Modeling Tourism Demand in Turkey (2008–2024): Time-Series Approaches for Sustainable Growth

Abstract

Tourism is a critical sector for economic growth and cultural exchange, particularly for destinations like Turkey, which consistently attracts millions of visitors annually. This study investigates the dynamics of tourism demand in Turkey between 2008 and 2024, with a focus on seasonality, long-term trends, and predictive modeling accuracy. Time-series data were analyzed, and the impacts of economic indicators and digital search trends were evaluated using SARIMA and SARIMAX models. The results demonstrate that the SARIMA models outperformed the SARIMAX models, highlighting the dominance of intrinsic seasonal patterns over external regressors, such as exchange rates and inflation. The findings emphasize that geographic proximity and cultural similarities drive consistent tourist flows, while behavioral data like Google Trends provide supplementary insights into demand shifts. However, economic variables showed limited short-term predictive power. These results underscore the importance of prioritizing time-series structures in forecasting frameworks while complementing them with behavioral indicators for enhanced accuracy. This study contributes to the literature by addressing a critical gap in understanding how various factors influence tourism demand in Turkey and offers practical implications for policymakers and tourism planners to optimize strategic planning and resource allocation, ensuring sustainable tourism growth. Future research should explore hybrid models that incorporate sentiment-driven data and cultural factors for more robust forecasting.

1. Introduction

Tourism has long been recognized as a critical driver of economic growth, cultural exchange, and sustainable development. Particularly for countries like Turkey, which consistently ranks among the top global destinations, understanding the dynamics of tourism demand is essential for strategic planning and sustainable growth [1,2]. Turkey’s tourism sector provides significant contributions to foreign exchange earnings, employment, and infrastructural development [3,4]. However, the volatility in international tourism flows due to economic, geopolitical, and health-related disruptions highlights the need for accurate forecasting models and deeper insights into demand patterns.

Building on the critical role of tourism in Turkey’s economic and infrastructural development, it is essential to explore how global advancements and innovative forecasting techniques can further enhance the sector’s resilience and strategic planning. Tourism stands as one of the most significant contributors to global economic development, fostering cultural exchange and societal progress. The tourism sector not only serves as a vital economic engine by generating significant revenues and employment opportunities but also plays a critical role in enhancing international connectivity and understanding. According to Cankurt and Subaşı (2016), tourism has become the largest global industry, offering substantial economic value and employment opportunities worldwide [5]. The sector’s resilience, even amidst geopolitical tensions and economic fluctuations, reflects its capacity to adapt to changing consumer preferences and external challenges.

Globally, tourism is not only a significant economic sector but also a social phenomenon shaping the cultural and international identity of nations. Tourism contributes to mutual understanding, fosters cross-cultural exchange, and serves as a platform for soft power diplomacy [6]. The resilience of the tourism industry during disruptions such as the COVID-19 pandemic underscores its importance as both an economic and social asset [7]. Moreover, advancements in forecasting methodologies, such as incorporating behavioral and digital data, have greatly enhanced our ability to analyze tourism demand. This study builds upon these perspectives to provide a comprehensive framework for understanding Turkey’s tourism demand dynamics.

2. Literature Review

Numerous studies have explored the determinants of tourism demand and forecasting methodologies. Ulucak et al. [1] used gravity models to analyze international tourism flows to Turkey, emphasizing economic size, income levels, and geographic proximity as key drivers. Peng et al. [8] conducted a meta-analysis of international tourism demand elasticities, examining how factors such as origin, destination, and modeling approaches influence elasticity estimates. Their findings highlight the significance of macroeconomic indicators, including income levels and relative prices, in shaping tourism demand. Similarly, Rosselló Nadal and Santana Gallego [9] provided an empirical review of gravity models in tourism demand modeling, highlighting that GDP, population, and distance are the predominant determinants of tourist flows. Additionally, Höpken et al. [10] leveraged big data and artificial neural networks (ANNs) to improve forecasting accuracy, demonstrating the advantages of integrating machine learning techniques. De Luca and Rosciano [11] highlighted the utility of Google Trends data to capture dynamic demand shifts, particularly during periods of economic uncertainty. Moreover, Zhang et al. [12] emphasized the benefits of using a decomposed deep learning approach, which significantly improved forecasting accuracy by isolating seasonal and trend components in tourism data. Wickramasinghe and Ratnasiri [7] demonstrated that incorporating disaggregated search data can enhance forecasts by reflecting real-time consumer interest at a regional level. Similarly, Cankurt and Subaşı [5] utilized data mining techniques to model and forecast multivariate time series in tourism, showcasing the importance of advanced methodologies in improving predictive performance. Li et al. [13] developed deep learning models to forecast tourism volume with high accuracy, emphasizing the role of search engine data and online behaviors. Their findings revealed that integrating behavioral indicators enhances the robustness of predictions.

