-
Notifications
You must be signed in to change notification settings - Fork 0
/
2023_12.qmd
822 lines (683 loc) · 33 KB
/
2023_12.qmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
---
title: "December 2023"
author:
- name: "Tomasz Woźniak"
url: https://github.com/donotdespair
affiliations:
- University of Melbourne
orcid: 0000-0003-2212-2378
execute:
freeze: auto
---
```{r date setup}
#| echo: false
# setup these
forecast_day = "2023-12-31" # this forecast was performed on
forecast_month = substr(forecast_day, 1, 7) # a string for the month the forecast was performed in
mfo = zoo::as.yearmon(forecast_month) # monthly forecast origin
wfo = as.Date(forecast_day) + 7 # weekly forecast origin (daily date)
```
```{r interest data}
#| echo: false
#| message: false
#| warning: false
###############################################################
# download daily interest rates
# from the RBA using package readrba
###############################################################
icr_dwnld = readrba::read_rba(series_id = "FIRMMCRTD") # Cash Rate Target
icr_tmp = xts::xts(icr_dwnld$value, icr_dwnld$date)
dates_tmp = xts::xts(as.Date(icr_dwnld$date), icr_dwnld$date)
by1m_dwnld = readrba::read_rba(series_id = "FIRMMBAB30D")
by1m_tmp = xts::xts(by1m_dwnld$value, by1m_dwnld$date)
by3m_dwnld = readrba::read_rba(series_id = "FIRMMBAB90D")
by3m_tmp = xts::xts(by3m_dwnld$value, by3m_dwnld$date)
by6m_dwnld = readrba::read_rba(series_id = "FIRMMBAB180D")
by6m_tmp = xts::xts(by6m_dwnld$value, by6m_dwnld$date)
by2y_dwnld = readrba::read_rba(series_id = "FCMYGBAG2D")
by2y_tmp = xts::xts(by2y_dwnld$value, by2y_dwnld$date)
by3y_dwnld = readrba::read_rba(series_id = "FCMYGBAG3D")
by3y_tmp = xts::xts(by3y_dwnld$value, by3y_dwnld$date)
by5y_dwnld = readrba::read_rba(series_id = "FCMYGBAG5D")
by5y_tmp = xts::xts(by5y_dwnld$value, by5y_dwnld$date)
by10y_dwnld = readrba::read_rba(series_id = "FCMYGBAG10D")
by10y_tmp = xts::xts(by10y_dwnld$value, by10y_dwnld$date)
###############################################################
# The following steps have to be performed because
# some of the downloaded rates have a gap from
# 16 May 2013 to 1 Jan 2022 as reported and discussed at:
# https://github.com/MattCowgill/readrba/issues/43
###############################################################
long_ou_tmp = na.omit(merge(by2y_tmp, by3y_tmp, by5y_tmp, by10y_tmp))
long_be = long_ou_tmp["/2013-05-16"]
long_af = long_ou_tmp["2022-01-01/"]
long_in_tmp = readxl::read_xls(path = "f02d.xls", skip = 10)
long_in = xts::xts(long_in_tmp[,2:5], as.Date(long_in_tmp$`Series ID`))
long_in = long_in["2013-05-17/2021-12-31"]
colnames(long_in) <- colnames(long_af)
short = na.omit(merge(icr_tmp, by1m_tmp, by3m_tmp, by6m_tmp))
long = rbind(long_be, long_in, long_af)
###############################################################
# construct a matrix (xts) of daily interest rates data
###############################################################
variables_all = na.omit(merge(short, long))
colnames(variables_all) = c("cash rate", "1m", "3m", "6m", "2y", "3y", "5y", "10y")
variables_all = variables_all[paste0("/",forecast_day)]
variables_weekly = xts::to.weekly(variables_all, OHLC = FALSE)
variables_monthly = xts::to.monthly(variables_all, OHLC = FALSE)
# variables_all[,c(1,5:8)]
# variables_all[,1:4]
# variables_weekly
# variables_monthly
###############################################################
# create a dummy for the interest raise regime
# from May 22 to June 23
# the dummies used for forecasting DO NOT include the hikes term
###############################################################
T = nrow(variables_monthly)
dm = xts::xts(as.matrix(rep(0, T)), zoo::index(variables_monthly))
dm["2022-05/2023-06"] = 1
colnames(dm) = "dum"
dmf = as.matrix(rep(0,12)) # turn off the 2022/23 hikes dummy
