1. INTRODUCTION

Export-led growth represents the application of a new consensus in developing countries. The argument focuses on the economic benefits of an open market using technological innovation, which promotes the creation of new products and competition in international markets. The success of the Southeast Asian export-oriented economies, known in the 1970s as New Industrialized Economies,¹² was the typical reference for this statement; this strategy aimed at increasing productive capacity by focusing on external markets appeared to be viable in other economies. However, with the Great Recession, contradictions started to emerge in export-led growth due to the contraction of demand in developed countries. This had a negative impact on the health of emerging markets due to their high dependence on this growth engine (^{Palley, 2012, p. 142}).

Meeting external demand can be seen as a way to produce new goods in sectors with strong productivity and growth (productive efficiency gains). This growth is not only a result of the traditional convergence hypothesis (^{Barro and Sala-I Martin, 1991} and ¹⁹⁹²; ^{Quah, 1993} and ¹⁹⁹⁶; ^{Aghion and Howitt, 2009}) as it is closely related to trade and trade openness. This gives way to the need to define strategies for transferring capital and human resources to sectors geared toward external markets, resulting in structural imbalances (^{Gaffard and Saraceno, 2008, pp. 1062-1063}). As such, investment redistribution among sectors is dual in nature, with the destruction of productive capacity in certain spaces and its creation in others.

Resolving these imbalances necessitates the feasibility of mobilizing financial resources to meet the demand for credit that such a redirection of productive capacity entails. The key point is that once the neo-export industrial policy has been defined, the viability of resource displacement it implies must be based on an ad hoc monetary and credit policy. In this case, this creates the problem of restricting financing of an export-led growth model. In the end, if the strategy is successful, the economy will have been led to a new equilibrium, after the rupturing and restructuring of the previous status quo, which means the reallocation of investment and an efficient credit system that supports the new growth model’s financing needs.

The neo-export strategy’s achieving a soft landing depends largely on the balances available for investment in conjunction with an efficient credit system to mobilize them towards new sectors of financing demand. Here is where the specific conditions of the analyzed space become relevant, i.e., those of an economy with a medium level of industrialization. This means the analysis of internal financing. The response to this, for five decades, has been the suggested ineffectiveness of the proposal of perfect markets from conventional theories on monetary and financial processes with a single interest rate or markets structured by term rates, when what was in effect was actually a fragmented market of real interest rates. Or they even worked off of a perfect substitution between real money balances and physical capital and, where convenient, used a complementarity approach to take on these variables (^{McKinnon, 1973, p. 3}).

The underlying issue is the importance of theorizing with basic assumptions which are “better suited to explaining the relationship between monetary processes and capital accumulation in the underdeveloped world” (^{McKinnon, 1973, p. 3}). Another aspect of the period’s strong tariff protectionism framework was proposing the convenience of opening the economy pari passu with the capital market’s liberalization to absorb imports’ external financing, which would lead to the dismantling of the protectionist scheme.

This was all within the framework of a financial repression understood as the weak depth of the banking system and great segmentation of access to credit, with its majority in a small number of large companies. This strongly limits the resources to the rest of an economy fragmented into a multitude of micro units with inefficient self-financing practices. As a result, one expects Foreign Direct Investment (FDI) and Bank Credit to the Industrial Sector (BCIS)¹³ to be factors in the productive sector’s expansion.

This article is structured in five sections, including the introduction: the second analyzes the link between trade openness and financial restrictions by compiling the seminal ideas on the subject of ^{McKinnon (1973}) and ^{Gurley and Shaw (1960}); the third takes on the characterization and measurement of financial constraints and their impact on the Mexican economy; the fourth section presents the methodology used to study the dynamics between the financial sector and the real sector; the fifth sets forth the results obtained for the Mexican economy in the period of 1995-2020. Finally, the conclusions are presented. These suggest that external financing and BCIS did not contribute to production’s growth level, as they do not show any relationship which would allow one to infer any causality from the financial sector to the real sector in the period studied.

2. TRADE OPENNESS AND FINANCIAL CONSTRAINTS

Trade openness is a type of structural change and can therefore integrate itself with technical progress as it entails the destruction of productive capacity and the construction of something new to replace it (^{Gaffard and Saraceno, 2008, p. 1063}). In other words, the coordination between the phases of building productive capacity and its utilization and between investment and consumption which characterizes the equilibrium are altered during the process of structural change. If the product that an economy is capable of generating depends on labor and the size of the capital stock, given the state of production techniques, if labor is omitted, the net production capacity increases in step with the increase in capital stock destined for investment. However, in order to make an efficient allocation of investment, the ad hoc financial systems which make this possible are necessary (^{Gurley and Shaw, 1960, p. 47}).

