forecasting - universitas...

ForecastingOutline:I. IntroductionII. Demand ManagementIII. Demand PatternsIV. Principles of ForecastingV. Qualitative TechniquesVI. Quantitative TechniquesVII. Forecast ErrorVIII. Monitoring the Forecast

Kuliah ke-9Rabu, 19 Nov 2008

Assignment: Kerjakan semua soal Forecasting (Chapter 8) - Tidak dikumpul

I. Introduction

Many factors influence the demand for firm’s products and services:

General business and economic conditionsCompetitive factorsMarket trendsThe firm’s own plans, such as promotions, advertising, pricing and product changes.

Characteristics of Demand

II. Demand Management

Demand management is the function of recognizing and managing all demands for productsDemand management includes:- Forecasting- Order processing- Making delivery promises- Interfacing between production planning

& control and marketplaces.

Demand management

Marketplace Demand Management

ProductionPlanning

Master ProductionSchedule

• Forecasting• Order processing

• Making delivery promises

III. Demand Patterns

Stable versus dynamicStable demand retains same general shape over timeDynamic demand tends to be erratic

Demand Patterns

Hypothetical historical demand pattern:• Trend• Seasonality• Random variation• Cycle.

Time Series Forecasting

AverageAverage

TrendTrend

Seasonal Seasonal

Cyclical Cyclical

Random Random

0 1 2 3 4 5

0 1 2 3 4 5

0 1 2 3 4 5

0 1 2 3 4 5

0 1 2 3 4 5

Dependent versus independent– Only independent demand needs to be

forecasted– Dependent demand should never be forecasted

Seat

Wheels

Handlebars

IV. Principles of Forecasting

ForecastsAre rarely 100% accurate over timeShould include an estimate of errorAre more accurate for product lines and familiesAre more accurate for nearer periods of time

Data Preparation and Collection

– Record data in terms needed for the forecast– Record circumstances relating to the data– Record demand separately for different

customer groups

Forecasting Techniques

• Extrinsic Techniques:– Based on external indicators– Useful in forecasting total company demand

or demand for families of products

• Intrinsic Techniques: use historical data to forecast.

V. Qualitative Techniques

Are based on intuition and informed opinionTend to be subjectiveAre used for business planning and forecasting for new products Are used for medium-term to long-term forecastingEx. Market Research, Delphi method, etc.

Example: Qualitative Methods

Grass Roots

Market Research

Panel Consensus

Executive Judgment

Historical analogy

Delphi Method

Qualitative

Methods

Strengths & Weaknesses Qualitative Methods

Type Characteristics Strengths WeaknessesExecutive opinion / Sales force composite

A group of managers meet & come up with a forecast

Good for strategic or new-product forecasting

One person's opinion can dominate the forecast

Market research

Uses surveys & interviews to identify customer preferences

Good determinant of customer preferences

It can be difficult to develop a good questionnaire

Delphi method

Seeks to develop a consensus among a group of experts

Excellent for forecasting long-term product demand, technological h d

Time consuming to develop

VI. Quantitative Forecasting

Based on historical data usually available in the companyAssume future will repeat past

Example: Historical data

Month SalesJanuary 92February 83March 66April 74May 75June 84July 84August 81September 75October 63November 91December 84January ?

Moving Average Forecasting

– Can be used to filter out random variation.– Longer periods smooth out random variation.– If a trend exists, it is hard to detect.– Manual calculations can be cumbersome

when dealing with more periods.

Simple Moving Average Problem (1)

Week Demand1 6502 6783 7204 7855 8596 9207 8508 7589 892

10 92011 78912 844

F = A + A + A +...+Ant

t-1 t-2 t-3 t-n

• Question: What are the 3-week and 6-week moving average forecasts for demand?

• Assume you only have 3 weeks and 6 weeks of actual demand data for the respective forecasts

500

600

700

800

900

1000

1 2 3 4 5 6 7 8 9 10 11 12

Week

Dem

and Demand

3-Week

6-Week

Plotting The Moving Averages and comparing them shows how the lines smooth out to reveal the overall upward trend in this example.

