**3. Methodology**

symmetric and asymmetric effects of exchanges rate on stock prices in Turkey is based on the perception, as expressed by [1, 2], that the stock prices can react positively or negatively to fluctuations in exchange rates. Determining the factors that cause movements in stock prices is very important and is of great interest to policy makers and investors. The role of exchange rate on stock prices is much more important for small open economies in particular emerging markets. There is no sufficient research evidence showing the links between foreign exchange rate and Turkish stock market. We believe that this study will fill the gap in the literature in

The rest of the chapter is organized as follows: Section 2 review the related literature; Section 3 describes data and methods applied; Section 4 presents empirical findings and discusses the implications of the analysis; and, finally, Section 5

The relationship between stock prices and exchange rates has been extensively studied by many researches. Some find positive association between the two [4, 5]

Studies on the relationship between exchange rate and stock prices in the literature can be summarized in different categories according to their empirical results. Firstly, there are some studies that find significant positive relationship between the two. For instance, the relationship between stock prices and exchange rates on financial, manufacturing, and services indices and fifteen sub-indices in Turkey investigated using Johansen cointegration test and the results show evidence that there is a long-run relationship among these indices and exchange rates. The results suggest that exchange rate exposure on financial and manufacturing industries have positive forex beta for the dollar exchange rate, but in terms of service industries there is negative forex beta [9]. A similar exercise undertook to investigate the effects of changes in foreign exchange on the stock returns on company level using panel data analysis. The results show evidence that changes in real exchange rate has positive and significant impact on stock returns in the manufacturing and trade

Secondly, there are some studies that find negative relationship between the two [6, 11]. For example, Akıncı and Küçükçayşı analyses the relationships between stock markets and exchange rates in 12 countries and finds that the exchange rate has negative effect on the stock market index [6]. Belen and Karamelikli investigates the causality between the exchange rates and stock returns in Turkey and finds no evidence supporting any causal relationship between the dollar exchange rate and the BIST-30 Index [11]. Tsai examine the relationship between stock price index and exchange rate in six Asian countries, namely Singapore, Thailand, Malaysia, the Philippines, South Korea, and Taiwan. Their results show that all countries in the study have negative the relationship between stock prices and exchange rates, which is in line with the portfolio balance effect [12]. Recently, the relationships between real exchange rate returns and real stock price returns in Malaysia, the Philippines, Singapore, Korea, Japan, the United Kingdom and Germany examined using dynamic conditional correlation (DCC) and multivariate generalized autoregressive conditional heteroskedasticity (MGARCH) models. The results show that there is a negative relationship between real exchange rate return

and real stock price return in Malaysia, Singapore, Korea and the UK [13]. Thirdly, there are some studies that find two-way causality between the exchange rate and stock prices [14]. For instance, Zeren and Koç examines the

others discover negative relations [6, 7] and even no relationship at all [8].

concludes the paper and provides policy implications.

*Linear and Non-Linear Financial Econometrics - Theory and Practice*

sectors between the years 2006–2014 [10].

this area.

**158**

**2. Literature review**

This study investigates symmetric and asymmetric effects of exchange rates on three major stock market indices in Turkey using four different models. Firstly, linear and nonlinear bivariate ARDL models are estimated where the exchange rates are the only determinant of stock prices. The linear models are used to capture the symmetric effects of exchange rate changes while the nonlinear models are applied to capture asymmetric effects of exchange rate changes on stock prices.

Following Pesaran et al. [20] and Shin et al. [21] we apply the following the bivariate model to account for cointegration between exchanges rate and stock prices in Turkey.

$$LnSP\_t = a + \beta LnEX\_t + \varepsilon\_t \tag{1}$$

where *a* is the drift component, *SPt* is the stock price index, *EXt* is the nominal effective exchange rate, and *ε<sup>t</sup>* is an error term. In order to estimate the short-run effects, the error correction form, proposed by Pesaran et al. [20] of the Eq. (1) can be written as follows:

As the interest rates are significant determinants of stock prices [25, 26], we use

