time series analytics
DESCRIPTION
Time Series AnalyticsTRANSCRIPT
Time Series AnalyticsPresented By:
Harsh Narula
Copyright © 2015 Accenture All rights reserved. 2
• Joined Accenture 5 years back (First company YAY!)
• Working on Data Warehousing domain since 2010
• Have expertise on SAP BODS, Microsoft SSIS, SSRS, SQL Server, Informatica, IDQ etc.
• Worked with Clients like Microsoft, Farmers Insurance, Intertek (Current)
• Microsoft certified Business Intelligence Developer
• Love Travelling, Eating, Jazz Music, Cooking, and Bartending (I make good cocktails, yes)
• Aspiring data scientist, working on a model to forecast solar radiation to help qualify future
sites for Solar Power Station Installation (Sounds cool right? It’s NOT! )
About Me
Copyright © 2015 Accenture All rights reserved. 3
Analytics – Key to Success
“With the rising complexity of global business, gut decisions and hunches no longer suffice. Successful responses to threats and opportunities now depend on rapid and smart execution. Let me state it plainly: Business analytics is the key to achieving these challenging objectives”-Jim Goodnight, CEO, SAS
“Human beings often make a long list of excuses not to be analytical, but there’s plenty of research showing that data, facts, and analysis are powerful aids to decision making, and that the decisions made on them are better than those made through intuition or gut instinct” – Tom Davenport, Jeanne G Harris and Robert Morrison ( Analytics at Work)
Copyright © 2015 Accenture All rights reserved. 4
Analytics –What is it? What does it do?
• Analytics is the science of analysis. It makes extensive use of data, statistical and
quantitative analysis, explanatory & predictive modeling, and fact-based
management to drive decision making
• Analytics may be used as input for human decisions or may drive fully automated
decisions
Copyright © 2015 Accenture All rights reserved. 5
Why Business is investing in Analytics?
Competitive Advantage
Gather insights into the target markets
Align customers to the right products
Adjust to the changing behavior and expectation of
new generation customers.
Help you understand customer’s buying patterns
Reduce the cost of doing business
Generate more revenue and profits
Acquire more customers
Helps you to know which customers are risk prone
and what are the ways to retain them.
Helps you to manage risks.
Increase value, execution into business processes.
Copyright © 2015 Accenture All rights reserved. 6
Business Impact of Actionable Insights
• Strategic Planning
• Consumer Management
• Product Management &
Development
Use Consumer Insights to drive Innovation & improve Customer Experience
Social Media Analytics
Consumer trend analysis
New Product Forecasting
CRM Analytics
Marketing
Sales & Account Planning
FinanceIncrease ROI on Spend
Business Impact Business Function Actionable Insights
Marketing Mix Analytics
Trade Spend Analytics
Variable Cost Estimation
Profit Modeling
Channel Management
Procurement,
Distribution & Logistics
Increase Collaborationw/ Channel & Suppliers
Demand Signal Repository
Returns Management
Partner Performance Analytics
Channel Profitability
• Manufacturing• Logistics & Distribution• Regulatory and Compliance• Energy Management
Achieve Operational Excellence
Traceability Analytics
Production data analytics
Carbon Management
Spare Parts Optimization
• Supply Chain• Manufacturing• Regulatory and Compliance
Optimize End-to-end Costs
Inventory Optimization
Warranty Analytics
SKU rationalization
Plant Efficiency
Copyright © 2015 Accenture All rights reserved. 7
What is Time Series?
Copyright © 2015 Accenture All rights reserved. 8
Introduction
• A Time Series is a sequence of data points, typically consisting of successive
measurements made over a time interval
• Time Series analysis comprises methods for analyzing time series data in
order to extract meaningful statistics and other characteristics of the data
Few time series patterns:
Copyright © 2015 Accenture All rights reserved. 9
Some Trends
Source: KD Nuggets Polls
Da
taty
pe
sU
se
d
Fo
reca
stin
g T
ech
niq
ue
Use
d
Copyright © 2015 Accenture All rights reserved. 10
Relevance of Time Series
• Energy and financial markets generates complex
data which need special treatment
• Uncertainty, randomness, seasonality, volatility,
and non-stationarity related to these markets
sometimes make it difficult to take appropriate
business decisions
• Electricity pool markets are said to be more
complex than stock markets
• The proposed course employs advanced data
analytics techniques for modeling and
forecasting of energy and financial markets
• These techniques can also be applied in areas
like marketing, sales etc.
