Push and pull determinants of the country risk premium for emerging economies: an econometric appraisal

Abstract This article aims to identify the main determinants of the country risk premiums CDS 5 Years and EMBI+ for eight emerging economies. Econometric estimations relied on autoregressive GMM (time series) and GMM-DIFF (panel data). The analysis period is 2003-2019 and depends on the country and the data availability (monthly and quarterly data). We have tested push (exogenous) and pull (country-specifics) regressors. The empirical results have shown that some push factors have significant effects, which indicates that the global financial and trade cycles play an essential role in determining emerging country risk premiums. However, those economies may mitigate global influences through some internal macroeconomic policies. In our models, the international reserves stock growth rate was the primary statistically significant pull variable, highlighting the importance of external sound accounts for emerging countries.


Introduction
Country risk premiums measures are essential in evaluating emerging economies' external sustainability.Those economies usually are more exposed to external shocks and international capital fl ow reversals than advanced economies.Standard metrics used as proxies for the country risk premiums are credit rating, the one classifi ed by Standard & Poor's, Moody's, and Fitch agencies, fi nancial vulnerability and currency indicators, external debt, default probability, and indexes such as CDS (Credit Default Swap) 1 and EMBI+ (Emerging Markets Bond Index Plus) 2 .
CDS is a security contract against assets credit risk negotiated bilaterally between the seller, usually a bank, and the purchaser.In this sense, the purchaser seeks protection against credit risks from the reference entity, i.e. the entity that issues the asset.Currently, CDS is the primary credit derivative in global terms (PIMCO, 2017).
EMBI+ is part of a family of indexes whose methodology was developed by the J.P. Morgan Chase bank in the 1990s.This index calculates the spread between the daily return of emerging sovereign bonds and U.S. (The United States) risk-free bonds with the same maturity and characteristics.The bonds must meet other requirements to be part of the index calculation (J.P. Morgan, 2018;2021).
This paper aims to identify the main determinants of the country risk premiums using CDS 5 Years and EMBI+ as indicators.We use time series and panel data methods and specifi cations suggested in the empirical literature for a sample of emerging economies from 2003 to 2019, depending on the country (time series models), and from 2008 to 2019 (panel data models).The panel data econometric strategy uses only the variables that have presented better statistical signifi cance in the time series models.The variables (regressors) will be both push (exogenous, external, global) and pull (country-specifi cs, domestic).The push and pull approach comes from the capital fl ows literature (Chuhan et al., 1993;Hannan, 2018;Naqiv, 2018).We hypothesize that some external variables play essential roles as determinants of the emerging country risk premium, while countryspecifi c variables can mitigate in some measure those exogenous effects.
We follow the suggestion of the Brazilian Central Bank (2020) that have classifi ed two groups of emerging countries as low and high-risk.We then selected Chile, Indonesia, and Russia (low-risk countries, according to that methodology) and Brazil, Colombia, Mexico, South Africa, and Turkey (high-risk countries) for our econometric proposals.The countries' sample is also based on data availability for monthly and quarterly frequency.
As far as we know, based on the literature review we have done, no papers have analyzed the countries of our sample.Also, the period from 2003 to 2019 covers almost all of the last two decades -a period of intense changes in the emerging integration in the fi nancial and trade markets.Finally, we believe that the combination of time series and panel data, running monthly and quarterly models, may be a vital sign of the robustness of our econometric results.Therefore, it contributes to the empirical literature on emerging country risk premiums determinants.
The paper is organized as follows: after this introduction, the next section presents a literature review of empirical works about country risk premiums determinants.Section 3 presents our econometric specifi cations' data, methodology, and results.Section 4 analyzes those models' results and the fi nal section contains the conclusions.

