Introduction

Once instability is understood as a theoretical possibility, then we are in a position to design appropriate interventions to constrain it

Hyman Minsky

For the classical school of thought, the exchange rate is just another price determined by the foreign exchange market, which reflects trade with the outside world and its flexibility allows to absorb the shocks that every economy tends to face. Therefore, in this approach, it is essentially an endogenous variable and it does not act as an economic policy instrument. Carrière‐Swallow et al. [1], based on a sample of 196 countries for the period 1990–2016, provide strong evidence that the exchange rate plays a key role of a shock absorber by reducing the response of output to changes in the terms of trade to which emerging countries are more sensitive.

Conversely, for development macroeconomics the active management of the exchange rate has a strategic role in promoting technical progress, investment and growth (Rapetti et al. [2, 3], Razmi et al. [4], Rodrik [5], Bresser-Pereira [6], Capriata and Flauzino [7], Gala [8], Goda and Priewe [9]). However, it has also been advised on the potential pitfalls and volatility of the nominal exchange rate and, thus, inflationary pressures coming from sistematic depreciations, [5, 10].

Since the 1995 Mexican Tequila Crisis, the vast majority of developing countries have avoided huge q depreciations because of their high inflationary costs and the deterioration in the living standards of the poorest share of the population.

For the New Developmentalism theory, managing the depreciated q is an important industrial policy tool that accelerates growth of emerging countries; China and India, for example, have experienced very rapid growth over the past decades. This is because they consider that the real exchange rate is not just another price of the economic system, but rather a determining variable of industrial policy and of economic development because it fulfills the Marshall-Lerner condition and avoids the Dutch Disease. This line of theory sustains that exchange rate appreciations have long-term de-industrializing and impoverishing effects, because it is believed that, insofar as inflation in emerging countries (particularly in Latin America) is higher than in the developed world, growth processes lead to exchange rate appreciations, which reduce demand for domestic goods in favor of goods of foreign origin and increase the use of imported capital against employment. So, if the exchange rate is used as an "inflation anchor" rather than a tool for growth, exchange rate appreciations that will occur will be similar to large currency inflows that cause Dutch Disease-associated effects and significant reductions in growth, Bresser-Pereira [6].

Thus, in this approach, the active management of q has growth effects due to the fact that, in the long run, there are redistributive gains, because, although inflation caused by depreciations in the short run deteriorates distribution, growth effects that it should generate outweigh the concentrating effects, Bresser-Pereira [6].

Due to the financial integration, there is an increasing volatility in the nominal exchange rate (E) and in q, so we seek to explore its main commercial and financial short- and long-run determinants. We claim that both factors are increasingly important to determine its short-terms movements and its long-run trajectory. That is why ee leave aside such macroeconomic determinants as fiscal and current account deficits.

We identify five main determinants: terms of trade (TOT), tradable to non-tradable price differentials (TNT, also known as the Balassa-Samuelson effect, BSE), interest rates differentials (ID), forward exchange rate (f) and risk premium (ρ), proxied by VIX international volatility that reflects the reward investors ask to hold a risky currency. While the former three have been extensively tested in empirical works (Clark and MacDonald [11], Ibarra [12], López Villavicencio and Raymond [13]), f and ρ have been left out of the picture. The joint consideration of these five factors is not common in the literature.

We argue that, while in the long run PPP theory tends to explain the path of the exchange rate, in view of the growing commercial and financial integration of the Mexican economy, financial variables have played an increasingly crucial role in determining movements in the nominal and real side of the currency. Foreign exchange markets, which operate with extremely high liquidity on a 24-hour basis and include spot and financial derivatives transactions, have gained immense importance as they have become the backbone of international financial wealth. For some years now, the Mexican peso has been used as a hedging currency for risks in currencies associated with lower liquidity or limited trading hours. According to BIS [14], the Mexican peso is the third most traded currency among emerging economies after the renminbi and the rupee. Its average daily trading volume during 2022 was US$114 billion dollars, well above US$1.6, US$0.16 and US$0.088 billion in exports, remittances and foreign direct investment, respectively.

In addition, it should be noted that 82% of transactions with Mexican pesos were carried out outside Mexico, 37% were in cash, and the rest (67%) were in financial derivatives, BIS [14]. All of this clearly indicates that the determination of the exchange rate in recent years has been the result of the appetite for the Mexican peso among the participants of the global foreign exchange market. On the other hand, we cannot underestimate the importance of the US monetary policy, which has generated massive and multiple financial and real effects around the world in recent decades.

