Data Description

Paper is Economic Costs of Conflict: A Case Study of the Basque Country, AER (2003)

The dataset contains information from 1955-1997 on 17 Spanish regions. It was used by Abadie and Gardeazabal (2003) to study the effect of terrorism in the Basque Country on GDP.

one treated region: the Basque country.
no ex-ante control group (they use the other Spanish regions as control groups).
dependent variable: GDP
treatment: terrorist activity
year where clear break in GDP trend: 1975
study the impact of terrorist activity on GDP growth, before and after 1975

The data set is full described here

Results of 2003 paper Effect reaches a max of \(-12 \%\), decreasing in the 1990s.

“the Basque Country per capita GDP takes values up to around \(12 \%\) below those of the synthetic control. The gap in per capita GDP seems to decrease at the end of the period, taking values around \(8 \%\) or \(9 \%\) percent in 1995-1997. Overall, Figure 1 suggests a \(10 \%\) loss in per capita GDP due to terrorism for the 1980’s and 1990’s.”

Our \(ARIMA(0,1,1)\) model matches these findings. See Figure “Treatment effect \(ARIMA(0,1,1)\) for \(log(DGP)\)”.

We find an effect of \(-10 \%\) on average for the period \(1976-1985\).
The maximum of effect is \(-13 \%\), which takes place in \(1981\).

Note 1 BIC and AIC both select \(ARIMA(0,1,1)\). Results of this test in the document. Test is “auto.arima”.

Note 2 Synth control needs \(T \rightarrow \infty\) as well as time series data with small variance.

Plot GDP for the treatment unit, 1955 - 1997.

Pre-treatment data: 1955-1975

Select data pre-1975 to fit an AR-type model. Plotting pre-1975 GDP and first differences.

##  [1] 3.853185 3.945658 4.033562 4.023422 4.013782 4.285918 4.574336
##  [8] 4.898957 5.197015 5.338903 5.465153 5.545916 5.614896 5.852185
## [15] 6.081405 6.170094 6.283633 6.555555 6.810769 7.105184 7.377892
## attr(,"tsp")
## [1]  2 22  1

##  [1] 3.853185 3.945658 4.033562 4.023422 4.013782 4.285918 4.574336
##  [8] 4.898957 5.197015 5.338903 5.465153 5.545916 5.614896 5.852185
## [15] 6.081405 6.170094 6.283633 6.555555 6.810769 7.105184

Detect which ARIMA to use for forecasting

Data used: GDP for 1955-1975.
Fit time series model to pre-treatment GDP to forecast post-1975 GDP

RESULTS: log(GDP) seems be to ARIMA(0,1,1)

## Series: log(ts_55_75) 
## ARIMA(0,1,1) with drift         
## 
## Coefficients:
##          ma1   drift
##       0.7208  0.0323
## s.e.  0.1649  0.0062
## 
## sigma^2 estimated as 0.0002988:  log likelihood=53.47
## AIC=-100.94   AICc=-99.44   BIC=-97.95
## 
## Training set error measures:
##                        ME       RMSE        MAE          MPE      MAPE
## Training set 0.0001193987 0.01600391 0.01168671 -0.001983195 0.7343174
##                   MASE       ACF1
## Training set 0.3544545 0.03806803

Fitting ARIMA for Forecasting

Fit ARIMA(0,1,1) to log(GDP)

Coefficients and in sample error measures:

summary(fit1)

## 
## Call:
## arima(x = log(ts_55_75), order = c(0, 1, 1))
## 
## Coefficients:
##          ma1
##       0.8580
## s.e.  0.1357
## 
## sigma^2 estimated as 0.0005912:  log likelihood = 45.29,  aic = -86.58
## 
## Training set error measures:
##                      ME       RMSE        MAE      MPE     MAPE      MASE
## Training set 0.01683819 0.02373116 0.01765185 1.014544 1.072991 0.5353754
##                    ACF1
## Training set -0.1397066

par(mfrow=c(1,2))
acf(resid(fit1), 
    main="ACF for ARIMA(0,1,1) log(GDP)")
pacf(resid(fit1), 
    main="PACF for ARIMA(0,1,1) log(GDP)")

Fit ARIMA(1,1,0) to log(GDP)

Coefficients and in sample error measures:

summary(fit2)

## 
## Call:
## arima(x = log(ts_55_75), order = c(1, 1, 0))
## 
## Coefficients:
##          ar1
##       0.8394
## s.e.  0.1076
## 
## sigma^2 estimated as 0.0004098:  log likelihood = 49.01,  aic = -94.02
## 
## Training set error measures:
##                       ME      RMSE        MAE       MPE      MAPE
## Training set 0.006022926 0.0197573 0.01237105 0.3744576 0.7693969
##                   MASE       ACF1
## Training set 0.3752103 0.02991256

par(mfrow=c(1,2))
acf(resid(fit2), 
    main="ACF for ARIMA(1,1,0) log(GDP)")
pacf(resid(fit2), 
    main="PACF for ARIMA(1,1,0) log(GDP)")

Fit ARIMA(1,1,1) to log(GDP)

Coefficients and in sample error measures:

summary(fit3)

## 
## Call:
## arima(x = log(ts_55_75), order = c(1, 1, 1))
## 
## Coefficients:
##          ar1     ma1
##       0.7442  0.3124
## s.e.  0.2026  0.4192
## 
## sigma^2 estimated as 0.0003961:  log likelihood = 49.3,  aic = -92.59
## 
## Training set error measures:
##                       ME       RMSE        MAE       MPE      MAPE
## Training set 0.006794403 0.01942433 0.01283329 0.4196213 0.7966188
##                   MASE       ACF1
## Training set 0.3892298 -0.1641544

par(mfrow=c(1,2))
acf(resid(fit3), 
    main="ACF for ARIMA(1,1,1) log(GDP)")
pacf(resid(fit3), 
    main="PACF for ARIMA(1,1,1) log(GDP)")

The economic impact of terrorism in the Basque country using data from Abadie and Gardeazabal (2003)

IB

November 5, 2016

Data Description

Pre-treatment data: 1955-1975

Detect which ARIMA to use for forecasting

Fitting ARIMA for Forecasting

Fit ARIMA(0,1,1) to log(GDP)

Fit ARIMA(1,1,0) to log(GDP)

Fit ARIMA(1,1,1) to log(GDP)

Actual, Fitted, and Forecasted log(GDP)

Computing the Treatment Effect