colnames(dmf) = "dum"
Tw = nrow(variables_weekly)
dw = xts::xts(as.matrix(rep(0, Tw)), zoo::index(variables_weekly))
dw["2022-05/2023-06"] = 1
colnames(dw) = "dum"
dwf = as.matrix(rep(0,55)) # turn off the 2022/23 hikes dummy
colnames(dwf) = "dum"
```
```{r cointegrating rank}
#| echo: false
#| eval: false
#| message: false
#| warning: false
#| results: hide
###############################################################
# perform cointegrating rank test to specify VECM models
###############################################################
library(vars)
# Johansen's cointegrating rank test
vecm_vma = ca.jo(variables_monthly, type = "trace", ecdet = "const", K = 5, spec = "transitory")
summary(vecm_vma) # r = 5, N = 8
vecm_vms = ca.jo(variables_monthly[,1:4], type = "trace", ecdet = "const", K = 5, spec = "transitory")
summary(vecm_vms) # r = 3, N = 4
vecm_vml = ca.jo(variables_monthly[,c(1,5:8)], type = "trace", ecdet = "const", K = 5, spec = "transitory")
summary(vecm_vml) # r = 4, N = 5
vecm_vwa = ca.jo(variables_weekly, type = "trace", ecdet = "const", K = 5, spec = "transitory")
summary(vecm_vwa) # r = 7, N = 8
vecm_vws = ca.jo(variables_weekly[,1:4], type = "trace", ecdet = "const", K = 5, spec = "transitory")
summary(vecm_vws) # r = 3, N = 4
vecm_vwl = ca.jo(variables_weekly[,c(1,5:8)], type = "trace", ecdet = "const", K = 5, spec = "transitory")
summary(vecm_vwl) # r = 4, N = 5
```
```{r forecasting}
#| echo: false
#| message: false
#| warning: false
###############################################################
# forecasting using VECM models
###############################################################
# idea:
# create many alternative VECM specifications with
# - various lag orders (pp below)
# - various data (all variables, only short- and long-run rates collected in vm)
# - various data sampling frequency (weekly and monthly)
# - model with the raise dummy dm and without
# the forecasts are weighted based on their capacity to model cash rate
###############################################################
library(vars)
###############################################################
# forecast with monthly data
###############################################################
vm = list(
variables_monthly,
variables_monthly[,1:4],
variables_monthly[,c(1,5:8)]
) # three types of monthly data
pp = c(3, 5, 7, 9, 11, 13, 15, 17) # selected lag orders
rr = c(5, 3, 4) # various cointegrating ranks for data systems
mm = 2 * length(pp) * length(vm) # count the number of models in this section
forecasts = array(NA, c(12, 3, mm)) # to be filled in with forecasts
loglik = rep(NA, mm) # to be filled in with value of cash rate-specific likelihood
f = 1 # it's used to fill in the outcomes
###############################################################
# the loop below estimates the models and predicts 12 months ahead
###############################################################
for (v in 1:3) {
for (p in pp) {
# estimation of models with raise dummy
###############################################################
vecm = ca.jo(vm[[v]], type = "trace", ecdet = "const", K = p, spec = "transitory", dumvar = dm)
var_cr = vec2var(vecm, r = rr[v])
var_pr = predict(var_cr, n.ahead = 12, ci = .68, dumvar = dmf)
forecasts[,,f] = var_pr$fcst$cash.rate[,1:3]
standardised_resid = var_cr$resid[,1]/sqrt(sum(var_cr$resid[,1]^2)/var_cr$obs)
loglik[f] = sum(dnorm(standardised_resid, log = TRUE))
f = f + 1
# estimation of models without the raise dummy
###############################################################
vecm = ca.jo(vm[[v]], type = "trace", ecdet = "const", K = p, spec = "transitory")
var_cr = vec2var(vecm, r = rr[v])
var_pr = predict(var_cr, n.ahead = 12, ci = .68)
forecasts[,,f] = var_pr$fcst$cash.rate[,1:3]
standardised_resid = var_cr$resid[,1]/sqrt(sum(var_cr$resid[,1]^2)/var_cr$obs)