Despite this major importance in the operation of foreign exchange markets, our econometric results show a significant, although still marginal, role of these variables.

While inflation targeting measures have been successful in maintaining price stability–compared to historical high inflation experienced in the 80’s–the floating exchange rate has increasingly been influenced by adjustments in international portfolios, especially in the aftermath of the Great Financial Crisis (GFC, 2008–2009).

In the early 2000s, the US federal funds rate decreased drastically (it plunged from 6.3% in 2000 to 1.1% in 2003), and both liquidity and sudden capital movements increased rapidly worldwide. Just after the bankruptcy of Lehman Brothers (September 15, 2008), with the first round of the so-called Quantitative Easing (QE) in November 2008, the Federal Reserve initiated a massive financial assets purchase program seeking to stabilize the economy and financial markets. The main purpose was to avoid at all cost a deflation like that in 1929–1932.

By the end of 2009, many other central banks adopted similar policies and the global macro-financial environment registered unprecedented levels of liquidity. As a result, emerging countries’ assets–such as the case of Mexico–became a target for the allocation of extra liquidity. According to Banco de México and INEGI data, non-residents holding government domestic securities amounted roughly to 1% of GDP in 2007 and this figure further jumped to 9% by the end of 2021 [15, 16]. Extraordinary monetary policies came to an end by 2012 and hawkish policies emerged and were followed by two years of adjustments in raw material prices, especially oil prices (West Texas Intermediate), which dropped from US$105 per barrel in June 2014 to US$30 in February 2016. The fall in oil prices did not attain the pre-2014 levels until March 2022 (US$108 per barrel [17]).

Our research begins in 2001.01, when the inflation targeting regime was adopted in Mexico, [18]. This allows us to capture an important phase of inflation reduction and control, and, therefore, a phase of homogeneous monetary policy. In this sense, while between 1995 and 2000 average inflation was 22.54% (with a standard deviation of 15.77) and the reference interest rate was 26.84% (with a standard deviation of 11.87), between 2001 and 2021 these figures attained 4.13% (1.18) and 6.03% (2.13), respectively.

Due to the increasing importance of international finance, we prove that q responds not only to the usual set of commercial fundamentals, but also to the unprecedented financial logic–i.e. expectations (proxied by f) and financial volatility (ρ, risk premium), proxied by VIX, which correspond to rational expectations, where f predicts the future movement in the nominal exchange rate MXN/USD. This theory has yielded mixed results in empirical studies [19]. According to Bush and López Noria [20], the currency risk premium for Mexico is strongly affected by international volatility.

Since we face a different order of integration among the five determinants of q, our ARDL Pesaran & Shin [21, 22] estimation finds a cointegrating vector only for 2009.01–2022.12, corresponding to QE as well as abundant international liquidity. Prior to 2009 and for the whole sample (2001.01–2022.12), our combination of commercial and financial determinants of q does not cointegrate. According to our econometric results, we suggest that TNT is the most important determinant of q in both the short and long run (-0.27 and -1.11, respectively), followed by TOT (-0.26 and -0.51). Besides, we find that our two financial determinants can alter q not only in the short run (as one would expect), but also in the long run, but sorpresilly in a marginal manner (0.0155 and 0.063 for f, and 0.009 and 0.036 for ρ, respectively). We also find high persistence effects of q on itself (0.75) over the four previous quarters.

In summary, using a combination of trade and financial variables, finding cointegration only for the period 2009.01–2022.12 and giving each factor its relative importance in the short and long terms are the main contributions of this work.

Section two reviews literature. Section three discusses theoretical issues of exchange rate determination. Section four depicts stylized facts (materials and methods) of the determinants of the MX peso–US dollar for 2001.01–2022.12. Section five addresses the econometric issues. Section six punctually presents the main results, and section seven concludes and gives further comments.

Literature review

The vast majority of literature addresses the determinants of the real exchange rate from the perspective of macroeconomic, trade or financial aggregates, without mixing them; so, we have decided to take an eclectic approach and make a joint analysis of the importance of the latter two for Mexico in the period 2001.01–2022.12.

Traditionally, macroeconomic fundamentals were considered to be the sole (main) determinants of q. The examples of models that test them are Clark & MacDonald’s [11] Behavioral Equilibrium Exchange Rate (BEER) model, Stein’s [23] Natural Real Exchange Rate (NATREX), and Williamson’s [24] Fundamental Equilibrium Exchange Rates (FEER).

