Linear Time Series

This project analyzes the time series of the Industrial Production Index (IPI) for the construction sector in France (Section F of the NAF classification), covering the period from January 1990 to February 2025. The goal is to model and forecast the series using ARMA/ARIMA models while addressing issues such as non-stationarity, outliers, and external shocks.

The GitHub repository is organized as follows:

.
├── assignment.pdf     # Project instructions and topic
├── code.R             # Full R code used for data processing, modeling, and forecasting
├── report.pdf         # Detailed project report with analysis and results
├── data.csv           # Dataset: Industrial Production Index (construction sector)
└── README.md          # Project documentation

Data

• Source: INSEE (Link to series)
• Frequency: Monthly
• Timeframe: January 1990 to February 2025 (422 observations)
• Preprocessing: Seasonally and working-day adjusted (CVS-CJO)
• Base year: 2021

Features

• Exploratory and graphical analysis of the series
• Stationarity testing (ADF, PP, KPSS)
• Detection and treatment of outliers
• Optimal ARIMA(p,d,q) model selection
• Forecasting with 95% confidence intervals
• Instantaneous causality analysis (Granger)

Installation

1. Clone the repository:

   git clone https://github.com/nayelsdk/time-series-project.git
   cd time-series-project

2. Install the required R packages:

   install.packages(c("forecast", "tseries", "fUnitRoots", "ggplot2", "tsoutliers", "ellipse", "astsa"))

Forecasting

• Horizon: March 2025 and April 2025
• Point forecasts:
  • March: 93.86 [85.77 ; 101.95]
  • April: 93.87 [84.99 ; 102.74]
• Visualization: Zoomed-in plot and 2D confidence ellipse

Dependencies

• R ≥ 4.0
• Packages: forecast, ggplot2, fUnitRoots, tseries, zoo, astsa, tsoutliers, ellipse, RColorBrewer, gridExtra, scales

Authors

• Ana-Sofia Cianfarani
• BENABDESADOK Nayel