Jonathan Whittington
**** **** ***** ***** *** **, Raleigh, NC, 27613
abl23a@r.postjobfree.com
Education
Masters Degree in Economics with an Emphasis on Predictive Analytics
North Carolina State University, Raleigh, NC 2010
Relevant Courses
• PhD course in Risk and Decision Theory
• PhD course in Econometric Theory focusing on variable selection
• Masters course in Applied Econometric Modeling which focused on model building and
validation for GARCH (for Portfolio analysis), LOGIT, PROBIT, TOBIT, Instrumental Variable,
ARIMA and Unobserved Component models using SAS
Technical Abilities
• SAS: SAS/MACRO and Development Environment Design, SAS/High-Performance Forecasting,
SAS/ETS, SAS/STAT, SAS/GRAPH, SAS/OR, SAS SQL, SAS DATA STEP, SAS ODS
• Programming Languages: SAS, C, R
• Analytical methods: Forecast Value Added Analysis, Decision Hierarchies, Time Series Models
(UCM, ARIMA, Holt-Winters), Applied Linear Models (LOGIT, PROBIT, TOBIT, Instrumental
Variable), Hypothesis Testing, Variable Selection
• Additional Software: Excel, PowerPoint, Word, GIS/ArcMap
Professional Experience
SAS Institute, Cary NC 2008-Present
Graduate Intern in Research and Development
Description
• Perform data mining, data cleaning, model development, model validation and software
development
• Develop software backend unit and end-to-end tests
• Train new team members on our model methodology and development environment
• Develop literature to assist team members in testing model quality and testing methodology
Accomplishments
• Investigated our modeling methodology to account for systematic errors in our forecasts and
presented my recommendations to management resulting in a 5-8% increase in forecast accuracy
• Wrote a program to automate tracking of our progress in terms of forecast accuracy as we
switched from one methodology to another over the course of several months.
• Designed a 50 page power point presentation for company consultants which explained our
modeling approach for both technical and non-technical audiences
• Wrote a 100 page technical document on our modeling methodology for offices in other parts of
the world to assist in testing and model validation
• Designed and wrote unit tests and end to end tests which increased the test coverage for our Retail
Solution from 60% to 80%
Examples of Model Specific Experience
• Task: Determine cause of systematic errors in demand forecasts for one of SAS’s customers
My Approach: I had to first ensure that data quality wasn’t the problem. Therefore I wrote a Macro
program which reverse engineered all of our customer’s data and added some random noise to the results
so the fit would not be perfect. Having eliminated the data quality issue I began examining all
components of the model which include attribute based regressions for determining price sensitivity as
well as time series components (combination of ARIMA and UCM) and other additional factors. After
inspecting the various residual plots and performing end to end tests to rule out problems with our
development code I determined one of our model components needed to take a logarithmic form in order
to normalize the errors.
Result: There was a 5-8% increase in forecast accuracy based on the MAPE before and the MAPE after
and all systematic bias was gone.
• Task: Determine cause of systematic errors in demand forecasts for one of SAS’s customers
My Approach: I had to first ensure that data quality wasn’t the problem. Therefore I wrote a Macro
program which reverse engineered all of our customer’s data and added some random noise to the results
so the fit would not be perfect. Having eliminated the data quality issue I began examining all
components of the model which include attribute based regressions for determining price sensitivity as
well as time series components (combination of ARIMA and UCM) and other additional factors. After
inspecting the various residual plots and performing end to end tests to rule out problems with our
development code I determined one of our model components needed to take a logarithmic form in order
to normalize the errors.
Result: There was a 5-8% increase in forecast accuracy based on the MAPE before and the MAPE after
and all systematic bias was gone.
• Task: Determine whether continuous progress is being made on our modeling methodology
My Approach (Currently in progress): I’m writing a Macro program that automates the tracking of our
model’s forecast accuracy. It will use a combination of measures such as R 2 of regression of forecast on
actual and MAPE to measure forecast accuracy and graph the percentage change over time. This will
allow us to determine if continued progress is being made and be able to locate model changes that were
not as effective.
• Task: Gather data and show empirical evidence for gun ownership’s effect on crime
My Approach: I gathered and cleaned public data from several government agencies and based on my
theory constructed a nonlinear, which I linearized, two ways fixed effects model relating changes in crime
to gun control laws while controlling for interactions with other effects. I used a proxy variable for
measuring gun ownership (ratio of suicides with guns to all suicides) which was readily available with
little measurement bias since almost all studies on this subject produce biased results due to the inability
to accurately measure this value. Using visual methods such as actual to predicted graphs and using
proper parameter tests I concluded right to carry laws have an effect on whether or not gun ownership
rates are correlated with crime rates. Paper and data is available at my site:
https://sites.google.com/site/makingdatamatta/home/da-papas
Result: I earned the highest grade awarded in class which was 98% for originality and methodology
• Task: Determine portfolio assignments based on market data
My Approach: I choose a GARCH(1,1) model to model market volatility and make a sound decision on
resource allocation between assets of different risk levels. Using modern portfolio theory and
optimization I determined the appropriate proportions to invest between risky assets so that for a given
willingness towards risk, which is usually teased out by financial consultants, the appropriate fraction of
wealth will be invested in U.S. treasury securities and the rest will be invested in the riskier market
portfolio.
Result: I earned a 96% on the project