For a detailed theoretical explanation of these models, we highly recommend MacDonald, [25].

From the theoretical standpoint, the family of models that rely on macroeconomic fundamentals accepts the concept of exchange rate equilibrium as the level that equilibrates the balance of payments. However, they accept that in the short run there may be some degree of misalignment (volatility) of the nominal exchange rate (E) due to shocks in the macroeconomic fundamentals and financial shocks. In this sense Alagidede & Ibrahim [26] argue that for small economies characterized by institutional failures (such as Ghana and Sub-Saharan Africa) but with a possibility to extend to emerging economies in the short run, about three-quarters of real exchange rate shocks are attributed to the real exchange rate itself, and the rest, to the factors such as government spending, money supply, foreign direct investment flows, terms of trade and output shocks. Market failures can occur even in large emerging economies such as China. Chen et al. [27] argue that uncertainty regarding economic policy (associated with government regulation and decisions) is a definitive factor in explaining exchange rate volatility.

In the long run, however, they claim the equilibrium should be restored through macroeconomic (inner) adjustments and policies. According to this family of models, in the long run macroeconomic fundamentals drive E–and, consequently, q–towards the equilibrium and the effect of financial elements runs only in the short term.

As we have previously stated, in this article we leave the above approach aside and focus exclusively on trade and financial aspects, which have become increasingly important, particularly in emerging countries.

The exchange rate affects commodity prices and the welfare of the poorest share of the population in emerging economies where exchange rate volatility is higher than in developed countries, Nor et al. [28]. Yépez & Dzikpe [29] state that in emerging countries the relative prices of tradable goods account for most of the volatility observed in the exchange rate and that world commodity prices explain around 30% of their fluctuations.

For 1980–2014 and using a sample of 17 Latin American countries, Ramírez [30] suggests that increases in productivity cause an appreciation of the real exchange rate. However, the pass-through is not complete since a 1.0% increase in productivity causes an appreciation of 0.73% in q.

Catalán [31] uses an ARDL model applied to Mexico to confirm that the Balassa-Samuelson Effect is validated for 1994.I-2018.IV, which in author’s terms rejects the monetary model.

Kassouri and Altıntaş [32] argue that in Africa the exchange rate response to terms of trade’ shocks is asymmetric, because in the long run the response is stronger to positive shocks than to the negative ones.

In contrast, the portfolio approach literature argues that exchange rate fluctuations are determined by the portfolio rebalancing which, in turn, is influenced by risk premium, uncovered interest parity and investment decisions, Aliber [33] and Dooley and Isard [34]. In the same vein, Brainard and Tobin [35], Branson [36], Tobin [37] and Backus et al. [38] consider the exchange rate from the point of view of assets portfolio optimization, the so-called portfolio approach. This approach considers that financial assets differ in their country-specific risk premiums, so they are not perfectly substitutable. Then, E is determined by the desire to diversify portfolio according to risk preferences and expectations. In other words, the appetite of financial asset holders will be crucial to the supply and demand of currencies and, therefore, to the value of E in the short term.

In the opposite vein, Bostan et al. [39] argue that, at least in the case of Romania, the exchange rate serves as a reference for financial markets because it reflects competitiveness and its evolution is fundamental since it is influenced by an array of factors and, at the same time, its disruptions affect external competitiveness, the real economy and financial markets.

With respect to financial determinants, Cardozo et al. [40] mention for the case of Colombia the importance of the forward exchange rate and, by estimating EGARCH and VARX-MGARCH models, they find evidence that changes in the real sector impact the forward exchange rate. However, that effect has not been stable between 2008 and 2015.

Cáceres [41] states that the Mexican peso and the Peruvian Sol are risky currencies and are commonly used to do carry trade, a financial strategy that consists in financing with currencies associated with low interest rates in order to invest in high yield currencies associated with the risk premium. This type of strategies can lead to abrupt depreciations. Capasso et al. [42] argue that there are important links between the exchange rate and monetary policy.

Materials and methods (stylized facts)

In order to find crucial regularities to be econometrically tested in the next section, in this section we briefly analyze the monthly evolution of the five main determinants of q here considered (January 2001 to December 2022).

