Forecasting Time Series: (2022–2023)

A data analysis and forecasting project aimed at identifying trends, seasonality, and making future predictions for a univariate time series dataset using Excel-based statistical techniques.

Problem Statement

To gain insights and forecast values for a given time series dataset spanning from 01-Sep-2022 to 01-Aug-2023. The analysis includes:

• Exploratory Data Analysis (EDA)
• Data cleaning and preprocessing
• Seasonality and trend detection
• Building two forecasting models
• Evaluating predictions using standard error metrics

Dataset Description

The time series contains:

• Date (format: dd-mm-yyyy)
• Value (numeric, unitless but assumed to reflect stock price-like behavior)
• Series ID (constant)

Solution Approach

The following steps were undertaken:

1. EDA: Summary statistics (mean, median, std dev etc) and visualizations (line charts, boxplots).
2. Data Cleaning:
   • Removed invalid negative stock values.
   • Filled missing values using FORECAST.ETS() with seasonality detected as 8 months.
   • Outlier detection using IQR on the residuals after seasonal decomposition.
3. Modeling:
   • Linear Trend + Seasonality
   • Linear Trend + Dummy Variables
4. Evaluation Metrics:
   • MAD, MAPE, MSE, and Bias

Key Visuals

• Line chart showing original values and cleaned data
• Forecast comparisons between linear and seasonal techniques
• Residual boxplot for outlier detection
• Forecasted vs actual values visualization

Modeling & Evaluation

Model 1: Linear Trend + Seasonal Index

Metric	Value
MAD	11.04
MAPE	59.00%
MSE	179.67
Bias	-0.02

Model 2: Linear Trend + Dummy Variables (Seasonality = 8)

Metric	Value
MAD	7.99
MAPE	42.46%
MSE	106.82
Bias	0.00

Forecast Output for Model 2 (Sep 2022 – Aug 2023)

Month	Forecast
Sep 22	56.51
Oct 22	62.45
Nov 22	63.55
Dec 22	72.70
Jan 23	82.30
Feb 23	85.45
Mar 23	84.77
Apr 23	74.11
May 23	63.84
Jun 23	69.78
Jul 23	70.87
Aug 23	80.03

Insights & Conclusions

• The dataset exhibits 8-month seasonality.
• There are cyclical patterns, but irregularities suggest external factors affect values.
• Dummy variable model performed better but still had noticeable prediction errors.
• A more advanced model (e.g., ARIMA, Prophet, ML-based regression) would likely improve performance.