Bai Lumsdaine Stock REStud 1998

These are replication files for Bai, Lumsdaine and Stock(1998) for their European data. BLS tests for a common break for a VAR (or similar linear system with common RHS variables). These examples all allow for a break in the intercept. Their methods also permit a break in the full coefficient vector, though that probably permits too much freedom to be useful.

Note that the results for the break date in the paper are off by one (due to a programming error) relative to the one that these programs produce. Also, the definition of dt(k) after their equation (2.1) isn't what was intended—it should be 0 for t<k and 1 for t>=k. That's the convention in the literature for dating breaks. Their calculations actually used this revised definition and that's what we use.

TABLE4_UNIVARIATE.RPF

Break analysis for univariate AR's (for all four variables, including the US). This includes a search over number of lags and over break point for the minimum BIC model.

TABLE4_BIVARIATE.RPF

Break analysis for the three pairs of European variables, looking for a break in the intercept of each bivariate VAR. Again, this searches both over lag length and break point.

TABLE4_TRIVARIATE.RPF

Break analysis for the trivariate VAR with the three European variables.

Only a few lines need to adjust to match the model. The calculations rely upon SWEEP to do most of the number-crunching. The two SWEEP instructions (one without the break variable(s), one with them) depend upon the model being estimated, as does the INQUIRE at the beginning to get the range. The DISPLAY at the end also needs to be adjusted. All the other calculations (for picking lag length, finding the break and computing the confidence interval) depend only upon the statistics produced by SWEEP.