Global Economic Policy Uncertainty and Financial Market Returns in Türkiye: Evidence from Stock and Exchange Rate Channels

1. Introduction

Global economic policy uncertainty has become increasingly relevant to financial market analysis because policy-related uncertainty can alter investor expectations, risk assessment, and international portfolio decisions. Changes in fiscal policy, monetary conditions, international trade arrangements, and regulatory direction may affect the pricing of financial assets, particularly in economies that are sensitive to global capital movements. The economic policy uncertainty framework developed by Baker, Bloom, and Davis provides a systematic basis for measuring policy-related uncertainty and examining its financial implications over time [1]. In the present study, the global version of this uncertainty measure is examined in relation to Turkish financial market returns.

The relationship between uncertainty and financial markets can operate through more than one channel. Uncertainty may encourage investors and firms to postpone decisions until the economic outlook becomes clearer, affecting investment conditions and market valuations [2]. In equity markets, greater policy uncertainty may be reflected in revised expectations regarding corporate performance and in changes in the risk premium required by investors. Policy uncertainty may therefore be associated with changes in equity return dynamics, especially during periods in which market participants face difficulty in evaluating future policy conditions [3]. At the international level, global financial conditions and shifts in risk appetite may also influence capital flows and financial prices across emerging markets [4].

Türkiye offers an appropriate setting for examining these relationships because its stock market and exchange rate provide two distinct financial channels through which global uncertainty may be reflected. The BIST 100 index represents the domestic equity market channel, while USD/TRY represents the foreign exchange channel. These markets need not respond identically to the same global uncertainty indicator. Equity returns may incorporate changes in investor expectations and risk pricing, whereas exchange rate returns may reflect a broader combination of external financing conditions, domestic macroeconomic developments, and currency market pressures.

Previous research has already examined the relationship between global economic policy uncertainty and Turkish financial variables. İlhan and Bağcı analysed the dynamic relationship between BIST 100, GEPU, bond interest rates, and the exchange rate using an ARDL bounds testing approach for the period from January 2013 to June 2024 [5]. Their study provides relevant evidence concerning the Turkish equity market and related financial variables. The present article differs in both focus and empirical design. Rather than modelling BIST 100 as the principal dependent variable within an ARDL framework, this study estimates an integrated Türkiye-specific vector autoregression system in which changes in GEPU, BIST 100 returns, and USD/TRY returns are treated as endogenous variables. This design permits direct comparison of predictive relationships across the equity and foreign exchange channels within the same dynamic model.

Accordingly, the objective of this article is to examine whether monthly changes in global economic policy uncertainty contain predictive information for Turkish stock market and exchange rate returns within an integrated VAR framework. The analysis uses monthly data from January 2010 to December 2020 and applies first logarithmic differences to GEPU_current, BIST 100, and USD/TRY. The empirical strategy combines stationarity testing, lag length selection, VAR estimation, Granger causality and block exogeneity testing, residual diagnostic evaluation, and an alternative lag specification for robustness assessment. The study focuses on predictive content within the estimated system and does not interpret statistically significant Granger relationships as evidence of structural economic causality.

The remainder of the article is organised as follows. Section 2 describes the data, variable measurement, and econometric procedures. Section 3 reports the empirical findings and robustness evidence. Section 4 discusses the interpretation of the results, and Section 5 concludes the article.

2. Materials And Methods

2.1 Data and Variable Measurement

The study employs monthly observations covering the period from January 2010 to December 2020. The empirical analysis is restricted to Türkiye and is constructed around three variables: global economic policy uncertainty, Turkish stock market performance, and the Turkish exchange rate against the United States dollar. This country-specific structure differs from broader cross-country frameworks by examining whether global policy uncertainty carries predictive information across two financial market channels within the same national setting.

Global economic policy uncertainty is represented by the GEPU_current series obtained from the Economic Policy Uncertainty database [6]. This series is used as the uncertainty indicator throughout the empirical analysis. Turkish stock market performance is measured by the BIST 100 index, while exchange rate dynamics are represented by the USD/TRY series. The financial market series were obtained from Investing.com [7][8]. USD/TRY is expressed as Turkish lira units per United States dollar; therefore, an increase in this variable indicates a depreciation of the Turkish lira against the dollar. The monthly dataset contains 132 observations in level form. Since the empirical analysis is conducted using first log differences, one observation is lost during transformation, resulting in 131 observations for the return series. The variables included in the Türkiye-specific system are therefore Δln(GEPU_current), Δln(BIST 100), and Δln(USD/TRY). Table 1 summarises the measurement and interpretation of these variables together with their respective data sources.

