Econometric Modeling - MEC

Best practices for designing an advanced analytics .... Similar to most forms of analytics and tools, there is no one-‐size-‐fits-‐all solution when it comes to ...
10MB Sizes 0 Downloads 77 Views
 Econometric  Modeling  

BEHIND  THE  BUZZ  –  WHAT  IS   ECONOMETRIC  MODELING  AND   HOW  IS  IT  DRIVING  TODAY'S   MARKETING  DECISIONS?  

December  2015  

WHY  DOES  ECONOMETRIC   MODELING  MATTER?  

What’s  all  the  buzz  about  and  why  should  I  pay   attention  

WHAT  IS  ECONOMETRIC   MODELING?  

A  topline  look  at  how  econometric  modeling  works  

TYPES  OF  MODELS  AND  THEIR   USES  

How  and  when  is  it  appropriate  to  use  MMM,   Brand  Equity  Modeling  and  Long  Term  Modeling?  

DEVELOPING  AN  ANALYTICS   ROADMAP  

How  to  make  the  most  of  the  data  available  now   while  also  planning  for  the  future  

GETTING  STARTED  –                                                       Best  practices  for  designing  an  advanced  analytics   3  GUIDING  PRINCIPLES   strategy   KEY  TAKEAWAYS  

Key  takeaways  for  brands  considering  econometric   modeling  

WHY  DOES   ECONOMETRIC   MODELING  MATTER?   PRECISE  BRAND-­‐SPECIFIC  LEARNINGS  AT  YOUR  FINGERTIPS   How  should  we  allocate  and  optimize  our  marketing  spend  across   all  products  and  channels?  How  can  we  forecast  how  a  budget   change  will  impact  our  sales  in  one  year?  Five  years?  Ten  years?   The  data  shows  us  that  awareness  behind  our  product  is   increasing  –  why  is  this  not  translating  to  sales?   As  marketers,  these  are  some  of  the  questions  we  are  faced   with  as  we  grapple  to  understand  how  to  optimize  marketing   spend.  While  we  often  work  against  assumptions  such  as   industry  norms  or  brand  comparisons  within  a  category  to   determine  how  to  allocate  budget  or  build  awareness,  looking   at  historic  and  competitive  data  is  not  always  enough.     Today,  more  is  being  demanded  of  marketers  and  with  that   comes  the  pressure  to  precisely  quantify  the  impact  of   marketing  decisions.    Econometric  modeling  is  key  to  finding   the  answers  to  these  questions,  particularly  when  brand-­‐ specific,  precise  recommendations  and  results  are  required.     This  Spotlight  On  goes  behind  the  scenes  to  discover  how   econometric  modeling  works,  how  to  choose  the  appropriate   model  based  on  common  questions,  and  how  to  quickly  start   leveraging  analytics  for  advantage.  

DATA  CONNECTIVITY   WILL   WHAT  IS   SAVE  US  ALL   ECONOMETRIC   MODELING?   WHAT  IS  ECONOMETRIC  MODELING?    ‘Econometric  modeling’  is  a  term  that  has  become  so  prevalent,  it’s  beginning  to  sound  like  a  buzzword,  or   industry  jargon,  used  a  to  make  learnings  and  results  seem  more  scientific.       The  fact  is  econometric  modeling  sounds  scientific  because  it  is.  Econometric  Modeling  leverages  the   variation  in  granular  data  to  statistically  tease  apart  the  impacts  of  marketing  activities  on  the  KPI  of   interest.  Today,  various  types  of  regression  analysis  are  implemented  in  econometric  modeling  within  the   industry,  including:  multivariate,  time  series,  vector  auto  regression  and  Bayesian  methods.         HOW  DOES  BASIC  REGRESSION  ANALYSIS  WORK?      Generally,  we  know  that  media  over  and  under  delivers  in  certain  markets,  and  the  amount  of  over/under-­‐ delivery  varies  when  examined  on  a  weekly  level.  Therefore,  when  looking  at  a  media  plan  on  a  DMA  level,   there  is  a  significant  amount  of  variation  in  the  executed  data.  Since  there  are  210  DMAs,  that  translates  to   210  data  points  per  week.  Most  models  are  built  with  three  years  (166  weeks)  of  data,  resulting  in  32,760   data  points  per  variable.  Therefore,  if  TV,  print,  radio,  OOH  and  banner  were  all  part  of  the  media  plan,  this   would  provide  up  to  five  variables  or  163,800  data  points