Money, income and causality in Argentina

Matías Rebozov 10607668 MSc in Economics Monetary Policy and Banking

University of Amsterdam

Money, Income and Causality in Argentina

Abstract

In this empirical study I investigate the relationship between money (MB, M1 and M2) and income (real GDP) in Argentina between 1987 and 2005. I employ Sims Test with the purpose of examining the causal order among these variables. The results show that the null hypothesis of unidirectional causality from MB and M1 to real GDP is rejected. However, the same hypothesis cannot be rejected for M2.

Table of Contents

I. Introduction ... 1

II. Literature Review ... 4

III. Methodology: Sims Test ... 9

IV. Sample Period Selection ... 12

V. Empirical Evidence ... 13

VI. Conclusions ... 18

References ... 19

Appendix ... 22

I. Introduction

The research question of this thesis states: Is there unidirectional causality among money supply (MB, M1 and M2) and income (real GDP) in Argentina during the period between 1987 and 2005? Before exposing the main results of this empirical study it is mandatory to answer: what is the actual meaning of the research question? And after that, why is this an interesting topic? Why is it relevant for this specific country?

Granger (1969) developed a statistical hypothesis test for determining causality between two variables. However, his interpretation is based entirely on predictability. This means causality is defined merely by the ability of one time series to forecast another time series. In other words, if some variable 𝑋 contains past information that helps to predict 𝑌, and if this information is not contained in any other variable, then 𝑋 is said to cause 𝑌.

Hence, restating my research question: Is there any information in the time series of MB, M1 and M2 for predicting or forecast changes in real GDP for Argentina during the period previously mentioned? The answer is yes, but only for M2.

In Argentina, between 1987 and 2005, there is statistical evidence to reject the null hypothesis of causality running one way from money to income for MB, M1 and real GDP. Nonetheless, the same hypothesis cannot be rejected for M2. Either way, changes in real GDP do not lead to changes in any of these monetary aggregates.

The results obtained are mainly based on an empirical technique employed for the first time by Christopher Sims. In his paper Money, Income, and Causality (1972), he introduced a methodological novelty for testing causality. With a two-variable distributed lag system he managed to demonstrate that causality ran from money to income in the postwar period for the US (1947-1969). In the words of Walsh (2010): “the past behavior of money helped to predict future GNP”.

To illustrate and simplify, with one past, present and future variable, the bivariate system has the following form:

𝑌𝑡 = 𝛽0+ 𝛽1𝑋𝑡−1+ 𝛽2𝑋𝑡+ 𝛽3𝑋𝑡+1+ 𝜇𝑡

𝑋𝑡 = δ0 + δ1𝑌𝑡−1+ δ2𝑌𝑡+ δ3𝑌𝑡+1+ 𝑒𝑡 1

Agreeing with the definitions given by Granger (1969), for affirming that causality runs in one way from 𝑋 to 𝑌, future observations of 𝑋 should have zero coefficients (𝛽3= 0).

Which means that 𝑋𝑡+1 does not explain current observations of 𝑌𝑡. Additionally, the second

regression with 𝑋 as a dependent and 𝑌 as an independent variable, future coefficient of 𝑌 have to be statistically significant (δ3≠ 0). This implies that future observations of 𝑌𝑡+1 are

linked with current values of 𝑋𝑡. Lastly, bidirectional causality holds if both coefficients, 𝛽3

and δ3 are significant, and no feedback1 is occurring among both variables if 𝛽3 = δ3 = 0.

But, why is this an interesting topic and why is it relevant specifically for Argentina? First of all, it is worth noting that the relationship between money supply and income has been the subject of discussion and debate among Monetarists and Keynesians. For the former, changes in income are induced largely by changes in money stock while according to the latter money supply reacts passively to fluctuations in income. In other words, the direction of causality runs from income to money for Keynesians and from money to income for Monetarists (Komura, 1982). In addition, these two schools have been also in conflict about the relationship between money and prices. Monetarists view inflation as a monetary phenomenon while Keynesians in contrast accept inflation as a real phenomenon caused by real factors (Stein, 1981).

However, as it is detailed in the next section, the results of several studies show that the theoretical framework is not consistent in most cases, so we cannot put an end to the discussion. “Some studies have established unidirectional causality running from income to money, and others from money to income. Some have established bidirectional causality while others have found no evidence of any causality” (Bengali, Khan and Sadaqat, 1999). Having said that, in order to understand how this topic is relevant for Argentina, it is essential to provide a brief review of its economic situation during the respective sample period. The data selected can be divided in to three different regimes: from 1987 to 1991, between 1992 and 2001, and from 2002 to 2005.

At the beginning of the first period, the monetary authority implemented a new currency unit together with a stabilization program to control the price rate acceleration (Dornbusch and De Pablo, 1995). The plan did not succeed and in 1989 and 1990 the economy was hit 1

“(…) feedback is occurring when 𝑋𝑡is causing 𝑌𝑡 and also 𝑌𝑡 is causing𝑋𝑡” (Granger, 1969).

2

by two episodes of hyperinflation with an annual rate of more than 3000% and 2300% respectively (Rapoport, 2010).

In 1992, the monetary instability led to a new fixed exchange rate regime where the currency was pegged to the US dollar. With a monetary policy heavily restricted, and an inflation rate converging to US levels, a sustained rate of real GDP growth was the result of the monetary reform until 1998 (Gerchunoff and Llach, 2004).

However, driven by the level of indebtedness, budget deficit and unemployment among other things, Argentina entered in recession in 1999. In 2001 the central bank prohibited any withdrawal from the banking system (Gerchunoff and Llach, 2004). In 2002 the real exchange devaluated to around 250% and the fiscal authority announced the public debt default (Heymann and Ramos, 2010).