Example: Weighted Moving Average

F = w A + w A + w A +...+w At 1 t-1 2 t-2 3 t-3 n t- n

w = 1ii=1

n

∑

While The Moving Average Formula implies an equal weight being placed on each value that is being averaged, The “Weighted Moving Average”permits an unequal weighting on prior time periods.

wt = weight given to time period “t”occurrence. (Weights must add to one.)

The formula for the moving average is:

Weighted Moving Average Problem (1) Data

Weights: t-1 .5t-2 .3t-3 .2

Week Demand1 6502 6783 7204

Question: Given the weekly demand and weights, what is the forecast for the 4th period or Week 4?

Note that the weights place more emphasis on the most recent data, that is time period “t-1”.

Weighted Moving Average Problem (1) Solution

Week Demand Forecast1 6502 6783 7204 693.4

F4 = 0.5(720)+0.3(678)+0.2(650)=693.4

Weights: t-1 .5t-2 .3t-3 .2

Weighted Moving Average Problem (2) Data

Weights: t-1 .7t-2 .2t-3 .1

Week Demand1 8202 7753 6804 655

Question: Given the weekly demand information and weights, what is the weighted moving average forecast of the 5th period or week?

Weighted Moving Average Problem (2) Solution

Week Demand Forecast1 8202 7753 6804 6555 672

F5 = (0.1)(755)+(0.2)(680)+(0.7)(655)= 672

Weights: t-1 .7t-2 .2t-3 .1

2. Exponential Smoothing

– Provides a routine method of updating item forecasts

– Works well for stable items– Is satisfactory for short-range forecasts– Detects trends, but lags them– Note: Exponential smoothing gives the same

results as a moving average but without the need to retain as much data and with easier calculations. It works well when dealing with stable items.

Exponential Smoothing Model

• Premise: The most recent observations might have the highest predictive value.

• Therefore, we should give more weight to the more recent time periods when forecasting.

Ft = Ft-1 + α(At-1 - Ft-1)α = smoothing constant

Exponential Smoothing Problem (1) Data

Week Demand1 8202 7753 6804 6555 7506 8027 7988 6899 775

10

• Question: Given the weekly demand data, what are the exponential smoothing forecasts for periods 2-10 using α=0.10 and α=0.60?

• Assume F1=D1

Week Demand 0.1 0.61 820 820.00 820.002 775 820.00 820.003 680 815.50 820.004 655 801.95 817.305 750 787.26 808.096 802 783.53 795.597 798 785.38 788.358 689 786.64 786.579 775 776.88 786.61

10 776.69 780.77

Answer: The Respective Alphas columns denote the forecast values. Note that you can only forecast one time period into the future.

Exponential Smoothing Problem (1) Plotting

500600700

800900

1 2 3 4 5 6 7 8 9 10

Week

Dem

and Demand

0.1

0.6

Note how that the smaller alpha the smoother the line in this example.

Exponential Smoothing Problem (2) Data

Question: What are the exponential smoothing forecasts for periods 2-5 using a =0.5?

Assume F1=D1

Week Demand1 8202 7753 6804 6555

Exponential Smoothing Problem (2) Solution

Week Demand 0.51 820 820.002 775 820.003 680 797.504 655 738.755 696.88

F1=820+(0.5)(820-820)=820 F3=820+(0.5)(775-820)=797.75

periods all for sales Averagesales average Period =index Seasonal

3. Seasonality

– Measures the amount of seasonal variation of demand for a product

– Relates the average demand in a particular period to the average demand for all periods

Quarter Average Quarterly Sales/100 Seasonal Index1 128/100 = 1.28 (Q1)2 102/100 = 1.02 (Q2)3 75/100 = 0.75 (Q3)4 95/100 = 0.95 (Q4)

Total = 4.00

Sales History

Year Quarter

Total

1 2 3 4 1 122 108 81 90 401 2 130 100 73 96 399 3 132 98 71 99 400

Average 128 102 75 95 400

Example: Developing Seasonal Sales Indexes

units 1004

400quarters all for sales Average ==

Seasonal Sales

Average Salesfor All Periods

VII. Forecast Error

BiasRandom variationMean Absolute Deviation (MAD)

