Journal homepage: https://resources.jaze.com.ng/index.

php/jaze/login

Article: https://bit.ly/JazeIvol.3ssue2 Journal of Arid Zone Economy 3(2): (2024) 11–24 ISSN: 2992–4952

Forecasting the Turkish Lira Exchange Rate against the U.S. Dollar

Fatima Muhammad1, Abdulkabir N. Adedeji2 and Mehdi Seraj1

1
Department of Economics, Near East University, Cyprus
2
Department of Economics, University of Maiduguri. Nigeria

Abstract
The link between exchange rates and macroeconomic variables have been
extensively studied in the literature. However, exchange rate forecasting
Keywords:
has long been a quest for academics and empiricists. This study forecasts
Exchange rate the Turkish Lira (TL) exchange rate in relation to US dollar, over a period
Macroeconomic variables 01/01/2010 to 03/04/2023, applying autoregressive integrated moving
Turkish Lira average (ARIMA) method. The out-sample data set used for prediction
US dollar covers 31/12/2022 to 31/12/2023. The Lira exchange rate is integrated of
COVID-19 order one, making ARIMA technique suitable for the analysis. We found
that Turkish Lira is volatile as there is no deviation between USD-YL
JEL Classification: exchange rate in the first quarter of 2023 and forecast is exact towards
O16; O11; H51 the end of the quarter. However, the deviation increased shortly towards
the second quarter of 2023. We affirm that, based on the findings for given
Corresponding author the trend between the two currencies, the exchange of TL against USD is
Abdulkabir N. Adedeji likely to depreciate in the next half period. This divergence, though not
abidatullah@gmail.com wide, could be attributed to the impacts Russian-Ukraine war and
Tel: +2348075510567 COVID-19 on Turkish economy.

1. Introduction efficient and informed decision making

The link between exchange rates and (Asadullah et al., 2021a; 2021b). As a result,
macroeconomic variables have been forecasting future trend exchange rate of local
extensively studied in the literature. However, currency against foreign currencies for
exchange rate forecasting has long been a quest mitigating economic loss, and for reaching
for academics and empiricists. Although a wide implications on the economy is critical. Clear
range of theoretical and empirical models has forecasting aids in the proper price of financial
been developed over the years to estimate and instruments as well as improved hedging.
predict exchange rate behaviour. In recognition Precise predictions are also used in value-at-
of the multidirectional benefits, academic risk models. Globally, countries adopt pegged
researchers devote considerable attention to or floating exchange rate systems, and the
projecting exchange rates. Exchange rate impacts on core macroeconomic indices
forecast increases precision and it can lead to overtly received considerable attention (Goh et
12 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

al., 2020). This is because the impact of the investigate the variables influencing the
exchange rate on these indicators is important Turkish Lira exchange rate.
for both developed and developing economies. The volatility trend of the Turkish Lira has
In the literature, the exchange rate may not been comprehensively investigated. This
have a positive significance on the economy has made it difficult to determine whether
and its benefits can only be attained if the local pegged or floating exchange rate system would
currency is competitive and stable to predict its be better for economic growth and
trend. The strength of a country's economy development in Turkey. A single non-linear
more often depends on the impacts of its model cannot adequately capture a time series
exchange rate. This relies on the ability of the variable that is highly volatile like Turkish
government, in a pegged system, to predict a Lira. This study applies nonlinear time series
growth-led exchange rate, while forces of models built on autoregressive integrated
demand and supply determine the currency rate moving average (ARIMA), developed by
changes in a floating exchange rate system. Grillenzoni (1993) and applied in Khashei et al.
Either of the systems is commonly seen in (2009). The ARIMA model in this study is
many emerging economies and its prediction is specified following Poon and Granger (2003).
seen to be an important tool for financial and ARIMA is seen to be not only an
management decision-making (Majhi et al., appropriate model to identify, estimate and
2009; Moosa, 2000; Hussain et al., 2019). Not forecast time series variables that are dynamic
only its unpredictability puts foreign currency in nature but also has the power to detect the
traders at risk but also the general economy and presence of nonlinear relationships in the
a challenge to many countries in the world. dynamics of the system Grillenzoni (1993). In
Turkey is not exempted as the country this regard, the method is applied for testing the
faces some economic issues, including a trade forecasting performance of the Turkey Lira-
imbalance and shortfall of foreign exchange U.S. dollar exchange rate to determine and
reserves, which could be traced to its exchange predict the exchange rate of the Turkey Lira
rate volatility. The Turkish Lira's exchange rate against the U.S. dollar over short and long
against the U.S. dollar has moved swiftly and horizons. Could short horizon accurately
drastically in recent years. The Turkish Lira's predict the Tukey Lira-U.S. dollar exchange
volatility was previously under control. The rate relationship? This question still remains
Turkish Lira was valued at 2.366 per U.S dollar unanswered. Hence, this study aims to forecast
in the first quarter of 2012, varied to 4.56 at the the exchange rate of the Turkish Lira against
end of the fiscal year 2017. The currency the U.S. dollar using the ARIMA model. This
depreciated by 2.2 units over the last decade, study provides insights into the trend between
indicating 93% depreciation of which average Turkish Lira against the U.S. dollar and
was about 13% per year. The Turkish Lira's forecasts the exchange rate between the two
value has been unpredictable and highly currencies. This may help the authority for
volatile since 2017. For example, the Lira making informed decision on exchange rate
exchange rate against the US dollar has policy. Such intervention may help for
continuously trending upward since the first stabilizing economic activities in the country.
quarter of 2020, the Turkish Lira reached an More importantly, this study suggests a
all-time high of 20.04 against the U.S. dollar convertibility rate that can reduce market risk
probably owing to decrease in export, and in banking, import and export, and currency
government transfer, and high credit volume and bureau de change trade, which may have a
and risk of non-payment of loans. However, the considerable impact on capital flows in Turkey.
Turkish Lira appreciated significantly the next
day and reverted to 12.84 (Dineri and Çütçü, 2. Literature review
2020). The variation in this situation is more The trade balance-exchange rate relationship
than 35% in a single day, forcing academics to can be seen through elasticity approach or the
13 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