Seasonality has been widely analyzed as a dominant factor influencing tourism patterns. Yabanci [2] emphasized that seasonality in Turkey’s tourism sector poses significant challenges related to capacity utilization, resource allocation, and economic stability. Forecasting techniques such as SARIMA and SARIMAX have been widely acknowledged for their effectiveness in capturing seasonality and trends in tourism demand data. Gil-Alana et al. (2020) further demonstrated that tourism demand time series exhibit long-term persistence, meaning that seasonal fluctuations and shocks can have prolonged effects on visitor arrivals, emphasizing the importance of seasonality in forecasting models [14]. SARIMA models, in particular, are highly regarded for their ability to analyze internal patterns without requiring external variables, making them robust against data fluctuations [5,15]. Saayman and Botha (2015) similarly concluded that SARIMA models outperform alternative approaches in capturing cyclical patterns in tourism demand, particularly in destinations with strong seasonal trends [16]. For example, Tang et al. (2016) demonstrated how the SARIMA model excels in explaining intrinsic seasonality patterns in tourism demand for China, particularly under stable exchange rate conditions [17]. Vergori (2016) also emphasized that forecasting errors increase in destinations with highly seasonal tourism flows, further underscoring the challenges in predicting demand in countries like Turkey [18]. Hu et al. (2021) highlighted the importance of accounting for seasonal and holiday patterns in tourism demand forecasts, demonstrating that integrating these factors improves short-term predictive accuracy [19]. On the other hand, SARIMAX models extend this framework by incorporating exogenous regressors, such as economic indicators, to better explain complex relationships between tourism demand and external factors [7,15]. Duro and Prats (2016) found that economic variables such as income levels and exchange rates had a limited short-term impact on tourism demand in Catalonia, reinforcing the findings of this study that intrinsic seasonal trends dominate forecasting accuracy [20]. Wickramasinghe and Ratnasiri (2020) emphasized that SARIMAX models outperform SARIMA when supplemented with regionally disaggregated Google search data, enhancing forecasting precision for destinations affected by external disruptions such as the COVID-19 pandemic [7].

Moreover, combining SARIMA and SARIMAX with machine learning and hybrid approaches has proven to increase accuracy significantly. For instance, Park et al. (2021) employed a SARIMAX model enriched with news discourse analysis to capture dynamic shifts in tourist sentiment and behavior, which yielded superior forecasting outcomes during periods of political and economic uncertainty [21]. Zhang et al. [12] applied deep learning techniques to decompose trends and seasonality, demonstrating significant improvements in forecast precision. Additionally, Volchek et al. (2018) highlighted the importance of integrating high-frequency Google Trends data into traditional SARIMA models, demonstrating notable improvements in short-term demand forecasting for attractions [15]. These advancements underscore the critical role of incorporating external behavioral and sentiment-driven data to refine tourism forecasting models and address the industry’s dynamic and multifaceted nature.

Recent research highlights the use of Google Trends data as a complementary tool for forecasting tourism demand [11,22,23,24,25,26]. By capturing real-time search behavior, Google Trends provides insights into tourists’ intentions and emerging patterns, enhancing model accuracy when combined with SARIMA and SARIMAX [7,10]. Such hybrid approaches leverage both structured time-series data and unstructured behavioral data to improve predictive performance [15,21]. Xiao et al. (2020) proposed an advanced EEMD-DBN (Ensemble Empirical Mode Decomposition–Deep Belief Network) approach, demonstrating that integrating Google Trends into forecasting models significantly improved prediction accuracy for inbound tourism to Shanghai [23]. Similarly, Park et al. (2017) used Google Trends to enhance short-term predictions of Japanese tourist inflows to South Korea, finding that models incorporating search query data performed better than traditional time-series approaches [24]. Additionally, Dinis et al. (2019) systematically reviewed the use of Google Trends in tourism and hospitality research, underscoring its value in demand forecasting [25]. Önder (2017) [26] showed that integrating Google Trends data, including web and image search indices, into forecasting models significantly improved prediction accuracy for major European destinations, further emphasizing the value of digital tools in tourism analytics [26]. Wickramasinghe and Ratnasiri [7] demonstrated the advantages of geographically disaggregated search data in refining regional forecasts, while Volchek et al. [15] emphasized the importance of higher-frequency search data for short-term predictions. Additionally, Park et al. [21] highlighted the utility of online news data for understanding shifts in tourism sentiment, illustrating the growing potential of digital and big data in forecasting efforts. Together, these studies underscore the importance of integrating digital behavioral indicators to enhance tourism demand forecasting models. Furthermore, recent studies indicate that social media data, particularly from platforms such as Twitter, can serve as valuable real-time indicators of tourism demand. Mendieta-Aragón et al. [27] found that integrating Twitter-based sentiment data into SARIMAX models improved the forecasting accuracy of tourism flows to Santiago de Compostela.

Studies have also addressed the role of economic indicators, such as exchange rates and inflation, in shaping tourism demand. Akarsu [3] examined the interrelationships between tourist arrivals, economic growth, and inflation in Turkey, finding that exchange rates and inflation impact tourism demand differently depending on geopolitical stability. Similarly, Bilgili et al. [28] highlighted that while exchange rates positively influence tourism demand, the effect of inflation can vary depending on the market segment and tourists’ sensitivity to price fluctuations. Kaymaz (2022) highlighted that macroeconomic factors, including exchange rates and unemployment, play a significant role in shaping tourism demand in Turkey [4]. Furthermore, Kan Tsui and Balli [29] demonstrated that macroeconomic variables, including marketing expenditures and exchange rate fluctuations, play a substantial role in forecasting international passenger arrivals, further validating the importance of economic stability for the tourism industry. These studies underscore that while economic indicators remain crucial for understanding tourism flows, their impact is often moderated by geopolitical and socio-cultural contexts. Additionally, Gavurova et al. [30] investigated the role of macroeconomic indicators, including GDP contribution, employment, and investment, in shaping tourism competitiveness across multiple countries. Their findings suggest that while GDP-related indicators are strong determinants of tourism demand, employment and investment indicators exhibit varying effects based on regional economic conditions. Nguyen et al. [31] further explored the influence of economic uncertainty on tourism consumption, highlighting that while uncertainty can reduce outbound tourism, it may simultaneously boost domestic tourism due to shifts in consumer spending behavior. Kristjánsdóttir et al. [32] emphasized the importance of integrated sustainability indicators in tourism, arguing that macroeconomic variables should be assessed alongside socio-environmental factors to develop more robust forecasting models.