During the last period the economic policy shifted from privatization, trade openness, deregulation and supply-side policies to a new floating exchange-rate regime and demand driven policies (Coremberg, 2013).

In sum, Argentina offers an economic history which is challenging for the monetary authority and its policy makers. In addition, the empirical results I attain agree with the effectiveness of monetary policy over real variables such as output. Hence, I consider the research and understanding of the topic of my thesis can be useful for an adequate monetary policy and its sustenance in the long run.

In the subsequent section, an overview of the literature is chronologically provided regarding causality tests in different countries through time (including other studies for Argentina). It also shows how each result depends on an individual economy. In section III the methodology used is detailed. Also, steps followed to apply Sims Test2 method. Why the data sample is until 2005 corresponds to section IV. The results and its implications are explained in depth in section V. To finish, conclusions and last remarks of theses outcomes are part of the last section.

2 _{Named like this for example by Guilkey and Salemi, 1982.}

3

II. Literature Review

As I mention in the previous section, the first paper that was published regarding this technique for testing causality was done by Christopher Sims in 1972. The author found statistical evidence to not reject the hypothesis of unidirectional causality running from money (MB and M1) to income (nominal GNP) with postwar data in the US. Moreover, the alternative hypothesis of income as an exogenous variable was rejected. His way of interpreting Granger causality is the following:

“If and only if causality runs one way from current and past values of some list of exogenous variables to a given endogenous, then in a regression of the endogenous variable on the past, current and future values of the exogenous variables, the future values of the exogenous variables should have zero coefficients”(Sims, 1972).

After the paper was published diverse articles appeared in circulation. This technique for testing causal ordering was useful for many other authors and applied to different countries. However, the outcomes of the tests are not similar between them. This takes us to be certain that the direction of causality depends on a number of factors that rest on specific features of each country. In other words, it is not possible to generalize Sims’ results to other economies.

Even though my work is focused on the Distributed Lag Methodology, as named by Sims (1972), the following articles try to show chronologically the variety of results among different countries, relating similar variables employing either Granger or Sims Test (Guilkey and Salemi, 1982).

In 1974, Barth and Bennett published “The role of money in the Canadian economy: an empirical test”, with similar goals as Sims but relating stock of money (M1 and M2) with both GNP and the Index of Industrial Production (IP). Their results suggested that M1 is explained by IP and that there is simultaneous causality between M1 and GNP. On the other hand, when relating M2 with IP and GNP, an absence of unidirectional causality was indicated. The argument for these results was that M1 excludes cash balances held by households. Hence, consisting predominantly of business firm balances, there is more sensitivity to output fluctuations. Likewise, at that time Canada was an open economy with

a fixed exchange rate regime, employing a different role for money than in the US. “Therefore, in contrast to Sims’ finding for the US, the monetarist contention that the money stock is the principal determinant of the level of economic activity is not supported by Canadian Data” (Barth and Bennett 1974) for the period 1957 to 1972.

Williams, Goodhart, and Gowland (1976) stated that there is not clear information indicating the causal order between money and income in the UK between 1958 and 1971. However, they allow for the possibility of a weak unidirectional causality from income to money, which completely differ from Sims’ outcomes. Some of the reasons for this were the structural differences between UK (smaller open economy with a fixed exchange rate during the observation period) and US.

Robert Barro (1977) derived that only unanticipated movements in money affect real economic variables like unemployment or output in the US for the period between 1941 and 1973. This author first published empirical studies considering causality for real variables. His results did not contrast with the conclusions obtained by Sims (1972)3_.

According to Komura (1982), it was only in the US that movements in money could cause fluctuations in income, given that they “can control its own money supply since the dollar serves as a reserve currency under a fixed exchange rate regime”. He did the same test as Sims but for Japan with observations for the period between 1955 and 1971 suggesting a bidirectional causal nexus: “feedback relationship between M2 and GNP” (Komura, 1982). Nevertheless, for the period between 1971 and 1980, Komura indicates causality running in one way from GNP to M2. As the author listed, one of the principal explanations of the diverse results obtained for the same country in different periods, is due to the modification in the exchange rate regime from fixed to flexible (even though this may not agree with the theoretical framework4).

In Argentina, one of the first empirical studies concerning causality among macro variables was done by Navarro and Rayó in 1983. They studied the relation between money (M1) and prices between 1956 and 1980. Their estimations showed no feedback among both variables. However, they also added that the causal relation among these variables might 3 _{See Barro (1982) for further details.}

4 _{See Komura (1982:33).}

5

change from one country to another or even within the same country (as in the case of Japan). Depending on the inflation rate, structural background of the fiscal organization, the opening of the economy, the exchange rate regime or just the monetary policy target (Navarro and Rayó, 1983).

Friedman and Kuttner (1992) analyzed for the US the empirical relationship between money and real income (also nominal income and prices). The time period selected was divided in three: 1960 to 1979, 1960 to 1990 and 1970 to 1990. They showed how including data after the eighties alters the results of older studies (like the one of Sims in 1972). For the first period, MB, M1 and M2 “contained information about future income movements”. However, extending the period eliminates this significance for MB and “beginning the sample in 1970 leaves only M1 significant among these three variables”. In other words, they showed that the relationship changes over time.

Bernanke and Blinder (1992), also examined for the US the effect of money supply over the real economy and the transmission mechanisms. They showed that “the interest rate on Federal funds is extremely informative about future movements of real macroeconomic variables”. They managed to demonstrate empirically, with a sample period from 1959 to 1989, that the funds rate is a good indicator of monetary policy actions. The main argument was that this indicator is “less contaminated by endogenous responses to contemporaneous economic conditions” than other variables.