loglik[f] = sum(dnorm(standardised_resid, log = TRUE))
f = f + 1
}
}
# compute the weights for forecasts from each of the models
# weights are proportional to marginal likelihood for the cash rate
###############################################################
ll = exp(loglik - max(loglik))
weights = ll/sum(ll)
# reweight the forecasts
###############################################################
forecasts_w = forecasts
for (i in 1:mm) {
forecasts_w[,,i] = weights[i] * forecasts[,,i]
}
# set up forecast date sequence
# this is used to assign the forecasts monthly periods
###############################################################
ym13 = mfo
ym1 = ym13 + 1/12 # the first forecasted period
ym2 = ym1 + 11/12 # the last forecasted period
s = seq(ym1, ym2, 1/12) # create yearmon sequence
###############################################################
# compute the pooled forecasts
# for VECM forecasting using monthly data
###############################################################
pooled_forecasts_m = apply(forecasts_w, 1:2, sum)
colnames(pooled_forecasts_m) = c("forecast", "lower", "upper")
pooled_forecasts_m = xts::xts(pooled_forecasts_m, s)
###############################################################
# forecast with weekly data
###############################################################
vw = list(
variables_weekly,
variables_weekly[,1:4],
variables_weekly[,c(1,5:8)]
) # three types of weekly data
rrw = c(7, 3, 4) # various cointegrating ranks for data systems
mmw = 2 * length(pp) * length(vw) # count the number of models in this section
forecastsw = array(NA, c(55, 3, mmw)) # to be filled in with forecasts
loglikw = rep(NA, mmw) # to be filled in with value of cash rate-specific likelihood
f = 1 # it's used to fill in the outcomes
for (v in 1:3) {
for (p in pp) {
# estimation of models with raise dummy
###############################################################
vecm = ca.jo(vw[[v]], type = "trace", ecdet = "const", K = p, spec = "transitory", dumvar = dw)
var_cr = vec2var(vecm, r = rrw[v])
var_pr = predict(var_cr, n.ahead = 55, ci = .68, dumvar = dwf)
forecastsw[,,f] = var_pr$fcst$cash.rate[,1:3]
standardised_resid = var_cr$resid[,1]/sqrt(sum(var_cr$resid[,1]^2)/var_cr$obs)
loglikw[f] = sum(dnorm(standardised_resid, log = TRUE))
f = f + 1
# estimation of models without raise dummy
###############################################################
vecm = ca.jo(vw[[v]], type = "trace", ecdet = "const", K = p, spec = "transitory")
var_cr = vec2var(vecm, r = rrw[v])
var_pr = predict(var_cr, n.ahead = 55, ci = .68)
forecastsw[,,f] = var_pr$fcst$cash.rate[,1:3]
standardised_resid = var_cr$resid[,1]/sqrt(sum(var_cr$resid[,1]^2)/var_cr$obs)
loglikw[f] = sum(dnorm(standardised_resid, log = TRUE))
f = f + 1
}
}
# compute the weights for forecasts from each of the models
# weights are proportional to marginal likelihood for the cash rate
###############################################################
llw = exp(loglikw - max(loglikw))
weightsw = llw/sum(llw)
# reweight the forecasts
###############################################################
forecastsw_w = forecastsw
for (i in 1:mmw) {
forecastsw_w[,,i] = weightsw[i] * forecastsw[,,i]
}
# set up forecast date sequence
# this is used to assign the forecasts weekly periods
###############################################################
sw = as.Date(rep(NA, 55))
sw[1] = wfo
for (i in 2:55) {
sw[i] = sw[i - 1] + 7
}
###############################################################
# compute the pooled forecasts
# for VECM forecasting using weekly data
# and transform them to monthly forecasts (I report monthly forecasts)
###############################################################
pooled_forecasts_ww = apply(forecastsw_w, 1:2, sum)
colnames(pooled_forecasts_ww) = c("forecast", "lower", "upper")