Literature review
In the last twenty years, there has been a relevant empirical production in Economics about the determinants of the emerging economies' country risk premiums.However, the theoretical aspects have not yet been well developed, and there is no theoretical paradigm to follow.For this reason, we start by analyzing some central results of the empirical literature, usually through time series and panel data applications.The empirical literature generally uses the concepts of international capital fl ows, the socalled push and pull debate that infl uences capital infl ows and outfl ows, and emerging economies' country risk premiums.We believe a critical (inverse) relationship exists between international capital fl ows to/from emerging economies and their country risk premiums.Calvo et al. (1993) were pioneers in analyzing the main drivers of capital infl ows and capital outfl ows to/from emerging countries.Based on that work, Chuhan et al. (1993), for the fi rst time in the literature, used the terms push and pull to denominate the factors that have an essential role in determining the capital fl ows to and from emerging economies.In short, push factors are related to external/global events such as monetary policy and economic growth in the world's most powerful economies, risk aversion by international investors, international oil prices, and so on.The pull factors are country-specifi c factors.They are related to domestic economic growth, international reserves stock, industrial production, monetary and fi scal policies, external debt, and so on.
Given the expected inverse relationship between capital fl ows and country risk premiums, we believe that the push and pull approach can also be adapted to analyze country risk premiums.Although simple, Koepke (2019) argues that this distinction is useful in economic literature.Hannan (2018) believes that the push and pull factors will continue to have an essential role in the capital fl ows literature.Aronovich (1999) conceptualized the country risk spread of emerging economies as [...] the compensation required by a foreign investor for assuming the risk of default implicit in a bond issued by a government i, which matures in n years and yields R in , when compared to the alternative return of purchasing a default risk-free bond of the same maturity ( R fn ), when compared to the alternative return of purchasing a default risk-free bond of the same maturity S in = R in + R fn (Ibidem, 1999, p. 466).
According to the author, that spread is useful because it describes the economic agents' perceptions of the fi nancial market about the long-term fundamentals of the economy.His empirical work analyzed Argentina, Brazil, and Mexico from June 1997 to September 1998.The author has found that positive variations in the default probability of the economies have increased external borrowing costs.Furthermore, the author has argued that the country risk spreads of the three countries in that period were superelastic concerning the long-term interest rate of The U.S. (Ibidem, 1999).García-Herrero and Ortíz (2005) assessed if the global risk aversion (GRA, proxy to the yield of USA corporative bonds with high relative risk) and some of its determinants, such as short and long-term interest rates and economic growth in the U.S., were responsible for impacting the sovereign spreads in a sample of nine Latin American countries from May 1994 to October 2003.The authors have used as proxies for the sovereign spreads the EMBI Global (Chile) and EMBI+ (other countries).They found a signifi cant positive relationship between GRA and Latin Ameri-can sovereign spreads.In contrast, U.S. economic growth and long-term interest rate (10-Year Treasury Bond Rate) had signifi cant negative effects.However, when the authors tested the econometric application with the short-term U.S. interest rate -Federal Fund Rate -the effect was immediate: when that interest rate has risen, the Latin American sovereign spread has risen also.Andrade and Teles (2006) developed a beta country risk model to assess the Brazilian country risk premium from January 1991 to December 2002.The authors found that the stock of international reserves was relevant only when Brazil had a fi xed exchange rate; when it fl oated, the coeffi cient associated with that variable lost signifi cance.Furthermore, fi scal variables (public debt and public sector primary net lending/borrowing) and the international oil price were insignifi cant in the author's beta model.Baldacci et al. (2008) empirically analyzed the main determinants of the country risk premium EMBI through panel data with 30 emerging countries from 1997 to 2007.To the authors, fi scal and political factors were relevant in the model: fi scal consolidation has contributed to limiting the emerging spreads; however, the authors found that the composition of the public expenditure matters: public investment, for example, presented a negative impact on the spreads while the fi scal position was sustainable and the fi scal defi cit did not become worse.On the other hand, political risks such as violence, expropriation, and instability have increased the country risk premiums of those countries.Rocha and Moreira (2010) developed a panel data approach with 23 emerging countries from 1998 to 2007.The authors aimed to assess the external (exogenous) and domestic determinants of the external vulnerability of those countries.The authors have used the VIX Index and the J.P. Morgan Domestic High Yield Spreads (H.Y.) as proxies for the global aversion to risk.The main fi nds of the paper were that those exogenous factors are relevant and produce different impacts on each economy: macroeconomic fundamentals are multipliers of those impacts.
The results support policies towards fi nancial liberalization, public debt management, consistent economic growth, development of the domestic fi nancial market, and improvements in governance indicators, especially the rule of law and regulatory quality (Ibidem, 2010, p. 181).Aidar and Braga (2020), through a principal component analysis, have shown that the fi nancial cycles in peripheral economies are subordinat-ed to the global fi nancial cycles.In a model with ten emerging countries from January 1999 to January 2019, the authors aimed to present the main drivers of the country risk premiums (EMBI+ and CDS) for that sample of countries.The push and pull approach was the center of the debate.The authors have argued that push factors such as VIX Index and the U.S. 5-Year T-Note Interest Rate (with a positive sign) and international oil price (with a negative sign) have played relevant roles as determinants of the country risk premiums.
Finally, the International Monetary Fund (IMF) developed a non-balanced panel data analysis in its Global Financial Stability Report (October 2019).The institution's researchers studied 71 countries, intending to explain the main determinants of the EMBI Global Index (proxy to the country risk premium) from 1996 to 2019.The model had exogenous variables (US BBB corporate spread, proxy to the global risk appetite, and external real GDP growth (one-year forward forecasts)).It also considered some country-specifi c variables: domestic real GDP growth and domestic CPI infl ation (one-year forward forecasts), current account, external debt, net issuance of foreign currency government debt, and foreign currency reserves, all as a percent of GDP.Domestic credit rating has interacted with the variable associated with global risk appetite.
In the results, the model has shown that domestic fundamentals are essential in explaining the sovereign spreads of those economies.For example, higher real GDP growth, lower infl ation, higher stock of international reserves, and lower external debt reduce sovereign spreads.Furthermore, countries with better credit ratings were less susceptible to external instabilities: Lower-rated issuers are more sensitive to global risk appetite.A 100 basis point increase in the US BBB corporate bond spread could widen spreads of B-rated EM bonds by more than 200 basis points, compared to only 50 basis points for A-rated EM issuers (IMF, 2019, p.14).
Based on this literature review, in the next section, we present the methodology and data of our empirical analysis.

