the software EViews 7. The theory says that the root must be outside the unit circle. For further information, see Enders (2010). 94 C.E. Gomes, M.H.A. Dias / EconomiA 21 (2020) 91–103 he proposed the VAR model, which is now widely used in econometrics because of its good predictive capacity and other inferences about the relation between series. According to Cavalcanti (2010), the wide use of VAR as an analysis tool is in part due to the possibility of analyzing the interrelations between the variables from a set of identification restrictions, allowing the estimation of the effect of a “shock” of this variable over the others, becoming an important analysis tool. Given this characteristic of VAR models, the vector is formed by a set of variables that are dependent on their lagged values and the lagged values of the other variables, allowing the relationships between them to be purely statistical (Johnston and Dinardo, 1997). Therefore, all variables are considered as explanatory factor of the others, that is, all variables are considered endogenous - this model assumes the existence of at least two series. After performing all these procedures, another tool of the VAR models is the impulse response function. This impulse response function has the objective of investigating the isolated performance of the variables in the face of some shock in the innovations or residues of the model. The variance decomposition of the prediction errors is a procedure that makes explicit the proportion of the movements generated in a variable due to the occurrence of a certain exogenous shock in itself and in the other variables over time. In other words, the variance decomposition provides information about the importance of a given variable in the changes of another variable in the model. For Enders (1995), the variance decomposition of the prediction errors aims to extract information about the degree of endogeneity of the variables, that is, it represents how much the variance of each variable can be explained by its own variance and how much is determined by the variance of other system variables. Then, response impulse functions and variance decomposition analysis are the main results used from the use of the VAR models. For Cavalcanti (2010), its wide use in VAR and SVAR models as an analysis tool is due to the possibility of analyzing the interrelations between the multiple variables from a set of constraints. In these cases, it is possible to estimate the effect of a “shock” of this variable on the other variables of the system. For Janeiro (2004), by applying the Cholesky decomposition, the ordering of the VAR may change the result, since the variable that occupies the first position has contemporaneous effects (i.e., it is the most exogenous) on the other variables of the system, while the variable that occupies the last position does not exert effects contemporaneously with the other variables (that is, it is more endogenous), being influenced by all other variables. Still in the ideas of Janeiro (2004), SVAR is an attempt to reduce the problem of lack of theoretical foundations through long run constraints, in order to verify if the data are consistent with one or another class of theoretical models. Structural VAR use economic theory to isolate the contemporary connection between variables and require an assumption of identification, allowing correlations to be interpreted. The SVAR models developed by Sims (1980) and Bernanke (1986) had as main proposal to model innovations using economic analysis, that is, the relationship between error predictions and structural innovations9with the objective of incorporating economic sense into the constraints that are imposed on the VAR model. 3. Econometric analysis for us private consumption from 1988 to 2014 The database for the econometric estimation was extracted from the World Bank, Federal Reserve Bank of St. Louis (FRED St. Louis), Organization for Economic Cooperation and Development (OECD) and Bank for International Settlements (BIS). The period used was from January 1988 to December 2014, that is, the periodicity used of the data was monthly. It should also be noted that the data were transformed into an index beginning in January 1988 (except for those already in index), in order to standardize the unit of measure and, consequently, to facilitate the interpretation of the data. Some data obtained did not have the desired periodicity and therefore were interpolated10 to become monthly data. All data were seasonal adjustment using the X12 method. The software used for the estimations was EViews 7 and the for Unit Root Tests, STATA 13. The variables used in this study were: 9See a formal presentation in Enders (1995). 