Fig 1 reports the evolution of q–calculated by Banco de México [15]–and its determinants. An increase in q represents depreciation and the opposite applies. According to our main hypothesis, the GFC marked two different paths, reason why we split the whole sample in two: 2001.01–2008.12 and 2009.01–2022.12. Ups and downs are more frequent and respond to major financial events. The normalization of monetary policy in the US, the outcome of the US presidential elections (Trump effect) as well as commercial tensions and NAFTA renegotiations provoked further depreciation and volatility (2014–2017). It is striking that, if we eliminate the strong Covid-related depreciation in 2020, the Mexican economy has exhibited a clear trajectory of exchange rate appreciation (until April 2023) that is definitely explained by the increase in id despite the strong deterioration of tot and the advance of tnt. Said phenomenon of a strong appreciation over the past three years would support our main hypothesis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Determinants of q, 2001M01-2022M12.

Note: own calculation, based on Banco de México [15], Bank for International Settlements [56], Bloomberg [44], BIS [14], FRED [17] and INEGI [16]. These sources are used for econometric estimates.

https://doi.org/10.1371/journal.pone.0286331.g001

The variables show a unit root behavior with the exception of id and ρ that behave as I(0), Table A1 in S1 Appendix. That is why the ARDL [21, 22] procedure is convenient.

Fig 2 clearly illustrates the strong relationship between id and ρ, which confirms the covered interest parity theory in explaining currency fluctuations, and the need to state a correct specification by eliminating id in the econometric estimation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Interest rate differentials and one-year forward rate, 2001.01–2022.12.

Note: In order to contrast visually, both series were normalized by their means and standard deviations.

https://doi.org/10.1371/journal.pone.0286331.g002

Econometric issues

ARDL models [21, 22] are linear time series models in which both dependent and independent variables are related historically and contemporaneously. These models have gained popularity in recent years as a method of examining robust cointegrating relationships between variables of different orders of integration. They are highly informative since they offer: a) an intertemporal dynamic estimation (given that they are also closely related to Vector Autoregressions), b) derivation of the long-run (equilibrium) relationship, and c) the conditional error correction.

Since we have a different order of integration and to avoid spurious regressions, by applying the Bound F-test, we can claim that the whole set of variables is cointegrated only for 2009.01–2022.12, Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Cointegration (Bound F-test) tests, 2001.12–2022.12.

https://doi.org/10.1371/journal.pone.0286331.t001

According to our approach, if y_t is the dependent variable (q) and x_j,t−z are explanatory variables, a general ARDL model can be represented as follows: (9) Where a₀, a₁, φ_i, and β_j,z are the coefficients associated to the constant, to a linear trend, to the lags of y_t and to the lags of regressors x_j,t−z, respectively for j = 1, …, k; ϵ_t are Gaussian innovations.

The model that minimizes the AIC is the ARDL (4,2,2,0,0,1,0). Tables 2 and 3 report short-run estimates. Since all variables are expressed in their logarithms, they evaluate constant elasticities.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Short-run deteminants of q, 2009.01–2022.12.

https://doi.org/10.1371/journal.pone.0286331.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Overall short-run estimates.

https://doi.org/10.1371/journal.pone.0286331.t003

Results

As expected, when both f and id are included, strong collinearity problems emerge, which caused id, q_t-3, q_t-4 and tnt_t-2 to become non-significant, hence, we eliminate id. The estimation confirms that there are important autoregressive processes that in all cases are significant at 1%, with the exception of q_t-4, which is significant at 5%. In addition, we test for omitted variable (H₀: id = 0), which cannot be rejected (F stat = 0.7118(0.4002)), and prove that f_t and f_t-1 are jointly redundant (H₀: f = 0; F stat = 20.55(0.0000)), which is clearly rejected, all of which reinforces our specification.

After eliminating non-significant regressors, we find that the constant and the linear trend (t) confirm the long-run depreciation trajectory over time.

According to all the correct specification statistics, our model fits the data well so we can fairly claim that it is a good approximation to Data Generating Process, Hendry [57]. To have a better picture of the above reported results, we summarize the overall short-run effects by summing up all coefficients for each variable and report them in Table 3. The strong (dynamic) persistence effect of q stands out in its own determination, which is above that of the other variables. Next in importance are tnt and tot and, finally, significant–albeit marginal–effects of ρ and f are notable.