Methodology and data
This paper developed time series and panel data econometric applications to verify the main determinants of the country risk premiums EMBI+ and CDS 5 Years for a sample of emerging economies.At fi rst, we ran time series models to select the main variables -both push and pull -that in the period 2003-2019, depending on the country, were more critical in that determination.Those variables were selected through the literature review (Section 2 above), but this procedure was mainly based on IFM (2019) and Aidar and Braga (2020).In this sense, the models proposed were the following: where t = 1,…,T; the number of observations depending on the country and the model, if it has monthly or quarterly data. 3In this sense, we have four models for each of the eight countries of our sample: two for monthly data and two for quarterly data, which totalizes 32 models.Because of their correlograms (autoregressive processes of order one), all the models have the dependent variables with one lag as regressors (Bueno, 2015).Figures 1 and 2 show the path of CDS 5 Years and EMBI+ indexes (basis points) for the countries in the sample. 4 As already accepted in the economic literature (Rezende, 2009;Lavoie, 2013;Serrano and Pimentel, 2017), a country issuer of its own currency cannot face a default on its public debt.In this sense, we do not consider internal fi scal variables relevant to the external solvency indicators.However, the possibility of a country with a fi scal expansion or monetizing its public debt may be assessed by international investors as a risk for the domestic infl ation rate.Although not necessarily representing a cost for the investor, this possible increase in the infl ation rate has adverse macroeconomic consequences, mainly in emerging economies, which can cause capital outfl ows.Also, we do not use external debt variables because of data unavailability for the needed frequency.We believe that the variable associated with the international reserves stock fulfi lls well that external issue.
3 See Table 2 on Appendices.The period of the models varies among countries basically because of data availability.4 End of period monthly data. (1) (2)  In this sense, we have selected the following variables for our econometric specifi cations ( 1) and ( 2): X t is a pull matrix with the following variables: GDP yearly growth rate (GDP_DOM_YOY), domestic indus-  We expect the coeffi cients associated with the dependent variables with one lag LN_CDS_5Y(-1) and LN_EMBI(-1), INF_YOY, LN_ INTEREST_5Y_US, and LN_VIX positively affect the dependent variables.More specifi cally, we expect an inertial process of the series LN_CDS_5Y and LN_EMBI over time.We also believe that an increase in the infl ation rate can cause a deterioration of the emerging country risk premium (IMF, 2019).An increase in the U.S. long-term interest rate may also trigger a fl ight to quality (international capital fl ow reversals) toward U.S. bonds and increase the emerging country risk (Aronovich, 1999;Aidar and Braga, 2020).VIX Index is a proxy for global turbulence in the U.S. fi nancial markets.A worse index may also increase the emerging country risk (Rozada and Yeyati, 2006;IMF, 2019;Aidar and Braga, 2020), mainly because of the above-referred fl ight to quality movement.
On the other hand, we expect that the coeffi cients associated with the variables GDP_US_YOY, IND_PROD_US_YOY, IND_PROD_YOY, IND_PROD_MANUF_YOY, RT4_LN_INT_RES and RT12_LN_INT RES, CA, GDP_DOM_YOY, and LN_OIL have signifi cant negative effects on those dependent variables. 8More specifi cally, we expect that the vari-ables associated with the external production, such as GDP_US_YOY and IND_PROD_US_YOY, proxies for the global economic performance, and GDP_DOM_YOY, IND_PROD_YOY, and IND_PROD_MANUF_ YOY, that represent the domestic economic growth, may contribute to lower the country risk of emerging markets (IMF, 2019).Additionally, we believe that the variables associated with the hoarding of international reserves and the current account are essential to reduce country-risk premiums because they improve the external accounts of the emerging economies, moving away, for example, from the balance of payments constraints.Finally, we expect an inverse relationship between the international oil price and the emerging country risk premium.As many emerging economies depend on international commodities markets, the lower the oil price, the lower the export revenues -mainly denominated in the U.S. dollar -absorbed by them.Therefore, there is a link between international oil prices and the capacity of emerging economies to deal with their external accounts and the global economic cycles (Aidar and Braga, 2020).
We ran the Generalized Method Of Moments (GMM) for each one of the models of our time series econometric specifi cations. 9We did it because GMM deals better with endogeneity problems, i.e., cov(u t , x t ) ≠ 0, serial correlation, and heteroskedasticity (Hansen, 1982;Wooldridge, 2001b).According to Wooldridge (2001a, pp. 50-51), endogeneity occurs because of omitted variables, measurement errors, or simultaneity. 10In our approach, we consider all of the pull variables as endogenous, and then we instrumentalize them; also, we consider all of the push variables as exogenous.A good instrument z t has to be valid in two cases: cov(u t , z t ) = 0 and cov(x t , z t ) ≠ 0. Thereby, we can be sure that the estimated GMM coeffi cients converge in probability to the true parameters, plim β ˆi = β i .11However, Deaton (2018) argues that it can be hard to fi nd instruments that fulfi ll the two hypotheses above.For this reason, we follow Johnston and DiNardo (1996), that suggest that lags of the independent variables may be used as instruments of the model, considering that those variables match the two cases mentioned above.It is important to highlight that many instruments, compared to the number of observations, may cause bias in the model, mainly if some of the instruments have a weak correlation with the potentially endogenous variables.In this case, we sought to be parsimonious and add a not very large proportion between instruments and the number of observations of the models.12The J-statistic was used as a test of overidentifying restrictions (Cragg, 1983), i.e., when the number of instruments is greater than the number of regressors of the true model (Hansen, 1982).It presents a test for the validity of the instruments.GMM also deals better with common issues in econometric estimations, serial autocorrelation and heteroskedasticity (Hansen, 1982;Newey-West, 1987).For this reason, we have applied the covariance HAC Newey-West matrix to the models to control those issues. 13able 1 summarizes the aggregate results of models ( 1) and ( 2), both month and quarter specifi cations, considering the adequacy of the coeffi cients to what we have hypothesized.In bold, we highlight the main variables that have presented expected signs on at least 50% of the specifi cations.In this sense, we have two push variables: LN_VIX and LN_OIL, and two pull variables: RT_12_LN_INT_RES and INF_YOY.Moreover, the dependent variables with one lag also have presented expected effects in all specifi cations we have tested, demonstrating the inertial character of the processes.
We then specifi ed a balanced panel data model with those variables that have presented better adequacy to the expected signs in the GMM autoregressive specifi cations.We ran a GMM-DIFF, as proposed by Arellano and Bond (1991), for the period from 2008 to 2019 with monthly and quarterly data.Such as in the time series models, the GMM method was chosen because it deals better with the endogeneity problem (Roodman, 2009).More common models like fi xed and random effects, which use ordinary least squares, present diffi culties in dealing with that problem and are not recommended for dynamic panel data.Other problems arise because we have a small sample of countries.According to Arellano (2002) and Roodman (2009), many instruments may cause problems to the GMM estimation, including the overidentifying J test.In this sense, we have limited the instruments to seven (since the number of countries of our sample is eight) and used the same strategy of the time series models: lagged variables as instruments, following Johnston and Di-Nardo (1996).Because of that, we had just four (static specifi cations) or fi ve (dynamic specifi cations) explanatory variables in the models that presented better adequacy to the expected effects in the time series models.We also transformed all the variables in growth rates concerning the previous period, month or quarter, to solve the panel data unit root issue.
14 Quarterly model for Russia took the growth rate from previous period for the dependent variable CDS 5 Years and for the regressors VIX Index, current account balance, international oil price, and the autoregressive variable.We did it to solve the unit root problem.