The linkages between money, real income and prices in Saudi Arabia between the years 1965 and 1993 were described by Mohammad Al-Jarrah (1996). “The results indicate that real income contributes significantly in explaining changes in money, while the reverse is not true. The evidence on the contribution of money in explaining prices change, however, is weak”. Al-Jarrah extended his analysis arguing that as many other developing countries, this type of findings corresponds to economies with a non-organized financial market and a fixed exchange rate regime.

Bengali, Khan and Sadaqat (1996) presented an article examining causation not only for M1, M2 and income, but also for the same monetary aggregates and the level of prices in Pakistan, for the period between 1972 and 1990. As in many other countries, causality ran

both ways from money to economic activity. Regarding the relation between money and prices, unidirectional causality was found going from the former to the latter.

With respect to some papers from the last decade, Chin Huat and Tai Wai (2000) did the same empirical study for Singapore with observations within 1975 and 1998. Unidirectional causality was found from GDP to M2 and M3, while a two-way causality was found for M1 and GDP. According to the author’s conclusion, the obtained results agree with the particularities of a small and open economy. “(..) external factors are more likely to explain output fluctuations than monetary policy” (Chin Huat and Tai Wai, 2000).

It is worth mentioning that Aguirre, Burdisso and Grillo (2006) did also test causality in Argentina for the money-income relation (M1, M2 and M3 with GDP). They got no feedback among these two variables and its components during 1993 and 2005. That is, money did not anticipate output performance, holding the same in the opposite. The only explanation given, was that the period selected corresponds to two different macroeconomic regimes. Even though this paper is worth to mention, it is not strictly comparable with my thesis for the following reasons: the main purpose is not to test Granger causality. The aim of this article is to estimate a money demand function in order to forecast monetary aggregates. Hence, times series differ on length, statistical tests are not the same, and the lag specification of the regressions has a dissimilar structure.

A recent paper examined the causal order in the money-income nexus for the Turkish economy with data over the period of 1987 to 2011 (Bozoklu, 2012). “A variable Granger-causes another variable, if including it in the information set will improve the forecast of the second variable”5. Regardless of the potential outcomes of this test and the one I employ, Bozoklu explains its importance for policy makers and the efficiency improvement in the policy rule. Simultaneous causality between money and income is the result of the test implemented for the Turkish economy during the sample period mentioned previously (Bozoklu, 2012).

In summation, I show that for many articles and different countries, money and income (and also real income) relationship depends essentially on the particular structures of a region in a certain period, in social and economic terms. Hence, there is no general rule for the 5 _{See Bozoklu, 2012:175, for further details.}

7

relation between these two variables. What is more, no conclusion can be obtained regarding the historical discussion among Keynesians and Monetarists.

III. Methodology: Sims Test

Quarterly data from 1987 until 2005 is used for this empirical test. All the time series are seasonally adjusted from the source and after that, measured in logarithms. Otherwise, two variables deseasonalized by different procedures could result in a spurious correlation among them (Sims, 1972).

Likewise, for doing statistical inference prefiltered data is needed to reduce serial correlation of the residuals (Pindyck and Rubinfeld, 1991). Thus, I filtered the observations with the same second order filter used by Sims (1972). This allows me to choose an appropriate value of 𝑘 to “create a non autocorrelated error structure” (Sims, 1972) for each regression:

(1 − 𝑘𝐿)2 _{= 1 − 2𝑘𝐿 + 𝑘}2_𝐿2 _{(0 < 𝑘 < 1)}

I employ the Durbin-Watson statistic for testing serial correlation of the disturbance terms and to check if the value of 𝑘 was correctly selected. “If the autoregression indicates the presence of some serial correlation, however, then change the value of 𝑘 and repeat steps” (Mehra, 1977). Also, I use the Jarque-Bera statistic for analyzing the normality distribution of the estimated error terms for each regression. Given the lack of observations for satisfying the test requirements, normality of distribution failed to appear in most of the cases.

Therefore, residuals lack of autocorrelation and heteroscedasticity. Normality of distribution depends on the disturbance of each regression.

It is noteworthy that a constant term, a linear trend term, and three seasonal dummies were included in each regression (apart from the leading and lagging values of the independent variable).

Regarding the methodology itself, the causality test I employ for this empirical study is a technique developed by Sims (1972) that can be summarized in the following bivariate system:

𝑔𝑑𝑝𝑡 = 𝛼 + � 𝛽𝑖𝑀𝑡−𝑖 𝑛 𝑖=−𝑝 + 𝜇𝑡 𝑀𝑡 = γ + � δ𝑗𝑔𝑑𝑝𝑡−𝑗 𝑛 𝑗=−𝑝 + 𝑒𝑡 (𝑀𝑡 = 𝑀𝐵𝑡, 𝑀1𝑡, 𝑀2𝑡)

Where 𝛼 and γ are constants, 𝛽𝑖 and δ𝑗 the coefficients on 𝑀(measured as 𝑀𝐵, 𝑀1 and

𝑀2) and 𝑔𝑑𝑝 (real GDP) respectively, and 𝜇𝑡 and 𝑒𝑡 the error terms.

𝑀𝐵 is the abbreviation for monetary base and it is composed of money in circulation plus reserves. On the other hand 𝑀1 is the currency held by public and current accounts in Pesos and US dollars, both public and private. Finally, 𝑀2 is the sum of 𝑀1 plus saving accounts in Pesos and US dollars, once more public and private6.