pooled_forecasts_ww = xts::xts(pooled_forecasts_ww, sw)
pooled_forecasts_wm = xts::to.monthly(pooled_forecasts_ww, OHLC = FALSE)
```
```{r data exchange rates and foreign ir}
#| echo: false
#| eval: false
#| message: false
#| warning: false
icr_dwnld = readrba::read_rba(series_id = "FIRMMCRTD") # Cash Rate Target
icr_tmp = xts::xts(icr_dwnld$value, icr_dwnld$date)
crt = xts::to.monthly(icr_tmp, OHLC = FALSE)
dates_tmp = xts::xts(as.Date(icr_dwnld$date), icr_dwnld$date)
# Trade-weighted Index May 1970 = 100
twi_dwnld = readrba::read_rba(series_id = "FXRTWI")
twi_tmp = xts::xts(twi_dwnld$value, twi_dwnld$date)
twi = xts::to.monthly(twi_tmp, OHLC = FALSE)
# US dollar trade-weighted index
twius_dwnld = readrba::read_rba(series_id = "FUSXRTWI")
twius_tmp = xts::xts(twius_dwnld$value, twius_dwnld$date, tclass = 'yearmon')
twius = xts::to.monthly(twius_tmp, OHLC = FALSE)
# Canada Target Rate
rcan_dwnld = readrba::read_rba(series_id = "FOOIRCTR")
rcan_tmp = xts::xts(rcan_dwnld$value, rcan_dwnld$date, tclass = 'yearmon')
rcan = xts::to.monthly(rcan_tmp, OHLC = FALSE)
# Euro Area Refinancing Rate
reur_dwnld = readrba::read_rba(series_id = "FOOIREARR")
reur_tmp = xts::xts(reur_dwnld$value, reur_dwnld$date, tclass = 'yearmon')
reur = xts::to.monthly(reur_tmp, OHLC = FALSE)
# Japan Policy Rate
rjap_dwnld = readrba::read_rba(series_id = "FOOIRJTCR")
rjap_tmp = xts::xts(rjap_dwnld$value, rjap_dwnld$date, tclass = 'yearmon')
rjap = xts::to.monthly(rjap_tmp, OHLC = FALSE)
# United Kingdom Bank Rate
rgbt_dwnld = readrba::read_rba(series_id = "FOOIRUKOBR")
rgbt_tmp = xts::xts(rgbt_dwnld$value, rgbt_dwnld$date, tclass = 'yearmon')
rgbt = xts::to.monthly(rgbt_tmp, OHLC = FALSE)
# United States Federal Funds Minimum Target Rate
rusa_dwnld = readrba::read_rba(series_id = "FOOIRUSFFTRMX")
rusa_tmp = xts::xts(rusa_dwnld$value, rusa_dwnld$date, tclass = 'yearmon')
rusa = xts::to.monthly(rusa_tmp, OHLC = FALSE)
foreign = na.omit(merge(
crt, twi, twius, rcan, reur, rjap, rgbt, rusa
))
library(vars)
vecm_f = ca.jo(foreign, type = "trace", ecdet = "const", K = 5, spec = "transitory")
# summary(vecm_f) # r = 1, N = 8
vecm_fr = ca.jo(foreign[,c(1,4:8)], type = "trace", ecdet = "const", K = 5, spec = "transitory")
# summary(vecm_fr) # r = 1, N = 8
vecm_feus = ca.jo(foreign[,c(1:3,8)], type = "trace", ecdet = "const", K = 5, spec = "transitory")
# summary(vecm_feus) # r = 1, N = 8
###############################################################
# forecast with monthly data
###############################################################
vm = list(
foreign,
foreign[,c(1,4:8)],
foreign[,c(1:3,8)]
) # three types of monthly data
pp = c(3, 5, 7, 9, 11, 13, 15, 17) # selected lag orders
rr = c(1, 1, 1) # various cointegrating ranks for data systems
mm = length(pp) * length(vm) # count the number of models in this section
forecasts = array(NA, c(12, 3, mm)) # to be filled in with forecasts
loglik = rep(NA, mm) # to be filled in with value of cash rate-specific likelihood
f = 1 # it's used to fill in the outcomes
###############################################################
# the loop below estimates the models and predicts 12 months ahead
###############################################################
for (v in 1:3) {
for (p in pp) {
# estimation of models without the raise dummy
###############################################################
vecm = ca.jo(vm[[v]], type = "trace", ecdet = "const", K = p, spec = "transitory")
var_cr = vec2var(vecm, r = rr[v])
var_pr = predict(var_cr, n.ahead = 12, ci = .68)
forecasts[,,f] = var_pr$fcst$crt[,1:3]
standardised_resid = var_cr$resid[,1]/sqrt(sum(var_cr$resid[,1]^2)/var_cr$obs)
loglik[f] = sum(dnorm(standardised_resid, log = TRUE))
f = f + 1
}
}