From Eq. (5), the coefficients estimate we get are the only long run effects. In

*<sup>δ</sup>kΔLnEXt*�*<sup>k</sup>* <sup>þ</sup>X*n*<sup>3</sup>

þ *λ*3*LnIPIt*�<sup>1</sup> þ *λ*4*LnIRt*�<sup>1</sup> þ *λ*5*LnVIXt*�<sup>1</sup> þ *ut* (6)

The Eq. (6) give short-run as well as long-run estimates in one step, where *λ*1, *λ*<sup>5</sup> are the long run parameters, Δ are the first difference operator, *n* and *q* are the optimal lag lengths for each variable, and *ut* is the usual White noise residuals. The estimates of coefficients attached to first-differenced variables gives the short-run effects while the estimates of *λ*2–*λ*5 normalized on *λ*1 give the long-run effects. In order for the long-run estimates to be valid, the F test proposed by Pesaran et al. [20] is applied to joint significance of lagged level variables *λ*<sup>1</sup> ¼ *λ*<sup>2</sup> ¼ *λ*<sup>3</sup> ¼ *λ*<sup>4</sup> in equation [6] as a sign of cointegration. The F test has obviously new critical values depending on whether variables in the model are I(0) or I [1], and whether the model contains an intercept and/or a trend. Once the cointegration established, the long-run effects of exchange rates, industrial productions, interest rates and volatility index on stock prices are captured by the estimates of *λ*<sup>2</sup> � *λ*<sup>5</sup> normalized on *λ*<sup>1</sup> . The short-run effects are gathered by the estimates of the coefficients of the first differenced variables such as the short-run effects of industrial production index on stock prices are determined by *θk*. The lag length of the first differences in Eq. (6) is chosen according to the Schwarz Bayesian

The nonlinear multivariate ARDL models are constructed to assess the asym-

*<sup>δ</sup>*1,*<sup>k</sup>ΔPOSt*�*<sup>i</sup>* <sup>þ</sup>X*<sup>n</sup>*<sup>3</sup>

Where the exchange rate is replaced by new generated POS and NEG variables. Thus the nonlinearity comes from the two new variables where POS refers appreciation of home currency and NEG refers depreciation of the home currency.

In this chapter both linear and nonlinear ARDL models are estimated for bivariate and multivariate models by using monthly data over the period of 2003 M1 to

*k*¼0

*<sup>δ</sup>*2,*<sup>k</sup>ΔNEGt*�*<sup>i</sup>* <sup>þ</sup>X*<sup>n</sup>*<sup>4</sup>

*ϑkΔLnVIXt*�*<sup>i</sup>* þ *λ*1*LnSPt*�<sup>1</sup> þ *λ*2*POSt*�<sup>1</sup> þ *λ*3*NEGt*�<sup>1</sup>

*k*¼0

*θkΔLnIPIt*�*<sup>i</sup>*

(7)

*k*¼0

*ϑkΔLnVIXt*�*<sup>k</sup>* þ *λ*1*LnSPt*�<sup>1</sup> þ *λ*2*LnEXt*�<sup>1</sup>

*θkΔLnIPIt*�*<sup>k</sup>*

the short term (overnight) interest rates as a broad measure of financing costs. However, the effects of on stock prices are ambiguous [27, 28]. And finally, considering the international effects and theoretical predictions [29, 30], the volatility

*The Impact of Exchange Rates on Stock Markets in Turkey: Evidence from Linear…*

order to infer the short-run effects, the Eq. (5) need to be rewrite in an error correction modeling format proposed by Pesaran et al. [20]. Therefore, we follow Pesaran et al.'s [20] bound testing approach and consider the following error-

*k*¼0

*k*¼0

Criteria (SBC) where we consider a maximum lag length of six.

*<sup>β</sup>kΔLnSPt*�*<sup>i</sup>* <sup>þ</sup>X*<sup>n</sup>*<sup>2</sup>

*<sup>φ</sup>kΔLnIRt*�*<sup>i</sup>* <sup>þ</sup>X*<sup>n</sup>*<sup>6</sup>

metric effects of exchange rate changes on stock prices as follows:

*k*¼0

*k*¼0

þ *λ*4*LnIPIt*�<sup>1</sup> þ *λ*5*LnIRt*�<sup>1</sup> þ *λ*6*LnVIXt*�<sup>1</sup> þ *ut*

index is included in the model.