model of imperfect substitutes. This assertion investigates the effect of interest and exchange
is based on the axiom that imported goods and rates volatilities on money demand in
domestic competing goods are imperfect Equatorial Guinea, Gambia, Nigeria and
substitutes based on the Marshall-Learner Uganda, using time series quarterly data,
condition (Boyd, Caporale, & Smith, 2001). covered the sample period 1993 to 2012,
The export-led growth theory states that a applied ordinary least squared. The author
favorable foreign exchange enables to import discovered that exchange rate did not have
capital goods into local economy which in turn significant effects on money demand in
increases production, consequently promotes developing countries owing to a possible
local economy. Generally, according to unstable economy in the continent. The author
Chimhore and Chivasa (2021), exchange rate is offered a policy recommendation that the
viewed as an important monetary policy monetary authorities in these countries should
instrument as it affects the decision-making of resort to inflation targeting monetary policy
investments in an economy. In addition, and employ the interest rate as the policy
Paksay, Genc and Semic (2018) stated that instrument. More recent, Jamal and Bhat
exchange rate volatility affects macroeconomic (2022) analyzed the nexus between the
variables such as interest rates, money supply, COVID-19 and the Rupee/USD, Pound/USD,
employment, inflation, and export growth. Yuan/USD, Real/USD, Lira/USD, and
Recently, Kayamo (2021) retraced exchange Euro/USD exchange rates using daily data
rate as nominal real exchange rate. The author spanned from March 11, 2020 to December 31,
defined nominal exchange rate as the price of a 2020, applied the panel ARDL model. They
country’s currency in terms of another country, confirmed that COVID-19 positively and
whereas he defined real exchange rate as significantly impact the sampled economies’
adjusted nominal exchange rate for relative exchange rates, and concluded that these
local price difference. He supported that economies might be the epicenters of the
exchange rate is an important macroeconomic COVID-19 crises which has a significant effect
variable that determines the country’s on the exchange rates in these epicenters.
economic growth and development as it often The combination of models to predict the
as significant impacts on aggregate output, exchange rate is wide spread in the literature
consumer price, and trade openness. Thus, and commonly used is the integrated
exchange rate is considered as driven engine of ARMA model, known as machine-learning
growth and development of any economy. model. Okoroaforh et al (2018) employed
In previous studies, some body of studies ARIMA model in forecasting inflation in
have examined the impact of exchange rate on Nigeria using annual data 1961 to 2016. They
different macroeconomic variables, using array found that ARIMA (1, 1, 1) is best for
of methods. For instance, Alsamara and Mrabet forecasting inflation in Nigeria. Out of sample
(2018) examined the asymmetric impacts of (2017-2020) is forecasted using the stochastic
foreign exchange rate on the demand for and dynamic method. The ARMA (1, 1)
money in Turkey, using quarterly data from forecast inflation rate in Nigeria for the year
1986:Q1 to 2014:Q4, and employed nonlinear 2017 to 2020 were found to be 16.4%, 16.7%,
ARDL for their analysis. They found that 16.5% and 16.4% on average respectively.
exchange rate has a strong relationship with Matroushi (2011) proposed a forecasting
money demand with appreciation effect was model that included two models. The study
found to be stronger than the depreciation hypothesized that ARIMA-MLP delivers better
effect. The study recommended that monetary results than other models, whether single or
policy decision makers should achieve more coupled, by combining ARIMA-ANN with
stable exchange rates to stabilize aggregate MLP.
price in Turkey. Among the studies that
focused on Africa, Nyumuah (2018)
14 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