10 The classic definition of Thiele apud Sadosky (1980) for interpolation is that it is able to read the values of a table. Stark (1979) says that interpolation is the mathematical process/method that allows the construction of intermediate values from a discrete set of known data. C.E. Gomes, M.H.A. Dias / EconomiA 21 (2020) 91–103 95 • Real Household Consumption (RHC): Expenditure on goods and services performed by households; values of 2010. This data is provided by the World Bank; • Inflation (CPI): Consumer Price Index. Data available at OECD; • Card Payments (CP): Total number of transactions made with credit card and debit card – Committee on Payments and Market Infrastructures: Red Book. Note: Except e-money. Data provided by the Bank for International Settlements (BIS); • Delinquency (DEL): Delinquency is the sum of unpaid loans (after ninety days) divided by total loans, considering the end of the period. An individual is considered to be in default when he or she has a debt past due for at least ninety days. This data comes from FRED St. Louis; • M1R (narrow money) “Real Balance”: includes currency, i.e. notes and coins, and overnight deposits – discounting inflation, i.e. M1 in real terms. Data available by OECD; • Productivity (PRODU): GDP divided by hours worked, that is, a measure of productivity. Data available by OECD. For empirical analysis of the theoretical model, which was developed for the stable state, the Euler equations were used because, through the first-order equations, it is possible to verify the behavior at the theoretical level of the variables used in the empirical test as explanatory variables and, thus, enable a robust empirical test. This technique was used by Cukierman et al. (1992), who carried out a study on seigniorage and political instability. In this study, the authors’ objective was to explain the importance of seigniorage in relation to other sources of government revenue – which is different between countries. To do so, they used a tax reform model, along with Euler’s equations (i.e., first-order equations) to perform the empirical tests. In this sense, the empirical test of the macroeconomic model of Dias (2002), carried out for the stable state, will also make use of Euler’s equations. Consider the proposal to identify the effects of shocks arising from variations in the quantity of money offered on the consumption in the US economy. For example, the theoretical model indicates that aggregate consumption depends on consumption itself over time, productivity, transaction costs and real balance, considering the technology stage. The higher the productivity and the currency, the higher the equilibrium consumption, given a technology stage, and the higher the transaction costs, the lower the long run equilibrium consumption. In addition, we intend to test the direction of causality among the variables involved in the system of equations of the structural vector autoregressive. In the case of real balance, or purchasing power of the currency, we would have to include an indicator of real income, currency growth and prices and transaction costs, in particular. The proof of the effects of this structural alteration can be done by means of the SVAR – Structural Vector Autoregressive model. Thus, our main long run SVAR system has the following specification for the consumption equation:11 yt= ⎡ ⎢ ⎢ ⎢ ⎢ ⎣ Δ(PRODU) Δ(M1R) Δ(DEL) Δ(CP) Δ(RHC) ⎤ ⎥ ⎥ ⎥ ⎥ ⎦ C = ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ C10 0 0 0 C2C30 0 0 C4C5C60 0 C7C8C9C10 0 C11 C12 C13 C14 C15 ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ et= ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ωt μt st st ηt ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ (1) In the case of the empirical SVAR model, the shocks could be of four types12 : i Shocks arising from productivity gains (␻t); ii Permanent (μt), changes in the amount of money offered; iii Permanent resulting from changes in transaction costs (st) and (st), for which we use indicators of use of credit/debit cards, such as shocks of reduction in transaction costs and indicators of delinquency for shocks of increase in transaction costs; 11 Other variant systems can be tested according to the statistical results of the estimated coefficients, based on the theoretical model, which may include number of firms for market structure of firms and market power. 12 In this case, we use only real variables. If we used nominal variables, inflation should appear, which would negatively impact the consumption and purchasing power of households. 96 C.E. Gomes, M.H.A. Dias / EconomiA 21 (2020) 91–103 Table 1 Signs expected for the empirical estimation. Variable Abbreviation Expected sign Real household consumption RHC Positive (+) Card payments CP Positive (+) Delinquency DEL Negative (−) Real balance M1R Positive (+) Productivity PRODU Positive (+) Source: Prepared by the authors. iv Shocks arising from consumption itself or preferences (␩t). For the currency purchasing power equation, a specification for the SVAR system would also be used based on the theoretical model. We tested two empirical specifications aiming to verify a robustness in the results.13 Thus, Specification 1 presents the most complete model, with all the variables of interest; while Specification 2 provides only one proxy for transaction cost, being the number of transactions with credit and debit cards. In Specification 1, the variable of interest is D1 RHC (real household consumption), and the other variables are: D1 CP (card payments), D1 DEL (delinquency), D1 M1R (real balance) and D1 PRODU (productivity). In Specification 2, the variable of interest is D1 RHC (real household consumption), and the other variables are: D1 CP (card payments), D1 M1R (real balance) and D1 PRODU (productivity). D1 RHC1= C1+ β1 k  j=1 D1 RHCt−j+ β2 k  j=1 D1 CPt−j+ β3 k  j=1 D1 DELt−j + β4 k  j=1 D1 M1Rt−j+ β5 