Geopolitical risks and external shocks also play a crucial role in determining tourism flows. Özcan [33] demonstrated that geopolitical risks significantly disrupt tourism demand in Turkey, leading to shifts in travel behavior and economic consequences. Akadiri et al. [34] highlighted that geopolitical tensions, particularly those tied to regional conflicts, negatively affect inbound tourism flows and overall economic stability. Wu et al. [35] underscored the impact of psychological distance on travel decisions, noting that perceived safety plays a critical role in shaping tourist preferences during periods of increased geopolitical tension. Similarly, Verma et al. [36] found that the psychological and physical distance associated with geopolitical instability can deter tourists from traveling to affected regions. Bilgili et al. [28] emphasized the short-term and long-term economic implications of terrorism and political unrest on Turkey’s tourism sector, further illustrating the complex interplay between geopolitical risks and tourism demand dynamics. Kayral et al. (2023) further emphasized the role of crises, such as the COVID-19 pandemic and the Turkey–Russia warplane crash, in shaping tourism demand. By incorporating hybrid forecasting models, their study highlighted how integrating exogenous factors like exchange rates and geopolitical disruptions can improve predictive accuracy and inform more resilient tourism strategies [37]. Dinçer et al. (2024) emphasized the devastating impacts of the COVID-19 pandemic on Antalya’s tourism sector, highlighting the critical role of crisis management strategies and government support in ensuring sustainability during and after the crisis [38].

Beyond economic drivers, Wu et al. [35] examined psychological distance and its impact on travel intentions, particularly in post-pandemic recovery scenarios. Psychological distance is particularly relevant for neighboring countries, where cultural and historical ties, shared traditions, and ease of travel reduce perceived barriers and foster stronger travel intentions. Similar findings have been reported in the tourism literature, where cultural and geographical proximity significantly influence travel behavior [39,40,41].

Unique to this study is the inclusion of data from 100 countries from five regions, providing a global perspective on tourism patterns in Turkey. Furthermore, it incorporates economic indicators and Google Trends data specific to each country, enabling a multi-dimensional analysis that evaluates both internal time-series structures and external drivers of tourism demand. This study also performs detailed country-level and region-based profiling, allowing for comparisons between countries with different economic and geographic characteristics. By leveraging the SARIMA and SARIMAX models, this study compares the predictive power of intrinsic seasonal patterns against external economic and behavioral factors. This comprehensive approach allows for insights into both global trends and country-specific dynamics, offering a robust framework for forecasting in tourism research.

3. Materials and Methods

3.1. Data

This study utilizes two datasets: one focusing on monthly tourist arrivals and the other capturing economic indicators and Google Trend data. Together, these datasets span from January 2008 to September 2024, enabling a comprehensive analysis of tourism trends and their economic drivers.

The first dataset consists of monthly tourist arrival data for Turkey, covering visitors from 100 countries from five key regions: Europe, America (North and South American Continents), Africa, the Commonwealth of Independent States (CIS), and Asia. The dataset comprises 201 observations, ensuring sufficient temporal coverage to capture trends and seasonal patterns. This dataset highlights the global scope of the analysis, enabling comparisons between regions and countries.

The second dataset contains economic indicators, including 21 exchange rates, 6 selected countries’ (Germany, Russia, the United Kingdom (UK), Bulgaria, Iran, and Georgia) monthly inflation rates, and Google Trends search interest data. Exchange rates and inflation data for Turkey were sourced from the Central Bank of the Republic of Turkey [42]. For regional analyses, the EUR exchange rate was used for Europe, while the USD exchange rate was applied for other regions to better capture region-specific economic dynamics. Inflation data for the six selected countries—Germany, Russia, the United Kingdom, Bulgaria, Iran, and Georgia—were collected from their respective central banks to ensure accuracy and comparability.

Google Trends data were included to capture demand signals and search interest related to Turkey as a tourist destination. For each country, search terms corresponding to “Turkey” in the native language of that country were selected, and the geographic location in Google Trends was set to the respective country. In regional analyses, the geographic filter was set to “worldwide” to capture broader global trends and interest. The data, normalized on a scale from 0 to 100 based on relative search popularity, were collected monthly from January 2008 to September 2024. This alignment with the temporal scope of the tourist arrival dataset ensured consistency. Aggregating the data to monthly intervals reduced noise and allowed for direct comparisons with tourism patterns.

3.1.1. SARIMA Model

The SARIMA (Seasonal AutoRegressive Integrated Moving Average) model is a widely used statistical technique for analyzing and forecasting time-series data that exhibit both trends and seasonal patterns. It extends the ARIMA (AutoRegressive Integrated Moving Average) model by incorporating seasonal components, making it particularly effective for datasets with periodic fluctuations.