Note: Δln(X_t) denotes the first logarithmic difference of variable X. USD/TRY is expressed as Turkish lira units per United States dollar.

2.2 Data Transformation

All variables were transformed before estimation in order to obtain comparable monthly change series and to avoid modelling financial variables in their raw level form. The transformation was implemented in two stages. First, the natural logarithm of each variable was calculated. Second, the first difference of the logarithmic series was computed. For a variable $X_t$ , the transformed series is defined as follows:

Applying the first logarithmic difference transformation reduces the number of usable observations from 132 monthly level observations to 131 transformed observations. These transformed series constitute the variables used in the subsequent stationarity tests and in the Türkiye-specific vector autoregression model.

2.3 Stationarity Testing

Before estimating the dynamic model, the stationarity properties of the transformed series were examined to avoid inference based on non-stationary variables. Since the empirical analysis is conducted using monthly first log-differenced series, the testing procedure was applied to Δln(GEPU_current), Δln(BIST 100), and Δln(USD/TRY).

Two complementary unit root tests were employed. First, the Augmented Dickey-Fuller (ADF) test was applied to examine whether each transformed series contains a unit root [9]. The test was estimated with a constant term, and the lag length was selected automatically using the Schwarz Information Criterion, with a maximum lag length of 12 to reflect the monthly frequency of the data. Second, the Phillips-Perron (PP) test was employed as an additional stationarity assessment [10]. The PP test was estimated with a constant term, using Newey-West automatic bandwidth selection and the Bartlett kernel. Its use is relevant because it applies a non-parametric correction that accommodates serial correlation and heteroskedasticity in the disturbance process.

For both tests, the null hypothesis states that the examined series contains a unit root. Rejection of the null hypothesis indicates that the transformed variable is stationary and can be included in the subsequent VAR analysis without further differencing. The numerical results of the ADF and PP tests are reported in the Results section.

2.4 Vector Autoregression Specification

A vector autoregression model was estimated to examine the dynamic relationships among global economic policy uncertainty, Turkish stock market returns, and Turkish exchange rate returns. The VAR approach provides a flexible multivariate framework in which each endogenous variable can depend on its own lagged values and on the lagged values of the other variables in the system [11]. This framework is appropriate for the present study because it allows dynamic relationships to be evaluated without imposing an a priori contemporaneous structural direction among the three variables.

Unlike a broader cross-country design, the present article estimates a single model for Türkiye. The endogenous variable vector is specified as follows:

The ordering of the variables places Δln(GEPU_current) first, followed by Δln(BIST 100) and Δln(USD/TRY). This ordering reflects the analytical focus of the study, in which global policy uncertainty is examined in relation to two Turkish financial market channels. The model is not interpreted as identifying structural shocks or strict economic causality. Instead, it provides a dynamic framework for evaluating whether lagged movements in one variable contain statistically significant predictive information for another variable within the estimated system.

2.5 Lag Length Selection

Selecting an appropriate lag length is necessary in VAR estimation because an insufficient number of lags may fail to capture relevant dynamic adjustment, whereas excessive lag inclusion may reduce degrees of freedom and weaken estimation efficiency. This consideration is especially important in the present study because the empirical model is estimated from monthly observations over a finite sample period.

The lag order selection procedure was conducted for the Türkiye specific system comprising Δln(GEPU_current), Δln(BIST 100), and Δln(USD/TRY). Given the monthly frequency of the dataset, a maximum lag length of 12 was considered in order to allow the selection process to evaluate dynamics extending over one year. The lag decision was assessed using the sequential modified likelihood ratio statistic (LR), the final prediction error criterion (FPE), the Akaike information criterion (AIC), the Schwarz criterion (SC), and the Hannan Quinn criterion (HQ).

The baseline VAR specification was selected by considering the overall evidence provided by these criteria together with the need to preserve an economical model structure for subsequent predictability testing. The selected lag order and the corresponding criterion values are reported in the Results section.

