the rows correspond to observations of the same observational unit, in different time periods.
we can not draw them randomly since the periods come på följande varandra. The gap is the same between them and it can be yearly/quarterly/monthly/daily and they are in fixed intervals
that we dont observe in between the intervals
we assume that time progresses smoothly and can therefore draw lines between the observations (dots) but in reality it could be very different in between.
They look different for each observation. We draw different univesres with their own counttries i and and evolution of GDP. We can therefore not look at one graph and assume the same for another.
we see time series as a collection of random variables, so we draw a new random variable for each time, dvs Y1 Y2 Y3 osv.
då the economics change in between years
predictions about the future
predicting the current period based on data that is available, but this data is often partially unobserved and observed.
stationarity and weak dependence
innan time series börjar, så ska vi tro att segmenen (innehållade ett antal t) har samma values. Detta assumption gör dem "identical". Vi vill lära oss om ett typical behaviour over time, men för att vi ska kunna göra det så krävs det att vi assume these typical behaviours within a segment to be the same over time and not change. Dvs the same economic rules är tillämpade i alla segment vi ska observera.
Vi säger att två segment ska ha samma distribution of Y1 Y2 Y3, så kallad joint distribution, innan vi startar time series
we assume segments A and B to be "almost" independet if they are far apart, dvs att det som händer i segment B inte ska bero på vad som hänt i segment A och att det ska finnas time dependence (serial correlation).
It ensures that we get new information och att våra SE blir mindre
de båda makes sure that we always learn about the same economic problem, dvs the economic rules persist. Med andra ord så repeatedly observes the same problem.
it makes sure that we get new information from the observations
moving from segment A to segment B, is like moving from one observation to the next as in cross section
det är svårt att motivera/verify imperically if a realized time series was generated by a process that was stationary and weakly dependent
how do the gdp growth today correlate with the gdp growth tomorrow
the covariance could potentially depend on time
independent sampling and they have to be identical, dvs the economics får inte ha changed
when picking samples, they have to be identiical and not depend on each other (no time dependence), hence stationarity
stationarity. It rules out that time series is systematically increasing or decreasing. Same for the variance
det visar att time series inte är stationary and that uncertainty increases over time
när observations of the same time series at different times are correlated
correlation between på två följande år. "tomorrow is built on what happens today"
a predictable pattern that persists over long time periods, exempelvis GDP
it can not be stationary. It consists of a part that is a fixed function of time (not good) and one that is due to random fluctuation
temporär trend som startar och slutar "randomly" and kan inte bli predicted from the time index
stochastic trends are incomatible with stationarity. Därför ger time series med stochastic trend inga rimliga resultat.
stochastic trends may cause very different results beroende på vilken tidsperiod vi tittar på. Det kan vara en väldigt dålig tidsperiod medan the overall trend is positive.
it is the difference between the original time series and its lag
it is the series of yesterdays values. It means that we loose one period baucase in period t=1, we cannot observe the previous periods value för att vi precis startat med vår time series
trending behaviour can be eliminated
man borde inte leta och analysera om det finns tecken av stichastiska trender
en time series that is stationary I(0) = Yt
det är en times series som blivit stationary (integrated of order zero) efter att man har gjort en first differentiating. I(1) = Yt
we have to differentiate at least twice. att göra det en gång räcker inte
SE will get bigger, as we have less new data the more serial correlation there is and the estimates becomes less percise which leads to higher SE
it accounts for serial correlation. It tells stata that the rows are not independent och den mäter serial correlation. Så våra SE blir därför större när vi använder Newey-West
use first differenciation => now stationary
change in Yt = Yt - Yt-1