# compute the weights for forecasts from each of the models
# weights are proportional to marginal likelihood for the cash rate
###############################################################
ll = exp(loglik - max(loglik))
weights = ll/sum(ll)
# reweight the forecasts
###############################################################
forecasts_f = forecasts
for (i in 1:mm) {
forecasts_f[,,i] = weights[i] * forecasts[,,i]
}
# set up forecast date sequence
# this is used to assign the forecasts monthly periods
###############################################################
ym13 = mfo
ym1 = ym13 + 1/12 # the first forecasted period
ym2 = ym1 + 11/12 # the last forecasted period
s = seq(ym1, ym2, 1/12) # create yearmon sequence
###############################################################
# compute the pooled forecasts
# for VECM forecasting using monthly data
###############################################################
pooled_forecasts_f = apply(forecasts_f, 1:2, sum)
colnames(pooled_forecasts_f) = c("forecast", "lower", "upper")
pooled_forecasts_f = xts::xts(pooled_forecasts_f, s)
```
```{r univariate monthly forecasting}
#| echo: false
#| message: false
#| warning: false
###############################################################
# forecast cash rate using univariate models
###############################################################
# idea:
# forecast using various models that differ by:
# - sampling data frequency (weekly and monthly cash rate)
# - ARMA model lag order (p_max and q_max) for conditional mean
# - various specifications of the conditional volatility equations:
# - GARCH model
# - GARCH-IN-MEAN model (time-varying risk premium)
# - GJR-GARCH model (leverage effect)
# - GJR-GARCH-IN-MEAN model
# The forecasts are pooled using weights
# based on their capacity to model cash rate
# Note: rugarch package requires bootstraping the forecasts to
# compute the forecasting standard errors
###############################################################
# monthly cash rate
###############################################################
crm = variables_monthly[,1]
# ARMA models maximum lag orders
###############################################################
p_max = 3
q_max = 3
# matrices to be filled in by forecasts
# and cash rate-specific likelihood
###############################################################
garchfm = array(NA, c(12, 3, 4, (p_max ) * (q_max + 1)))
loglik = matrix(NA, 4, (p_max ) * (q_max + 1))
loglik_sr = matrix(NA, 4, (p_max ) * (q_max + 1))
for (i in 1:p_max) {
for (j in 0:q_max) {
# arma-garch forecasting
###############################################################
garch_ms = rugarch::ugarchspec(
variance.model = list(model = "sGARCH"),
mean.model = list(armaOrder = c(i, j)),
distribution.model = "norm"
)
garch_mf = rugarch::ugarchfit(garch_ms, crm, solver = "hybrid")
garch_mfor = rugarch::ugarchboot(garch_mf, n.ahead = 12,
method = "Partial", n.bootpred = 5000)
garchfm[,,1, (q_max + 1)*(i - 1) + j + 1] = cbind(garch_mfor@forc@forecast$seriesFor,
t(apply(garch_mfor@fseries, 2, HDInterval::hdi , credMass = .68)))
loglik[1, (q_max + 1)*(i - 1) + j + 1] = -sum(garch_mf@fit$log.likelihoods)
# arma-garch-in-mean forecasting
###############################################################
garch_ms = rugarch::ugarchspec(
variance.model = list(model = "sGARCH"),
mean.model = list(armaOrder = c(i, j), archm = TRUE),
distribution.model = "norm"
)
garch_mf = rugarch::ugarchfit(garch_ms, crm, solver = "hybrid")
garch_mfor = rugarch::ugarchboot(garch_mf, n.ahead = 12,
method = "Partial", n.bootpred = 5000)
garchfm[,,2, (q_max + 1)*(i - 1) + j + 1] = cbind(garch_mfor@forc@forecast$seriesFor,
t(apply(garch_mfor@fseries, 2, HDInterval::hdi , credMass = .68)))
loglik[2, (q_max + 1)*(i - 1) + j + 1] = -sum(garch_mf@fit$log.likelihoods)
# arma-gjr-garch forecasting
###############################################################