*DOI: http://dx.doi.org/10.5772/intechopen.96068*

*<sup>Δ</sup>LnSPt* <sup>¼</sup> *<sup>α</sup>* <sup>þ</sup>X*n*<sup>1</sup>

*<sup>Δ</sup>LnSPt* <sup>¼</sup> *<sup>α</sup>* <sup>þ</sup>X*<sup>n</sup>*<sup>1</sup>

*k*¼1

þX*<sup>n</sup>*<sup>5</sup> *k*¼0

**4. Empirical findings**

**161**

*k*¼1

þX*<sup>n</sup>*<sup>4</sup> *k*¼0

correction forms of multivariate model respectively:

*<sup>β</sup>kΔLnSPt*�*<sup>k</sup>* <sup>þ</sup>X*n*<sup>2</sup>

*<sup>φ</sup>iΔLnIRt*�*<sup>k</sup>* <sup>þ</sup>X*<sup>n</sup>*<sup>5</sup>

$$\Delta L n \text{SP}\_t = a + \sum\_{k=1}^{n1} \beta\_k \Delta L n \text{SP}\_{t-k} + \sum\_{k=0}^{n2} \delta\_k \Delta L n \text{EX}\_{t-k} + \gamma\_1 \ln \text{SP}\_{t-1} + \gamma\_2 \ln \text{EX}\_{t-1} + u\_t \tag{2}$$

By now, we basically assume that exchange rate changes have symmetric effects on stock prices, but it might be possible that the effects could be asymmetric. In order to assess whether exchange rate changes have asymmetric effects on stock prices, we decompose the exchange rates into its positive and negative partial sums. For example, there might be differences between increases and decreases of the short-run interest rates. The partial sum of positive values is computed by replacing negative values with zeros as *POS*<sup>þ</sup> *<sup>t</sup>* ¼ *Ln*EX<sup>þ</sup> *<sup>t</sup>* <sup>¼</sup> <sup>P</sup>*<sup>t</sup> <sup>j</sup>*¼<sup>1</sup>ΔLn*EX*<sup>þ</sup> *<sup>j</sup>* ¼ P*<sup>t</sup> <sup>j</sup>*¼<sup>1</sup> *max* <sup>Δ</sup>LnEX*j*, 0 � �, and the partial sum of negative values are computed by replacing positive values with zeros as *NEG*� *<sup>t</sup>* ¼ *Ln*EX� *<sup>t</sup>* <sup>¼</sup> <sup>P</sup>*<sup>t</sup> <sup>j</sup>*¼<sup>1</sup>ΔLn*EX*� *<sup>j</sup>* ¼ P*<sup>t</sup> <sup>j</sup>*¼<sup>1</sup> *min* <sup>Δ</sup>LnEX*j*, 0 � � where <sup>Δ</sup>*EX*<sup>þ</sup> *<sup>j</sup> is the* positive sum of changes in exchange rates, and Δ*EX*� *<sup>j</sup>* is the negative sum of changes in exchange rates.

The *LnEX* in Eq. (2) is replaced by new generated POS and NEG variables in the nonlinear ARDL models as follows:

$$LnSP\_t = a + \beta POS\_t + NEG\_t + \varepsilon\_t \tag{3}$$

Thus, the error correction form of the Eq. (3) takes the following form with POS and NEG variables.

$$\begin{split} \Delta \text{LnSP}\_{t} &= a + \sum\_{i=1}^{n1} \beta\_{i} \Delta \text{LnSP}\_{t-i} + \sum\_{i=0}^{n2} \delta\_{1,i} \Delta \text{POS}\_{t-i} + \sum\_{i=0}^{n3} \delta\_{2,i} \Delta \text{NEG}\_{t-i} + \lambda\_{1} \text{LnSP}\_{t-1} \\ &+ \lambda\_{2} \text{POS}\_{t-1} + \lambda\_{3} \text{NEG}\_{t-1} + u\_{t} \end{split} \tag{4}$$

