structural break test

[Felipe Damasceno]

Hey guys,

I'm using stata 14 with windows 10 OS.

I have monthly interest rate time series and I want to perform structural break tests. Here is a pick of my data:

* Example generated by -dataex-. To install: ssc install dataex
clear
input double cdi float date2
3.47502515853439 420
3.23747318136127 421
4.40822213344707 422
4.21989885894041 423
4.27035403040215 424
 4.0497660786883 425
  4.009614654801 426
3.80662814999979 427
3.25288849103103 428
3.05600706704963 429
2.83796446026445 430
2.72550349506859 431
2.55935041299287 432
2.30581216982331 433
2.19757310217217 434
2.03348327026813 435
2.00466479832968 436
1.93996682979749 437
1.91278301600952 438
1.95025875933725 439
1.88372583558607 440
1.85560423645026 441
1.79387600451737 442
1.78967665426328 443
1.74049527185354 444
1.66034599557952 445
1.62470624337958 446
1.65510520353813 447
1.57871843132162 448
1.59238410179436 449
1.60553388552447 450
 1.5803722520175 451
1.58019624278565 452
1.68303558145024 453
2.98316800452982 454
 2.9138698731435 455
2.66625622357042 456
2.18029774380719 457
2.18029774380719 458
1.69483567170077 459
1.62767375670729 460
1.59693587246705 461
1.69396814819309 462
1.47358235925017 463
2.49255186573782 464
2.92534461047926 465
2.57557115019433 466
2.37801715168904 467
2.17419437215498 468
2.34922802814348 469
 3.2852338217789 470
2.27554363625533 471
1.95953778682965 472
1.63477980545372 473
1.62259300068981 474
1.54639985680503 475
1.46773563562628 476
1.37445271016801 477
1.37248738368485 478
1.58229431971624 479
1.44130380359162 480
1.44052411652278 481
1.43885736764087 482
1.28378980455175 483
1.48812188122252 484
1.38519727696782 485
 1.3014905675353 486
1.39507199891251 487
1.21703320204545 488
1.27949761525665 489
1.21561913102959 490
  1.193770275508 491
1.25859933997241 492
 1.0095694067533 493
 1.2499886797849 494
1.18034804702087 495
1.33438498465388 496
   1.27290929307 497
 1.5008741213786 498
1.60122014919186 499
1.32292744970999 500
1.53402090400612 501
 1.3932996137014 502
1.39360372125745 503
 1.5318887559342 504
1.24746854986277 505
1.36994352303814 506
1.48295046097884 507
1.40365668456837 508
1.30973662318032 509
end
format %tm date2

To run the structural break test I'm using the following code:

regress cdi date2, robust
estat sbknown, break(tm(possible break date))

or

regress cdi date2, robust
estat sbsingle

The problem is that when I'm testing a specific date, I reject the Null hypothesis for almost every date. Let's take 1996m5 (the middle of decrease tendency). The result is:

Wald test for a structural break: Known break date

Number of obs = 96
Sample: 1995m1 - 2002m12
Break date: 1996m5
Ho: No structural break

chi2(2) = 58.2749
Prob > chi2 = 0.0000

Exogenous variables: date2
Coefficients included in test: date2 _cons

or if use the search code I found 2000m9

Test for a structural break: Unknown break date

Number of obs = 96

Full sample: 1995m1 - 2002m12
Trimmed sample: 1996m4 - 2001m10
Estimated break date: 2000m9
Ho: No structural break

Test Statistic p-value
-----------------------------------------------
swald 119.4136 0.0000
-----------------------------------------------
Exogenous variables: date2
Coefficients included in test: date2 _cons

and disconsider the big spike on 1997m11.

Am I thinking this wrong? Does anyone have a better idea to perform this tests?

[Phil Bromiley]

I assume you've set your data as a time series.

I'm not sure what is going on, but one thing to do is to run the regressions for the two parts and look at the parameters.

What happens if you only work with data after November 1997? Maybe that extreme value is creating unusual parameters for the each of the comparison regressions on the first part of the data and so make the tests significant.