1991 calendar(weekly) 1 5 allocate 1995:12:30 open data e:\winrats\bosnia.wk1 data(format=wks,org=obs) [the first step to doing an intervention analysis is to decide the event you will be testing. My time series is Serbia's behavior towards Bosnia from 1991 to 1995 weekly data. I want to see what impact the initial bombings by NATO in the first week of Feb. 1994 had on S's behavior towards B.] [The next decision to make is the functional form of the impact - my theory is that the bombing may have an impact that lasts for quite some time. So, I add a dummy variable to my data set - named "BOMB". It is coded as a "0" before the first week of Feb, and a "1" from the first week of Feb to the end of the series. You could specify a pulse function as an alternative hypothesis, suggesting the event has a sharp but very fast influence on behavior.] [Here is the dummy variable:] statistics bomb Statistics on Series BOMB Weekly Data From 1991:01:05 To 1995:12:30 Observations 261 Sample Mean 0.38314176245 Variance Standard Error 0.48708640741 SE of Sample t-Statistic 12.70790 Signif Level Skewness 0.48353 Signif Level Kurtosis -1.77990 Signif Level
0.237253 Mean 0.030150 (Mean=0) 0.00000000 (Sk=0) 0.00152035 (Ku=0) 0.00000001
[From prior examination of the SB series, I know it is an AR1. Here is the command to estimate the intervention model:] boxjenk(ar=1,constant,iterations=50,input=1) sb / res #bomb 0 1 [The input=1 command tells RATS you have a supplementary variable to add - the "#bomb 0 1" - this tells RATS you want the First order transfer function. I believe a "#bomb 0" would calculate a zero-order transfer function. I add the "iterations" command - RATS defaults at 20 and you will often exceed this doing intervention models. 50 is arbitrary - pick whatever you like!] Dependent Variable SB - Estimation by Box-Jenkins Iterations Taken 12 Weekly Data From 1991:01:12 To 1995:12:30 Usable 260 Observations Degrees of Freedom 256 Centered R**2 0.307990 R Bar **2 0.299880 Uncentered R**2 0.503602 T x R**2 130.937 Mean of Dependent Variable -30.54730769 Std Error of Dependent Variable 48.75588186 Standard Error of Estimate 40.79558051 Sum of Squared Residuals 426055.52367 Durbin-Watson Statistic 1.965955 Q(36-1) 33.395474 Significance Level of Q 0.54565608 Variable Coeff Std Error T-Stat Signif ******************************************************************************* 1. CONSTANT -23.26932924 6.19612763 -3.75546 0.00021422 2. AR{1} 0.51791608 0.05363665 9.65601 0.00000000 3. N_BOMB{0} -35.98429668 15.86756453 -2.26779 0.02417584 4. D_BOMB{1} -0.93838048 0.05136221 -18.26986 0.00000000 [So, as of the first small NATO bombings, Serb behavior towards Bosnians drops
immediately by nearly 36 points. RATS assigns the 0 term - or the pre-intervention series - the N for numerator. Over time, this impact lessens - the denominator term is negative: 1-D_BOMB = 1.93. Calculating the asymptotic change = -18.5, suggesting that a large negative impact will arise from the bombings, then lessen over time.] [Note that the denominator is close to the bounds of inver...