The result is that the restructuring of productive sectors for neo-export specialization will depend largely on trade openness and the free flow of capital. Restructuring can take on different forms and “is likely to have distributional impacts-both in the short term as a consequence of adjustment costs and in the long term as a result of permanent changes in relative [productive] factor demands,” as ^{Rodrik (1998, p. 6)} points out. In other words, the imbalances of the new strategy are the flip side of productive efficiency gains, which the author dramatically emphasizes with “No pain, no gain!”

3. SOME STYLIZED FACTS ON THE REAL AND FINANCIAL SECTORS IN MEXICO

As ^{Venegas et al.explain (2009, p. 256)}, the financial reforms in Mexico took place in the late 1970s, which coincided with the creation of the Mexican IPC Stock Index,¹⁸ in parallel with the country’s entry into the market of negotiable government bonds, liberalization of banks, a reduction in the requirements placed on banks’ reserves, the non-intervention on certain interest rates, the opening of the capital account, as well as the devaluation of the peso. All of the above made it possible to replace national intermediation in deposits with international operations between the late 1970s and early 1980s.

The aims were to bring about less friction in the workings of the capital and money markets, achieve greater efficiency and competitiveness and favor greater integration with international financial markets. The deregulation process therefore focused on the gradual elimination of credit controls and restrictions imposed on the financial sector, thus promoting the entry of foreign portfolio investment. This decision sought to develop the public and private sectors by promoting new means of financing (^{Cabello, 1999, p. 216}).

A consultation of references in historical order after this moment will yield different results. ^{Bandiera et al. (2000}), for example, upon finding a long-term negative relationship between savings and financial liberalization in Mexico, point out that financial repression had a positive but small effect on the financial system’s development by driving savings’ interest rates. Along the same lines, ^{Venegas et al. (2009, p. 280)} found empirical evidence a decade later suggesting that financial development had a positive, albeit small, influence on economic growth in Mexico in the period of 1961-2007 and that financial repression tends to slow down growth. This is due to this effect being inversely proportionate to financial development. However, they found no short-term effects between repression, financial development and the level of economic activity.

According to ^{Kindelberger (2012, pp. 233-295)}, this supports an important factor with regards to most economies worldwide. In the last hundred years the increase in the credit supply was generated by the banks and resulted in a multitude of new financial instruments favoring a systematic development for reducing transaction costs, such as maintaining liquidity and monetary balances.

Along the same lines, ^{De la Cruz and Alcántara (2011, pp. 32-33)} showed that the total credit granted by commercial banks tends to increase the level of production; however, by disaggregating the destination of credit by sector, only that destined for services and consumption has a positive effect on Mexico’s economic activity. They also find a bidirectional causal relationship between consumer credit and production levels as a close interaction of banking processes encouraging aggregate demand prevails, in spite of the financial sphere having a weak relationship with the country’s generation of added value. Meanwhile, ^{Clavellina (2013, pp. 31-32)} offers up empirical evidence that suggests a negative causality of total credit in the determination of product. This could be explained by the low quality of financial intermediation which in turn triggers non-optimal results. These then affect the Mexican economy’s performance by directing a large part of its resources to unproductive activities such as consumption and the public sector.

By estimating equations for cointegration and for Mexico’s common cycle, ^{Portal and Feitó (2014, p. 92)} were able to determine the presence of a positive long-term relationship between commercial banking’s sectoral loans and activity levels in the secondary and tertiary sector. This suggests that commercial banks’ credit allocation to sectors of economic activity positively influence these sectors’ growth for the period of 1995-2012.

Furthermore, ^{Landa (2019, p. 50)} identified in the Mexican economy an inverse relationship regarding the scope commercial banking, an issue arising from highly targeted credit and an insufficient savings rate; Landa also identified earnings which are driven by the stock market. This suggests that the satisfactory levels of liquidity provided by the stock market’s segment is related to financial stability and production’s growth rate.

In spite of the financial market’s liberalization policies in the late 1980s, such as the elimination of legal reserves, non-selectivity in credit policies, liberalization of active and passive interest rates, international banking’s operating in the country, authorization of financial groups and the privatization of commercial banking, among others, in 1994 ^{Venegas et al. (2009, pp. 256-258)} verified that new financial-business groups in larger segments controlled 50% of the assets. Today, the seven largest banks control 78.34% of the assets in the system. Of these, five are foreign and account for 62.85% of the sector.¹⁹

In terms of the banking system’s depth,²⁰ Mexico is known as a country lagging behind other Latin American countries: 27% of its population has a bank account and approximately 90% of daily transactions in the country are made in cash (^{Herrera, 2019, March 12}); credit to the non-financial private sector is around 42% of the GDP, well below the average of the five largest countries in Latin America (72% of GDP), and even further from the 143% average of emerging countries (^{Savedra, 2019, August 12}). Specifically, in 2019 in Mexico’s non-financial private sector, the ratio of total financing to GDP was close to 40%, while in Chile it was over 170% (^{Bank of Mexico} [BdeM]²¹ , 2019).