Secondly, linear and nonlinear multivariate ARDL models are estimated where industrial production index (IP), volatility index (VIX) and interest rates (IR) are used as a determinants of stock prices in Turkey. In order to account the effect of these variables on stock prices we employ a linear multivariate model of Moore & Wang [22] and Bahmani-Oskooee & Saha [23] as follows:

$$
\ln \text{SP}\_t = \mathfrak{a} + \beta \text{LnEX}\_t + \gamma \text{LnPI}\_t + \delta \text{LnIR}\_t + \theta \text{LnVI}X\_t + \varepsilon\_t \tag{5}
$$

where *IPIt* is an index of industrial production, *IRt* is the short term (overnight) interest rates, *VIXt* is a measure of stock market volatility index and *ε<sup>t</sup>* is an error term. The coefficient sign of *β* could be positive or negative depending on the firm's international competitiveness and production costs due to depreciation in exchange rates. When firms gain international competitiveness, they export more and thus exchange rate affects stock prices positively. However, increased costs due to depreciation in exchange rate are expected to affect stock prices negatively. Since there is a common consensus that economic activities affect stock prices positively [23, 24], the industrial production index is used as a proxy for measuring economic activity. Thus, we can expect stock prices to increase through increasing industrial production. Thus, we can expect the coefficient sing of *γ* to be positive.

*The Impact of Exchange Rates on Stock Markets in Turkey: Evidence from Linear… DOI: http://dx.doi.org/10.5772/intechopen.96068*

As the interest rates are significant determinants of stock prices [25, 26], we use the short term (overnight) interest rates as a broad measure of financing costs. However, the effects of on stock prices are ambiguous [27, 28]. And finally, considering the international effects and theoretical predictions [29, 30], the volatility index is included in the model.

From Eq. (5), the coefficients estimate we get are the only long run effects. In order to infer the short-run effects, the Eq. (5) need to be rewrite in an error correction modeling format proposed by Pesaran et al. [20]. Therefore, we follow Pesaran et al.'s [20] bound testing approach and consider the following errorcorrection forms of multivariate model respectively:

$$\begin{split} \Delta Ln \text{SP}\_{t} &= \alpha + \sum\_{k=1}^{n1} \beta\_{k} \Delta Ln \text{SP}\_{t-k} + \sum\_{k=0}^{n2} \delta\_{k} \Delta Ln \text{EX}\_{t-k} + \sum\_{k=0}^{n3} \theta\_{k} \Delta Ln \text{IP}I\_{t-k} \\ &+ \sum\_{k=0}^{n4} \rho\_{i} \Delta Ln \text{IR}\_{t-k} + \sum\_{k=0}^{n5} \theta\_{k} \Delta Ln \text{VI} \text{X}\_{t-k} + \lambda\_{1} \text{LnSP}\_{t-1} + \lambda\_{2} \text{LnEX}\_{t-1} \\ &+ \lambda\_{3} \text{LnPI}\_{t-1} + \lambda\_{4} \text{LnIR}\_{t-1} + \lambda\_{5} \text{LnVI} \text{X}\_{t-1} + u\_{t} \tag{6} \end{split}$$

The Eq. (6) give short-run as well as long-run estimates in one step, where *λ*1, *λ*<sup>5</sup> are the long run parameters, Δ are the first difference operator, *n* and *q* are the optimal lag lengths for each variable, and *ut* is the usual White noise residuals. The estimates of coefficients attached to first-differenced variables gives the short-run effects while the estimates of *λ*2–*λ*5 normalized on *λ*1 give the long-run effects. In order for the long-run estimates to be valid, the F test proposed by Pesaran et al. [20] is applied to joint significance of lagged level variables *λ*<sup>1</sup> ¼ *λ*<sup>2</sup> ¼ *λ*<sup>3</sup> ¼ *λ*<sup>4</sup> in equation [6] as a sign of cointegration. The F test has obviously new critical values depending on whether variables in the model are I(0) or I [1], and whether the model contains an intercept and/or a trend.

