I've seen several posts this week around this topic, but I cannot find a right answer, probably because time series is something very artisan! I'm doing a task about finding causality between surface temperature and sea level in a specific zone. The thing is to explain the relation of the process through a CCF and Transfer Function. I have modelled my series with much care as I could. In R, I took 1 difference and logarithms for $x_t$ and only logarithms for $y_t$ (this one is really seasonal - sea level!) The best models arisen from auto.arima had been :

for $x_t$

ARIMA(2,0,1) with zero mean 

         ar1     ar2      ma1
      0.4138  0.2498  -0.9517
s.e.  0.0526  0.0511   0.0231

sigma^2 estimated as 0.006168:  log likelihood=513.04
AIC=-1018.09   AICc=-1018   BIC=-1001.6

and $y_t$

ARIMA(1,0,1)(0,1,2)[12] with drift 

         ar1     ma1     sma1    sma2   drift
      0.5607  0.4494  -0.8262  0.1173  -5e-04
s.e.  0.0493  0.0545   0.0508  0.0490   1e-04

sigma^2 estimated as 0.0009074:  log likelihood=922.19
AIC=-1832.38   AICc=-1832.19   BIC=-1807.81

Then, I got this graph from ccf(x$residuals, y$residuals), which I am triyng to interpret: enter image description here


a) In this example, at $k$ = 4, $x_{t + k}$ is significantly correlated with $y_t$? My hypotesis is that there is backfeed but I'm not sure of that.

b) How I can calculate and express the transfer function with the information above? (e.g with MTS::tfm1)

Thank you in advance. All useful, interesting and helpful comments will be rewarded!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.