^{Téllez and Venegas (2019}) have likewise found, based on different panel data specifications, that the determinants of Mexico’s financial system’s intraregional depth are as follows: the current state of the law or institutions; banking competition and regulation; formal employment, as well as the propensity to save and financial education. Therefore, the low degree of financial intermediation is conditioned, as ^{Herman and Klemm (2017}) point out, on the sector’s structural aspects. While there has been credit growth in the last decade, driven by increased supply, there appears to be some lag in credit cycles relative to business cycles.

^{Fukuda (2019}), with the use of Vector Error Correction Models (VECM), similarly finds empirical evidence that suggests globalization and financial deepening in Mexico, understood as the proportion of internal credit granted by banks to the private sector, have a negative influence on economic growth. According to ^{López and Basilio (2016, pp. 227-228)}, the deregulation of the financial system caused a contraction in credit destined for productive activities and revealed commercial banking’s speculative and rentier nature. The latter in turn obstructs intermediation between productive activities and financing.

In summary, in theoretical terms it is assumed that the dynamic of economic activity levels is that BCIS and FDI should not behave differently in the economy’s periods of recession or recovery, given that allocation in the financial sector is efficient and the resources channeled from abroad encourage production level growth. However, in the case of economies such as Mexico, the dynamics of bank credit have a structural restriction as they are mainly aimed at import- and export-related activities. In addition, the aforementioned data reflects a situation of financial repression and fragmentation which liberalization policies should have done away with after operating for four decades.

With the aim of answering these questions, we will first study the dynamics between the financial sector and the real sector, using the Markov methodology for regime change to analyze the short- and long-term relationships exhibited by BCIS, FDI and the production level in terms of which phase of the business cycle they find themselves in and the probability of remaining there. Second, a VECM will be used to study how these variables react to long-term equilibrium.

4. METHODOLOGY

Two models were used to capture the dynamics of the activity level of the economy, FDI and BCIS: the first is a Markov-Switching model, with which the average duration of recessions and recoveries is obtained to determine if the aforementioned variables can be synchronized within the period studied.

Along these lines, ^{Hamilton (1994, pp. 677-684)} points out that if a process changed in the past, it could clearly change again in the future, and that this should be taken into account when drawing up a forecast. Moreover, regime change should not be considered the result of a perfectly foreseeable deterministic event. Quite the contrary, regime change is itself a random variable. Therefore, a complete time series model would include a description of the law of probability which governs the nature of change μ₁, μ₂.

These observations suggest that the process could be considered to be influenced by an unobserved random variable st*, which will be called the state or regime in which the process was on date t. If st*=1 , then the process is in regime 1, while st*. = 2 means that the process is in regime 2, therefore it can be expressed as:

Where μst* can be μ₁ when st* = 1 and μ₂ when st* = 2. This makes a description of the time series process for the unobserved variable st* necessary; as st* only takes discrete values (in this case st* is either 1 or 2). The proposed model for a discrete value random variable is a Markov chain where s _t is assumed to be a random variable that can only assume an integer value 1, 2, ...., N. Assuming that the probability of s _t being equal to the specific value j depends on the past only via the most recent value s _t-1:

According to ^{Mejía (2000, pp. 392-395)}, the state s _t is not directly observed, therefore the probability that the process is in state 1 on date t is conditional on the data observed up to date t. This algorithm can be considered a formalization of the statistical identification of inflection points in a time series with the filter and, specifically, the softest probabilities used to identify periods of contraction and expansion.

The second model is a VECM and, according to ^{Pesaran (2015}), it allows one to determine the variables’ trajectory based on their long-term equilibrium if there are different dynamic specifications to represent these equilibria, as is the case of the autoregressive distributed lag model (ARDL). This model will be effective, regardless of whether the regressors are integrated to the order of I(0) or I(1), or are co-integrated with each other, meaning that the appropriate lag for each variable can be included (^{Pesaran et al. , 2001}). As the ARDL model is a single equation dependent on its own lags and those of the explanatory variables, which need to at least be weakly exogenous (^{Enders, 2014}), we opted for the VECM as the latter allows one to study the dynamics between the three variables as it is a multiple equation system.