2.6 Granger Causality and Block Exogeneity Testing

Granger causality and block exogeneity Wald tests were applied after estimating the VAR model in order to examine directional predictability among the variables included in the Türkiye-specific system. In the Granger testing framework, one variable contains predictive information for another when its lagged values improve the explanation of the second variable beyond the information already contained in that variable’s own past values [12]. The procedure is therefore appropriate for assessing predictive linkages among Δln(GEPU_current), Δln(BIST 100), and Δln(USD/TRY) within the estimated VAR system.

Within this framework, three sets of predictive relationships are of direct relevance. First, the analysis examines whether past changes in global economic policy uncertainty contain predictive information for Turkish stock market returns. Second, it evaluates whether changes in global economic policy uncertainty provide comparable predictive information for USD/TRY returns. Third, the model allows the investigation of possible feedback predictability from Turkish financial market returns to the global uncertainty indicator and between the two domestic financial market variables.

For each dependent variable in the estimated system, the null hypothesis states that the lagged coefficients of the excluded variable are jointly equal to zero. Rejection of the null hypothesis indicates statistically significant predictive content from the excluded variable to the dependent variable within the VAR system. The Wald test results are reported through chi-square statistics and corresponding probability values.

The interpretation of these tests is deliberately restricted to predictive relationships. A statistically significant Granger causality result does not, by itself, establish structural causality or demonstrate a direct economic transmission mechanism. This qualification is particularly important when considering possible feedback from Turkish market returns to GEPU_current, since the latter is a global uncertainty measure rather than an indicator determined by Turkish market conditions alone.

2.7 Diagnostic Evaluation and Robustness Check

The adequacy of the baseline VAR specification was evaluated through a sequence of diagnostic tests. First, the stability condition of the estimated system was examined using the roots of the characteristic polynomial. A VAR model is considered dynamically stable when all characteristic roots lie inside the unit circle. This condition is necessary before interpreting the predictive relationships identified within the system.

Second, residual serial correlation was assessed through VAR residual serial correlation LM tests. The tests were conducted through five lags in order to determine whether residual dependence remained after estimating the baseline model. The null hypothesis in these tests is that no serial correlation is present at the examined lag or jointly across the specified lag interval.

Third, the distributional properties of the residuals were evaluated through multivariate residual normality tests based on Cholesky orthogonalization. The tests report skewness, kurtosis, and Jarque-Bera statistics for the individual residual components and for the system jointly. These results are used to identify whether departures from normality are general across the system or concentrated in a particular financial market equation.

Fourth, residual heteroskedasticity was examined using the VAR residual heteroskedasticity test with cross terms. This test evaluates whether the variance and covariance structure of the residuals displays systematic dependence on the included regressors. The diagnostic evidence is used to qualify the interpretation of the estimated predictive relationships and to assess whether the baseline model exhibits major residual specification problems.

In addition to the baseline VAR model, an alternative VAR(2) specification was estimated as a robustness check. This alternative specification was used to assess whether the main predictive findings depend exclusively on the one-lag structure selected for the baseline analysis. For the VAR(2) model, Granger causality and block exogeneity results were re-examined, and the stability condition and residual serial correlation properties were evaluated. The robustness analysis is therefore focused on the persistence of the principal predictive relationships under an alternative dynamic specification.

3. Results

3.1 Descriptive Statistics

Table 2 reports the descriptive statistics for the three return series included in the Türkiye-specific model. Each transformed series contains 131 monthly observations. Among the variables examined, Δln(GEPU_current) displays the greatest dispersion, with a standard deviation of 0.198292, compared with 0.066952 for Δln(BIST 100) and 0.043522 for Δln(USD/TRY). The mean values are positive for all three variables, although their magnitudes remain relatively small on a monthly basis.

The distributional properties differ noticeably across the financial market series. Δln(USD/TRY) exhibits pronounced positive skewness of 1.762159 and a kurtosis value of 13.10257, indicating a distribution characterised by asymmetry and substantial fat-tailed behaviour. The Jarque-Bera test strongly rejects normality for this series, with a probability value of 0.0000. In contrast, the normality hypothesis is not rejected for Δln(GEPU_current) or Δln(BIST 100), whose probability values are 0.2747 and 0.3194, respectively. These preliminary results indicate that exchange rate returns contain more extreme distributional features than stock market returns during the sample period; however, they do not establish any directional predictive relationship among the variables.