Despite the fact that exchange rate the optimal model in forecasting China’s GDP
depreciations appear to have a substantial over the period 2020 and 2021. They predicted
influence on resolving balance-of-payments in that the value between the forecast GDP and
the economy. It has been deduced that the real GDP over 2020 and 2021 was actually
relative success in achieving this goal depends small. They affirmed that ARIMA model can
on a number of factors on the country's better reflect the development trend and
competitive capacity to meet any shift in accurate for short-term forecast.
demand for and supply of locally produced Similarly, Ma et al (2018) explored the
goods (Boateng et al., 2020). Perhaps, the most appropriateness of dynamic models for
critical condition for resource-dependent forecasting GDP in China and they offered
economies is that exchange rate depreciation some unique results of the appropriateness of
does not guarantee their trade balance. In other dynamic models. The authors pushed their
perspective, exchange rate is seen to have a argument that while basic economic theories
significant effect on stock market in open are undeniably, traditional economic models
economies. And that its volatility affects the may not be efficient enough to provide reliable
stock prices as it influences the cash inflows predicting outcomes for dynamic variables.
and outflows and foreign competitiveness of The traditional economic models are built on a
firms. Arguably, a growth-led exchange rate linear assumption but poorly performs if non-
often attracts cash flows of foreign currency linear trend exists. The authors found that
into local economy, and as such, forecasting ARIMA (1, 1, 0) displayed a good dynamic
such exchange rate is required to predicting the model fit as the fitted values and the residual of
inflow in the future which better predict by a the model correspond to the actual data,
moving-average technique. In Adeniyi and showing that forecast value was closed to the
Kumeka’s (2019) study, the symmetry and actual value.
asymmetry of the exchange rate-stock price Using ARIMA, Wang et al. (2016)
nexus for 54 firms listed on the Nigerian Stock projected currency parity from 2010 to 2013,
Exchange (NSE) were examined. Employing the procedure was completed using US/Euro
asymmetric Auto Regressive Distributed Lag rates. The results show that the linked models
(ARDL) model, using daily data for the period chosen outperformed the linear model. In
from December 12, 2001 to December 8, 2017, Shahriari’s (2011) and Nouri et al.’s (2011)
they found that exchange rate movements have studies, it was discovered that mixed
impacts on stock prices in many of the firms techniques generate better-
under the study. This supports the notion that a predicting outcomes than separate models.
change in the local exchange rate is beneficial Lam et al. (2008) and Altavila and Grauwe
and makes locally produced commodities and (2008) conducted similar studies with various
make it more competitive in the foreign local currencies. The analysis revealed that
market. Since exchange rate depreciation using mixed models generates better results
causes import costs to be highly correlated with than using a single model. In Dunis and Chen’s
local outputs, and that, this may enhance local (2006) study, 16 models simulated and the
currency competitiveness, reduce imports and results demonstrated that all the models were
increase exports. not successful but the results showed that
Traditionally, random walk models were mixed model outperforms all other models.
often applied for predictions with out-of- The volatility and different time series were
sample methodologies. However, Ince and incorporated in the mixed model, implying that
Trafalis (2006) applied ARIMA to forecast a traditional model will perform low when series
broad range of currency exchange rates and in not stationary.
found that the random walk models were less In sum, a body of studies have worked on
efficient than ARIMA models. More recently, combining ARIMA, machine learning, and
Liu et al. (2022) selected ARIMA (1, 1, 1) as time series models for forecasting exchange
15 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

rates. However, little is known about the trend where ∆ 𝐷.𝐸𝑋𝐶𝑡 denotes D-th differenced of
of Turkish Lira exchange rate against US. USD-YL exchange rate, ∆ 𝐷.𝐸𝑋𝐶𝑡−1 denotes
Dollar and their forecast behavior, one of its past values of the series, 𝜀𝑡 represents white
kind that has not been given much attention in error noise, assume to be constant, an
the literature. This may be necessary for uncorrelated process with mean zero, 𝜀𝑡−1
making informed decisions that may help for represents the past values of the error term. 𝛼
economic growth of the country. Thus, this is constant, ∅1 and 𝜑1 are parameters to be
study used univariate time series of Turkish estimated. The lag operator of the ARIMA (p,
Lira and applied non-Linear ARIMA technique d, q) can be written as:
to predict behavior of the currency. Further,
this research applies a grid search to choose the ∅ ∗.(𝐿)𝐸𝑋𝐶𝑡 = ∅(𝐿)(1𝐿) 𝐷.𝐸𝑋𝐶𝑡
best ARIMA model outperform independent
models for the forecast. = 𝛼 + 𝜑(𝐿)𝜀𝑡 2