Table 4 reports the long-term equilibrium relation, since the cointegration error coefficient (ECM) is correct and significant and indicates that 25% of a shock occurred today is absorbed within the next month in the cointegrating relationship.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Long-run estimates.

https://doi.org/10.1371/journal.pone.0286331.t004

In general terms, estimates in Table 4 confirm the short-run outcome at 1% in all cases. The same order of importance of the determinants of q is observed, and short-term proportions are maintained, although it stands out that all parameters are much higher, and the elasticity of tnt is now above unity.

Our approach and results are largely consistent with Catalán [31] who sustains that exchange rate fluctuations cannot be explained in a conventional manner through the supply and demand approach, which would be applied to any good, because economic agents’ perception of the value of each currency is highly relevant.

Similarly, our results, in part, coincide with the PPP approach [31], which we verified based on the depreciation trend of the real exchange rate with the coefficient of 0.0020.

Finding a period of cointegration is very important because it is not a general condition, as in the case of Canada [31]. The importance of this result lies in that this evidence supports the presence of the Balassa-Samuelson Effect in the long run, which implies the rejection of the monetary model [31] mainly because of the relationship between tradable and non-tradable goods. Yépez & Dzikpe [29] testify to this feature by mentioning that in emerging countries the relative prices of tradable goods account for most of the volatility observed in the exchange rate, which is approximately 30%.

With respect to tot, which we place in the second place of relevance, Ramírez [30] shows that increases in productivity lead to an appreciation of the real exchange rate with an elasticity of 0.73.

Finally, we mention the effect of financial variables (which we recover as the forward exchange rate and the risk premium). Cardozo et al. [40] analyze the forward exchange rate for Colombia and conclude that, despite being an important variable in explaining the exchange rate, its relevance declined in the 2008–2015 period. This stands in contrast to Cáceres [41] who points out that the Colombian peso lacks the importance that the Peruvian Sol and the Mexican peso have in international markets to apply carry trade strategies. Therefore, these financial variables are more relevant for the case of Mexico. The latter result reinforces the position of the Mexican Peso as the third most traded currency among emerging markets [14].

Concluding remarks and further comments

We study and estimate short- and long-run determinants of the real bilateral (MEX-US) exchange rate (q) for the Mexican economy (2001.01–2022.12). We define five variables (in logs) as their main drivers: terms of trade (tot), price differentials (tnt), interest rate differentials (id), forward exchange rate (f) and risk premium (ρ).

Given the different order of integration, an ARDL model (4,2,2,0,0,1,0) was estimated, which confirms cointegration and short- and long-run effects only for 2009.01–2022.12.

We estimate short–and long-run interactions. In both cases, the Balassa-Samuelson Effect (tnt) turned out to be the main determinant of q (-0.27 and -1-11 for the short and long run), followed by tot (-0.126 and -0.51), f (0.005 and 0.063), and ρ (0.009 and 0.030), respectively. All in all, an increase in domestic relative prices, as well as an improvement in terms of trade, lead to q appreciation, while increases in f and ρ generate depreciation. Due to high collinearity, interest rate differentials (id) are not statistically significant and were, thus, excluded from estimations, following a detailed econometric procedure.

The results corroborate our hypothesis that, in the aftermath of the Great Financial Crisis and coincidently with the abundance of international liquidity, the depreciation/appreciation trajectory of q has been gradually influenced by financial determinants.

The current international environment is still characterized by an unprecedented level of liquidity, generated in response to the coronacrisis. Therefore, investors around the globe continue reallocating “cheap money” basically to emerging markets. In this continuously expanding financial environment, currency volatility in all countries is to be expected as a norm, rather than a signal of unusual events. Indeed, not only asset managers have disposable money, but also government, private firms and households’ debt issuance have been growing exponentially. The research of short-run gains of financial investors jointly with the possible mismanagement of debt (both public and private) could generate dangerous dynamics, which often lead to a well-known result: the currency crisis, which strongly affects real variables, which, in turn, depress the living standards in the long run.

Although financial determinants are second in importance for the whole estimation period, they are very likely to continue gaining relative weight in the current context of high liquidity and uncertainty and high appreciation of the Mexican peso. This seems to be related to the increased appetite for the Mexican Peso over the last three years, which is clearly associated with the increase in the forward rate and the interest rate differential.

Although the determinants of the long-term exchange rate respond to PPP, in the short-term financial instruments have a significant impact on it, so it is highly probable that as long as Mexico’s reference rate remains 6 points higher than the US rate (as it has occurred until April 2023), it is expected to ensure a relatively stable nominal and real exchange rate and to continue appreciating with a downward effect on inflation.

Supporting information