Dynamic specifi cation:
Static specifi cation: where Y it represents both dependent variables, the growth rate of the indexes CDS 5 Years and EMBI+; Y it-1 defi nes the autoregressive variables.
The regressors are the growth rate of the international reserves stock (RT1_LN_INT_RES) and the infl ation rate (INF_QOQ) (pull variables).The growth rate of the VIX Index (RT1_LN_VIX) and the international oil price growth rate (RT1_LN_OIL) are the push variables.i = 1,…,8 eight countries) and t = 1,…,T (1.152 observations for monthly models, from January 2008 to December 2019, and 384 observations for quarterly models, from 2008.Q1 to 2019.Q4).All the variables were transformed by their natural logarithm, except infl ation µ i represents country specifi cs effects and u it is the error term.
It is worth mentioning that the GMM-DIFF method, taking the fi rst difference of the variables, rules out those variables that are time-invariant (Baltagi, 2005).In our models, there are no estimations for intercept terms and country specifi cs effects then.
Tables 2 and 3 summarize the GMM-DIFF results.In the next section, we present some considerations about our results.Two of the eight models had problems with the AR(2) Arellano-Bond Serial Correlation Test: the dynamic quarterly model for the dependent variable CDS 5 Years and the dynamic monthly model for the dependent variable EMBI+ have rejected the null hypothesis of the test (p-value < 0,10).Arellano and Bond (1991) propose a test for the hypothesis that there is no second-order serial correlation for the disturbances of the fi rst-differenced equation.This test is important because the consistency of the GMM estimator relies upon the fact that E[∆v it ∆v it-2 ] = 0 (Baltagi, 2005, p. 141).
The fi rst-order serial correlation AR( 1) is expected by the construction of the test and it is not an issue. (3)