k  j=1 D1 PRODUt−j+ μt1(2) D1 RHC2= C2+ β6 k  j=1 D1 RHCt−j+ β7 k  j=1 D1 CPt−j+ β8 k  j=1 D1 M1Rt−j + β9 k  j=1 D1 PRODUt−j+ μt2(3) In which: C1is the constant of Specification 1; C2is the constant of Specification 2; β1, β2, β3, β4and β5are the parameters estimated for Specification 1; β6, β7, β8and β9are the parameters estimated for Specification 2; μt1is the error of Specification 1; μt2is the error of Specification 2. Table 1 lists the expected results for the empirical tests of the long run SVAR model. It is worth mentioning that these results are based on the results of the theoretical model of Dias (2002). In relation to the unit root tests, they showed that all the variables of the model are stationary in the first difference. Initially it was verified whether the series was correlated or not. When a series is autocorrelated, it is indicated to use the ADF test, but when it is not, the test used should be DF. However, the DFGLS test, which is considered more robust than the ADF test, because it makes use of the generalized least-squares regression method that corrects the residues. 13 The objective of checking these two empirical specifications was to try to validate if the “complete” estimation would remain with the same results, in other words, if changing the composition of the variables in the model, the results would remain the same. C.E. Gomes, M.H.A. Dias / EconomiA 21 (2020) 91–103 97 Table 2 Results of empirical specifications. Variables Specification 1 Specification 2 Real Household Consumption (D1 RHC) 0.594002* (0.0000) 0.770089* (0.0000) Payment Cards (D1 CP) 0.424918* (0.0000) 0.273304* (0.0000) Delinquency (D1 DEL) −0.169784* (0.0000) – Real Balance (D1 M1R) 0.446674* (0.0000) 0.108076** (0.0190) Productivity (D1 PRODU) 0.451023* (0.0000) 0.114608** (0.0138) Source: Prepared by the authors from EViews 7. Observations: Statistical significance level: * = 1%; ** = 5%. Values in parentheses are p-values. Included remarks: Specification 1: 317; Specification 2: 317. 3.1. Empirical specifications According to the variables used in the VAR model, the two empirical specifications will be presented. The first specification (considered the most complete) will be the one analyzed, and this analysis is applicable (extended) to the second specification. After verifying the stationarity of the series, it is necessary to verify the interrelations between the variables. For this, we must analyze the causal relation in the Granger sense. This type of causality – in the Granger sense – is a test of temporal precedence, and some temporal precedence relationships – based on 10% statistical significance – were found to configure the VAR “universe”. Then, it was verified that the optimal number of lags was 6 in the VAR model. Afterwards, the robustness analyses were implemented. For this, the VAR stability conditions were checked, if the residuals were autocorrelated, normal, homoscedastic and if they were stationary (for the model as a whole and also for the equation of interest – consumption). In relation to the stability of the model, all the roots characteristic of the VAR was located within the unit circle, indicating that the model meets the conditions of stability. That is, this test shows the inverse of the unit root and, therefore, the roots must remain inside the unit circle and be smaller than one in module.14 Residues of the regression, whose variable of interest is “D1 RHC”, were not correlated, i.e. there was no autocorrelation present. According to the Jarque–Bera test for normality of residues, it rejected the null hypothesis that residues (including together) follow a normal distribution. However, based on the property of large samples of the Central Limit Theorem, it can be considered that asymptotically the residues follow a normal distribution.15 Despite the non-normality of the residues, they are considered homoscedastic and the tests indicated that they (the residues of the VAR as a whole and the equation of interest, consumption) are stationary. 3.1.1. Results of empirical specifications Nevertheless, the main interest of the present study is to verify the long run empirical validation (Stationary State) of the theoretical model of Dias (2002), that is, to investigate the effects of monetary policy shocks on agent consumption in the long run. For this, the results of the Structural Vector Autoregressive (SVAR) Model are presented. According to Sims (1986), it is possible to verify if the effects of explanatory variables remain in the long run. Thus, Table 2 presents the results of the two empirical specifications. Analyzing the results demonstrated in the SVAR in Table 2 (for Specification 1), it is possible to perceive a significant long run relationship between the variables. The column of Specification 1 (as well as Specification 2) presents the empirical result including all the variables of the system, all being statistically significant, confirming empirically the results of the theoretical model of Dias (2002), and all the signs were as expected – see Table 1.16 The variables that showed positive effects on consumption in the first difference were: consumption – D1 RHC - (0.594002), number of transactions with credit/debit cards – D1 CP – (0.424918), currency – D1 M1R – (0.446674) and productivity – 14 The theory of this test predicts that the roots need to be outside the unit circle and be larger than one in module. However, as the software used here (EViews 7), it presents the inverse of the unit root and, therefore, it must be inside the unit circle and be smaller than one in module. 