3. Methodology where ∅ ∗.(𝐿)𝐸𝑋𝐶𝑡 is non-stationary AR

The US-TL exchange rate daily series data, operator polynomial with unit roots. The
covering January 01, 2010 April 03, 2023, ∅(𝐿)(1 − 𝐿)𝐷 eliminates the unit root process,
were collected for forecasting the trend of the where ∅(𝐿) is a stationary degree p AR lag
Turkish Lira (YL) exchange with US dollar operator polynomial, and that 𝜑(𝐿) is an
(USD). The data were obtained from Central invertible degree q MA lag operator
Bank of the Republic of Türkiye. We employ polynomial. The analysis of these models are
autoregressive integrated moving average in five steps. First step, the stationarity level of
(ARIMA) for forecasting US-TL exchange the Lira exchange rate is identified, determined
rate. The method is commonly used in social by using autocorrelation function (ACF) and
sciences and it is art than science. Following partial autocorrelation function (PACF)
Ma et al. (2018), we specify autoregressive graphs. PACF measures collations between
integrated moving average (ARIMA), observations and its lags, while ACF measures
proposed by Box and Jenkins (1970), for the correlation between observations. ACF and
analysis. The ARIMA model has three PACF are used in this study to confirm the
specifications. The first specification is appropriateness of the model’s parameters and
autoregressive model (AR(p) that determines their coefficients are calculated to determine
the number of lag observations in the model. the decay and geometric trend. We equally
The second specification is the Integration (I) applied Augmented Dickey-Fuller (ADF) to
that indicates the order of single integer, or the test the random root of the data and identified
number of times the raw observations are its order of stationarity. Second step, we
differenced. The third specification is the determining the order of integration of the
Moving Average (MA(q)) model that indicates variable. Third step, as the Lira exchange rate
the size of the moving average, also referred to is volatile, we expect that the data are a non-
order of moving average. ARIMA model is stationary sequence, then we transformation
often seen to be parsimonious and can be used the data into first difference. The number of
for forecasting when the series is stationary. differences shows the order of integration.
ARMA (p,q) series that is integrated after Fourth Step, once the values of AR(p) and
being first differenced is denoted by ARIMA MA(q) are determined, we estimate ARMA
(p,D,q). The ARIMA (p,D,q) process can be model. The ACF and PACF graphs are
modeled as: carefully studied to determine the number of
ACF and PACF factors with remarkable
∆ 𝐷.𝐸𝑋𝐶𝑡 = 𝛼 + ∅1 ∆ 𝐷.𝐸𝑋𝐶𝑡−1 + 𝜀𝑡 significant level in selection of models. After
completing the process, we estimated eight (8)
+𝜑1 𝜀𝑡−1 1 models out of which choose the best model
16 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

with highest number of significant coefficients,

highest value of adjusted R2 and lowest values 4. Results and discussion
of Akaike info criterion (AIC) and Schwarz 4.1 Stationarity test
criterion (SBICI). Fifth step, we diagnose the The time sequence, diagram in Figure 1a,
optimal fitted by testing the residual. Then, we shows that the YL- USD exchange rate is
forecast YL- USD exchange rate between increasing with time, and has highest spike in
December 2022 and December 2023. The the first sixty days in 2023. This indicating that
choice for these periods characterized high the series is non-stationary sequence. When the
volatility of USD-YL exchange rate and period data is first differenced, the series become
where Turkish economy, like other economies stationary, as shown in Fig. 1b.
globally, is recovering from COVID-19
pandemic.

We equally applied the Augumented-Dickey- 4.2 ARIMA Model Identification

Fuller (ADF) method to further carry out unit We use E-views software, version 10, to
root test. The hypothesis that YL- USD has unit compute ACF and partial PACF graphs for
root, or has a random work, cannot be rejected USD-YL series are plotted in Table 2a and 2b.
at all levels of significance. Therefore, this We observed, Table 2a, that the ACF
series is non-stationary, this is shown in a coefficients of the USD-YL sequence show
graph, Fig. 1b. The results show that YL- USD presence of serial correlation of the series as all
increases with time. However, the first order lags, 1 to 36, are significant and PACF
difference of the series is stationary, Fig 1b. coefficients are significantly outside the
Thus, this data series belongs to a stationary standard error bands, exponential decay, and
series. The unit root test results show that the both the ACF and APCF do not have the same
p-value is less than 0.05, that is, it is significant pattern. Thus, we affirm that the series is not
at the level of 5%. Since the variable is stationary. After taken the first difference, the
stationary at the first difference (I(1)), the data series appears to be stationary. This is shown in
series has no unit root, that is, USD-YL Table. 2b, the Turkish currency appears to be
belongs to a stationary series. ARMA model significantly non-zero when the lag order is 1,
can be established. and basically in the confidence bound when the
lag order is greater than 2, so q lag is taken to
be 2. The PACF factor is significantly non-zero
when the lag order is equal to 1, and it is also
17 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

significantly different from zero when the lag subjective, to establish a more accurate model,
order is 8, so p lag is taken to be 8. This shows the range of values of p and q is appropriately
that the model is not AR or MA but an ARIMA relaxed, and multiple ARMA (p, q) models are
model. Considering that the judgment is very established.