Empirical analysis
Our econometric approaches, both time series and panel data, have tested some push and pull variables to analyze the main determinants of the country risk premiums for a sample of emerging economies.At fi rst, we computed all the results from time series models.In the previous section, we have neither exhibited the individual models for each of the eight countries nor the coeffi cients that the GMM models estimated.Table 1 just summarized the most crucial information about those estimated models.
In that table, we have the degree of adequacy of each one of the independent variables concerning the coeffi cient sign we have expected.Push and pull variables such as GDP_DOM_YOY, IND_PROD_YOY, GDP_US_YOY, and CA have demonstrated poor suitability (in all models that were tested, more than 50% had insignifi cant coeffi cients).Other variables such as IND_PROD_MANUF_YOY, IND_PROD_US_YOY, and LN_INTEREST_5Y_YOY have demonstrated mixed results according to the signs of the expected coeffi cients.
However, econometric models developed by Nogués and Grandes (2001), Afonso (2003), and FMI (2019) have found economic growth as an essential factor that improves emerging economies' country risk premiums.We believe that U.S. GDP growth and U.S. industrial production growth did not match the function of being good proxies for international economic growth.Perhaps we should have used another proxy weighting the participation of other relevant economies such as Germany, China, France, and others that have a great economic relationship with the countries of our sample.GDP and domestic industrial production growth rates had poor suitability in the models proposed.As GDP is an aggregate variable, we believe this feature may affect its impacts on emerging country risk premiums, which are daily variables.Additionally, it is possible that industrial production is not a good proxy for monthly economic performance.It is important to highlight that the services sector is the most important in most economies worldwide.
Furthermore, econometric estimations by Aronovich (1999), Arora and Cerisola (2001), Nogués and Grandes (2001), González-Rozada and Yeyati (2008), Dailami et al. (2008), Aidar andBraga (2020), andHartelius et al. (2008) have found evidence that a rise in the U.S. interest rate can cause increases in the emerging country risk premiums.For Aronovich (1999), emerging economies' spreads are superelastic to the long-term U.S. interest rate.Dailami et al. (2008) fi nds that the relation between U.S. monetary policy and emerging country risk is positive.Still, the countries with moderate debt levels are generally less impacted by the U.S. interest rate movements.Aidar and Braga (2020, p. 99) argued: "The empirical exercise suggests that an increase in the interest rate associated with the 5-Year T-Note coincides with a higher perception of risk captured by the fi rst principal component".In our estimations, using the variable Market Yield On U.S. Treasury Securities at 5-Year Constant Maturity, only 28,1% of the models have demonstrated evidence of a signifi cant positive relationship between that interest rate and emerging country risk premiums.García-Herrero and Ortíz (2005), in turn, found a positive and instantaneous relationship between the U.S. short-term interest rate and the emerging sovereign spread.In future works, we should test the real interest differential -short and long terms -between emerging economies and the United States.It may be more relevant in our context.
In the case of the autoregressive independent variables tested in the 16 specifi cations, all of them had the expected positive sign.It shows the inertial character of the series, as their correlograms have already demonstrated.In other words, the current level of the dependent variables depends in great measure on their previous levels.
Push variables LN_VIX and LN_OIL coeffi cients estimated also had the expected signs.VIX Index has presented signifi cant positive coeffi cients in all 32 monthly and quarterly models.It shows that global economic turbulence impacts risk perception in the emerging world.The international oil price, in turn, has demonstrated signifi cant negative coeffi cients, as expected, in 3/4 of the monthly and quarterly specifi cations.The economic dependence of emerging economies on commodities and international export markets explains the importance on the risk perception of those economies (Aidar and Braga, 2020).Our models captured it.
In this sense, the VIX Index and international oil price were the main push variables we found through time series specifi cations.This situation emphasizes the relevant role some global factors play in emerging country risk premiums pricing.The role of international liquidity, captured in those push variables, implies that there is a common cause for the country risk premium dynamics, as noted by Aidar and Braga (2020).Although 2020 data was not included in our sample, we can use the fi rst months that followed the outburst of the COVID-19 pandemic to illustrate that joint movement.Figure 3 shows that the country risk premiums, measured by the CDS 5 Years, increased in all our sample countries.
On the other hand, the coeffi cient signs of the primary pull variables were as expected: the infl ation rate, with positive effects, and the growth rate of the international reserves stock (monthly models), with negative effects.Our results for both variables align with IFM (2019).Still, they contradict Andrade and Teles' (2006) study about the Brazilian economy because the authors have argued that the international reserves stock was relevant in explaining the country risk premium only for fi xed exchange rate periods.However, according to the Assessing Reserve Adequacy methodology by IMF ( 2021), all the countries in our sample have fl oating exchange rates.
In this sense, the time series models have suggested that lower infl ation and a growing stock of international reserves are the main pull variables that can mitigate some effects of the global fi nancial cycles on emerging country risk premiums.
Static and dynamic GMM-DIFF panel data estimations were produced out of the time series results using the main variables verifi ed in those estimations.In this sense, for both dependent variables, we have tested as independent variables: autoregressive variables (dynamic models), two push regressors (growth rates of the VIX Index and international oil price), and two pull regressors (international reserves stock growth rate and the infl ation rate).
The results were similar in all eight models we estimated.For the dependent variable associated with the CDS 5 Years, neither monthly nor quarterly models have demonstrated signifi cant positive effects in the coeffi cient related to the autoregressive regressors.However, for the dependent variable EMBI+, it happened as expected.Furthermore, both dynamic and static, monthly and quarterly estimations, have demonstrated the same results: push variables VIX Index (positive effects) and international oil price (negative effects) have played important roles in explaining the emerging economies' country risk premiums for the reasons discussed above.
Accumulating international reserves is an important economic tool to reduce the country risk premium and deal with the exogenous shocks from the international markets, like those from variations in the VIX Index and international oil price.It is worth mentioning that in all models, the coeffi cients estimated for the international reserves variable were larger than those associated with the push variables: considering the models that did not present problems with the AR(2) statistics, the coeffi cients ranged from - .It suggests the great relevance of accumulating international reserves in lowering the emerging country risk premiums since it acts as a fi nancial backing for futures market transactions and safety against capital outfl ows (fl ight to safety or fl ight to quality).
Contrary to most of the time series results, the infl ation rate concerning the previous period was insignifi cant in all models we have tested.The panel data models did not capture the effects of the rising prices, as they were captured through the time series models.
In this sense, besides the inertial characteristic of both dependent variables, our GMM-DIFF estimations have demonstrated that the movements of the VIX Index, the international oil price, and the growth rate of the international reserves stock played essential roles as drivers of the emerging economies' country risk premiums movements throughout the last two decades.

