📈 Gold Price Prediction in KSA [2020-2026]

Time-Series Forecasting using Hybrid ARIMA-GARCH Model

📄 Read Full Documentation

📋 Project Overview

This project predicts the daily price of 24K Gold (per Gram) in Saudi Arabia (KSA) for the year 2026.

Gold prices are inherently non-stationary and exhibit "volatility clustering" (periods of high risk followed by calm). Standard models like Linear Regression or simple ARIMA fail to capture this changing risk. This project solves that problem by implementing a Hybrid ARIMA-GARCH model:

• ARIMA: Models the trend (mean).
• GARCH: Models the volatility (variance/risk).

The final output provides a 365-day forecast with a Risk-Adjusted 95% Confidence Interval, helping retailers and investors quantify market risk.

---

🎯 Objectives

1. Data Extraction: Ethically scrape historical data (2020–2025) from gold.sa using reverse-engineering techniques.
2. EDA: Analyze trends and confirm high volatility (IQR: ~66 SAR).
3. Modeling: Build a time-series model that handles non-stationarity.
4. Forecasting: Generate a 2026 prediction with risk bounds.

---

🛠️ Technology Stack

• Data Handling: Pandas, Numpy, JSON
• Web Scraping: Requests (with ethical delays to avoid throttling)
• Modeling: statsmodels (ARIMA), arch (GARCH)
• Visualization: Matplotlib, Seaborn

---

📂 Repository Structure

• 01_Gold_Extraction.ipynb: Scrapes data from gold.sa and saves it to CSV.
• 02_Exploratory_Data_Analysis.ipynb: Visualizes trends and volatility.
• 03_Model_Prediction.ipynb: Builds the ARIMA-GARCH model and generates the 2026 forecast.
• gold_sa_historical_data_2020_onwards.csv: The historical dataset used for training.

---

📊 Key Results

1. Trend Analysis

The data shows a clear upward trend from 2020 to 2025, driven by global economic factors. The market is non-stationary and required differencing ($d=1$).

2. Volatility Clustering

The analysis confirmed that gold volatility is not constant. High-risk periods cluster together, which justified the use of GARCH(1,1) over a simple Random Walk model.

3. Final Forecast (2026)

The model predicts the following for the next 365 days:

Metric	Value	Interpretation
Prediction Horizon	365 Days	Full year 2026 forecast
Avg. Predicted Price	525.49 SAR/g	Expected Mean Price
Lower Bound (95%)	438.01 SAR/g	Best-Case / Low-Risk Scenario
Upper Bound (95%)	612.96 SAR/g	Worst-Case / High-Risk Scenario

---

⚠️ Challenges & Solutions

Challenge	Solution
Dynamic Data	Standard scraping failed. Reverse-engineered network requests to find the JSON endpoint.
Rate Limiting	Implemented time.sleep(7) to respect server limits and avoid 403 errors.
Model Flatlining	Basic ARIMA(0,1,0) failed to capture risk. Integrated GARCH(1,1) to model volatility.

---

⚙️ How to Run

1. Clone the repository:

   git clone [https://github.com/YourUsername/Gold-Price-Prediction-KSA.git](https://github.com/YourUsername/Gold-Price-Prediction-KSA.git)

2. Install dependencies:

   pip install -r requirements.txt

3. Run the Notebooks: Start Jupyter Notebook and run the files in order: 01 -> 02 -> 03.

---

📄 License & Disclaimer

Data Source: Gold.sa Disclaimer: This project is for educational purposes only. Financial decisions should not be made based solely on this model.