Table. 1a. ACF and PACF of non-stationary series

Autocorrelation Partial Correlation AC PAC Q-Stat Prob
|******| |******| 1 0.786 0.786 2049.8 0.000
|**** | *| | 2 0.588 -0.080 3196.0 0.000
|**** | |*** | 3 0.583 0.387 4324.8 0.000
|**** | | | 4 0.581 0.011 5445.2 0.000
|**** | |** | 5 0.580 0.237 6563.1 0.000
|******| |***** | 6 0.771 0.670 8534.8 0.000
|******* |******| 7 0.963 0.785 11616. 0.000
|******| ***| | 8 0.770 -0.357 13587. 0.000
|**** | | | 9 0.579 -0.064 14702. 0.000
|**** | | | 10 0.575 -0.026 15802. 0.000
|**** | | | 11 0.574 0.014 16895. 0.000
|**** | | | 12 0.575 0.064 17993. 0.000
|******| |* | 13 0.763 0.124 19931. 0.000
|******* |** | 14 0.954 0.318 22961. 0.000
|******| *| | 15 0.764 -0.198 24904. 0.000
|**** | | | 16 0.575 -0.014 26007. 0.000
|**** | | | 17 0.571 -0.016 27092. 0.000
|**** | | | 18 0.568 -0.015 28166. 0.000
|**** | | | 19 0.569 0.033 29245. 0.000
|***** | | | 20 0.756 0.037 31149. 0.000
|******* |* | 21 0.945 0.165 34125. 0.000
|***** | *| | 22 0.755 -0.180 36026. 0.000
|**** | | | 23 0.568 -0.012 37103. 0.000
|**** | | | 24 0.563 -0.021 38163. 0.000
|**** | | | 25 0.562 0.024 39216. 0.000
|**** | | | 26 0.564 0.037 40277. 0.000
|***** | | | 27 0.748 -0.012 42145. 0.000
|******* |* | 28 0.935 0.102 45066. 0.000
|***** | *| | 29 0.747 -0.111 46933. 0.000
|**** | | | 30 0.562 0.012 47991. 0.000
|**** | | | 31 0.558 -0.007 49032. 0.000
|**** | | | 32 0.556 0.010 50067. 0.000
|**** | | | 33 0.557 0.008 51107. 0.000
|***** | | | 34 0.740 0.002 52943. 0.000
|******* | | 35 0.926 0.067 55815. 0.000
|***** | *| | 36 0.740 -0.072 57650. 0.000
18 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

Table 1b. ACF and PACF of stationary series

Autocorrelation Partial Correlation AC PAC Q-Stat Prob
| | | | 1 -0.003 -0.003 0.0200 0.888
|* | |* | 2 0.119 0.119 36.734 0.000
| | | | 3 -0.043 -0.043 41.635 0.000
| | | | 4 0.025 0.011 43.232 0.000
| | | | 5 -0.062 -0.053 53.260 0.000
| | | | 6 -0.011 -0.017 53.558 0.000
*| | | | 7 -0.069 -0.055 65.935 0.000
*| | *| | 8 -0.099 -0.103 91.315 0.000
*| | *| | 9 -0.093 -0.081 113.95 0.000
| | | | 10 0.009 0.024 114.16 0.000
| | | | 11 -0.000 0.012 114.16 0.000
| | | | 12 0.006 -0.008 114.26 0.000
| | | | 13 0.020 0.011 115.30 0.000
|* | | | 14 0.074 0.063 129.75 0.000
| | | | 15 -0.021 -0.036 130.86 0.000
| | | | 16 -0.009 -0.044 131.07 0.000
| | | | 17 0.029 0.025 133.22 0.000
| | | | 18 -0.001 -0.001 133.22 0.000
| | | | 19 0.013 0.017 133.67 0.000
| | | | 20 -0.029 -0.027 135.86 0.000
| | | | 21 -0.012 -0.010 136.25 0.000
*| | | | 22 -0.067 -0.046 147.96 0.000
| | | | 23 0.024 0.027 149.51 0.000
| | | | 24 -0.021 -0.018 150.65 0.000
| | | | 25 -0.016 -0.030 151.35 0.000
|* | |* | 26 0.102 0.122 178.35 0.000
| | | | 27 0.008 0.005 178.53 0.000
*| | *| | 28 -0.099 -0.141 204.12 0.000
| | | | 29 -0.019 -0.022 205.04 0.000
| | | | 30 0.017 0.036 205.80 0.000
| | | | 31 -0.001 -0.008 205.80 0.000
| | | | 32 0.012 0.009 206.19 0.000
| | | | 33 -0.053 -0.059 213.46 0.000
| | | | 34 -0.050 -0.037 219.96 0.000
| | | | 35 -0.033 -0.006 222.87 0.000
| | | | 36 0.012 -0.009 223.23 0.000

The order with ARIMA lags (2, 1, 8) and (8, 1, model for better prediction and forecast. As a
2) in autoregressive moving average pre- rule of thumb, a high Adjusted R-squared, and
estimation are performed on the processed low AIC, SC values and standard error (S.E.)
sample data. In grid search for parsimonious of regression are criteria for selecting the best
model, we also captured all significant lags: model (Grillenzoni, 1993). Generally, the
ARIMA (8, 1, 9), ARIMA (9, 1, 8), ARIMA (9, larger the Adjusted R-squared (coefficient of
1, 26), ARIMA (26, 1, 9), ARIMA (26, 1, 28), determination), and the smaller the AIC value
and ARIMA (28, 1, 26) as they possess and the SC value, and the residual variance, the
information about the differenced series and better the corresponding ARIMA (p,D,q)
they captured for identifying parsimonious model.
19 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

4.3 Model estimation model Equation 1 specified above with

Having affirmed that the series is stationary ARIMA lag operators and the results are
and integrated of order one, I(1), as shown by presented in Table 1 below.
ACF and PACF coefficients, we estimate the

Table 2. Estimation results of the ARIMA model with lags identified

The coefficients in all the eight models we substantially different, in the same range, with
estimated are statistically significant at the 0.01 other models. This indicates that model 5 can
significance level except constant. The fifth be chosen as a tentative model and used for
column, ARIMA model 5, appears to be forecasting. The suitability of the model is
parsimonious, for having highest 𝑅̅2 value and tested. The diagnostic of the residual and
the model’s AIC and SBIC values are not inverse roots of AR and MA are presented
below.