Once the cointegration established, the long-run effects of exchange rates, industrial productions, interest rates and volatility index on stock prices are captured by the estimates of *λ*<sup>2</sup> � *λ*<sup>5</sup> normalized on *λ*<sup>1</sup> . The short-run effects are gathered by the estimates of the coefficients of the first differenced variables such as the short-run effects of industrial production index on stock prices are determined by *θk*. The lag length of the first differences in Eq. (6) is chosen according to the Schwarz Bayesian Criteria (SBC) where we consider a maximum lag length of six.

The nonlinear multivariate ARDL models are constructed to assess the asymmetric effects of exchange rate changes on stock prices as follows:

$$\begin{split} \Delta Ln \text{SP}\_{t} &= \alpha + \sum\_{k=1}^{n1} \beta\_{k} \Delta Ln \text{SP}\_{t-i} + \sum\_{k=0}^{n2} \delta\_{1,k} \Delta POS\_{t-i} + \sum\_{k=0}^{n3} \delta\_{2,k} \Delta N \text{ESG}\_{t-i} + \sum\_{k=0}^{n4} \theta\_{k} \Delta Ln \text{IP}\_{t-i} \\ &+ \sum\_{k=0}^{n5} \rho\_{k} \Delta Ln \text{IR}\_{t-i} + \sum\_{k=0}^{n6} \theta\_{k} \Delta Ln \text{VI} \text{X}\_{t-i} + \lambda\_{1} \text{LnSP}\_{t-1} + \lambda\_{2} \text{POS}\_{t-1} + \lambda\_{3} \text{NEG}\_{t-1} \\ &+ \lambda\_{4} \text{LnPI}\_{t-1} + \lambda\_{5} \text{LnIR}\_{t-1} + \lambda\_{6} \text{LnVI}\_{t-1} + u\_{t} \tag{7} \tag{7} \end{split}$$

Where the exchange rate is replaced by new generated POS and NEG variables. Thus the nonlinearity comes from the two new variables where POS refers appreciation of home currency and NEG refers depreciation of the home currency.

## **4. Empirical findings**

In this chapter both linear and nonlinear ARDL models are estimated for bivariate and multivariate models by using monthly data over the period of 2003 M1 to

where *a* is the drift component, *SPt* is the stock price index, *EXt* is the nominal effective exchange rate, and *ε<sup>t</sup>* is an error term. In order to estimate the short-run effects, the error correction form, proposed by Pesaran et al. [20] of the Eq. (1) can

By now, we basically assume that exchange rate changes have symmetric effects on stock prices, but it might be possible that the effects could be asymmetric. In order to assess whether exchange rate changes have asymmetric effects on stock prices, we decompose the exchange rates into its positive and negative partial sums. For example, there might be differences between increases and decreases of the short-run interest rates. The partial sum of positive values is computed by

*<sup>j</sup>*¼<sup>1</sup> *max* <sup>Δ</sup>LnEX*j*, 0 � �, and the partial sum of negative values are computed by

*<sup>j</sup>* is the negative sum of changes in exchange rates. The *LnEX* in Eq. (2) is replaced by new generated POS and NEG variables in the

Thus, the error correction form of the Eq. (3) takes the following form with POS

Secondly, linear and nonlinear multivariate ARDL models are estimated where industrial production index (IP), volatility index (VIX) and interest rates (IR) are used as a determinants of stock prices in Turkey. In order to account the effect of these variables on stock prices we employ a linear multivariate model of Moore &

where *IPIt* is an index of industrial production, *IRt* is the short term (overnight) interest rates, *VIXt* is a measure of stock market volatility index and *ε<sup>t</sup>* is an error term. The coefficient sign of *β* could be positive or negative depending on the firm's international competitiveness and production costs due to depreciation in exchange rates. When firms gain international competitiveness, they export more and thus exchange rate affects stock prices positively. However, increased costs due to depreciation in exchange rate are expected to affect stock prices negatively. Since there is a common consensus that economic activities affect stock prices positively [23, 24], the industrial production index is used as a proxy for measuring economic activity. Thus, we can expect stock prices to increase through increasing industrial