Following Granger’s (1969) definition, unidirectional causality running from M to gdp holds if ∑ 𝑛𝑖=−𝑝𝛽𝑖 is not statistically significant (𝑖 < 0) and ∑ 𝑛𝑗=−𝑝δ𝑗 is statistically significant (𝑗 <

0). Stated in another way, when gdp is regressed on future, present and past values of M, then future values of M are not significant, while in the reversed regression future values of gdp are significant. From Sims (1972), this implies exogeneity of M in the money-income relationship. On the other hand, causality will run one way from gdp to M if the opposite happens. In that case, M will be endogenous and gdp exogenous. Moreover, bidirectional causality holds if both ∑ 𝑛𝑖=−𝑝𝛽𝑖 and ∑ 𝑛𝑗=−𝑝δ𝑗 are statistically significant (𝑖; 𝑗 < 0) which

means both variables are “endogenous to each other, and neither one can be treated as exogenous in explaining the other” (Komura, 1982). Lastly, there is no feedback among these two variables when ∑ 𝑛𝑖=−𝑝𝛽𝑖 and ∑ 𝑛𝑗=−𝑝δ𝑗 are not significant (𝑖 < 0 and 𝑗 < 0) and

no conclusion regarding causation can be done.

However, applying F-tests for defining causal direction it is not the only factor that matters. Sims (1972) suggested something called economic significance, where “the absolute size of the coefficients is important regardless of the F value” and “coefficients that are large from the economic point of view should not be casually set to zero no matter how statistically not 6 _{Official definitions from: Aguirre, H., Burdisso, T. & Federico, G., 2006, Towards an estimation of money} demand with forecasting purposes, Investigaciones Económicas, Central Bank of Argentina.

10

significant they are”. Hence, applying F-tests for defining causal direction simply demonstrates that unidirectional causality is a possibility.

In practical terms, if the sum of future coefficients in absolute values is bigger than the sum of past coefficients in absolute values too, then future coefficients are economically significant.

IV. Sample Period Selection

Official Consumer Price Inflation Index (CPI), provided by the National Statistics Institute (INDEC) started to hide inflation since 2007 (an intervention that also continuous today, August 2014). Given that the methodology applied to calculate CPI is different before and after 2007, I decide to use trustworthy data. Hence, time series are not updated in order to get methodologically consistent information. “Several academic and private analysts have estimated that the actual CPI has been considerably higher than the one reported on the official series” (Coremberg, 2013). Even more important is that this kind of intervention has also influenced other economic indicators like real GDP. Thus, without a consistent estimation of the inflation rate, official real GDP loses reliability.

GDP distortions driven by problems with official statistics happened in many other countries. For example, China (Madison and Wu, 2008), Greece and Chile (Garcia and Freyhoffer, 1970). In the latter case, official inflation rate was underestimated for the last part of the sixties. The Argentinian case and its distortions has also been studied and reported by Heymann and Ramos (2010), Cavallo (2012), Damill and Frenkel (2013), and Coremberg (2013), among others.

Therefore, I choose to regress from official sources like INDEC but before the intervention, rather than splicing time series prepared with different methodologies and therefore risking my inferences.

On the other hand, the monetary aggregates data is provided by the Central Bank of Argentina. Information is given at the end of the month. Thus, I calculate a three months average to get quarterly data.

V. Empirical Evidence

For Table 1, Table 2 and Table 3, the first column corresponds to the lag structure specification. The letter L means Lag and the letter F stands for Future. Thus, for example the first row, “4L, 2F” implies four lags and two future values. The next column shows the number of observations used for each particular regression. In the third one, K is the parameter that belongs to the adjustment of the second order filter. The coefficient of determination R-sq. indicates how well data fits the regression. D-W is the Durbin and Watson statistic for testing serial correlation of the residuals. When D-W is around two, residuals are not serially correlated. J-B is the Jarque and Bera statistic for checking the normality of distribution of the errors. The column pertinent to F-Ratio indicates if future values are significant or not. F-P is the difference (in absolute values) between the sum of future and past coefficients, which points out the economic significance. The last column shows with arrows the direction of causality among the variables involved in each table in statistical terms. A two way arrow specifies bidirectional causality and one way arrow means unidirectional causality.

Table 1 shows the results obtained from the causality test applied to the real GDP-MB relation.

As we can see in the last column, both tables with four and eight past lags show bidirectional causality at 1% significant level in almost all cases. The only exception corresponds to the regression with four lags and eight future values where MB appears to cause gdp.

Outcomes of Sims Test concerning the causality relationship between M1 and gdp are shown in Table 2. In this case, for both lag structures with four and eight past values, bidirectional causality can be seen among these variables at 1% significant level in all the regressions. K R-sq. D-W J-B F-Ratio F-P K R-sq. D-W J-B F-Ratio F-P 4L, 2F 64 0.54 0.6612 1.996 0.94 8.00*** -0.03 0.59 0.7925 2.026 1.64 20.8*** 0.46 MB↔gdp 4L, 4F 62 0.48 0.7416 1.921 0.39 8.27*** 0.08 0.58 0.8014 2.009 1.31 11.86*** 1.67 MB↔gdp 4L, 6F 60 0.46 0.7874 2.024 0.29 7.82*** 0.12 0.57 0.8055 1.945 1.30 8.16*** 1.81 MB↔gdp 4L, 8F 58 0.42 0.8388 1.972 0.20 1.74 0.16 0.56 0.8015 1.948 3.67 5.34*** 2.35 MB→gdp K R-sq. D-W J-B F-Ratio F-P K R-sq. D-W J-B F-Ratio F-P 8L, 2F 64 0.56 0.7377 1.967 0.72 7.57*** -0.03 0.54 0.8504 1.964 6.86** 17.9*** -0.07 MB↔gdp 8L, 4F 62 0.53 0.7781 1.989 0.30 7.00*** 0.04 0.54 0.8505 1.995 5.67* 9.39*** 0.86 MB↔gdp 8L, 6F 60 0.47 0.8337 1.969 1.04 8.73*** 0.10 0.54 0.8540 1.984 4.52 6.51*** 0.83 MB↔gdp 8L, 8F 58 0.49 0.8522 2.020 2.36 8.25*** 0.11 0.53 0.8448 1.927 3.89 4.83*** 1.15 MB↔gdp ***: Significant at 0.01 level. ***: Significant at 0.05 level. ***: Significant at 0.10 level. 1987I-2005IV

