Transcript New Data Sources

NEW DATA SOURCES
2 Cases
• Varying Data Sources
• Simulation-Based
OBJECTIVES
■
■
■
needed to understand factors
affecting volume performance
needed to quantify impact of
each factor
possible forecast of immediate
future performance
 volume decomposition
 quantified impact of each
factor
 effects of short/long term
activities (promo, media)
Classes of Data
Internal

Sales by Subcategories
Government
 economic indicators ( GNP, inflation, P-$, unemployment, etc)
 weather ( temperature, rainfall , humidity)
Competitive




Retail Audit
Consumer confidence
Media TARP
Events/Promotions
APPROACH
Pre- analysis
Estimate of Base Volume
Volume Decomposition
Effect of Short term activities
Simulations
PRE-ANALYSIS
– assigned dummy variables for events (
month of occurrence)
Group of Indicators
 Economy
 Weather
applied data reduction technique to
address problem on short time series and
no. parameters > no. of observations
used principal components per group of
indicators to come up with composite
scores
 Humidity
 Rainfall
 Media
 Distribution
 Promotion
 Events
 Pricing
 Competitor Act.
Predictive Analytics
All factors affect volume but of varying degrees
Economy
Weather
Media
Distribution
Promotion
Events
Pricing
Competitor
Activities
SALES
Publication
Monte Carlo Simulation
■ Stochastic methods to generate new configurations of a system
of interest – simulation of a phenomena
■ Monte Carlo: importance sampling or systems at equilibrium.
– Start: initial configuration of the system
■ can be data-based random variable generation
– Change the configuration
■ acceptance/rejection of changes
Monte Carlo Simulation
■ Given a data-generating mechanism
– Example: drawing colored balls from an urn, input-output
model, adaptive sampling, etc.
– model of the process you wish to understand
– produce new samples of simulated data, replicate current
data
– examine results of those samples
– may also amplify this procedure with additional
assumptions
Monte Carlo Simulation
■ Computer Simulation/Monte Carlo Models
– Not solved by mathematical analysis but are used for
numerical experimentation.
– Goal of Numerical Experimentation: Answer questions of
real world (What if-sensitivity analysis)
– Purpose of Sensitivity Analysis
■
Validation of the model
– Would the customers exhibit similar credit
behavior?
– Are their credit behavior similar?
Simulation
■ Big Data
Big Data
■ Machine
Learning/Modeling
Leading Indicators
■ Simulation
Validation
■ Estimation
■ Validation
Model-Based Estimation
Simulation Procedures (Resampling)
■ Construct a simulated universe
– composition similar to the universe whose behavior we wish to
describe and investigate.
■ Specify the procedure that produces a pseudo-sample
– simulates the real-life sample in which we are interested
– specify procedural rules by which the sample is drawn from the
simulated universe (purposive sampling)
■ Describe: if several simple events must be combined into a composite
event
■ Calculate the probability of interest
– estimate parameters
– test hypothesis
– Based on tabulation of outcomes of the resampling trials.
THANK YOU.