• No results found

Improving Demand Forecasting

N/A
N/A
Protected

Academic year: 2021

Share "Improving Demand Forecasting"

Copied!
14
0
0

Loading.... (view fulltext now)

Full text

(1)

2nd July 2013

John Tansley - CACI

(2)

Overview

The ideal forecasting process:

 Efficiency, transparency, accuracy

Managing and understanding uncertainty:

 Limits to forecast accuracy, including the Poisson limit

The Forecast Value Add approach:

 From simple to more complex models

Types of forecasting model:

 Econometric, quantitative, and combined

Types of quantitative models:

 Time series, explanatory, combined

Case study 1:

 Improved call volume forecasting for financial services debt management

Case study 2:

(3)

The ideal forecasting process

 Goal of forecasting process:

 Provide the best possible forecast, given the information available

 ‘Best’ means:

 Efficiency:

 Automate data feeds as much as possible

 Transparency:

 Understandable (avoiding black box or overly complicated Excel)

 Accuracy:

 Self explanatory!

(4)

Managing and understanding uncertainty

 Best possible accuracy is outside the control of the analyst  Factors that affect accuracy:

 Lack of all necessary information:

 Only have access to limited data  Problem changes rapidly over time

 Inaccuracies in known information:

 Inaccurate data

 Incorrect mental model of the business problem

 Fundamental limits to accuracy

(5)

Poisson accuracy limit

 When forecasting counts, there is a fundamental limit of achievable

accuracy – the Poisson limit

08:00:00 09:00:00 10:00:00 11:00:00 12:00:00 13:00:00 14:00:00 15:00:00 16:00:00 17:00:00 18:00:00

100 calls in a 10h day - equally spaced

Call

08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00

100 calls in a 10h day - random

(6)

Poisson accuracy limit

 When forecasting counts, there is a fundamental limit of achievable

accuracy – the Poisson limit

08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00

1000 calls in a 10h day

Call 113 107 109 87 105 96 88 103 110 82 0 50 100 150 08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00

(7)

Poisson accuracy limit

 Demand in a particular time period or location is generally distributed

according to a universal distribution – the Poisson distribution.

 The spread of this distribution is around the square root of the demand volume

 Understanding this limit helps to

 Set reasonable accuracy expectations Demand

Mean Spread Spread (%)

9 3 33%

25 5 20%

64 8 13%

100 10 10%

(8)

0 20 40 60 80 100 120 140 160 180 0 5 10 15 20 E rr o r

Forecast error versus model complexity

Error Best model

Naïve model performance Poisson limit performance

The Forecast Value Add approach

Start simple

Additional complexity is only worth it if it increases accuracy

Can measure by how much each incremental step improves the

forecast

No point adding complexity if forecast error increases

(9)

Types of forecasting models

Econometric Models

• Manually built

models

• Small datasets (if

any)

• Manual setup

• Based on business

knowledge

Bayesian

Econometric Models

• Manual model

structure

• Model parameters

set from user

constraints and

data

• Based on both

business

knowledge and

data

Quantitative Models

• Automatic models

• Larger datasets

• Little user control

over parameters

• Based on data

• Examples:

Regression,

Decision Trees

Knowledge

(10)

Types of quantitative models

 .

Time input1 input2 input3 Target

1 2 3 4 explanatory time series combined

Regression, Decision Trees, Neural Networks

Weekly profile, moving

average, ARIMA, Exponential smoothing,

ARIMA with drivers,

Decomposition Forecasting Use drivers, add insights

Good for trends

(11)

Case study 1 – Improving the forecasting process

Improved call volume forecasting for the debt management function

 CFS were creating forecasts in large Excel sheets, populated manually  CFS had a desire to improve process, and remove single man dependency  Solution:

 Software: statistical forecasting models  Automation of data feeds

 Results:

 Reduced single man dependency

 Immediately showed increased speed (from 2.5 to 1 day) and

transparency

(12)

Case study 2 – Long term demand forecasting

 30 year water demand forecasting for a large water board

 Yearly forecasts across 10-20 geographical areas, and 10-20

business sectors

 A few years’ worth of demand, economic and weather data

 Approach: Bayesian Econometric Models

 Allows the model structure to be set by the user  Model parameter estimates are set by the user

 Model parameters are then calibrated on existing data

 Result:

 Forecasts that combine the best business knowledge and the data  Parameters can be set by the user if needed, or dictated solely by

(13)

Wrap-up

 A large number of techniques are currently available for forecasting,

the key is choosing right technique for right problem

 A good forecasting approach should add insight as well as accuracy  Forecast Value Add approach: keep it as simple as possible

 Key is to keep on top of models – keep them understandable and easy

to update

(14)

References

Related documents

For prerequisites and setup instructions, see “ Setting Up Network Printing ” in the Novell Small Business Suite Installation and Administration manual!. Set up

discusiones y malentendidos alrededor de su obra también habría sido responsable el propio Althusser, se afirmaba, en tanto sus recortes y aumentos en las ediciones de sus

Comparison to the Omni-Optimizer While the Omni-optimizer uses the same SBX crossover operator as DIOP, it uses an adaption of polynomial mutation with η m = 20 [4] instead of

unopened flowers on the plant. ACS and ACO mRNA levels in pedicels of open and unopened Plectranthus flowers kept in 16 h light: 8 h dark, 24, 48, 72 and 96 h dark.* marks

Europe France, Centre: Eure-et-Loir, Chartres Tour FW View from cathedral grounds.. 20,

SAR and bioactivity results indicated that carboxamide group on C-3 and hydrogen and/or fluorine at C-6 of the benzimidazole scaffold are important for PARP-1 enzyme

In addition to the frequent and the subjective Bayesian views on the meaning of probability, there is a third established view, which attributes non-frequentist but

Through the use of portable HDMI pattern generator MODEL 72-7480, you are able to use 48 timings and 34 patterns, and operate it continuously for 6~8 hours after the battery has