*<sup>δ</sup>*1,*<sup>i</sup>ΔPOSt*�*<sup>i</sup>* <sup>þ</sup>X*<sup>n</sup>*<sup>3</sup>

*ln SPt* ¼ *a* þ *βLnEXt* þ *γLnIPIt* þ δ*LnIRt* þ θ*LnVIXt* þ *ε<sup>t</sup>* (5)

*<sup>t</sup>* ¼ *Ln*EX<sup>þ</sup>

*<sup>t</sup>* ¼ *Ln*EX�

*δkΔLnEXt*�*<sup>k</sup>* þ *γ*<sup>1</sup> ln *SPt*�<sup>1</sup> þ *γ*<sup>2</sup> ln *EXt*�<sup>1</sup> þ *ut*

*<sup>t</sup>* <sup>¼</sup> <sup>P</sup>*<sup>t</sup>*

*<sup>t</sup>* <sup>¼</sup> <sup>P</sup>*<sup>t</sup>*

*<sup>j</sup> is the* positive sum of changes in exchange

*LnSPt* ¼ *a* þ *βPOSt* þ *NEGt* þ *ε<sup>t</sup>* (3)

*i*¼0

*<sup>j</sup>*¼<sup>1</sup>ΔLn*EX*<sup>þ</sup>

*<sup>j</sup>*¼<sup>1</sup>ΔLn*EX*�

*δ*2,*<sup>i</sup>ΔNEGt*�*<sup>i</sup>* þ *λ*1*LnSPt*�<sup>1</sup>

(4)

*<sup>j</sup>* ¼

*<sup>j</sup>* ¼

(2)

be written as follows:

*<sup>Δ</sup>LnSPt* <sup>¼</sup> *<sup>α</sup>* <sup>þ</sup>X*n*<sup>1</sup>

P*<sup>t</sup>*

P*<sup>t</sup>*

**160**

rates, and Δ*EX*�

and NEG variables.

*<sup>Δ</sup>LnSPt* <sup>¼</sup> *<sup>α</sup>* <sup>þ</sup>X*<sup>n</sup>*<sup>1</sup>

*k*¼1

replacing negative values with zeros as *POS*<sup>þ</sup>

replacing positive values with zeros as *NEG*�

*<sup>j</sup>*¼<sup>1</sup> *min* <sup>Δ</sup>LnEX*j*, 0 � � where <sup>Δ</sup>*EX*<sup>þ</sup>

nonlinear ARDL models as follows:

*i*¼1

*<sup>β</sup>iΔLnSPt*�*<sup>i</sup>* <sup>þ</sup>X*<sup>n</sup>*<sup>2</sup>

þ *λ*2*POSt*�<sup>1</sup> þ *λ*3*NEGt*�<sup>1</sup> þ *ut*

Wang [22] and Bahmani-Oskooee & Saha [23] as follows:

*i*¼0

production. Thus, we can expect the coefficient sing of *γ* to be positive.

*<sup>β</sup>kΔLnSPt*�*<sup>k</sup>* <sup>þ</sup>X*n*<sup>2</sup>

*Linear and Non-Linear Financial Econometrics - Theory and Practice*

*k*¼0

2018 M12 for three major stock market indices in Turkey. The results of short and long-run estimates of both linear and nonlinear for the bivariate and multivariate models are reported in **Tables 1** and **2**. Each of the tables consist of three panels: Panel A reports the short run estimates, Panel B reports the long-run estimates and the diagnostic statistics are then reported in Panel C. To ensure one of the requirements of Pesaran et al.'s [20] method that the variables could be I(0) or I [1] but not I [2], we use the traditional Augmented Dickey-Fuller (ADF) tests on levels as well as the first differenced variables. The lag order of the ADF test statistics is determined by the Akaike Information Criterion (AIC) and the results show that there are no I [2] variables.