-2

2 -4

0 -6

-2

-4

-6
10 12 14 16 18 20 22

Residual Actual Fitted

Fig. 1a. Residual stability test Fig. 1b. Inverse root of AR/MA

Fig 1a shows that the null hypothesis of testing diagnostics tests, though perfect ARIMA
the residuals are white noise cannot be rejected, model is difficult, we proceed and forecast
indicating that the model is stable. Also, the using this model.
AR and MA roots, Fig. 1b, are lie inside the
circle, implying that the estimated model 5 is
stable. Since the model passed all the
20 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

5. Forecasting results showing that the ARIMA ((9,1,26) model fits

Model 5 was used for forecasting with 320 the data well in the held-out sample. In
iterations and 2588 observations. We predict addition, the time series of the actual USD-YL
the USD-YL exchange rate trend using data exchange rate, in-sample forecasts and the
points from January 2020 to December 2023 residuals are presented in Fig. 2a and Fig.2b.
and this out-sample data was evaluated using Both the dynamic and static forecast models
Root mean square error (RMSE) and Mean fall within the 95% confidence interval,
absolute error (MAE) metrics. The dynamic accounts for forecast uncertainty. This implies
RMSE and MAE values are given as 0.044 and that the ARIMA (9,1,26) model is adequate for
0.038, respectively. The values are quite small, USD-YL exchange rate forecasting.

20.8
Forecast: US_YLF
Actual: US_YL
20.4
Forecast sample: 12/31/2022 12/31/...
Adjusted sample: 1/12/2023 12/31/2023
20.0
Included observations: 354
Root Mean Squared Error 0.044108
19.6
Mean Absolute Error 0.038334
Mean Abs. Percent Error 0.203259
19.2
Theil Inequality Coefficient 0.001169
Bias Proportion 0.755319
18.8
Variance Proportion 0.217526
Covariance Proportion 0.027154
18.4
I II III IV Theil U2 Coefficient 6.922245
Symmetric MAPE 0.202986
2023

USD_TRYF ± 2 S.E.
Fig. 2a Dynamic forecast

18.94
Forecast: US_YLF
18.92 Actual: US_YL
18.90 Forecast sample: 12/31/2022 12/31/...
Adjusted sample: 1/12/2023 3/12/2023
18.88
Included observations: 36
18.86 Root Mean Squared Error 0.004891
18.84 Mean Absolute Error 0.004091
Mean Abs. Percent Error 0.021709
18.82
Theil Inequality Coefficient 0.000130
18.80 Bias Proportion 0.026224
18.78 Variance Proportion 0.000012
Covariance Proportion 0.973764
18.76
19 26 2 9 16 23 2 9 Theil U2 Coefficient 0.771412
Symmetric MAPE 0.021710
M1 M2 M3

US_YLF ± 2 S.E.
Fig. 2b. Static forecast
The prediction of the forecast series and the deviation between USD-YL exchange rate in
actual series are the same over the years of the first quarter of 2023 and forecast is exact
observations, though not exact but the future towards the end of the quarter. However, the
trend. We forecast for 365 days over a period deviation started to increase shortly when
31/12/2022 to 31/12/2023, Fig. 3. There is no Turkish economy enters the second quarter of
21 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

2023. The exchange rate of YL against USD is volatility of Turkish Lira, measured by stigma
likely to be depreciated more in the half period square, we first observe the estimated
of the year. This divergence, though not wide, coefficients which are all significant and
could be somewhat attributed to the impacts of positive, providing evidence of long memory
Russian-Ukraine war and COVID-19 on and persistence in the volatility processes.
Turkish economy. Turning to the analysis of

Fig 3. USD-TL Forecast

This result can be interpreted that the forecast COVID-19 pandemic on the country's
upward rising in the exchange rate in Turkey economy and the economic uncertainty in
may be traced to the thin openness of the general. The increase in the government
economy and dependence of the economy on expenditure in Turkey especially on healthcare
imports, among other factors caused by the during COVID-19 in 2020 coupled with tax
COVID-19 pandemic, which reflect on the holidays, decrease in revenues, fall in
current inflation rate in the country. This result aggregate production, decrease in tourism
is in line Kayamo (2021) who found that revenues, shrinkage in trade volume, among
exchange rate has impact on inflation and the others, possibly cause the exchange rate to rise
impact is more relatively reflect in consumer in Turkey. In addition, according to Alsamara
prices, and that, the increase in imports tends to & Zouhair Mrabet (2018) and Dineri and Çütçü
increase the rate of inflation. The same result (2020), the increase in the exchange rate in
found by Baraka et al (2022) that exchange rate Turkey negatively affects investment
depreciation deteriorates trade balance. expectations which slows the economy. The
Contrarily, Nigusse et al. (2019) found that, monetary authority of Turkish, thus, need to
among others, soaring of exchange rate reduces pay an urgent attention to its exchange rate and
consumer prices. However, the role of the target a float rate for sustaining and stabilizing
authority is crucial in determining the degree of the economy of the country. Possibly, such
the effect of exchange rate fluctuations. action will more liberalize the economy and
Nonetheless, our results support the Dineri and bring the exchange rate back to a point of
Çütçü’s (2020) findings who pointed that the equilibrium and control the rand in the face of
possible reason for increase in exchange rate in volatility of Turkish Lira.
Turkey is the consequences of weak The diagnostics of the forecast ARIMA
macroeconomic policies, adverse effects of model is presented in Table 3. It shows that the
22 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

ACF and PACF of the residuals fall within noise. Hence, it can be concluded that the
95% confidence limits and the residual is white model is adequate and valid.