1. Forecast Error: Bias

Month Forecast Actual

Monthly Cumulative Monthly Cumulative

100 110

235

355

480

610

720

-

200

300

400

500

600

-

1 100 110

2 100 125

3 100 120

4 100 125

5 100 130

6 100 110

Total 600 720

Forecast and actual sales with bias:

2. Forecast Error: Random Variation

Month Forecast Actual Variation (error)

1 100 105 5

2 100 94 -6

3 100 98 -2

4 100 104 4

5 100 103 3

6 100 96 -4

Total 600 600 0

Forecast and actual sales without bias:

Forecasts Can Be Inaccurate in Two WaysBias Random Variation

Cumulative sales may not be the same as forecast

Sales will vary plus and minus about the average

Month Forecast Actual Variation Forecast Actual Variation

1 100 90 -10 100 105 +5

2 100 125 +25 100 94 -6

3 100 120 +20 100 98 -2

4 100 125 +25 100 104 +4

5 100 120 +20 100 103 +3

6 100 110 +10 100 96 -4

Cumulative Total 600 690 +90 600 600 0

Bias exists since cumulative variation is not zero.

There is no bias since cumulative variation is zero.

MAD = A - F

n

t tt=1

n

∑ 1 MAD 0.8 standard deviation1 standard deviation 1.25 MAD

≈≈

• The ideal MAD is zero. That would mean there is no forecasting error.

• The larger the MAD, the less the desirable the resulting model.

MAD Problem Data

Month Sales Forecast1 220 n/a2 250 2553 210 2054 300 3205 325 315

Question: What is the MAD value given the forecast values in the table below?

MAD Problem Solution

MAD = A - F

n=

404

= 10t t

t=1

n

∑

Month Sales Forecast Abs Error1 220 n/a2 250 255 53 210 205 54 300 320 205 325 315 10

40

Note that by itself, the MAD only lets us know the mean error in a set of forecasts.

Normal Distribution with Mean=0 and MAD=1

Error berada pada +-1 MAD 60%,; +- 2 MAD 90%; +-3 MAD 98%

60%

90%

98%

VIII. Evaluasi Forecast

Seberapa besar simpangan forecast?

Bagaimana kita mengevaluasi ini?

Forecasts

Demands

Evaluasi Forecast

Forecast error:

Mean Absolute Deviation

Mean Squared Error

Mean Absolute Percent Error 100/)/1(

)/1(

)/1(

1

1

2

1

×⎥⎦

⎤⎢⎣

⎡=

=

=

−=

∑

∑

∑

=

=

=

n

iii

n

ii

n

ii

ttt

DenMAPE

enMSE

enMAD

FAe

Tracking Signal Formula

• The TS is a measure that indicates whether the forecast average is keeping pace with any genuine upward or downward changes in demand.

• Depending on the number of MAD’s selected, the TS can be used like a quality control chart indicating when the model is generating too much error in its forecasts.

• The TS formula is:

TS =RSFEMAD

=Running sum of forecast errors

Mean absolute deviation

Tracking Signal

• Monitor, hitung tracking signal

• Range yang baik jika –4 ≤ TS ≤ +4

( )MAD

ForecastActual∑ −=Signal Tracking

Contoh TS: Diketahui MAD item adalah 2. Triggerkurang lebih 4. Pada periode ke berapa review dilakukan?

Period Forecast Actual Deviation Cumulative Deviation TS

5 2.5

1 100 96

2 100 98

3 100 104

4 100 110

Contoh TS: Diketahui MAD item adalah 2. Triggerkurang lebih 4. Pada periode ke berapa review dilakukan?

Period Forecast Actual Deviation Cumulative Deviation TS

5 2.5

1 100 96 -4 1 0.5

2 100 98 -2 -1 -0.5

3 100 104 4 3 1.5

4 100 110 10 13 6.5

A Plot of The Tracking Signals

Monitoring Forecast Akurasi

Monitor simpangan forecast setiap periode untukmelihat apakah masih dalam batas kendali

0

-10

10

Fore

cast

Err

or

Forecast PeriodLower Limit

Upper Limit

MAD25.1=σ

MAD25.1=σ