Copyright © 2015 Accenture All rights reserved. 11
Time Series Forecasting can be done through various
methodologies:
1. Moving Averages
2. Regression
3. Auto Regression Moving Averages (ARIMA)
4. Generalized Auto Regressive Conditional
Heteroscedasticity (GARCH, eGARCH, mGARCH,
sGARCH etc.)
5. Cointegration and Vector Auto Regression
Time Series Techniques
Copyright © 2015 Accenture All rights reserved. 12
• The n-period moving average builds a forecast by averaging the observations in the most recent n periods:
where xt represents the observation made in period t, and At denotes the moving average calculated after making the observation in period t.
What is Moving Average?
At = (xt + xt–1 + … + xt–n+1) / n
13
A statistician can have his head in an oven and his feet
in ice, and he will say that on the average he feels fine.
Copyright © 2015 Accenture All rights reserved. 14
Time Series Forecasting can be done through various
methodologies:
1. Moving Averages
2. Regression
3. Auto Regression Moving Averages (ARIMA)
4. Generalized Auto Regressive Conditional
Heteroscedasticity (GARCH, eGARCH, mGARCH,
sGARCH etc.)
5. Cointegration and Vector Auto Regression
Time Series Techniques
Copyright © 2015 Accenture All rights reserved. 15
• Regression analysis is concerned with the study of the relationship
between one variable called explained or dependent variable (y) and
one or more other variables called independent or explanatory
variables (x1, x2……xn)
Y = f (x1, x2……xn)
What is Regression?
Copyright © 2015 Accenture All rights reserved. 16
How Regression is Done?
Theory: y = f(x1, x2 …..x)
Mathematical model of theory; y = B1 + B2 x2 + B3 x3 … + Bn xn (Deterministic)
Data collection on y, x2 , x3 …. (time series, cross-sectional, panel
Estimation of sample b1, b2 ….bn through OLS (min RSS)
Hypothesis testing (t- stat Ho : B2 = 0, F- stat Ho : B2 = B3 =… Bn = 0 )
Diagnostic Checking (errors free from M-C, H-C, A-C & e ~ N (0, σ2)
Use estimated model for information extractions and/or forecasting
Explanatory Power of the Model (R2 / Adj R2 )
17
Problems with Linear
Regression:
1. Only looks at Linear Relationships
2. Only looks at the mean of the
Dependent Variable
4. It is Sensitive to Outliers
5. Data must be Independent
Copyright © 2015 Accenture All rights reserved. 18
Time Series Forecasting can be done through various
methodologies:
1. Moving Averages
2. Regression
3. Auto Regression Moving Averages (ARIMA)
4. Generalized Auto Regressive Conditional
Heteroscedasticity (GARCH, eGARCH, mGARCH,
sGARCH etc.)
5. Cointegration and Vector Auto Regression
Time Series Techniques
Copyright © 2015 Accenture All rights reserved. 19
Why ARIMA, GARCH, and VAR
• Traditional estimation
methods look only
towards the past
values, average it, and
forecast it.
• However, they do not
account for volatility,
trends, and
seasonality etc.
• These models can
take into account all
these factors, and
produce accurate
results.
Copyright © 2015 Accenture All rights reserved. 20
Forecasting With Time Series Models
• Two important features:
• Uses historical data for the phenomenon we wish to forecast
• We seek a routine calculation to apply to a large number of cases and that may be automated, without relying on qualitative information about the underlying phenomena
• Short-term forecasts are often used in situations that involve forecasting many different variables at frequent intervals
• We separate the base component, forecast the base, and apply seasonality and trends to the forecasted series
Base Series
Trends
Seasonality
Copyright © 2015 Accenture All rights reserved. 21
How We Forecast
Copyright © 2015 Accenture All rights reserved. 22
How We Forecast
Copyright © 2015 Accenture All rights reserved. 23
How We Forecast
Time
Co
nsu
mption o
f C
oke
• Remove Trends
• Remove Seasonality
• Achieve White Noise
• Forecast Stationary
Series
• Apply Trends and
Seasonality back
Copyright © 2015 Accenture All rights reserved. 24
Summary
Time
Series
Is
Pure
Science
Copyright © 2015 Accenture All rights reserved. 25
Questions