Note: The reported series are monthly first logarithmic differences. Δln(BIST 100) represents Turkish stock market returns, while a positive value of Δln(USD/TRY) indicates depreciation of the Turkish lira against the United States dollar.

3.2 Unit Root Test Results

The stationarity properties of the three transformed series were examined using the Augmented Dickey-Fuller and Phillips-Perron unit root tests. The results are reported in Table 3. For each variable, both tests reject the null hypothesis of a unit root at the 1 per cent significance level. The ADF test statistics are -16.73495 for Δln(GEPU_current), -11.84043 for Δln(BIST 100), and -10.48492 for Δln(USD/TRY), with probability values of 0.0000 in each case.

The Phillips-Perron results confirm the ADF evidence. The PP test statistics are -26.55356 for Δln(GEPU_current), -13.44112 for Δln(BIST 100), and -10.61022 for Δln(USD/TRY), and each corresponding probability value is 0.0000. Taken together, the two testing procedures provide consistent evidence that all return series included in the Türkiye-specific model are stationary.

These findings support the estimation of the VAR model using the transformed variables in their existing form. No additional differencing is required, and the subsequent dynamic analysis is therefore conducted with Δln(GEPU_current), Δln(BIST 100), and Δln(USD/TRY).

Note: The ADF tests include a constant, with lag length selected automatically according to the Schwarz Information Criterion and a maximum of 12 lags. The Phillips-Perron tests include a constant and apply Newey-West automatic bandwidth selection using the Bartlett kernel. The null hypothesis is that the examined series contains a unit root. All null hypotheses are rejected at the 1 per cent significance level.

3.3 Lag Length Selection and Baseline Model Specification

The appropriate lag order for the Türkiye specific VAR system was determined before estimating the baseline model. Since the data are monthly, the lag selection procedure evaluated up to 12 lags. Table 4 presents the results obtained from the sequential modified likelihood ratio statistic, the final prediction error criterion, the Akaike information criterion, the Schwarz criterion, and the Hannan Quinn criterion.

The selection criteria do not indicate a completely uniform lag choice. The final prediction error criterion, the Akaike information criterion, and the Hannan Quinn criterion identify one lag as the preferred specification. In contrast, the Schwarz criterion selects lag zero, while the sequential modified likelihood ratio statistic selects lag five. Since the central purpose of the article is to examine lagged predictive relationships among global policy uncertainty, Turkish stock returns, and exchange rate returns, a zero lag structure would not provide an appropriate basis for the required dynamic testing. At the same time, using five lags would substantially increase the number of estimated parameters in a relatively limited monthly sample.

Accordingly, VAR(1) is adopted as the baseline specification. This choice is supported by three of the reported criteria and preserves a parsimonious dynamic structure for subsequent Granger causality and block exogeneity testing. The baseline model is estimated using 130 observations after adjustment for the single lag.

Note: The endogenous variables are Δln(GEPU_current), Δln(BIST 100), and Δln(USD/TRY). The exogenous variable is a constant term. The lag selection procedure is based on 119 common observations because the assessment evaluates lags up to 12. An asterisk indicates the lag selected by each criterion. LR denotes the sequential modified likelihood ratio statistic; FPE denotes final prediction error; AIC denotes Akaike information criterion; SC denotes Schwarz criterion; and HQ denotes Hannan Quinn criterion.

3.4 Granger Causality and Block Exogeneity Results

Table 5 presents the Granger causality and block exogeneity Wald test results for the baseline VAR(1) specification. The findings reveal a clear difference between the equity and foreign exchange channels in Türkiye. When Δln(BIST 100) is treated as the dependent variable, the null hypothesis that Δln(GEPU_current) does not contain predictive information is rejected at the 5 per cent significance level. The chi-square statistic is 5.961110, with a probability value of 0.0146. This result indicates that lagged changes in global economic policy uncertainty contain statistically significant predictive information for Turkish stock market returns within the estimated system.