15 Further details in Wooldridge (2006). 16 See page 10. 98 C.E. Gomes, M.H.A. Dias / EconomiA 21 (2020) 91–103 D1 PRODU – (0.451023); while the variable that presented a negative sign for household consumption was: delinquency – D1 DEL – (−0.169784). The results bring the inertia of US household consumption in the long run, and corroborate Carroll and Summers (1991) that the hypothesis of permanent income may not apply to the case of the North American economy, in the sense that it does not is the only relevant variable for the determination of private consumption in the USA. Productivity positively influences consumption and makes the family more available, thus increasing its purchasing power and, possibly, its quality of life, since it may have more leisure time or more resources available, since its productivity has increased. Regarding the use of cards and delinquency (proxies for the fall and increase in transaction costs, respectively), the use of cards positively change the relation of consumption of the households since they tend to facilitate at the time of the exchanges, besides that use of this tool makes the transactions faster, it is not necessary for the individual to move to a physical store, being able to make purchases through the computer, without leaving home and use this “time saving” in other activities, mainly leisure (properly speaking), since the time spent on purchases is discounted from leisure time. Another point that needs to be highlighted is that this facility helps the individual to protect from the transaction cost known as “shoe leather costs” because he can search, for example, through the internet, to buy goods/services at lower prices. These results are in agreement with the results found by Gross and Souleles (2002), that the use of cards positively affect consumption. In addition, the greater the use of the card, the lower the transaction costs and the greater the ease of exchanges, which translates into a greater consumption capacity. Delinquency negatively affect households’ consumption, since they need to “regularize” their past due debts, that is, they need to separate part of the budget to pay off past due debts and, as a result, their available income for consumption decreases. Once debits are regularized, families can get credit again so that they can return to their “normal” consumption patterns in the future. The individual considered in default is deprived of some tools that can expedite the consumption – such as cards, for example – that is, these tools have the purpose of reducing transaction costs, leaving more time for leisure (other activities), since the individual in delinquency incurs costs of “shoe leather costs” to protect himself from inflation. According to Romer (2012), one of the hypotheses of permanent income presupposes that interest rates, both for loans and for savings, are the same. However, the interest rates that households pay on card debt or some other type of delinquency are higher than savings rates, and not all individuals are able to lend more at any interest rate. Liquidity constraints can increase savings in two ways. The first, and most obvious, is the fall in consumption, and the second, as approached by Zeldes (1989), refers to binding restrictions, even if they are not binding, in the future they may be and consequently reduce consumption. The real balance positively affects household consumption, this possibly occurs because of the adjustments of expectations that families make, in addition to which, more money in circulation in the economy tends to increase the price level (ceteris paribus), causing families to increase the speed of consumption in order to mitigate the loss of purchasing power with inflation. Another factor is that the more money individuals have, the more bargaining power they have, that is, they can bargain for the value paid for goods (McCallum, 1989). In addition, it is evident that the real balance does not appear as neutral in the North American economy, since it is able to change the relations of consumption. Regarding the robustness tests, the results obtained were the same as the VAR. Therefore, according to the results found here, it is clear that other variables, besides income, are determinants of the consumption function in the long run, besides evidencing the importance of the mechanisms to reduce transaction costs and, finally, the ownership of the currency is highly important at the time of making exchanges. 3.1.2. Impulse response functions Figs. 1 and 2 show the impulse response functions of the SVAR model with a single and accumulated shock, respectively, considering a time horizon of 50 months.17,18 17 “Shock1” means “D1 RHC”; “Shock2” means “D1 CP”; “Shock3” means “D1 DEL”; “Shock4” means “D1 M1R”; “Shock5” means “D1 PRODU”. 18 The analysis of impulse response and variance decomposition of Specification 2 followed the same pattern presented in Specification 1. Therefore, the same analyses can be applied and, therefore, we chose not to present these results.