garch_ms = rugarch::ugarchspec(
variance.model = list(model = "gjrGARCH"),
mean.model = list(armaOrder = c(i, j)),
distribution.model = "norm"
)
garch_mf = rugarch::ugarchfit(garch_ms, crm, solver = "hybrid")
garch_mfor = rugarch::ugarchboot(garch_mf, n.ahead = 12,
method = "Partial", n.bootpred = 5000)
garchfm[,,3, (q_max + 1)*(i - 1) + j + 1] = cbind(garch_mfor@forc@forecast$seriesFor,
t(apply(garch_mfor@fseries, 2, HDInterval::hdi , credMass = .68)))
loglik[3, (q_max + 1)*(i - 1) + j + 1] = -sum(garch_mf@fit$log.likelihoods)
# arma-gjr-garch-in-mean forecasting
###############################################################
garch_ms = rugarch::ugarchspec(
variance.model = list(model = "gjrGARCH"),
mean.model = list(armaOrder = c(i, j), archm = TRUE),
distribution.model = "norm"
)
garch_mf = rugarch::ugarchfit(garch_ms, crm, solver = "hybrid")
garch_mfor = rugarch::ugarchboot(garch_mf, n.ahead = 12,
method = "Partial", n.bootpred = 5000)
garchfm[,,4, (q_max + 1)*(i - 1) + j + 1] = cbind(garch_mfor@forc@forecast$seriesFor,
t(apply(garch_mfor@fseries, 2, HDInterval::hdi , credMass = .68)))
loglik[4, (q_max + 1)*(i - 1) + j + 1] = -sum(garch_mf@fit$log.likelihoods)
}
}
# models that failed to estimate obtain -Inf log-likelihood
###############################################################
loglik[is.na(loglik)] = -1000000
# compute the weights
###############################################################
llm = exp(loglik - max(loglik))
weightsm = llm/sum(llm)
# reweight the forecasts
###############################################################
forecastsm = garchfm
forecastsm[is.na(forecastsm)] = 0
for (i in 1:dim(garchfm)[3]) {
for (j in 1:dim(garchfm)[4]) {
forecastsm[,,i,j] = weightsm[i,j] * forecastsm[,,i,j]
}
}
###############################################################
# pool the weighted monthly forecasts using univariate models
###############################################################
pooled_forecasts_garchm = apply(forecastsm, 1:2, sum)
colnames(pooled_forecasts_garchm) = c("forecast", "lower", "upper")
pooled_forecasts_garchm = xts::xts(pooled_forecasts_garchm, s)
```
```{r univariate weekly forecasting}
#| echo: false
#| message: false
#| warning: false
# weekly cash rate series
###############################################################
crw = variables_weekly[,1]
# ARMA maximum lag orders for weekly arma-garch forecasting
###############################################################
p_max = 4
q_max = 1
# to be filled in with forecasts and
# cash rate-specific likelihoods
###############################################################
garchfw = array(NA, c(55, 3, 4, (p_max ) * (q_max + 1)))
loglik = matrix(NA, 4, (p_max ) * (q_max + 1))
for (i in 1:p_max) {
for (j in 0:q_max) {
# arma-garch forecasting
###############################################################
garch_ms = rugarch::ugarchspec(
variance.model = list(model = "sGARCH"),
mean.model = list(armaOrder = c(i, j)),
distribution.model = "norm"
)
garch_mf = rugarch::ugarchfit(garch_ms, crw, solver = "hybrid")
garch_mfor = rugarch::ugarchboot(garch_mf, n.ahead = 55,
method = "Partial", n.bootpred = 5000)
garchfw[,,1, (q_max + 1)*(i - 1) + j + 1] = cbind(garch_mfor@forc@forecast$seriesFor,
t(apply(garch_mfor@fseries, 2, HDInterval::hdi , credMass = .68)))
loglik[1, (q_max + 1)*(i - 1) + j + 1] = -sum(garch_mf@fit$log.likelihoods)
# arma-garch-in-mean forecasting
###############################################################
garch_ms = rugarch::ugarchspec(
variance.model = list(model = "sGARCH"),
mean.model = list(armaOrder = c(i, j), archm = TRUE),
distribution.model = "norm"
)
garch_mf = rugarch::ugarchfit(garch_ms, crw, solver = "hybrid")
garch_mfor = rugarch::ugarchboot(garch_mf, n.ahead = 55,
method = "Partial", n.bootpred = 5000)