Table 3. Diagnostic check of ARIMA (9,1,26) ACF and PACF

Autocorrelation Partial Correlation AC PAC Q-Stat Prob

**| . | **| . | 1 -0.207 -0.207 1.6757

.|. | .|. | 2 0.048 0.005 1.7676
.|. | .|. | 3 0.027 0.039 1.7972 0.180
. |*. | . |*. | 4 0.146 0.166 2.7050 0.259
.*| . | .*| . | 5 -0.162 -0.109 3.8691 0.276
.|. | .|. | 6 0.012 -0.059 3.8755 0.423
.|. | .*| . | 7 -0.061 -0.083 4.0521 0.542
.*| . | .*| . | 8 -0.114 -0.159 4.6865 0.585
. |*. | . |*. | 9 0.132 0.140 5.5675 0.591
.*| . | .|. | 10 -0.073 -0.014 5.8501 0.664
.|. | .|. | 11 0.023 0.029 5.8778 0.752
. |*. | . |*. | 12 0.075 0.100 6.1999 0.798
.*| . | .*| . | 13 -0.084 -0.153 6.6205 0.829
. |*. | . |*. | 14 0.160 0.179 8.2155 0.768
.*| . | .*| . | 15 -0.104 -0.106 8.9178 0.779
.|. | .|. | 16 0.015 -0.035 8.9324 0.835
Note: Q-statistic probabilities adjusted for 2 ARMA terms

the rand in the face of volatility of Turkish Lira.

6. Conclusion and recommendation And as such, this may equally strengthen the
In this study we apply ARIMA model forecast, balance of trade and international trade open of
a relatively mixed time series prediction Turkey, the multiplier effect of which may spur
method commonly used, to predict the future economic growth and development of the
trend of Turkish Lira against US dollar over a country. Second, a flexible exchange rate may
period 31/12/2022 to 31/12/2023. This paper is be planned as this may sustain Turkish
first of it kinds that forecast Turkish Lira economy and stabilize the consumer food
exchange rate to fill the sparse of Lira prices in particular rather than following
economic history. The USD-YL exchange rate restrictive exchange rate policy.
matched in the first quarter of 2023 and
forecast is exact towards the end of the quarter. References
However, we affirm that the deviations is large Asadullah, M., Bashir, A., & Aleemi, A. R.
when forecasting time scale is long but the (2021). Forecasting exchange rates: An
deviation starts to decay shortly when Turkish empirical application to Pakistani
economy enters the second quarter of 2023. rupee. The Journal of Asian Finance,
This could be partly attributed to the impacts of Economics and Business, 8(4), 339-347.
Russian-Ukraine war and COVID-19 on Asadullah, M., Uddin, I., Qayyum, A., Ayubi,
Turkish economy. S., & Sabri, R. (2021). Forecasting Chinese
Yuan/USD via combination techniques
The policy recommendations as inferred in this during COVID-19. The Journal of Asian
study are: first, the Turkish monetary authority Finance, Economics and Business, 8(5),
should target a float exchange rate system. 221-229.
Such action may possibly bring the exchange Alsamara, M. and Mrabet, Z. (2018).
rate back to a point of equilibrium and control Asymmetric impacts of foreign exchange
23 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