The corresponding evidence for the exchange rate channel is substantially different. When Δln(USD/TRY) is the dependent variable, the exclusion of Δln(GEPU_current) produces a chi-square statistic of 0.208394 and a probability value of 0.6480. The null hypothesis cannot therefore be rejected. Under the baseline specification, changes in global economic policy uncertainty do not provide statistically significant predictive information for USD/TRY returns. The contrast between these two findings suggests that the predictive relevance of global policy uncertainty is not uniform across Turkish financial market channels.

The results also indicate feedback predictability from Turkish stock market returns to the global uncertainty indicator. In the equation where Δln(GEPU_current) is the dependent variable, the exclusion of Δln(BIST 100) is rejected with a chi-square statistic of 10.00663 and a probability value of 0.0016. This result is interpreted cautiously as feedback predictability within the VAR system. It does not imply that movements in the Turkish stock market structurally determine global economic policy uncertainty.

No statistically significant predictive relationship is identified from Δln(USD/TRY) to either Δln(GEPU_current) or Δln(BIST 100), with probability values of 0.9974 and 0.9511, respectively. A weak indication is observed from Δln(BIST 100) to Δln(USD/TRY), where the probability value is 0.0955. Since this relationship is significant only at the 10 per cent level, it is not treated as a principal finding of the baseline analysis and will be reconsidered in the robustness assessment.

Note: The results are based on the baseline VAR(1) specification estimated with 130 observations. The null hypothesis is that the excluded variable does not contain predictive information for the dependent variable within the estimated system. Statistical significance indicates Granger predictive content and should not be interpreted as evidence of structural economic causality. The result from Δln(BIST 100) to Δln(GEPU_current) is therefore interpreted as feedback predictability only.

3.5 Diagnostic Test Results

The diagnostic results for the baseline VAR(1) specification are summarised in Table 6. The stability condition is satisfied, since all roots of the characteristic polynomial lie inside the unit circle. The largest modulus is 0.272131, indicating that the estimated dynamic system is stable.

The residual serial correlation LM tests provide no evidence of autocorrelation at the 5 per cent significance level. The null hypothesis of no serial correlation is not rejected at any individual lag from one to five. Similarly, the joint test covering lags one through five produces a probability value of 0.1703, indicating that residual serial dependence does not represent a material specification problem for the baseline model.

The residual normality evidence is less favourable. The joint Jarque-Bera test rejects the null hypothesis of multivariate normality, with a probability value of 0.0000. This rejection is driven primarily by the residuals of the Δln(USD/TRY) equation, which display pronounced positive skewness and substantial leptokurtosis. In contrast, normality is not rejected individually for the residual components associated with Δln(GEPU_current) and Δln(BIST 100).

Finally, the residual heteroskedasticity test with cross terms does not reject the null hypothesis of homoskedastic residuals. The joint test reports a probability value of 0.6835. Taken together, the diagnostic evidence indicates that the baseline model is dynamically stable and does not exhibit residual serial correlation or detectable heteroskedasticity. Nevertheless, the rejection of multivariate normality, concentrated in the exchange rate equation, requires cautious interpretation of system-level inference.

Note: The diagnostic tests are based on the baseline VAR(1) specification estimated with 130 observations. For the residual serial correlation LM test, the reported result corresponds to the joint test covering lags one through five. Residual normality is assessed using Cholesky orthogonalization. The rejection of joint residual normality is driven primarily by the Δln(USD/TRY) residual component. LRE* denotes the Edgeworth expansion corrected likelihood ratio statistic.

3.6 Robustness Check under the Alternative VAR(2) Specification

An alternative VAR(2) specification was estimated to examine whether the principal findings of the baseline model depend on the use of a single lag. Table 7 reports the Granger causality and block exogeneity results obtained under this alternative specification, together with the relevant stability and residual serial correlation evidence.

The principal predictive relationship identified in the baseline model remains statistically significant under VAR(2). In the equation for Δln(BIST 100), the exclusion of Δln(GEPU_current) produces a chi-square statistic of 7.246182 and a probability value of 0.0267. This finding is consistent with the baseline VAR(1) result, where the corresponding probability value is 0.0146. Accordingly, the evidence that global economic policy uncertainty contains predictive information for Turkish stock market returns does not depend exclusively on the one-lag specification.