Lag Form Obs. gdp=f(MB) MB=f(gdp) Causality

Table 1. Sims Test

1988I-2005IV

Lag Form Obs. gdp=f(MB) MB=f(gdp) Causality

For Table 3, the outcomes of tests between M2 and gdp differ from the ones described in Table 1 and Table 2 (MB and M1). In this last case, there is mixture of results. When increasing the number of future values in the regression, causality tends to be bidirectional and not unidirectional, for both sample periods (1989I-2005IV and 1988I-2005IV) in statistical terms. However, it turns out that from an economic point of view, looking at the F-P columns, it can be observed that in all cases unidirectional causality is running from M2 to gdp. K R-sq. D-W J-B F-Ratio F-P K R-sq. D-W J-B F-Ratio F-P 4L, 2F 64 0.53 0.7015 1.995 1.35 3.73** -0.08 0.60 0.8737 1.958 0.17 14.42*** 1.51 M1↔gdp 4L, 4F 62 0.49 0.7382 1.924 2.13 4.32*** 0.02 0.58 0.8949 1.995 0.99 12.33*** 3.02 M1↔gdp 4L, 6F 60 0.44 0.8052 1.911 0.39 5.6*** 0.12 0.58 0.8963 1.941 2.78 8.22*** 3.04 M1↔gdp 4L, 8F 58 0.42 0.8405 1.901 0.98 5.27*** 0.17 0.57 0.9000 1.964 1.01 6.12*** 3.61 M1↔gdp K R-sq. D-W J-B F-Ratio F-P K R-sq. D-W J-B F-Ratio F-P 8L, 2F 64 0.56 0.7532 1.911 1.90 4.75** -0.09 0.57 0.9066 1.967 0.10 13.45*** 1.16 M1↔gdp 8L, 4F 62 0.49 0.7856 1.852 4.72* 4.49*** 0.01 0.55 0.9227 1.932 0.12 11.4*** 2.55 M1↔gdp 8L, 6F 60 0.43 0.8484 1.872 1.65 6.33*** 0.11 0.56 0.9223 1.996 1.51 7.56*** 2.48 M1↔gdp 8L, 8F 58 0.42 0.8739 1.895 0.35 5.43*** 0.14 0.55 0.9228 1.923 1.16 6.09*** 2.93 M1↔gdp 1987I-2005IV

Lag Form Obs. gdp=f(M1) M1=f(gdp) Causality

***: Significant at 0.01 level. ***: Significant at 0.05 level. ***: Significant at 0.10 level.

1988I-2005IV

Lag Form Obs. gdp=f(M1) M1=f(gdp) Causality

Table 2. Sims Test

I analyze various lag structures given that Sims Test results may change according to their specification (Mehra, 1977). Strictly speaking, this empirical test is sensitive to the lag form. As indicated before, the sample period I use for this empirical test contains three different regimes: from 1987 to 1991, between 1992 and 2001, and from 2002 and 2005. The first one is represented by a flexible exchange rate, very high inflation and a tight fiscal situation. The second one is composed of a pegged exchange rate, stabilized prices at international levels and deteriorating fiscal accounts. And the last part of the sample selected combines a floating exchange rate, less complex fiscal commitments but a deteriorating inflation rate. Not to mention the targets for monetary policy, the degree of openness of the economy and how both changed through the entire sample period (1987-2005).

Thus, it is difficult to justify with certainty why there is a difference between the results I obtain regarding the monetary aggregates chosen and the real GDP. More specifically, why M2 is considered to empirically cause real GDP and on the other hand why MB and M1 are considered to be part of an endogenous relation with real GDP. Nevertheless, it is important to remark that in all cases income is not exogenous in its relation with money.

K R-sq. D-W J-B F-Ratio F-P K R-sq. D-W J-B F-Ratio F-P 4L, 2F 60 0.54 0.6458 1.964 2.79 0.13 -0.26 0.59 0.8790 1.985 7.01** 8.36*** 2.00 M2→gdp 4L, 4F 58 0.54 0.6775 1.958 2.62 1.06 -0.24 0.57 0.8804 1.940 3.51 6.14*** 2.32 M2→gdp 4L, 6F 56 0.49 0.7185 1.938 3.33 2.75** -0.16 0.55 0.9122 1.929 3.25 4.71*** 2.32 M2↔gdp 4L, 8F 54 0.43 0.8242 1.910 3.18 3.89*** -0.11 0.55 0.8821 1.967 4.87* 3.69*** 2.29 M2↔gdp K R-sq. D-W J-B F-Ratio F-P K R-sq. D-W J-B F-Ratio F-P 8L, 2F 60 0.55 0.6848 1.989 1.04 0.16 -0.31 0.55 0.9088 2.046 8.22** 8.12*** 2.10 M2→gdp 8L, 4F 58 0.55 0.6990 1.929 0.69 1.76 -0.33 0.52 0.9145 1.976 4.08 6.36*** 2.60 M2→gdp 8L, 6F 56 0.51 0.7555 1.940 0.01 3.59*** -0.27 0.51 0.9191 1.963 3.81 4.56*** 2.54 M2↔gdp 8L, 8F 54 0.47 0.8172 1.995 0.06 3.51*** -0.23 0.52 0.9045 2.010 3.49 3.47*** 2.59 M2↔gdp Causality ***: Significant at 0.10 level. ***: Significant at 0.01 level. ***: Significant at 0.05 level.