Concluding remarks
Based on the empirical literature, mainly on works by IMF (2019) and Aidar and Braga (2020), this paper presented a model with two different econometric approaches to evaluate the main drivers of the country risk premium for a group of emerging economies in the last two decades.In the time series models, we have found that the two main external or push variables were the VIX index and the international oil price.The fi rst variable had a positive or direct effect on emerging country risk premiums; the second, in turn, had a negative or inverse effect on those premiums.Furthermore, the pull variables that stood out were the growth rate of international reserves stock (negative effects) and the infl ation rate (positive effects).
In the panel data GMM-DIFF approach, the push variables related to the VIX Index and international oil price kept playing the same role as determinants of the emerging country risk premiums.However, among the country-specifi c variables we have selected for the panel data models, the growth rate of the international reserves stock and the infl ation rate concerning the previous period, only the fi rst demonstrated negative signifi cant effects on the emerging country risk premiums.We highlight the large coeffi cients estimated for that variable, mainly in the CDS 5 Years panel data models, which explain the importance of accumulating international reserves for emerging economies.International investors can consider it as a sign of external sound accounts of the emerging economies and a necessary condition for an economy growing without the balance of payments constraints.The infl ation rate, in turn, was insignifi cant in all eight models we tested.
Although 2020 data was not included in our sample, we can interpret what happened with CDS 5 Years and EMBI+ during the COVID-19 pandemic based on our fi ndings.In the fi rst four months of 2020, emerging economies' country risk premiums measured by CDS 5 Years and EMBI+ increased in all our sample countries -an expected result given our models.According to FRED Economic Data, VIX Index increased by 34.7 points from January to March 2020, the period when the fi rst impacts of the pandemic started to be globalized.From January to April 2020, the international oil price decreased, in nominal terms, by $ 40.26.The effect of the reversal of international liquidity, mainly through the VIX Index and the international oil price, was sizeable in the emerging country risk premiums.Between January and March 2020, except for Mexico and Russia, all countries lost international reserves to deal with the pandemic economic impacts.However, the effect on the international reserves stocks was not so strong according to IMF.Throughout 2020, the most impacted country in terms of international reserves stock was Chile, which lost almost 8 billion dollars.According to our econometric results, it was another force contributing to elevating the country risk premium at the beginning of the pandemic.
In a fi nancialized world, we conclude that emerging economies are exposed to global shocks, which can be refl ected in their country risk spreads.Besides, country-specifi c variables such as the positive growth rate of the international reserves stock (mainly) and the low infl ation rate may act as a buffer for those external shocks.In this sense, we expect that our econometric fi ndings may contribute to the empirical literature about the determinants of emerging economies' country risk premiums.

