CODEWITHSUNDEEP
r
bstockton
bstockton

Reputation: 577

Incorporating one autoregressive term and a moving average in a regression

I am doing a fixed effects regression and am having a problem with autocorrelation, to deal with this I am doing ARIMA modeling using the forecast, lmtest, and plm packages. My data is general panel data, looks like this, I am trying to do some ARIMA modeling but am having a hard time incorporating autoregressive terms and moving averages into a fixed effects regression using the plm package. Here is my attempt.

world_hour_fix = 
    plm(WBGDPhour ~ broadband + resourcerents + education, 
        data = hourframe, model = "within")

auto.arima(world_hour_fix$residuals)

# Series: world_hour_fix$residuals 
# ARIMA(1,0,1) with zero mean     
# 
#     Coefficients:
#       ar1     ma1
#       0.403  0.3135
# s.e.  0.138  0.1586
# 
# sigma^2 estimated as 0.4901:  log likelihood=-175.54
# AIC=357.09   AICc=357.23   BIC=366.4

auto.arima(world_fix$residuals)

How do I incorporate one autoregressive term and a moving average of one into my regression?

Upvotes: 6

Views: 7031

Answers (1)

Charlie
Charlie

Reputation: 2851

I economics, we often don't try to do ARIMA modeling with panel data. Instead, we use (quasi-) difference-in-difference estimation. If you aren't worried about non-stationarity, which it sounds like you aren't, then this paper by Bertrand, Duflo, and Mullainathan, "How Much Should We Trust Differences-in-Differences Estimates?", compares different means of taking autocorrelation into account for panel data. They find that the block bootstrap and HAC standard errors tend to work well.

Upvotes: 5

Related Questions