M17 - Tutorial Time series and Panel data Flashcards
within-transformation
eliminating alpha by sustracting from each variable for each unit the mean over time
- -> averages the equation over time for each person
- -> no individual interferences
Time series analysis
- def
- time series on … or … variables
- serves to
= time-ordered set of values of a variable x which was measured at several points in time
–> time-series on one/more variable/-s
- serves to:
- -> describe the ytemporal change of a variable
- -> forecasting
Procedure of Time series
- Visualization of the time series (are there trends, structural breaks?)
- Formulation of the model
- Estimation of the model (which one has the highest R², but not solely base on it)
- Preparation of the forecasts
- Assessment of the forecasting quality
Formulation adn Estimation of time series model
- basic principle
time series decomposition
- trend component
- cyclic component
- random component
why does SPSS drop one dummy?
-
if all would be included, there would be perfect collinearity
–> the last dummy is captured by the constant, so all others refer to this dummy
–> if alpha exists: if there is noa lpha, we can include all dummys
Forecast error
the difference between the actual or real and the predicted or forecast value of a time series
–> the bigger k = the mor einto the future i look, the higher the forecasting error
Projection interval
Area in which the unknown value will lie with a certain probability
With 95% confidence are the values for YT+k within the predicted value 1 and value 2 in the period T+k
- 95% is not 100%
- there could have been exogenous shocks
Panel data
- def
- two types
- two formats
= contain for the same observation units data for several points in time
- balanced panel: all obs units have been included in all time perdios
- unbalanced panel: obs are not available for all obs units for all time periods
- wide: each line = one obs unit, time-varying variables are stored in separate columns
- long: each column = one variable ; each obs is integrated several times in the data set, set IDs
disadv
- panel mortality
- data collection takes time
- high costs
Mannheim Innovation Panel
- why do they bother to do the calls for non-respondents?
- innovation behaviour of german economy in any industries
- if there is a reason why companies are not responding, the results might be baised
Pooled regression
- the result of pooled regr is biased, because…
- the disturbance term is correlated with …
- 2 regression models:
- the result of pooled rgr is biased, because it does not consider heterogeneity
- the disturbance term is correlated with the dependent variable (violation of an OLS assumption)
- models:
–> fixed effects
–> random effects
are there any unobservable influences that correlate with the IV?
what does within-transformation do?
fixed effects disappear
fixed effects: individual heterogeneity
Hausman test
- def
- tests if …
- H0
to assess whether to use fixed or random effects estimation
–> tests if the unobserved effects alphai are independent of the explanatory variables xj
- H0: alphai is distributed independently of Xj
- -> if rejected: RER will be subject to unobserved heterogeneity–> use FER (not randomly /dep of Xj)
- -> if not rejected: FER and RER are consistent, but FER will be inefficient –> use RER (randomly /indep of Xj)
Adv of FE model
the influences of all atributes that are constant over time are considered, irrespectve of whether or not they are observed
Herfindahl index
measure of concentration/ fragmentation
–> large H : large fragmentation (Monopol)
variable time in time series analysis
time can be seen as proceeding completely steadily and unaffected of any other events