Now, if the variables are co-integrated (that is, they have a long-term relationship), they will share the same stochastic trends and therefore cannot deviate too much. This is how the error correction representation reacts to long-term deviation from the equilibrium (^{Enders, 2014}). We followed the proposal of ^{Kilian and Lütkepohl (2017}) in incorporating the concept of cointegration, represented in its simplified form in the VECM equation as:

Where y _t-1 is the vector of the variables in levels (not stationary) and Π is an r rank non-singular matrix. The range of Π is therefore referred to as the cointegration range of process y _t . If this matrix r rank K × K can be broken down as a product of two matrices, then the VECM equation (^{Kilian and Lütkepohl, 2017}) can be rewritten as follows:

Where ß' is the cointegration matrix and α is the adjustment rate matrix, which makes the error correction term explicit (aß'y _t _ ₁ ). In other words, if it is possible to decompose the Πγ _t - ₁ matrix, the use of VECM as explained by ^{Loría (2007, p. 275)} allows one to combine “economic theory [by finding stable long-term relationships established by the theory] and the statistical adjustment of the imbalance that can exist in the short-term.”

5. RESULTS

This section analyzes the main empirical regularities which characterize the Mexican economic cycle, BCIS and FDI with quarterly data in the period of 1995-2020²² which were seasonally adjusted and then adjusted to 2013 prices; in the case of FDI, the series was seasonally adjusted and an index based on 2013 was constructed. The cyclical patterns of GDP and BCIS were evaluated using ^{Hamilton’s (2018}) proposed methodology; meanwhile, in order to ascertain whether the variables described above can be synchronized within recessions and recoveries, a Markov-switching probability model was elaborated with the methodology Hamilton set forth (1994) and used to obtain each phase’s average duration. Furthermore, a VECM was used to determine whether BCIS, FDI and GDP have any long-term relationship.

After determining the recovery and recession phases of GDP and BCIS, they were used to study the amplitude and duration of each (see Table 1). When GDP was in a recessionary period, it presented an average duration of 8.99 quarters, with a cyclical component average of -.05523 and a standard deviation of 0.04020; BCIS showed an average duration of 24.90 with an average of -0.1355 and a standard deviation of 0.12721; FDI had an average duration of 1.899 with an average of -0.6276 and a standard deviation of 0.2597. In recovery phases, GDP presented an average duration of 25.93 quarters and an average of 0.01921; FDI had an average duration of 28.28 with an average of 0.07562 and BCIS showed an average duration of 22.52 and an average component of 0.1384.

Based on the empirical regularities found, asymmetry can be inferred in the two phases that characterized the business cycle as none of three series presented any symmetry in duration or amplitude during contractionary phases. In absolute terms, GDP showed an average for the cyclical component lower than the BCIS and FDI averages, a situation which repeated itself during the recovery periods. Empirical evidence implies a different dynamic and, to verify this, Markov-chains were used to model phases in order to identify the probability of each of the three-time series staying in a state of recession or recovery.

Figure 1 shows the GDP’s and BCIS’s probability of transitioning from the economic cycle; FDI is excluded as it is in recovery for most of the study period. BCIS had three phases of recession: the first, of a prolonged magnitude, in periods of recession it could be assumed to have been contemporaneous to the economic cycle and was countercyclical during episodes of recovery; in the second and third phases it was contemporaneous to the Mexican economic cycle; however, these results do not allow any conjecture to be made since so much of the period analyzed is in recession.

Source: Created by the authors.

Figure 1 Cyclical component of GDP and probability to transition 1997-2020. 

The evidence for BCIS and FDI reports the presence of asymmetric behaviors in each of the phases of the economic cycle for BCIS and FDI with respect to their magnitude and duration. This was corroborated with the values obtained from the final matrices. The probability of remaining in the recovery phase or transitioning to it in the next period is, in the case of GDP 88.52%, 82.04% for BCIS and 84.666% for FDI, thereby demonstrating a different dynamic between GDP, BCIS and FDI in the recovery phases of the economic cycle. In phases of recession, the probability of remaining in it in the following period was 67.42% for GDP, while FDI has a much lower probability at 7.35% and 83.83% for BCIS. Empirical evidence demonstrates synchronization between the national economy, BCIS and FDI in phases of recovery or recession to be non-existent.

As no empirical evidence was found to suggest the existence of any synchronization between the variables analyzed, we proceeded to estimate a VECM to try to capture the effects of BCIS and FDI on the economic activity level, which would imply a positive relationship between the two. Thus, the Augmented Dickey-Fuller (ADF), Phillips-Perron (PP) and Kwiatkowski, Phillips, Schmidt and Shin (KPSS) tests were used to corroborate the series’ order of integration.