garchfw[,,2, (q_max + 1)*(i - 1) + j + 1] = cbind(garch_mfor@forc@forecast$seriesFor,
t(apply(garch_mfor@fseries, 2, HDInterval::hdi , credMass = .68)))
loglik[2, (q_max + 1)*(i - 1) + j + 1] = -sum(garch_mf@fit$log.likelihoods)
# arma-gjr-garch forecasting
###############################################################
garch_ms = rugarch::ugarchspec(
variance.model = list(model = "gjrGARCH"),
mean.model = list(armaOrder = c(i, j)),
distribution.model = "norm"
)
garch_mf = rugarch::ugarchfit(garch_ms, crw, solver = "hybrid")
garch_mfor = rugarch::ugarchboot(garch_mf, n.ahead = 55,
method = "Partial", n.bootpred = 5000)
garchfw[,,3, (q_max + 1)*(i - 1) + j + 1] = cbind(garch_mfor@forc@forecast$seriesFor,
t(apply(garch_mfor@fseries, 2, HDInterval::hdi , credMass = .68)))
loglik[3, (q_max + 1)*(i - 1) + j + 1] = -sum(garch_mf@fit$log.likelihoods)
# arma-gjr-garch-in-mean forecasting
###############################################################
garch_ms = rugarch::ugarchspec(
variance.model = list(model = "gjrGARCH"),
mean.model = list(armaOrder = c(i, j), archm = TRUE),
distribution.model = "norm"
)
garch_mf = rugarch::ugarchfit(garch_ms, crw, solver = "hybrid")
garch_mfor = rugarch::ugarchboot(garch_mf, n.ahead = 55,
method = "Partial", n.bootpred = 5000)
garchfw[,,4, (q_max + 1)*(i - 1) + j + 1] = cbind(garch_mfor@forc@forecast$seriesFor,
t(apply(garch_mfor@fseries, 2, HDInterval::hdi , credMass = .68)))
loglik[4, (q_max + 1)*(i - 1) + j + 1] = -sum(garch_mf@fit$log.likelihoods)
}
}
# models that fail to estimate correctly obtain -Inf log-likelihood
###############################################################
loglik[is.na(loglik)] = -1000000
# compute the weights
###############################################################
llm = exp(loglik - max(loglik))
weightsw = llm/sum(llm)
# reweight the forecasts
###############################################################
forecastsw = garchfw
forecastsw[is.na(forecastsw)] = 0
for (i in 1:dim(garchfw)[3]) {
for (j in 1:dim(garchfw)[4]) {
forecastsw[,,i,j] = weightsw[i,j] * forecastsw[,,i,j]
}
}
###############################################################
# pool the weighted weekly forecasts from univariate models
###############################################################
pooled_forecasts_garchw = apply(forecastsw, 1:2, sum)
colnames(pooled_forecasts_garchw) = c("forecast", "lower", "upper")
pooled_forecasts_garchw = xts::xts(pooled_forecasts_garchw, sw)
pooled_forecasts_garchwm = xts::to.monthly(pooled_forecasts_garchw, OHLC = FALSE)
```
```{r pool forecasts}
#| echo: false
#| message: false
#| warning: false
###############################################################
# pool all the forecasts
# from VECM and ARMA-GARCH models for weekly and monthly data
###############################################################
# pooled_forecasts = (1/5) * (pooled_forecasts_m + pooled_forecasts_wm + pooled_forecasts_garchm + pooled_forecasts_garchwm + pooled_forecasts_f)
pooled_forecasts = (1/4) * (pooled_forecasts_m + pooled_forecasts_wm + pooled_forecasts_garchm + pooled_forecasts_garchwm)
# save to file
###############################################################
zoo::write.zoo(pooled_forecasts, sep = ",", file = paste0("forecasts/",forecast_month,".csv"))
```
```{bash}
#| echo: false
#| eval: false
# copy the forecast to archive
###############################################################
cp forecasts/2023-12.csv forecasts-backup/2023-12.csv
```
> Have a look at my end-of-year forecasts prepared to answer the RBA cash rate survey by [finder.com.au](https://www.finder.com.au/rba-cash-rate).
## Cash rate forecasts
The figure below presents the monthly cash rate series starting from January 2010, with the forecasts reported from January 2024 to December 2024 as the forecast mean and the 68% forecasting intervals.