Table 3. Sims Test

1988I-2005IV

Lag Form Obs. gdp=f(M2) M2=f(gdp) Causality

1989I-2005IV

Lag Form Obs. gdp=f(M2) M2=f(gdp)

These results might be hard to use when doing theoretical interpretation but might be useful for monetary policy design.

VI. Conclusions

For this study I have collected information regarding the relationship between money and income for a period of approximately twenty years. This evidence has concerned quarterly data of real GDP and monetary aggregates as MB, M1 and M2. Furthermore, I have applied the Test of Sims in order to determine the direction of causality among these variables. Outcomes have been presented intending to contribute to the empirical research and investigation of the Money-Income nexus in Argentina, during the period corresponding to 1987 and 2005. The main conclusions of this work have been summarized already in different ways. To be concise, changes in income do not lead to changes in money supply, other things being equal. That is, there is no information in the time series of real GDP predicting changes in the monetary aggregates employed for the test. Additionally, changes in M2 appears to lead to changes in income, while MB and M1 relates with real GDP in a bidirectional way.

As I have exposed in the introduction, Argentina’s economy has a volatile business cycle. Therefore, I consider the correct understanding of the money-income general trend behavior can be useful to the monetary authorities and its policies.

However, it is important to mention that for analyzing long run relationships, this particular sample period has turned out to be challenging, due to the economic instability that Argentina has exhibited during the sample period. Henceforth, long-run values of the estimated coefficients might not correspond with the economic theory and thus, might not be helpful for theoretical interpretation.

It is also worthy to be mentioned that the research on the topic could be improved from official and non-official sources without the intervention of the National Statistics Institute. For example, considering the history of inflation in Argentina, studying the causal relationship between the money stock and the level of prices, or between the former and its impact on other real variables like agricultural and industrial production.

Even more, not only for this country but for the region itself.

References

Aguirre, H., Burdisso, T. & Federico, G., 2006, Towards an estimation of money demand with forecasting purposes, Investigaciones Económicas, Central Bank of Argentina.

Al-Jarrah, M., 1996, Money, Income, and Prices in Saudi Arabia: A cointegration and causality analyzing, Pakistan Economic and Social Review, Department of Economics, University of the Punjab.

Barro, R., J., 1977, Unanticipated Money Growth and Unemployment in the United States, The American Economic Review, American Economic Association.

Barth J., R. & Bennett J., T., 1974, The role of money in the Canadian economy: an empirical test, The Canadian Journal of Economics,Canadian Economics Association.

Bengali, K., Khan, A., H. & Sadaqat, M., 1999, Money, Income, Prices, and Causality: The Pakistani Experience, The Journal of Developing Areas.

Bernanke B., S. & Blinder A., S., 1992, The Federal Funds Rate and the Channels of Monetary Transmission, The American Economic Review, American Economic Association.

Bozoklu, S., 2012, Money, Income, and Causality: An examination for the Turkish Economy, Department of Economics, Istanbul University.

Cavallo, A., 2012, Online and Official Price Indexes: Measuring Argentina’s Inflation, Massachusetts Institute of Technology.

Coremberg, A., 2013, Measuring Argentina’s GDP growth: Just stylized facts, University of Buenos Aires.

Damill F. & Frenkel R., 2013, La economía argentina bajo los Kirchner: una historia de dos lustros, La Década Kirchnerista, Editorial Sudamericana.

Dornbusch, R. & De Pablo, J.C., 1995, The Austral Plan, National Bureau of Economic Research.

Friedman B., M. & Kuttner K., N., Money, Income, Prices, and Interest Rates, The American Economic Review, American Economic Association.

García, J. & Freyhoffer, H., 1970, La tasa efectiva de inflación en Chile entre 1961 y 1968 y el comportamiento de los agentes económicos, Instituto de Economía y Planificación, Universidad de Chile.

Gerchunoff, P. & Llach, L., 2004, El ciclo de la ilusión y el desencanto: un siglo de políticas economicas argentinas, Ariel Publications.

Granger, J., W., 1969, Investigating Causal Relations by Econometric Models and Cross-spectral Methods, Econometrica, The Econometric Society Publications.

Guilkey D., K. & Salemi M., K., 1982, Small Sample Properties of Three Tests for Granger-Causal Ordering in a Bivariate Stochastic System, The Review of Economics and Statistics, The MIT Press.

Heymann, D. & Ramos, D., 2010, An incomplete transition. Inflation and macroeconomic policies in Argentina after Convertibility, Political Economy Journal of Buenos Aires.

Hut, O., C. & Wan Tai Wai, D., 2000, Money, Output and Causality: The Case of Singapore, ASEAN Economic Bulletin, Institute of Southeast Asian Studies.

Komura, C., 1982, Money, Income, and Causality: The Japanese case, Southern Economic Journal, Southern Economic Association.

Maddison, A. & Wu H., 2008, Measuring China’s Economic Performance, World Economics. Mehra, Y., P., 1977, Money, Wages, Prices, and Causality, Journal of Political Economy, The University of Chicago Press.

Navarro, M., A. & Rayó, A., R., 1983, Precios, Causalidad y Dinero en Argentina, Revista Económica, La Plata.

Pindyck, R., S. & Rubinfeld, D., L., 1991, Econometric Models and Economic Forecasts, McGraw-Hill.

Rapoport M., 2000, Historia económica, política y social de la Argentina, 1880-2000, Editorial Macchi.

