🌍 Forecasting Tourism Flows: Predictive Analytics (Mexico–USA)

📖 Project Overview

This project focuses on forecasting short-term U.S. tourist arrivals to Mexico using real-world tourism and macroeconomic data.
The objective is to help tourism stakeholders anticipate seasonal demand, optimize resource allocation, and make data-driven strategic decisions.

The analysis compares several forecasting models to identify the most accurate and reliable approach for predicting quarterly tourism flows. Emphasis is placed on combining statistical soundness, interpretability, and real-world applicability.

---

🧩 Data Sources

• dataTour.Rdata – Contains quarterly tourism flow data between 20 destination countries and their five main origin countries.

  • Each destination includes:
    • 5 origin country columns
    • A Total column
    • A Difference column representing arrivals from other origins

• IMFdata.Rdata – Provides macroeconomic indicators for 46 countries, such as:

  • GDP growth
  • Purchasing Power Parity (PPP)
  • Inflation rates
  • Other key economic factors

These datasets enable the exploration of both seasonal patterns and economic influences on tourism flows.

---

⚙️ Methodology

Multiple forecasting models were developed and compared to evaluate short-term prediction accuracy:

1. Seasonal Naïve Model – Baseline model capturing consistent quarterly seasonality.
2. Exponential Smoothing (ETS) – Captures level, trend, and seasonal effects (notably ETS(A,Ad,A)).
3. Autoregressive Lag Model (AR) – Models temporal dependencies in tourist arrivals.
4. Lag Model with Seasonal Dummies – Enhances AR models with quarter-specific seasonal adjustments.
5. Macroeconomic Regression Model – Incorporates U.S. GDP growth and Mexico’s PPP to link economic indicators with tourism flows.

Evaluation Metrics

• Root Mean Squared Error (RMSE)
• Mean Absolute Error (MAE)

Models were trained and validated using historical data up to 2018, with 2019 serving as the out-of-sample test period.

---

📊 Key Findings

• The Seasonal Naïve model achieved the lowest RMSE and MAE across both validation and test datasets.
• The ETS(A,Ad,A) model effectively captured trends but showed mild overfitting on the test period.
• Autoregressive and Lag + Dummy models provided good fit during training but lacked generalization power.
• The Macroeconomic regression model offered valuable interpretability but produced higher errors for short-term predictions.

✅ Conclusion

Tourism flows between the U.S. and Mexico show strong, stable seasonal patterns.
The Seasonal Naïve model is the most reliable, interpretable, and operationally efficient method for short-term forecasting.

ETS and regression models remain useful for long-term trend monitoring and policy scenario analysis, providing context on how economic changes influence tourism.

---

🧠 Tools & Technologies

• Language: R
• Techniques: Time Series Forecasting (ETS, AR, Regression)
• Evaluation Metrics: RMSE, MAE
• Validation Strategy: Train–Validation–Test Split
• Visualization: R packages for time series diagnostics and forecasting visualization

---

📈 Practical Applications

This analysis supports:

• Forecasting seasonal tourism demand between the U.S. and Mexico
• Optimizing resource allocation (staffing, infrastructure, marketing)
• Integrating macroeconomic indicators for strategic planning
• Enhancing data-driven decision-making in tourism management