Forecasting GDP growth Torsten Lisson (contact: [email protected]) Emanuel Gasteiger (contact: [email protected]) Note: This talk was given at the class „Economic Forecasting“ of Prof. Robert. M. Kunst at the University of Vienna, Austria on January the 9 th 2007.
40
Embed
Forecasting GDP growth - Zentraler Informatikdiensthomepage.univie.ac.at/robert.kunst/070107_efc.pdf · Forecasting GDP growth Torsten Lisson ... We use public data of the Austrian
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Note: This talk was given at the class „Economic Forecasting“ of Prof. Robert. M. Kunst at the University of Vienna, Austria on January the 9th 2007.
Exhibit 2
Introduction
Model-free forecast
Model-based univariate forecast
Model-based multivariate forecast
Discussion of results
Exhibit 3
Basic idea to this study steams from the Keynesian economy
How close are our forecasts of model-free and model-based procedures to forecasts ofleading research institutions?
!
Y = C +G + I + (EX " IM)
How do forecasts develope if we use the components of the GDP according to a Keynesianeconomy?
Exhibit 4
We use public data of the Austrian GDP and its components onquarterly basis
Data source: Statistical Office of the European Communities (EUROSTAT)http://ec.europa.eu/eurostat
Data set: Country: Austria Period: 1988q01 to 2006q03 (75 observations) Variables:
GDP (Yt) Household and non-profit sector consumption (Ct) Government expenditures (Gt) Gross investment (It) Exports of goods and services (EXt) Imports of goods and services (IMt)
Unit: mio. EURO fixed prices (base year is 1995)
Exhibit 5
GDP over time shows clear cyclical patterns and trending
Cyclical patterns:
q1 to q2 ⇑
q2 to q3 ⇑
q3 to q4 ⇑
q4 to q1 ⇓
Trends: More ressources Technological progress (Inflation in nominal
GDP)
Exhibit 6
The time series of GDP growth rate appears to be stable butcyclical patterns remain
!
ˆ y t =(Yt "Yt"1
)
Yt"1
The growth rate:
Exhibit 7
Introduction
Model-free forecast
Model-based univariate forecast
Model-based multivariate forecast
Discussion of results
Exhibit 8
We choose the Holt-Winters Seasonal Smoothing method
Why Holt-Winters method?
Which parameter shall one choose?
Which form shall one choose?
Exhibit 9
Recall the Holt-Winters method
Multiplicative version: Additive version:
!
Lt="(X
t# S
t#s) + (1#")(Lt#1 + T
t#1)
!
Tt
= "(Lt# L
t#1) + (1#")Tt#1
!
St= "
Xt
Lt
+ (1# ")St#s
!
X
^
N (h) = LN
+ TNh + S
N +h"s
!
Lt="
Xt
St#s
+ (1#")(Lt#1 + T
t#1)
!
Tt
= "(Lt# L
t#1) + (1#")Tt#1
!
St= "
Xt
Lt
+ (1# ")St#s
!
X
^
N (h) = (LN
+ TNh)S
N +h"s
Note: Stata derives the starting value from the mean of the first half of the samples’ observations by default
Exhibit 10
We evaluate the procedures by the predicted mean squarederror (PRMSE)
!
ˆ Y t
!
Yt
88q1 05q3!
PRMSE =1
n( ˆ Y
i
i= 72
n
" #Yi)
2
06q3
used observations forprediction
Evaluation method:
benchmarkobservations
forecast
predictions
Idea: we want a good forecast, not the best model fit
Note: For all further analysis we use the sub-sample of 71 observations
Exhibit 11
Evaluation results favour the additive (0.3; 0.3; 0.3) method