Neas-Seminars

Time Series, Introduction


http://33771.hs2.instantasp.net/Topic8600.aspx

By NEAS - 12/1/2009 10:49:44 AM

Time Series, Introduction

 

(The attached PDF file has better formatting.)

 

Updated: December 1, 2009

 

This course cover ARIMA modeling for time series.

 


           Many candidates have taken college regression courses, but they lack the time series component needed for VEE credit.  You can combine the on-line time series course with your college regression course to fulfill VEE requirements.  The SOA web site has full instructions and the required forms; you must complete the paperwork to get credit.

           Candidates who have not take a college regression course can take the time series course with the regression analysis course. The courses cover complementary material.


 

 

Time series vs regression analysis

 

Jacob: How do time series models differ from regression analysis?

 

Rachel: Suppose we want to forecast next year’s interest rate.

 

We might use a regression equation linking the interest rate to various economic variables, such as Federal Reserve policy, inflation rates, GNP, the current account balance (foreign trade and investment), and other political and economic matters.  This may be impractical:

 


 

           We don’t know the effects of these items on interest rates, and we don’t know the ideal regression equation or parameters.  Economists disagree; there is no consensus.

           Even if we knew the relations and parameters, we don’t know the future values of most of these economic variables. We presume that Federal Reserve Board (FED) decisions affect interest rates, but we don’t know what the FED will do next year.

           Even if we thought that the FED will take certain action next year, the uncertainty in this prediction makes the regression analysis of limited use.


 

 

Instead, we predict next year’s interest rate from the past pattern of interest rates.  In many situations, this time series model gives as good an estimate as most regression analyses.  The time series model also gives a distribution of interest rates next year and confidence intervals on the realized interest rate.

 

Jacob: Simple time series don’t work for cyclically fluctuating variables. Suppose we model auto insurance losses month by month.  More accidents occur in winter months, when roads are icy, than summer months, when days are longer (so less night-time driving) and weather is better.  Can we predict December losses from the past half-year losses?

 

Rachel: A large part of this course deals with seasonal and other cyclical fluctuations in the time series.  We use autocorrelation functions throughout this course.  Selecting the proper time series is the hardest part; fitting parameters to the selected model is easier.