The SARIMA model handles non-stationary data by applying differencing to remove trends and seasonal variations, ensuring that the data become stationary before modeling. It combines three key components, namely, autoregression (AR), moving average (MA), and differencing (I), along with seasonal counterparts for each component. These elements work together to model temporal dependencies, seasonal cycles, and random noise in the data.

Due to its flexibility, SARIMA is commonly applied in fields such as economics, sales forecasting, climate studies, and energy consumption analysis, where understanding both long-term trends and seasonal effects is critical for making accurate predictions [43]. The model structure is as follows:

$$Y_t=(1-\varphi \left(B\right)){(1-B^s)}^D{(1-B)}^d{Y}_t+(1+\theta \left(B\right))(1+\mathsf{\Theta }\left(B^s\right))e_t$$

(1)

• Parameters:

• B: lag operator;
• d: degree of differencing (trend component);
• s: seasonal period;
• D: degree of seasonal differencing;
• $\varphi \left(B\right)$: autoregressive (AR) coefficients;
• $\theta \left(B\right)$: moving average (MA) coefficients;
• $\mathsf{\Theta }\left(B^s\right)$: seasonal MA coefficients;
• $e_t$: white noise.

3.1.2. SARIMAX Model

The SARIMAX (Seasonal AutoRegressive Integrated Moving Average with eXogenous variables) model is an extension of the SARIMA model that incorporates external variables (exogenous predictors) to improve forecasting accuracy. It is particularly useful when external factors, such as economic indicators, weather patterns, or policy changes, influence the time-series data.

The SARIMAX model builds upon the SARIMA framework by adding a regression component that accounts for the effects of exogenous variables. This allows the model to capture not only the internal dynamics of the time series but also the impact of external influences.

Key components of the SARIMAX model include autoregression (AR), moving average (MA), and differencing (I) terms, along with their seasonal counterparts. Additionally, it incorporates regression coefficients for exogenous variables to model their effects directly. It is especially valuable in scenarios where external variables provide predictive power beyond the intrinsic structure of the time series. The model is represented as [44]

$$Y_t={\beta }_0+{\beta }_1{X}_t+(1-\varphi \left(B\right)){(1-B^s)}^D{(1-B)}^d{Y}_t+(1+\theta \left(B\right))(1+\mathsf{\Theta }\left(B^s\right))e_t$$

(2)

Additional components:

• $X_t$: exogenous variables (e.g., inflation, exchange rates, or Google Trends data).
• ${\beta }_0,{\beta }_1$: regression coefficients for external predictors.

3.2. Model Selection and Evaluation

To measure the models’ prediction ability, the data are divided into “Train” (up to the end of 2022) and “Test” (from 2023 to 2024 September). The rationale here lies in ensuring robust model training and future forecasting. Using data up to 2022 for the training set allows the model to learn from long-term trends, seasonal patterns, and external shocks, such as the COVID-19 pandemic, while preserving the most recent observations as a Test set. This division provides a realistic evaluation of the model’s predictive performance, as it simulates forecasting future values without prior knowledge of those periods. Moreover, this method ensures that the model is trained on the most representative historical data, reflecting both normal and disruptive conditions. In this study, the parameters of the SARIMA and SARIMAX models are identified by selecting the configurations that yield the lowest error terms on the Test data, ensuring optimal model performance and predictive accuracy. All model assumptions were checked, and only models that did not violate these assumptions were selected. However, to avoid excessive length, this process is not included in detail in this paper as there are too many models. The performance metrics included are as follows:

• Root Mean Square Error (RMSE);
• Mean Absolute Error (MAE);
• Mean Absolute Percentage Error (MAPE);
• Theil’s U statistic.

Logarithmic transformations were applied to all models with the aim of reducing variance and improving predictive performance. So, at first, the log number of visitors was used instead of the original observations. However, the transformed data yielded considerably poor predictions, indicating that logarithmic transformation is not an appropriate approach for this dataset.

4. Results

This section presents the analysis of monthly tourist arrivals in Turkey between 2008 and 2024, focusing on seasonality, long-term trends, and the predictive performance of the SARIMA and SARIMAX models. The results emphasize the role of clear time-series structures and evaluate the forecasting accuracy based on error metrics. Figure 1 illustrates the monthly tourist arrivals in Turkey from 2008 to 2024. Strong seasonality is evident, with consistent peaks mid-year, reflecting Turkey’s attractiveness as a summer destination. The series also displays long-term growth, interrupted briefly by sharp declines, such as during 2020, likely due to external shocks like the COVID-19 pandemic. Due to seasonality, SARIMA and SARIMAX seem suitable modeling choices here.

Table 1. Comparison of forecasting performance metrics between SARIMA and SARIMAX models.

Metric	SARIMA	SARIMAX
RMSE	437,254.2	672,901.0
MAE	351,647.9	512,397.5
MAPE (%)	9.18	16.91
Theil’s U	0.5143	1.0319

shows the results of the SARIMA and SARIMAX models on Test data. These results show that SARIMA has lower errors across all metrics, especially in terms of RMSE and MAE, which are critical indicators of forecast accuracy. The MAPE for SARIMA is also significantly lower, suggesting better predictive performance relative to actual values. Additionally, none of the explanatory variables (USD, Turkey’s inflation, and Google Trend data) included in the SARIMAX model provide statistically significant contributions to the predictions. This lack of meaningful impact from additional variables highlights the robustness and simplicity of the SARIMA model, making it a more suitable choice for forecasting in this context. Consistent with the error metrics, Figure 2 also shows that the SARIMA outperforms SARIMAX, as its predictions on Test data are closer to the observed values than those of SARIMAX.