As the three series are stationary²³ in their first difference, we proceed to choose the optimal number of lags so we may present the VECM model to be estimated. According to the information criteria of Akaike, Hannan-Quinn and the final error prediction, we concluded that two lags should be included in the model.²⁴

We then verified the existence of at least one cointegration equation using the Johansen test. The eigenvalues and the graphs themselves confirm the existence of at least one cointegrating equation. The results of this test are presented in Table 2.

The model estimated using the information obtained from the optimal number of lags, the order of integration of the time series and the Johansen test are as follows:

Where Y _i is the GDP logarithm, Cr _i is the BCIS logarithm and IE _i is the FDI index logarithm. The results of the VECM estimation (1) are presented in Tables 3, 4 and 5. In the case of the GDP equation, the coefficient of the first FDI and BCIS lag are negative without being statistically significant; only the lag itself is statistically significant. In the case of the FDI equation, only the FDI lag turned out to be significant; as for the BCIS equation, the GDP lag is significant as is the bank credit lag.

The model is stable, and furthermore, the serial correlation and heteroscedasticity tests show that the model does not present any problems. The stationarity of the errors in each equation was also checked and the residuals of the three equations were found to be stationary with trend and constant (see Tables A3 and A4 of the Statistical Appendix). The impulse response functions of the estimated model are presented in Figure 2.

Source: Created by the authors.

Figure 2 VECM impulse response functions. 

Empirical evidence obtained through impulse response functions suggests the BCIS’s two standard deviation shock’s depressive effect on production level growth, as well as its contractionary effect on FDI. On the other hand, the short-term recessionary effect of FDI on GDP and BCIS is evident. Meanwhile, GDP shocks have a contractionary effect on FDI and BCIS in the short-term. In turn, Table 6 presents the GDP variance decomposition, from which we can conclude that in period 10 GDP variability explains itself at 85.53%, meanwhile FDI is at 7.91% and BCIS at 6.54%.

In addition to the above, the Granger causality test was performed to determine whether the events in past periods could provide information on the events that are currently transpiring (see Table 7). This is how the results obtained allow us to reject the null hypothesis, which itself establishes that the GDP growth rate does cause, in the Granger test, the BCIS growth rate. At the same time, the BCIS growth rate causes the FDI growth rate. However, no causal relationship of FDI to GDP or BCIS to GDP was found as the test statistic obtained is not significant.

6. CONCLUSIONS

Financial liberalization and trade openness were implemented, in accordance with the Washington Consensus, as part of a structural adjustment meant to overcome the 1980s’ Latin American Debt Crisis (^{Williamson, 1991}).

With this framework, an estimate was made for the relationship between the BCIS, FDI and Mexico’s economic activity during the period of 1995-2020 using two proposals. The first was a spatial representation of state using a Markov-Switching model which showed that FDI in Mexico was in recovery for most of the period studied. In the case of BCIS, it showed three phases of recession: the first has a prolonged amplitude which could be assumed as contemporaneous to the economic cycle in periods of recession and countercyclical in periods of recovery, while in the second and third phases it would be considered as contemporaneous with Mexico’s economic cycle. However, these results did not allow us to make any conjecture on the series’ behavior since so much of the period analyzed was in recession.

The second proposal used a VECM to determine the dynamics between GDP, BCIS and FDI. From the Johansen test it was concluded that there is a long-term relationship between the variables. The estimates made suggest, through impulse response functions, that shocks in FDI and BCIS can cause a contractionary dynamic in economic activity levels; however, the variance decomposition showed that BCIS and FDI explain about 14.50% of GDP variability in the tenth period.

Furthermore, the Granger causality test suggests a unidirectional behavior going from GDP to BCIS or FDI as there is no causality from BCIS or FDI towards GDP. These results suggest that BCIS or FDI do not encourage the productive sector’s growth after the transition which the Mexican economy underwent in the early 1990s. This highlights the lack of credit depth as it is linked to large companies’ export activities, when in the country the overwhelming majority of economic units are micro, small and medium enterprises that find it hard to meet bank credit requirements.

The paradox is that, although the liberalization process that took place in the early 1980s met most of the established theoretical guidelines, its implementation did not fully permeate the economic system as the financial fragmentation and repression to be overcome keenly survived. Since we saw no relationship which allows us to infer any causality from the financial to the real sector in the period studied, this means that the empirical evidence found does not suggest external financing and BCIS contributed to the growth of production levels.