```{r forecast plot}
#| echo: false
# this is used for a forecast figure
###############################################################
ym12 = zoo::as.yearmon("2010-1") # first data point for the plot - do not change
s3 = seq(ym13, ym2, 1/12) # create yearmon sequence
s2 = seq(ym12, ym2, 1/12) # create yearmon sequence
datainforecast = as.vector(vm[[2]][(dim(vm[[2]])[1] - (length(s2) - 12 - 1)):dim(vm[[2]])[1], 1])
last_point = datainforecast[length(datainforecast)]
cols = c("darkorchid4","mediumorchid1","mediumorchid2","mediumorchid3","hotpink1","hotpink2","hotpink3","hotpink4")
ci1_tmp = col2rgb(cols[2])
ci2_tmp = col2rgb(cols[4])
ci1 = rgb(ci1_tmp[1], ci1_tmp[2], ci1_tmp[3], 100, maxColorValue = 255)
ci2 = rgb(ci2_tmp[1], ci2_tmp[2], ci2_tmp[3], 100, maxColorValue = 255)
plot(x = s2, y = c(datainforecast, pooled_forecasts[,1]), main = "Cash rate forecast",
type = "l", ylab = "[%]", xlab = "time",
ylim = range(pooled_forecasts, datainforecast), bty = "n",
lwd = 1, col = cols[1]
)
polygon(x = c(s3, s3[13:1]),
y = c(last_point,as.vector(pooled_forecasts[,2]), as.vector(pooled_forecasts[,3])[12:1], last_point),
col = ci1, border = ci1)
lines(x = s2, y = c(datainforecast, pooled_forecasts[,1]), lwd = 2, col = cols[1])
abline(v = ym13, col = cols[6], lty = 3)
```
The table below makes the numerical values presented in the figure more accessible.
```{r forecast table}
#| echo: false
options(knitr.kable.NA = '')
# pooled_mq = merge(pooled_forecasts,pooled_forecasts_q)
# colnames(pooled_mq) = c("monthly", "lower", "upper", "quarterly", "lower", "upper")
# knitr::kable(as.matrix(pooled_mq), caption = "Monthly and quarterly cash rate forecasts", digits = 2)
pooled_mq = merge(pooled_forecasts)
colnames(pooled_mq) = c("monthly", "lower", "upper")
knitr::kable(as.matrix(pooled_mq), caption = "Monthly and quarterly cash rate forecasts", digits = 2)
```
## Survey answers
Based on the forecasts above, and the analysis of forecasts from individual models, I formed the following survey answers:
**When you think the RBA will change the cash rate?**
| | Jan 2023 | Feb 2024 | Mar 2024 | Apr 2024 | May 2024 | Jun 2024 | Jul 2024 | Aug 2024 | Sep 2024 or later |
|--------|--------|--------|--------|--------|--------|--------|--------|--------|--------|
| Increase | | | | | | | | | |
| Hold | ✓ | ✓ | | | | | | | |
| Decrease | | | | | | | | | |
**Why do you think this?**
> Happy New Year! My forecasts hardly contain the current cash rate value within the forecast interval, with most of the forecast density mass indicating downward movements. My reading of these results is that the RBA will HOLD the cash rate at the current value to push the inflation down more. I expect the first interest rate cuts by mid-2024. My forecasts are available at https://forecasting-cash-rate.github.io/
**Do you think the cash rate has peaked?**
> YES
**When do you think the RBA will first cut the cash rate?**
> June 2024
## Forecasting system
My forecasting system for November is based on the cash rate target and government bond yields at various maturities as presented in the figure below.
```{r data plot yields}
#| echo: false
plot(
x = index(variables_weekly),
y = as.vector(variables_weekly[,2]),
main = "Australian interest rates at various maturities",
type = "l",
ylab = "yield [%]",
xlab = "time",
ylim = range(variables_weekly),
bty = "n",
lwd = 1,
col = "mediumorchid1"
)
for (i in 3:8) {
lines(
x = index(variables_weekly),
y = as.vector(variables_weekly[,i]),
col = cols[i]
)
}
lines(
x = index(variables_weekly),
y = as.vector(variables_weekly[,1]),
col = "darkorchid4",
lwd = 2
)
legend(
"topright",
legend = colnames(variables_weekly),
col = cols,
lwd = c(2, rep(1, 7)),
bty = "n"
)
```
The system consists of over two hundred of models. Half of them are are models of weekly and the other half of monthly data.
Vector Error Correction models for weekly and monthly series with different model specification parameters. Univariate models for the cash rate capture complex patterns of data persistence using autoregressive moving average equation extended by time-varying volatility equation - a GARCH model. Part of the models include the leverage effect and/or time-varying risk premium.
The forecasts are pooled in two stages. Firstly, the models are weighted in four sub-groups VECM vs. ARMA and weekly vs. monthly data models proportionally to their cash rate forecasting ability. Such four pooled forecasts are equally weighted in the second stage to provide monthly forecasts reported above.