Figure 3 illustrates the 10-year forecasts of monthly tourist arrivals in Turkey from 2025 to 203 based on SARIMA (1,1,0)(1,1,0)₁₂. The forecasts reveal a clear seasonal pattern, with peaks observed during the summer months, indicative of the high-tourist season, and troughs during the winter months, reflecting lower tourist activity. The annual total number of visitors is projected to increase steadily, from approximately 53.8 million in 2025 to 72.6 million by 2035, suggesting a long-term growth trend and highlighting both seasonality and the overall upward trajectory in tourism. The seasonal variation in the data underlines the importance of the summer months as a critical period for the tourism industry, likely driven by favorable weather, school holidays, and the appeal of Turkey’s coastal destinations. Conversely, the lower activity during the winter months suggests the potential for further development of year-round tourism opportunities, such as promoting winter sports, cultural tourism, or wellness retreats.

Figure 4 demonstrates the time-series plots of the top six countries sending tourists to Turkey between 2008 and 2024, and it reveals distinct patterns influenced by geography and seasonality. The annual average number of visitors from each country is as follows: Germany (4,535,084), Russia (4,127,855), the UK (2,364,622), Bulgaria (1,874,252), Iran (1,691,208), and Georgia (1,444,034). The first three countries, Germany, Russia, and the UK, exhibit clear seasonal patterns in tourist arrivals, with significant peaks during the summer months. This suggests a reliance on pre-planned holidays, facilitated largely by air travel, as none of these countries share a land border with Turkey. Their travel behavior reflects the influence of vacation periods and favorable weather, contributing to higher seasonal variability. However, Bulgaria, Iran, and Georgia show less pronounced seasonality, characterized by more steady and consistent visitor flows throughout the year. The primary factor influencing this pattern is their geographical proximity and shared land borders with Turkey, enabling frequent, short-term visits via road or rail transportation. This accessibility supports cross-border travel throughout the year, reducing the dependence on seasonal trends.

A detailed analysis of these six countries is now undertaken, beginning with Germany, to examine the patterns of tourist arrivals and to evaluate the forecasting performance of the models. The results of the SARIMAX and SARIMA models for German tourist arrivals in Turkey highlight notable differences in predictive performance. The SARIMA (1,1,0)(1,1,0)₁₂ model outperforms the SARIMAX (1,1,0)(1,1,0)₁₂ model, as reflected in both visual and numerical evaluations. Visually, Figure 5 demonstrates that the SARIMA model forecasts (green line) align more closely with the observed values (red line) compared to the SARIMAX model forecasts (blue line). This suggests that the inclusion of external regressors in the SARIMAX model does not substantially improve prediction accuracy. Numerically, the SARIMA model yields a lower RMSE (70,399.9) and MAE (50,904.09) than the SARIMAX model (RMSE = 101,003.8 and MAE = 81,152.37). The SARIMA model also exhibits a lower MAPE (11.06%) compared to the SARIMAX model (18.44%), indicating better overall predictive accuracy and smaller relative errors. Furthermore, Theil’s U statistic for the SARIMA model (0.3944) is significantly lower than that of the SARIMAX model (0.5719), confirming its superior forecasting performance.

These findings, combined with previously discussed p-values, suggest that external regressors (EUR, Turkey’s inflation, Germany’s inflation, and Google Trends Germany score) are statistically insignificant and may not provide additional explanatory power. Consequently, the simpler SARIMA model appears to be more effective, implying that seasonal and autoregressive components alone capture the primary patterns in the data without the need for external variables.

Having analyzed German tourist arrivals, the focus is now shifted to Russian tourists to determine whether comparable patterns are observed and to assess the influence of exogenous variables on forecasting performance. The estimation results of the SARIMAX model for Russian tourist arrivals in Turkey provide insights into the significance of both autoregressive components and external regressors. Among the time-series components, the moving average (MA1) and seasonal moving average (SMA1) terms demonstrate statistical significance, with p-values of 0.0390 and <0.001, respectively. These results highlight the importance of accounting for short-term fluctuations and seasonal patterns in modeling tourist arrivals. In contrast, the seasonal autoregressive (SAR1) component is not statistically significant (p = 0.4443), suggesting that the seasonal autoregressive structure may not contribute meaningfully to the model’s explanatory power.

Regarding the external regressors, only the Google Trends Russia data are statistically significant (estimate = 3386.5, p = 0.0078), indicating a potential relationship between online search trends and tourist arrivals. However, other external variables, including the RUB exchange rate (p = 0.6815), Russia’s inflation rate (p = 0.9365), and Turkey’s inflation rate (p = 0.4289), fail to demonstrate significance. These findings imply that economic indicators may not exert a direct or immediate influence on Russian tourist flows within the given model structure.

Figure 6 compares the performance of the SARIMA and SARIMAX models, revealing that the inclusion of these external regressors does not substantially improve predictive accuracy. This further reinforces the dominance of seasonal and autoregressive patterns over macroeconomic variables in explaining short-term variations.

