Series de tiempo

Forecasting Time Series: An Introduction

Overview of Time Series Forecasting

• The session introduces a new forecasting model called time series forecasting, emphasizing the need for theoretical understanding before practical application.

• The goal of time series forecasting is to decompose historical data into its components: baseline, trend, and seasonality.

• A retrospective process is applied to identify the best-fitting model for the dataset while minimizing forecast error.

Assumptions in Time Series Forecasting

• Key assumptions include that historical data has a baseline, trend, and seasonality; non-volatility indicates a clear trend.

• Structural changes (e.g., company mergers or expansions) can disrupt trends but do not prevent forecasting.

Practical Application of Time Series Forecasting

Steps in Forecasting Sales Levels

• The focus shifts to predicting sales levels for a company using Excel's risk simulator tool.

• Users are guided through selecting options in Excel for time series analysis, including distribution selection and seasonal adjustments.

Methodologies and Model Selection

• The software automatically selects methodologies based on historical data distribution; four quarters are set as seasonal parameters.

• Different methodologies are presented, with the multiplicative Holt-Winters method identified as the best fit for the data.

Understanding Holt-Winters Methodology

Components of Holt-Winters Model

• The multiplicative Holt-Winters method decomposes data into three factors: alpha (baseline), beta (trend), and gamma (seasonality).

• Two models exist within this methodology: additive and multiplicative. The additive model combines components rather than decomposing them.

Analysis Results

• Software analysis reveals that the multiplicative Holt-Winters model is optimal based on historical distribution.

• Parameters such as seasonality are established based on user input; results show an automatic adjustment by the software using specific alpha values.

Final Insights from Time Series Analysis

Summary of Findings

• Historical data inputs lead to adjusted forecasts displayed graphically alongside calculated predictions.

Error Metrics in Forecasting

Key Error Measures

• The Root Mean Square Error (RMSE) is introduced as a primary measure, calculating the square root of the deviation between forecasted data and actual data.

• The Mean Absolute Percentage Error (MAPE) is discussed, which provides a statistical measure of relative error expressed as an average percentage, making it suitable when forecasting errors are closely related to their percentage.

• The U-statistic of Tail is mentioned as another important metric that assesses the credibility of model forecasts; if this statistic is less than one, it indicates that the forecasting method yields statistically better estimates than mere guessing.

Visualization Insights