What to do with Time Series Data?

[Buffy Summer]

Hi,

This might seem a very simple question, but I feel kinda confused.
I have data on Stock prices and interest rates. Lets assume that I am trying to estimate the effect of interest rates on stock prices (it's more complicated than that but let's keep it this way).

What's the correct way to analyze these data?

1) graph the data and see if there is a trend
3) look at the time correlation using command "corrgram"
2) do a Unit Root test using dfuller to check if the data are stationary (if the data are not than look at the difference)
3)if they are do whatever model you are looking at (in my case is VAR because y depends on it's privies values and also on x that depends on its previous values and y)
4) Check and correct for heteroskedastisity and autocorrelation if any.

What Am I missing? Why should I care about find the right q of AR(q) using the command varsoc? Why is that important?

Thank you so much!

Buffy

[Buffy Summer]

Hi,

This might seem a very simple question, but I feel kinda confused.
I have data on Stock prices and interest rates. Lets assume that I am trying to estimate the effect of interest rates on stock prices (it's more complicated than that but let's keep it this way).

What's the correct way to analyze these data?

1) graph the data and see if there is a trend
3) look at the time correlation using command "corrgram"
2) do a Unit Root test using dfuller to check if the data are stationary (if the data are not than look at the difference)
3)if they are do whatever model you are looking at (in my case is VAR because y depends on it's privies values and also on x that depends on its previous values and y)
4) Check and correct for heteroskedastisity and autocorrelation if any.

What Am I missing? Why should I care about find the right q of AR(q) using the command varsoc? Why is that important?

Thank you so much!

Buffy

No one?

[Andrew Musau]

Looks like you need to take a course in time series analysis!

First of all, you would be concerned about stationarity in a time series context because of spurious correlation: you may get significant relationships between variables where none exist.

In summary: Graphing will just give you a visual - which might not be very reliable. A Dickey Fuller/ Phillips Perron test is a necessary step in specifying a VAR/ VECM model (more formal and reliable than looking at a graph).

Autoregressive lags are to account for serial correlation (which may go back several lags). If the lag length is too small, the remaining serial correlation will bias your tests. On the other hand, if too large, the power of the test will suffer (generally, you would rather have a larger lag (than optimal) compared to a smaller one).

Time series analysis is a huge topic, so get a good book to grasp it well. A suggestion is James Hamilton's Time Series Analysis