Lecture Notes: Time Series Analysis and Forecasting III.
Forecast Accuracy
Introduction Definition
Forecasting is essential for managerial decision-making, such as Forecast accuracy measures how close the forecasted values are to the
production planning, inventory management, and budgeting. It involves actual outcomes.
predicting future outcomes based on historical data. Forecasts can be
qualitative (based on judgment and expertise) or quantitative (based on Key Formulas and Metrics
numerical data and models). This chapter focuses on quantitative
forecasting methods, particularly time series analysis. Forecast Error (et)
Difference between actual and forecasted values.
I. Understanding Time Series Mean Forecast Error (MFE)
Average of forecast errors; indicates bias (under- or over-
Definition forecasting).
Mean Absolute Error (MAE)
A time series is a sequence of data points collected or recorded at Average of the absolute forecast errors.
regular time intervals (daily, weekly, monthly, etc.). Example: Monthly Mean Squared Error (MSE)
sales, daily temperatures, annual profits. Average of squared forecast errors.
Mean Absolute Percentage Error (MAPE)
Expresses forecast accuracy in percentage terms.
Purpose Useful for comparing forecasts across different scales.
To identify patterns in past data and extend them to forecast future IV. Forecasting Methods
values.
1. Naïve Forecasting
II. Time Series Patterns
Uses the most recent observation as the forecast for the next
1. Horizontal (Stationary) Pattern period.
Simple but reacts quickly to changes.
The data fluctuates around a constant mean. Best for short-term forecasts with stable data.
No upward or downward trend.
Example: Weekly gasoline sales that stay relatively constant.
2. Moving Average
Stationary time series: Statistical properties (mean and variance)
do not change over time.
Forecast is the average of the most recent k observations.
2. Trend Pattern Formula:
Smooths random fluctuations.
A smaller k responds faster to changes; a larger k provides
A consistent long-term upward or downward movement in data.
smoother forecasts.
Caused by long-term factors like technology, population growth, or
changing preferences.
Example: Annual increase in bicycle sales. 3. Weighted Moving Average
3. Seasonal Pattern Assigns different weights to recent data (more weight to newer
values).
Regular fluctuations that repeat over a specific period (e.g., Formula:
quarters or months). Sum of weights = 1.
Example: Umbrella sales peaking during rainy seasons.
4. Exponential Smoothing
4. Trend and Seasonal Pattern
Gives exponentially decreasing weights to older observations.
Combines a general trend with seasonal variations. Formula:
Example: Smartphone sales increasing annually but varying by α\alphaα = smoothing constant (0 ≤ α ≤ 1).
quarter. Small α → smoother, slower response; large α → faster
response.
5. Cyclical Pattern
V. Choosing a Forecasting Method
Long-term oscillations above or below a trend line lasting more
than a year. Horizontal data → Moving average or exponential
Often linked to economic or business cycles. smoothing.
Trend data → Linear trend projection or regression.
Seasonal data → Seasonal models with indices or
decomposition.
Both trend and seasonality → Trend-seasonal regression Steps in Linear Trend Forecasting
models.
Term Definition
A prediction of future events based on
Forecast
data or analysis.
A set of observations measured at
Time Series
successive points in time.
Data whose mean and variance do not
Stationary Time Series
change over time.
Long-term increase or decrease in data
Trend
values.
Repeating pattern over a fixed period,
Seasonality
usually within a year.
Long-term fluctuations tied to business or
Cyclical Pattern
economic cycles.
Difference between actual and predicted
Forecast Error
values.
MAE (Mean Absolute Error) Average of absolute forecast errors.
Forecast Evaluation
MSE (Mean Squared Error) Average of squared forecast errors.
MAPE (Mean Absolute Average of absolute percentage forecast
Once the trend equation is developed, the accuracy of the forecasts
Percentage Error) errors.
can be measured using MAE, MSE, or MAPE, just like in moving
Average of recent data points used for averages or exponential smoothing.
Moving Average
forecasting.
Average with assigned weights to each VI. Seasonality
Weighted Moving Average
past observation.
Forecasting method that gives more Definition
Exponential Smoothing
weight to recent data.
Determines the weight given to recent Seasonality refers to regular, predictable variations in a time series that
Smoothing Constant (α)
observations. repeat at fixed intervals — usually within a year.
Examples:
Lecture Notes (Part II): Linear Trend Projection and
Seasonality Ice cream sales peak in summer. Electricity use increases during
hot or cold months. Retail sales spike in December.
V. Linear Trend Projection
Identifying Seasonality
Linear Trend Projection is a quantitative forecasting method that fits a
straight line to time series data to model and predict trends over time. To detect seasonality, examine data over multiple years. A time series
plot often reveals periodic highs and lows that repeat annually,
quarterly, or monthly.
Types of Seasonal Models
1. Seasonality Without Trend
Used when data fluctuates around a constant level (no overall upward or
downward trend).
Steps:
[Link] the average value for each season (e.g., each month or
quarter).
[Link] the seasonal index for each period:
Seasonal Index=Overall Average/Average for Season
[Link] forecast, multiply the average or trend value by the seasonal index.
Example:
If average quarterly sales = 100 units and the index for Q2 = 1.2,
then forecasted Q2 sales = 100 × 1.2 = 120 units.