Rapoport, M., 2010, Una revisión histórica de la inflación argentina y de sus causas, Aportes de Economía Política en el Bicentenario de la Revolución de Mayo, http://www.mariorapoport.com.ar/.

Sims, C., A., 1972, Money, Income, and Causality, The American Economic Review,American Economic Association.

Stein, J., 1981, Monetarist, Keynesian, and New Classical Economics, The American Economic Review, American Economic Association.

Walsh, C., E., 2010, Monetary Theory and Policy, Third Ed., The MIT Press, pp. 1-31.

Williams, D., Goodhart, C., A., E. & Gowland D., H., 1976, Money, Income, and Causality: The U.K. Experience, The American Economic Review, American Economic Association.

Appendix

Table 4 lists the detailed profile of each regression in Table 1. The column Lag illustrates the extensions of lags used, understanding that negative lags correspond to leading values. All coefficients and their level of significance are presented as well. The bottom row restates the economic significance. The same description applies for Table 5 and Table 6. Both correspond to Table 2 and Table 3 respectively.

-8 -0.0760** 0.6283 -0.0744** 0.2860 -7 -0.0485** 0.2915 -0.0277 0.0801 -6 -0.0535** -0.0314 0.0775 -0.1236 -0.0649** -0.0492** -0.1441 -0.2227 -5 -0.0704** -0.0233 0.4930 0.2906 -0.0439 0.0018 0.7300 0.7501 -4 -0.0687** -0.0503* -0.0170 0.9812* 0.8024 0.6581 -0.0612** -0.0535* -0.0352 0.8614 0.6609 0.6354 -3 -0.0671** -0.0281 -0.0159 0.2633 0.1861 0.0267 -0.0409 -0.0069 0.0084 0.1035 0.0669 -0.0178 -2 -0.0715 -0.0624 -0.0239 -0.0256 -0.4566 -0.5553 -0.6318 -0.5634 -0.0813** -0.0778* -0.0497 -0.0610* -0.2489 -0.4451 -0.5176 -0.4162 -1 -0.1012*** -0.0626 -0.0518 -0.0558 2.9505*** 2.8600*** 2.6653*** 2.7825*** -0.0824** -0.0526 -0.0385 -0.0363 2.9258*** 2.8686 2.5954*** 2.6337*** 0 0.0005 0.0399 0.0336 0.0179 -0.2143 -0.6814 -0.5650 -0.5494 -0.0054 0.0155 0.0148 -0.0054 -0.3813 -0.7907 -0.6381 -0.6587 1 0.1609*** 0.1672*** 0.1560*** 0.1416*** -0.9980** -0.9034** -0.8870* -0.7332 0.1643*** 0.1744*** 0.1695*** 0.1731*** -0.7956 -0.7172 -0.8331 -0.7615 2 0.0302 0.0166 0.0103 0.0007 0.3148 0.3433 0.3968 0.4435 -0.0071 -0.0132 -0.0259 -0.0338 0.7070 0.7578 0.8259 0.7943 3 -0.0206 -0.0279 -0.0339 -0.0345 -1.5658*** -1.5563*** -1.5316*** -1.6169*** -0.0009 0.0035 -0.0039 0.0077 -1.0448** -1.0916** -1.1383** -1.1472** 4 0.0275 0.0226 0.0219 0.0253 0.2129 0.2410 0.2353 0.2679 -0.0055 -0.0114 -0.0169 -0.0225 0.7183 0.7509 0.8219 0.8820 5 0.0636*** 0.0641*** 0.0630*** 0.0696*** -0.2226 -0.1922 -0.0536 -0.0834 6 -0.0301 -0.0371 -0.0367* -0.0444** -0.7294** -0.7735** -0.8301** -0.8787** 7 -0.0039 -0.0024 -0.0049 -0.0008 -0.6971* -0.5874 -0.5335 -0.5524 8 0.0135 0.0127 0.0154 0.0121 -0.6786** -0.6741* -0.8194** -0.8300 F-P -0.0253 0.0824 0.1239 0.1603 0.4578 1.6739 1.8059 2.3520 -0.0301 0.0419 0.0979 0.1125 -0.0658 0.8612 0.8314 1.1519 ***: Significant at 0.01 level. ***: Significant at 0.05 level. ***: Significant at 0.10 level.

Table 4. Detailed Lag Profile of the Regressions

Lag _gdp=f(MB) 1988I-2005IV _{MB=f(gdp) gdp=f(MB)} 1987I-2005IV _MB=f(gdp)