Following these results, the case of UK tourist arrivals is examined next to evaluate the role of exogenous variables before proceeding to analyze other key source countries. Figure 7 illustrates the observed and forecasted values for the number of UK visitors using SARIMA (1,1,0)(0,1,1)₁₂ and SARIMAX (1,1,0)(0,1,1)₁₂ models. Both models appear to capture the seasonal patterns and trends effectively, although SARIMAX forecasts align more closely with observed values during peak and trough periods, suggesting the inclusion of exogenous variables improves prediction accuracy. The coefficient test results from the SARIMAX model provide insights into the statistical significance of model parameters. The autoregressive (AR1) term (estimate = 0.3175, p < 0.001) and the seasonal moving average (SMA1) term (estimate = −0.7769, p < 0.001) are highly significant, indicating a strong dependency on previous observations and seasonal effects. Among the exogenous variables, the GBP exchange rate (estimate = 20,043, p = 0.023), the UK’s inflation (estimate = 22,373, p = 0.0026), and Google Trends UK data (estimate = 638.51, p = 0.048) are statistically significant predictors, demonstrating their positive influence on visitor numbers. On the other hand, the coefficient for Turkey’s inflation is −3856.6 with a p-value of 0.085, indicating an insignificant negative relationship with the number of UK visitors.

Given their geographical proximity to Turkey, Bulgaria, Iran, and Georgia are analyzed together to explore whether similar patterns emerge and to assess the influence of seasonal dynamics and exogenous variables on forecasting tourist arrivals. The analysis of Bulgaria, Iran, and Georgia reveals distinct patterns and model performance when forecasting tourist arrivals to Turkey (Figure 8). For Bulgaria, the SARIMA (1,1,1)(0,1,1)₁₂ model demonstrates superior predictive accuracy, as indicated by a lower RMSE (52,508.79 vs. 60,651.02) and Theil’s U statistic (1.1663 vs. 1.4844) compared to the SARIMAX (1,1,1)(0,1,1) model. Although the SARIMAX (1,1,1)(0,1,1) incorporates exogenous variables, only Bulgaria’s inflation rate (estimate = −12,454, p = 0.01809), Turkey’s inflation (estimate = 3998.8, p = 0.04783), and Google Trends Bulgaria data (estimate = 2975.3, p < 0.001) are statistically significant predictors. However, the overall predictive performance remains inferior to that of SARIMA, suggesting that seasonal and autoregressive components dominate the dynamics of Bulgarian tourist arrivals, with limited added value from external factors.

For Iran and Georgia, a similar pattern emerges, where SARIMA models outperform SARIMAX models based on error metrics. In Iran, SARIMA (1,1,0)(1,1,0)₁₂ produces lower RMSE (42,061.35 vs. 106,692.8) and Theil’s U statistic (0.7323 vs. 1.7663), highlighting its robustness in capturing seasonal trends without reliance on external regressors. While Google Trends Iran data (p = 0.02118) in SARIMAX (1,1,1)(1,1,1)₁₂ show significance, other economic indicators, such as inflation and exchange rates, fail to demonstrate meaningful impacts. Similarly, Georgia’s SARIMA (1,1,1)(0,1,1)₁₂ model achieves lower RMSE (24,570.7 vs. 50,383.12) and Theil’s U statistic (1.4347 vs. 3.0489). Although Google Trends data (p = 0.02532) in SARIMAX exhibit some significance, the overall performance suggests that simpler models leveraging seasonal patterns provide more reliable forecasts. These findings reinforce the dominant role of fundamental time-series structures over exogenous variables in predicting tourist arrivals from neighboring countries.

Building on these findings, a broader analysis of regional visitor patterns to Turkey offers additional insights into the influence of geographic and economic connections on tourist arrivals. Figure 9 presents distinct patterns in visitor trends in Turkey, segmented by continents and the Commonwealth of Independent States (CIS) between 2008 and 2024. Europe continues to dominate, with an annual average of approximately 17.1 million visitors. The CIS region follows as the second-largest source, contributing around 8.4 million visitors per year. This highlights the growing influence of CIS countries, possibly driven by geographic proximity and trade relations. Asia also demonstrates a significant presence, with an annual average of 5.4 million visitors. Visitor numbers from America (North and South American Continent) and Africa remain relatively lower, with averages of 1.1 million and 824,000, respectively. These figures may be influenced by greater distances, higher travel costs, and limited direct connections.

In this study, the analysis was extended to include visitor data from these regions. For each dataset, both the SARIMA and SARIMAX models were employed to capture trends and seasonal patterns. In the SARIMAX models, USD exchange rates and Turkey’s inflation rates were incorporated as external regressors to assess their impact on visitor numbers. The SARIMAX models did not yield statistically significant coefficients for these predictors across any of the regional datasets. This indicates that fluctuations in USD exchange rates and Turkey’s inflation rates may have a limited influence on visitor flows from these regions. Furthermore, the SARIMA models, which exclude exogenous variables, consistently outperformed their SARIMAX counterparts in terms of predictive accuracy. These results suggest that seasonal patterns and autoregressive components are more critical for forecasting visitor arrivals than external macroeconomic variables in these cases. This finding highlights the dominant role of intrinsic seasonal trends over economic factors for explaining variations in tourism demand across diverse regions.