IND_PROD_US_YOY*
U.S. industrial production growth rate (%) concerning the same month of the previous year.Monthly models.Proxy for the U.S. monthly economic growth.

LN_INTEREST_5Y_US*
Natural logarithm of the Market Yield On U.S. Treasury Securities at 5-Year Constant Maturity.End of period for monthly and quarterly models.

LN_OIL*
Natural logarithm of the international oil price (Brent crude).End of period constant prices for monthly and quarterly models.

LN_VIX*
Natural logarithm of the VIX Index, end of the period.Monthly and quarterly models.

RT1_LN_CDS_5Y**
Growth rate (%) of the country risk premium CDS 5 Years natural logarithm concerning the previous period.

RT1_LN_EMBI**
Growth rate (%) of the country risk premium EMBI+ natural logarithm concerning the previous period.

RT1_LN_INT_RES**
Growth rate (%) of the international reserves stock natural logarithm concerning the previous period.

FRED Economic Data and OECD
RT1_LN_VIX** Growth rate (%) VIX Index natural logarithm concerning the previous period.

Figure 1
Figure 1 CDS 5 Years country risk premium

Figure 2
Figure 2 EMBI+ country risk premium

Figure 3
Figure 3 CDS country risk premiums from October 2019 to December 2020 series models.** Panel data models.

Table 1
Summary of time series models 14 DEPENDENT VARIABLES: LN_CDS_5Y and LN_EMBI

Sign of the coef- fi cient as expected Sign of the coeffi - cient diff erent from the expected Insignifi cant
Note: Prob.< 0.10.