The absence of significant predictive evidence in the exchange rate channel is also preserved under the alternative model. When Δln(USD/TRY) is treated as the dependent variable, the exclusion of Δln(GEPU_current) produces a probability value of 0.8572, compared with 0.6480 under VAR(1). Thus, neither specification provides evidence that monthly changes in global economic policy uncertainty contain statistically significant predictive information for USD/TRY returns during the sample period.

Feedback predictability from Δln(BIST 100) to Δln(GEPU_current) remains statistically significant under VAR(2), with a probability value of 0.0040. As in the baseline analysis, this result is interpreted only as predictive feedback within the estimated system and not as structural evidence that the Turkish equity market determines global economic policy uncertainty. In contrast, the weak indication from Δln(BIST 100) to Δln(USD/TRY) observed under VAR(1) does not persist in the alternative specification, where the probability value rises to 0.2502.

The alternative VAR(2) system also satisfies the stability condition, since all characteristic roots remain inside the unit circle and the largest root modulus is 0.368241. In addition, the residual serial correlation LM tests do not reject the null hypothesis of no serial correlation through lag five. The joint test covering lags one through five reports a probability value of 0.1250. These findings support the use of VAR(2) as a robustness assessment and confirm that the central distinction between the equity and foreign exchange channels remains evident under an alternative dynamic specification.

Note: VAR(1) is the baseline specification, while VAR(2) is estimated as an alternative dynamic specification for robustness assessment. The reported probability values for the predictive relationships are derived from Granger causality and block exogeneity Wald tests. Statistical significance indicates predictive content within the estimated VAR system and does not establish structural economic causality. Stability is satisfied when all characteristic roots lie inside the unit circle. The reported residual serial correlation result refers to the joint LM test covering lags one through five.

4. Discussion

The empirical evidence identifies a clear difference between the two Turkish financial market channels examined in this study. Changes in global economic policy uncertainty contain statistically significant predictive information for BIST 100 returns under the baseline VAR(1) specification, and this result remains evident under the alternative VAR(2) model. In contrast, no corresponding predictive relationship is detected for USD/TRY returns under either specification. The central finding is therefore not that global policy uncertainty is reflected uniformly across Turkish financial markets, but that its predictive relevance during the sample period is concentrated in the equity market channel.

The equity market evidence reported in this article can be considered alongside the Turkish findings of İlhan and Bağcı [5], although the two studies address different forms of market adjustment. Using an ARDL bounds testing framework for January 2013 to June 2024, they report that the short-run relationship between GEPU and BIST 100 is weak or insignificant, while a negative relationship becomes evident in the long run. The present study does not estimate a long-run equilibrium relationship and therefore does not seek to confirm or reject that conclusion. Instead, it shows that lagged monthly changes in GEPU_current contain statistically significant predictive information for BIST 100 returns within a Türkiye-specific VAR system over January 2010 to December 2020. Read together, the findings indicate that the relevance of global policy uncertainty for the Turkish equity market may depend on the return measure, the time horizon, and the econometric framework used.

The significance of the stock market result can be understood in relation to the forward-looking character of equity pricing. Stock market returns may respond to shifts in international uncertainty through revisions in expected profitability, investment conditions, risk appetite, and the required compensation for holding risky assets. Within the present analysis, the BIST 100 appears to incorporate information associated with changes in global policy uncertainty in a manner that is statistically detectable in subsequent monthly returns. Nevertheless, the VAR framework used here identifies predictive information rather than a structural causal mechanism. The result therefore indicates a dynamic statistical relationship that is relevant for market interpretation, without establishing that changes in GEPU_current mechanically determine Turkish equity returns.

The absence of statistically significant predictive evidence for USD/TRY returns requires equally careful interpretation. It does not imply that the Turkish lira is insulated from international uncertainty or that exchange rate movements are unaffected by global financial conditions. Rather, the result indicates that, within the monthly sample period and the specific three-variable VAR system estimated in this article, past changes in GEPU_current do not add statistically significant predictive information for subsequent USD/TRY returns. Exchange rate movements may reflect a wider set of influences, including domestic monetary conditions, inflation expectations, country-specific risk developments, external financing pressures, and market interventions, which are not separately identified within the present specification.

