Chapter 7: Forecasting

2007 Pearson Education

Forecasting Techniques

Qualitative and judgmental Statistical time series models Explanatory/causal models

Qualitative and Judgmental Methods

Historical analogy comparative analysis with a previous situation Delphi Method response to a sequence of questionnaires by a panel of experts

Indicators and Indexes

Indicators measures believed to influence the behavior of a variable we wish to forecast Leading indicators Lagging indicators Index a weighted combination of indicators Indicators and indexes are often used in economic forecasting

Time Series

A time series is a stream of historical data Components of time series Trend Short-term seasonal effects Longer-term cyclical effects

Example of a Time Series

Statistical Forecasting Methods

Moving average Exponential smoothing Regression analysis

Simple Moving Average

Average random fluctuations in a time series to infer short-term changes in direction Assumption: future observations will be similar to recent past Moving average for next period = average of most recent k observations

Example: Moving Average Forecast With k = 3

Time Series Data and Moving Averages

Excel Tool: Moving Averages

Tools > Data Analysis > Moving Average

Enter range of data
Enter value of k Select output options Select options

Excel Results

Caution: chart aligns forecasts for next period with current period data

Weighted Moving Average

Weight the most recent k observations, with weights that add to 1.0 Higher weights on more recent observations generally provide more responsive forecasts to rapidly changing time series

Error Metrics and Forecast Accuracy

Mean absolute deviation (MAD)

A
MAD =
i =1

Ft

n
n

Mean square error MSE = (MSE)

A t
i =1

Ft

Mean absolute percentage error (MAPE)

MAPE =
i =1

At Ft At n

Exponential Smoothing

Exponential smoothing model:

Ft+1 = (1 a )Ft + aAt = Ft + a (At Ft )

Ft+1 is the forecast for time period t+1, Ft is the forecast for period t, At is the observed value in period t, and a is a constant between 0 and 1, called the smoothing constant.

Excel Tool: Exponential Smoothing

Tools > Data Analysis > Exponential Smoothing

Enter data range
Damping factor = 1 - a Select output range and options

Exponential Smoothing Example

Exponential Smoothing Forecasts (a = 0.6)

Forecasting Models With Linear Trends

Double Moving Average Double Exponential Smoothing Based on the linear trend equation

Ft k at bt k

Double Moving Average

Mt = [ At-k+1 + At-k+2 + At]/k
Dt = [Mt-k+1 + Mt-k+2 + Mt]/k

at = 2Mt Dt
bt = (2/(k-1))[Mt Dt]
Use aT and bT in the linear trend equation to forecast k periods beyond period T:

FT k aT bT k

Example Calculations

Double Moving Average Forecasts

Double Exponential Smoothing

at = ayt + (1-a) (at-1 + bt-1) bt = (at at-1) + (1-)bt-1
Initialize:

a1= A1 b1 = A 2 A 1

Forecasting Models With Seasonality

Additive model

Ft k at S t s k

Multiplicative model

Ft k at S t s k

Additive Seasonality

Level and seasonal factors:

at = a ( At - St-s ) + (1-a) at-1 St = (At - at ) + (1-) St-s

Forecast for next period

Ft 1 at S t s 1

Initialization
as =

A /s
t 1 t

at = as t = 1,2,s
St = At - at t = 1,2,s

Example of Additive Seasonal Model

Models for Trend and Seasonality

Holt-Winters Additive Model Holt-Winters Multiplicative Model

Holt-Winters Additive Model

Smoothing equations:
at = a ( At - St-s ) + (1-a) (at-1 + bt-1) bt = (at at-1) + (1-)bt-1 St = (At - at ) + (1-) St-s

Forecast for period t + 1:

Ft 1 at bt S t s 1

Initialization
bt = bs , for t = 1,2,s bs = [ (As+1 A1)/s + (As+2 As)/s + .(As+s As)/s] / s
Initial values for level and seasonal factors are the same as in the additive seasonal model.

CB Predictor

Excel Add-In for forecasting

CB Predictor Method Gallery

CD Predictor Output: Methods Table

Durbin-Watson statistic: check for autocorrelation; a value of 2 indicates no autocorrelation

Theils U statistic: comparison to nave forecast. U<1 indicates that forecasting result is better than guessing; U=1, about the same; U>1, worse than guessing

CD Predictor Output: Results Table

CD Predictor Output: Chart and Forecast Values

Regression Trend Lines for Forecasting

Nonlinear Trend Line for Forecasting

Incorporating Seasonality into Regression Models

Use ordinal variables. Example:

Gas Usage = 0 + 1Time + 2 January + 3 February + 4 March + 5 April + 6 May + 7 June + 8 July + 9 August + 10 September + 11 October + 12 November

The forecast for December of the first year will be 0 + 1(12). The forecast for January (Time = 1) would be 0 + 1(1) + 2(1).

Data Matrix

Regression ANOVA Results

Regression Forecast Results

Causal Forecasting Models

Causal models incorporate independent variables such as economic indexes or demographic factors that may influence the time series.

Model

Sales = 0 + 1Week + 2 Price/Gallon Sales = 39406.69 + 508.67 Week 16463.20 Price/Gallon