5. Discussion

This section discusses the key findings of the study, highlighting patterns in tourism demand, the effectiveness of forecasting models, and the role of external factors in influencing tourist arrivals. The results emphasize the importance of seasonality and geographic proximity while also revealing the contributions of economic indicators and digital search data.

The findings of this study demonstrate the dominant influence of seasonality on tourist arrivals in Turkey. SARIMA models consistently outperformed SARIMAX models, underscoring the effectiveness of time-series approaches in capturing patterns without relying heavily on external regressors. Seasonality continues to act as a primary driver, reflecting predictable patterns driven by cultural, climatic, and institutional factors, such as school holidays and public events. These findings are consistent with the multi-methodical analysis conducted by Yabanci [2], which highlighted seasonal effects as critical determinants of tourism demand in Turkey. Similarly, Çekim and Koyuncu [45] emphasized the importance of seasonality in explaining tourism patterns, where seasonal variations and weather conditions significantly influenced tourist inflows. Önder [26] further supports this perspective, demonstrating that monthly data, when combined with Google Trends indices, effectively capture seasonal fluctuations and improve forecasting accuracy.

In comparing the performance of the SARIMA and SARIMAX models, this study highlights the superior forecasting ability of SARIMA, particularly in contexts where exogenous variables such as exchange rates, inflation, and Google Trends data do not provide statistically significant contributions. This finding is consistent with Gil-Alana et al. (2020), who demonstrated that tourism demand exhibits strong persistence, meaning that seasonal fluctuations play a dominant role in shaping visitor arrivals, whereas economic shocks tend to have only transitory effects [14]. Our finding also aligns with findings by Tang et al. [17], who observed that, while exchange rates can influence tourism demand, their impact often diminishes in stable tourism markets, where intrinsic seasonal trends dominate. Similarly, Volchek et al. [15] found that SARIMA models alone are effective in capturing seasonal dynamics, particularly in markets with established tourism cycles. Saayman and Botha (2015) further reinforced that SARIMA models outperform alternative forecasting methods in highly seasonal destinations, as they effectively capture cyclical variations without requiring additional external variables [16]. However, our finding contrasts with Wickramasinghe and Ratnasiri [7], who demonstrated that SARIMAX models incorporating disaggregated Google Trends data significantly improved forecasting accuracy for Sri Lanka. This divergence may stem from regional differences in the relevance of digital data and the planning behaviors of tourists. Vergori (2016) also highlighted that seasonality introduces significant forecasting uncertainty, making it crucial to select models that accurately capture periodic fluctuations without over-relying on economic indicators [18].

Geographic proximity emerged as another critical determinant, with neighboring countries exhibiting more stable and consistent flows of visitors. Countries such as Bulgaria, Iran, and Georgia benefited from ease of access, visa agreements, and cultural affinities, resulting in reduced travel barriers and higher frequencies of short-term visits [4]. Conversely, distant countries, such as Germany and the UK, displayed greater reliance on organized travel packages and exhibited sharper seasonal peaks, particularly during summer vacations [2]. Moreover, McKercher and Mak [46] demonstrated that land-bordering countries contribute to more stable and frequent tourism flows, reinforcing the importance of accessibility and travel costs in shaping demand patterns. Recent studies further emphasize the influence of geographic proximity and psychological factors, such as economic, social, and cultural similarities, in stabilizing visitor flows [1,33]. Psychological distance, encompassing cultural and political differences, also affects tourists’ length of stay and destination choices [19,33,47]. Reduced transaction costs, such as language similarities and visa policies, often play a more significant role than natural and cultural attractions in attracting international tourists [3,33].

The inclusion of economic indicators, such as exchange rates and inflation, in SARIMAX models did not yield statistically significant improvements in forecasting accuracy, confirming the results of [3]. These findings emphasize the robustness of SARIMA models, which simplify forecasting by focusing on internal data structures. Despite the intuitive link between economic indicators and tourism demand, recent research suggests that exchange rates and inflation may have limited short-term impacts, especially in mature tourism markets [26,45]. Önder [26] emphasizes that tourists often make travel decisions well in advance, reducing their sensitivity to short-term price fluctuations. Athari et al. (2021) further highlighted that political risks and unfavorable economic conditions, such as high inflation, significantly deter international tourism arrivals, underscoring the complex interplay between macroeconomic factors and tourism demand [48]. Studies also highlight that wealthier tourists, who form a significant portion of long-distance travelers, may be less price-sensitive, reducing the overall impact of exchange rate variations [11].

Additionally, Gavurova et al. [30] demonstrated that tourism competitiveness is driven by macroeconomic indicators such as GDP contribution, employment, and investment. Their findings suggest that while GDP-linked metrics have a clear impact on tourism demand, the role of employment and investment may vary based on regional economic conditions. Similarly, Nguyen et al. [31] analyzed the influence of economic uncertainty on tourism consumption, revealing that while uncertainty discourages outbound travel, it can simultaneously stimulate domestic tourism as consumers reallocate expenditures.