A further result concerns the statistically significant predictive relationship from BIST 100 returns to GEPU_current. This finding remains visible in both VAR(1) and VAR(2). Since GEPU_current is a global measure of economic policy uncertainty, the result should not be interpreted as evidence that developments in the Turkish equity market determine global uncertainty. A more defensible interpretation is that movements in Turkish equity returns may contain information associated with wider uncertainty conditions already developing in international markets, or that both variables respond dynamically to information not separately modelled in the estimated system. Accordingly, this relationship is treated as feedback predictability rather than structural causality.

The robustness assessment reinforces the main conclusion of the study. Increasing the dynamic specification from one lag to two lags does not eliminate the predictive relationship from GEPU_current to BIST 100 returns, while the absence of a significant relationship from GEPU_current to USD/TRY returns is preserved. Moreover, both specifications satisfy the stability condition and do not exhibit statistically significant residual serial correlation through five lags. These results indicate that the distinction between the equity and exchange rate channels is not merely an artefact of the baseline lag choice.

The diagnostic evidence nevertheless introduces an important qualification. Although the baseline VAR(1) model is stable and shows no detectable residual serial correlation or heteroskedasticity, the hypothesis of joint residual normality is rejected. This rejection is attributable primarily to the USD/TRY equation, whose residuals display pronounced asymmetry and fat-tailed behavior. This pattern is consistent with the descriptive characteristics of the exchange rate return series, which also exhibits substantial non-normality. The findings relating to the foreign exchange channel must therefore be interpreted with particular care, while the principal equity market result remains supported across both estimated lag specifications.

Taken together, the results contribute to the analysis of uncertainty and financial markets by showing that a common global uncertainty indicator need not exhibit the same predictive relevance across different financial channels within the same emerging economy. In the Turkish case examined here, the evidence is stronger and more persistent for stock market returns than for exchange rate returns. This distinction is important for empirical research because it cautions against treating equity and currency markets as interchangeable indicators of uncertainty transmission. It is also relevant for investors and market analysts, since the financial variable most responsive in predictive terms may depend on the channel through which changing international uncertainty is reflected in domestic market behavior.

5. Conclusion

This study examined whether changes in global economic policy uncertainty contain predictive information for Turkish stock market and exchange rate returns within an integrated financial market system. Using monthly observations from January 2010 to December 2020, the analysis employed first log-differenced series for GEPU_current, BIST 100, and USD/TRY. The empirical strategy was based on stationarity testing, a Türkiye-specific VAR model, Granger causality and block exogeneity tests, diagnostic evaluation, and an alternative VAR(2) specification for robustness assessment.

The findings indicate that global economic policy uncertainty carries statistically significant predictive information for Turkish stock market returns. Under the baseline VAR(1) specification, lagged changes in GEPU_current significantly predict Δln(BIST 100), and this result remains statistically significant when the model is re-estimated under VAR(2). In contrast, no significant predictive relationship is identified from Δln(GEPU_current) to Δln(USD/TRY) in either specification. The results therefore do not support a uniform financial market response to global policy uncertainty in Türkiye. Instead, they indicate that predictive relationships differ across market channels, with more persistent evidence observed in the equity market than in the foreign exchange market.

The study contributes to the empirical literature by examining Turkish stock and exchange rate returns jointly within a single national VAR system rather than treating the two channels separately or embedding Türkiye within a broader cross-country model. This approach provides a direct comparison of whether the same global uncertainty indicator carries similar predictive information for two important domestic financial market variables. The findings suggest that the BIST 100 may reflect information associated with changing global policy uncertainty more clearly in monthly return dynamics than USD/TRY during the period considered.

Several limitations should be acknowledged. The analysis is based on monthly data and a sample ending in December 2020, and therefore does not capture subsequent episodes of global and domestic financial stress. The model uses GEPU_current as the sole uncertainty indicator and does not separately identify domestic policy uncertainty, monetary policy conditions, inflation expectations, or other country-specific determinants of Turkish market returns. In addition, residual normality is rejected in the baseline system, primarily because of the distributional properties of USD/TRY residuals. Future research may extend the analysis using more recent data, alternative uncertainty indicators, higher-frequency observations, or modelling approaches designed to account more directly for volatility clustering, non-normality, and possible regime-dependent behaviour in Turkish financial markets.

Acknowledgement

The authors received no external financial support for the preparation of this article.

Conflict Of Interest

The authors declare that there is no conflict of interest regarding the publication of this article.