Forecasting Macroeconomic Variables using Artificial Neural Network and Traditional Smoothing Techniques


Journal of Applied Finance & Banking, vol.3, pp.73-104, 2013 (Peer-Reviewed Journal)

  • Publication Type: Article / Article
  • Volume: 3
  • Publication Date: 2013
  • Journal Name: Journal of Applied Finance & Banking
  • Journal Indexes: IBZ Online, ABI/INFORM, EconLit
  • Page Numbers: pp.73-104
  • Istanbul University Affiliated: Yes


For  many  years,  economists  have  been  using  statistical  tools  to  estimate  parameters  of
macroeconomic models. Forecasting plays a major role in macroeconomic planning and it
is  an  essential  analytical  tool  in  countries’  economic  strategies.  In  recent  years,
researchers  are  developing  new  techniques  for  estimation.  Most  of  these  alternative
approaches  have  their  origins  in  the  computational  intelligence.  They  have  the  ability to
approximate  nonlinear  functions,  parameters  are  updated  adaptively.  In  particular,  this
research  focuses  on  the  application  of  neural  networks  in  modeling  and  estimation  of
macroeconomic  parameters.  Neural  networks  have  received  an  increasing  amount  of
attention  among  macroeconomic  forecasters  because  of  the  ability  to  approximate  any
linear and  nonlinear relationship  with a reasonable  degree  of accuracy. Turkey  is  one  of
the  European  Union  candidate  countries  such  as  Iceland,  Montenegro,  Serbia  and  The
Former  Yugoslav  Republic  of  Macedonia.  In  this  study  eight  macroeconomic  indicators
including  gross  domestic  product  (volume,  NGDPD),  gross  national  savings
(NGSD_NGDP),  inflation  (average  consumer  prices,  PCPI),  population  (LP),  total
investment  (NID_NGDP),  unemployment  rate  (LUR),  volume  of  exports  of  goods  and
services  (TX_RPCH),  volume  of  imports  of  goods  and  services  (TM_RPCH)  were  used
for  forecasting.  As  analysis  tools,  classical  time  series  forecasting  methods  such  as
moving  averages,  exponential  smoothing,  Brown's  single  parameter  linear  exponential
smoothing,  Brown’s  second-order  exponential  smoothing,  Holt's  two  parameter  linear
exponential smoothing and  decomposition  methods applied to  macroeconomic  data. The
study  focuses  mainly  on  the  applicability  of  artificial  neural  network  model  for
forecasting  macroeconomic  parameters  in  long  term  and  comparing  the  artificial  neural
network’s  results  with  the  Traditional  Time  Series  Analysis  (Smoothing  &
Decomposition  Techniques).  To  facilitate  the  presentation,  an  empirical  example  is
developed to forecast Turkey’s eight important macroeconomic parameters. Time Series
statistical theory and methods are used to select an adequate technique, based on residual analysis.  Turkey  will  celebrate  the  100th  anniversary  of  its  foundation  in  2023.  Policies
and implementations targeted for raising economic position.