Moreover, inflation and exchange rates may act as indirect influencers, affecting broader economic stability and disposable income rather than directly shaping tourism demand. While economic indicators are important for long-term planning, their predictive power diminishes in short-term forecasting, where non-economic drivers like safety concerns and cultural events dominate [26,45]. Agiomirgianakis et al. [49] found that exchange rate volatility negatively affects tourism flows, particularly during prolonged instability, while Bilgili et al. [28] demonstrated that geopolitical risks, such as terrorism, often moderate the impact of economic variables on tourism demand. Similarly, Akadiri et al. [34] emphasized that regional conflicts could overshadow macroeconomic factors, directly reducing tourist arrivals. In Turkey’s case, the dominance of intrinsic seasonal patterns over economic indicators aligns with these findings, suggesting that behavioral and sentiment-driven variables, such as web-traffic data and climate indicators, could complement economic factors in improving forecasting accuracy [11]. This highlights the need for multi-dimensional forecasting models that integrate behavioral, economic, and geopolitical data to capture the complex dynamics of tourism demand effectively. Kristjánsdóttir et al. [32] further argued that sustainability indicators should be integrated with economic variables to provide a more holistic understanding of tourism trends.

The integration of Google Trends data provided insights into specific countries, such as the UK, Bulgaria, Russia, and Iran, where digital search behavior correlated with tourist inflows. This supports the prior findings of studies [11,23,24,25,26,45], which demonstrated that incorporating big data sources, such as web search traffic, into forecasting models can significantly enhance predictive accuracy. Volchek et al. [15] further confirmed that high-frequency search engine data improve short-term demand predictions, especially in dynamic markets. Similarly, Wickramasinghe and Ratnasiri [7,23] highlighted the role of disaggregated search data in refining regional forecasts, while Park et al. [21] emphasized the value of combining online news data with seasonal models to capture shifts during periods of instability. However, the effectiveness of digital data integration varies across models and contexts. Zhang et al. [12] found that advanced decomposition techniques in deep learning models improve accuracy compared to traditional methods. Xiao et al. [23] also demonstrated that hybrid models like EEMD-DBN outperform SARIMA models when Google Trends data are included, highlighting the benefits of machine learning in capturing non-linear relationships. While integrating digital data sources has shown promise, future research should focus on optimizing hybrid models by combining search trends, social media engagement, and online news sentiment. This multi-source approach could further enhance forecasting accuracy, particularly in times of economic uncertainty or crisis-induced travel fluctuations.

Finally, the results of this study highlight the challenges of improving forecasting accuracy using data transformations. Specifically, the log transformation, which is often employed to stabilize variance and reduce heteroscedasticity, did not enhance predictive performance in this analysis. Instead, it resulted in poorer forecasts compared to models applied to untransformed data. This outcome aligns with prior findings by Lütkepohl and Xu [50], who emphasized that the effectiveness of log transformation depends largely on its ability to stabilize variance. They noted that when the transformation fails to achieve this stabilization, it may degrade forecast accuracy rather than improve it. Their study suggested that log transformations are more suitable for datasets exhibiting exponential growth or multiplicative errors, conditions not strongly observed in our dataset.

6. Conclusions

This study provides valuable insights into the dynamics of tourism demand in Turkey between 2008 and 2024 by evaluating the forecasting performance of the SARIMA and SARIMAX models. The findings underscore the critical role of seasonality and geographic proximity in shaping tourism demand, with SARIMA models consistently outperforming SARIMAX models in predictive accuracy. This highlights the robustness of time-series models in capturing intrinsic patterns without heavily relying on external regressors such as exchange rates and inflation.

From a practical perspective, the results emphasize the importance of leveraging SARIMA models for short- and medium-term tourism forecasting, allowing policymakers and industry practitioners to allocate resources more efficiently during peak and off-peak seasons. For example, understanding seasonal patterns enables tourism stakeholders to optimize staffing, marketing campaigns, and infrastructure planning, ensuring a balance between supply and demand.

The integration of Google Trends data, although yielding mixed results, demonstrates the potential of behavioral indicators in forecasting models. While economic variables showed limited short-term predictive power, digital search data provided supplementary insights, particularly for countries with high internet penetration rates. This suggests that tourism planners could monitor real-time search trends to capture early signals of demand shifts and adapt their strategies accordingly. For instance, sudden increases in search interest could inform marketing decisions or resource allocations in targeted markets.

Academically, this study contributes to the literature by addressing the research gap on the combined impact of seasonality, economic indicators, and digital search behavior on tourism demand. The findings challenge the assumption that economic variables significantly influence short-term tourist arrivals, instead advocating for a stronger focus on intrinsic time-series patterns and behavioral data. Moreover, this study underscores the need for hybrid forecasting approaches that integrate sentiment-driven and digital data with traditional time-series models to enhance forecasting accuracy.

Future research should expand on these findings by exploring the role of cultural and psychological factors, which were beyond the scope of this study. Additionally, further exploration into the applicability of advanced machine learning and hybrid models could provide a deeper understanding of the complex dynamics in tourism demand forecasting. For industry practices, tourism businesses can leverage insights from digital search behavior, such as Google Trends data, to refine marketing strategies and tailor offerings to specific markets. Real-time tracking of search trends could enable businesses to anticipate demand spikes, optimize pricing strategies, and enhance customer targeting efforts. Moreover, integrating predictive analytics into operational decision-making can support resource allocation, workforce planning, and supply chain optimization, particularly in response to seasonal fluctuations.

In conclusion, this study not only offers theoretical contributions by advancing the understanding of tourism demand dynamics but also provides actionable insights for practitioners to enhance strategic planning and ensure sustainable tourism growth. By bridging the gap between academic research and industry application, this study supports the development of evidence-based policies that cater to the evolving needs of the tourism sector.