-8 -0.0601 0.7649 -0.0613 0.3917 -7 -0.1171* 0.0409** -0.0941* -0.0020 -6 -0.0949* -0.0347 -0.1971 -0.3965 -0.1162** -0.0603 -0.2079 -0.2361 -5 -0.1037* -0.0737 0.5499 0.4738 -0.0846 -0.0558 0.5811 0.5280 -4 -0.0649 -0.0231 0.0391 1.2001*** 1.1466** 0.9653* -0.0640 -0.0209 0.0314 1.1802*** 1.0996** 1.1015** -3 -0.0506 -0.0059 -0.0157 0.1954 0.0531 -0.0184 -0.0450 0.0081 0.0014 0.0536 -0.0354 -0.1042 -2 -0.0426 -0.0371 0.0185 0.0369 -0.3203 -0.3497 -0.3215 -0.2514 -0.0615 -0.0597 -0.0069 0.0118 -0.1461 -0.1948 -0.2424 -0.1743 -1 -0.103* -0.0831 -0.0800 -0.0691 2.5137*** 2.4303*** 2.1818*** 2.2829*** -0.0816* -0.0613 -0.0565 -0.0487 2.4626*** 2.3741*** 2.1307*** 2.2206*** 0 0.0462 0.0841 0.1067* 0.0956 0.0816 -0.4230 -0.3705 -0.3699 0.0325 0.0698 0.0955 0.0813 0.0573 -0.4196 -0.3524 -0.3916 1 0.2055*** 0.2104*** 0.2097*** 0.1745*** -0.7433** -0.5615* -0.5673* -0.4386 0.2262*** 0.2329*** 0.2271*** 0.1973*** -0.7393* -0.5454 -0.6413* -0.5308 2 0.0435 0.0363 0.0257 0.0116 0.8098** 0.8166** 0.8152** 0.8116* -0.0009 -0.0064 -0.0202 -0.0289 0.9297** 0.9543** 0.979** 0.9202** 3 -0.0243 -0.0325 -0.0545* -0.0598* -0.8835*** -0.8589*** -0.8278*** -0.8968*** 0.0110 0.0052 -0.0224 -0.0262 -0.3229 -0.2964 -0.3253 -0.3575 4 0.0042 -0.0028 -0.0111 0.0012 0.1367 0.1464 0.2034 0.2765 -0.0251 -0.0313 -0.0376 -0.0279 0.4372 0.4318 0.5021 0.5567 5 0.0576** 0.0547* 0.0478 0.0429 0.0721 0.1044 0.1774 0.1400 6 -0.0312 -0.0361 -0.0304 -0.0287 -0.5364** -0.608** -0.6445** -0.6355** 7 -0.0234 -0.0254 -0.0338 -0.0332 -0.8199*** -0.7608** -0.6972** -0.7580** 8 0.0192 0.0242 0.0344* 0.0367** -0.1762 -0.1479 -0.1938 -0.1330 F-P -0.0833 0.0245 0.1192 0.1670 1.5130 3.0187 3.0363 3.6142 -0.0902 0.0121 0.1121 0.1436 1.1609 2.5450 2.4820 2.9273

Table 5. Detailed Lag Profile of the Regressions

Lag _gdp=f(M1) 1988I-2005IV _{M1=f(gdp) gdp=f(M1)} 1987I-2005IV _M1=f(gdp)

***: Significant at 0.01 level. ***: Significant at 0.05 level. ***: Significant at 0.10 level.

-8 -0.1183*** 0.4153 -0.0448 0.2317 -7 -0.0678** -0.0038 -0.0908** -0.0588 -6 -0.1061** -0.0723** 0.2523 0.0729 -0.107** 0.0371 0.117 0.0200 -5 -0.0405 0.0204 0.6391 0.5603 -0.0115 -0.0460 0.6798 0.7084 -4 -0.0701 -0.0611 -0.0270 0.4866 0.2971 0.1804 -0.0720 -0.0633 0.1226** 0.558 0.3256 0.2566 -3 0.0328 0.0821 0.0966* -0.0015 -0.3109 -0.2672 0.0663 0.1162** -0.0190 0.019 -0.2720 -0.2392 -2 -0.0181 -0.0306 -0.0011 0.0150 0.8385** 0.6753 0.6363 0.5892 -0.0283 -0.0468 -0.0368 0.0546 0.7775* 0.5641 0.5602 0.5789 -1 -0.0211 0.0095 0.0260 0.0423 1.4619*** 1.5653*** 1.3883*** 1.4025** -0.0125 0.0212 0.0359 -0.0151 1.4105*** 1.5049*** 1.3397** 1.3235** 0 -0.0253 -0.0378 -0.0268 -0.0418 -0.3981 -0.6360 -0.6373 -0.5910 -0.0245 -0.0565 -0.0439 0.0003 -0.3625 -0.6059 -0.5670 -0.5495 1 0.1798** 0.1996*** 0.2037*** 0.1965*** -0.0351 0.0076 0.0388 0.0500 0.1797*** 0.2003*** 0.2142*** -0.0523 -0.1497 -0.1510 -0.1313 -0.1020 2 0.1371** 0.135** 0.1153* 0.1097 0.4741 0.5201 0.5397 0.5782 0.0869 0.0876 0.0696 0.2107*** 0.6210 0.7303 0.7988* 0.7669* 3 0.0157 0.0140 -0.0026 -0.0133 -0.3001 -0.3445 -0.2980 -0.2979 0.0175 0.0139 0.0216 0.0778 0.0523 0.0284 0.0128 0.0024 4 -0.0319 -0.0459 -0.0594** -0.0762*** 0.1584 0.2183 0.3013 0.3302 0.0110 -0.0052 -0.0049 0.0177 0.5064 0.5774 0.6092 0.6200 5 0.0850** 0.0948** 0.0938** -0.0442 0.1707 0.2022 0.1645 0.1354 6 -0.0067 -0.0047 -0.0218 0.0583 -0.4281* -0.4901 -0.5132** -0.4958** 7 -0.0084 0.0000 -0.0097 -0.0306 -0.5043** -0.4858* -0.3770 -0.3435 8 -0.0176 -0.0261 -0.0262 -0.0021 -0.3596 -0.3628 -0.3495 -0.3494 F-P -0.2615 -0.2442 -0.1563 -0.1057 2.0031 2.3243 2.3204 2.2891 -0.3065 -0.3293 -0.2702 -0.2340 2.0966 2.5974 2.5359 2.5872 M2=f(gdp) ***: Significant at 0.01 level. ***: Significant at 0.05 level. ***: Significant at 0.10 level.

Table 6. Detailed Lag Profile of the Regressions

Lag _gdp=f(M2) 1989I-2005IV _{M2=f(gdp) gdp=f(M2)} 1988I-2005IV