rate on the demand for money in Turkey: Grillenzoni, C. (1993). Multilinear models
new evidence from nonlinear ARDL. with time-varying parameters. Statistica,
International Economics and Economic 53(2), 207–230.
Policy, 1-32. https://doi.org/10.6092/issn.1973-
https://doi.org/10.1007/s10368-018-0421-y 1- 2201/935.
16. Hogan, L. I., Urban, P. J., & Anh, V. V. (1985).
Adeniyi, O. and Kumeka, T. (2019): Exchange A Vector Autoregressive Forecasting
Rate and Stock Prices in Nigeria: Firm- Model of the US / A Exchange
Level Evidence. Journal of African Rate. Australian Journal of
Business, 1-30. Management, 10(2), 47-65.
DOI:10.1080/15228916.2019.1607144. Hussain, I., Hussain, J., Ali Khan, A., & Khan,
Boateng, A., Claudio-Quiroga, G. and Gil- Y. (2019). An analysis of the asymmetric
Alana, L. A. (2020): Exchange rate impact of exchange rate changes on GDP
dynamics in South Africa, Applied in Pakistan: application of non-linear
Economics, 1-14. ARDL. Economic research-Ekonomska
DOI:10.1080/00036846.2019.1688245 1- istraživanja, 32(1), 3094-3111.
14. Ince, H., & Trafalis, T. B. (2006). A hybrid
Boyd, D., Caporale, G. M., & Smith, R. (2001). model for exchange rate
Real exchange rate effects on the balance prediction. Decision Support
of trade: cointegration and the Marshall– Systems, 42(2), 1054-1062.
Lerner condition. International Journal of Kandil, M., Berument, H., & Dincer, N. N.
Finance & Economics, 6(3), 187-200. (2007). The effects of exchange rate
Dunis, C., & Miao, J. (2006). Volatility filters fluctuations on economic activity in
for asset management: An application to Turkey. Journal of Asian
managed futures. Journal of Asset Economics, 18(3), 466-489.
Management, 7(3/4), 179. Kayamo, S. E. (2021) Asymmetric impact of
Dineri, E. and Çütçü, I. (2020). The COVID- real exchange rate on inflation in Ethiopia:
19 Process and the Exchange Rate a non-linear ARDL approach. Cogent
Relation: An Application on Turkey. Economics & Finance, 9(1), 1986931.
ResearchSquare, p. 1-19. DOI: 10.1080/23322039.2021.1986931.
https://doi.org/10.21203/rs.3.rs-49026/v1 Khashei, M., Bijari, M., & Ardali, G. A. R.
Chen, H., Liu, L., Wang, Y., & Zhu, Y. (2016). (2009). Improvement of auto-regressive
Oil price shocks and US dollar exchange integrated moving average models using
rates. Energy, 112, 1036-1048. fuzzy logic and artificial neural networks
Chimhore, M. and Chivasa, S. (2021). The (ANNs). Neurocomputing, 72(4-6), 956-
Effects of Exchange Rates on Zimbabwe’s 967.
Exports. Journal of Economics and Kutan, A. M., & Wyzan, M. L. (2005).
Behavioral Studies, 13(4), 8-16. Explaining the real exchange rate in
Faust, J., Rogers, J. H., Wang, S. Y. B., & Kazakhstan, 1996–2003: Is Kazakhstan
Wright, J. H. (2007). The high-frequency vulnerable to the Dutch
response of exchange rates and interest disease?. Economic Systems, 29(2), 242-
rates to macroeconomic 255.
announcements. Journal of Monetary Kühl, M. (2008). Excess Comovements in the
Economics, 54(4), 1051-1068. Foreign Exchange Market with an
Goh, T. S., Henry, H., & Albert, A. (2021). Application to the Euro-GBP-USD triplet.
Determinants and prediction of the stock Lam, L., Fung, L., & Yu, I. W. (2008).
market during COVID-19: Evidence from Comparing forecast performance of
Indonesia. The Journal of Asian Finance, exchange rate models. Available at SSRN
Economics and Business, 8(1), 1-6. 1330705
24 Muhammad et al./Journal of Arid Zone Economy 3(1): (2024) 11–24

Ma, L., Hu, C., Lin, R. and Han, Y. (2018). Okoroafor, D. O. K., Adeniji, S. O. and
ARIMA model forecast based on EViews Olasehinde, T. (018). Estimating and
software. IOP Conference Series: Earth Forecasting the Impact of Inflation on
and Environmental Science. 1-9. Economic Growth in Nigeria Using
doi:10.1088/1755-1315/208/1/012017. Threshold Analysis. Journal of Applied
Mahesh, K. T. (2005). Forecasting exchange Statistics. 9(1): 1-22.
rate: A univariate out of sample approach Liu, Y., Chen, X., Wang, Q., Zhu, L. and Zhou,
(Box-Jenkins Methodology). The IUP Z. (2022). Forecast and Analysis of
Journal of Bank Management, 4, 60-74. National GDP in China Based on ARIMA
Majhi, R., Panda, G., & Sahoo, G. (2009). Model. Academic Journal of Science and
Efficient prediction of exchange rates with Technology, 3(2): 78-83.
low complexity artificial neural network Poon, S. H., & Granger, C. W. J. (2003).
models. Expert systems with Forecasting volatility in financial markets:
applications, 36(1), 181-189. A review. Journal of economic
Moosa, I. A., & Korczak, M. (2000). The role literature, 41(2), 478-539.
of fundamentalists and technicians in Uddin, I., Siddiq, Q. U., Tabash, M. I.,
exchange rate determination. Economia Qayyum, A., & Asadullah, M. (2021).
Internazionale/International Forecasting Malaysian ringgit against US
Economics, 53(1), 97-106. dollar; individual models vs combined
Nouri, M., & Navidi, H. (2013). The exchange models. Academy of Accounting and
rate risk and non-oil export in Financial Studies Journal, 25(6), 1-10.
Iran. Quarterly journal of economic Shahriari, M. (2011). A combined forecasting
growth and development research, 3(9), approach to exchange rate
70-59. fluctuations. International Research
Nyumuah, F. S. (2018). The Impacts of Interest Journal of Finance and Economics, 79,
Rate and Exchange Rate Volatilities on the 112-119.
Demand for Money in Developing Taylor, J. B. (2001). The role of the exchange
Economies. International Journal of rate in monetary-policy rules. American
Economics and Finance, 10(3): 56-69. economic review, 91(2), 263-267.
Van der Ploeg, F. (2011). Fiscal policy and
Dutch disease. International economics
and economic policy, 8, 121-138.