i
Mas e Deg ee P og am in
S a is ics and In o ma ion Managemen
Mul i-coun y Analysis o Unemploymen Ra e Nowcas ing
Du ing Co id-19 Wi h Sea ch Que y Da a
A hu Hen ique Fe nandes Campos
Disse a ion
p esen ed as pa ial equi emen o ob aining he Mas e Deg ee P og am in S a is ics and In o ma ion Managemen
NOVA In o ma ion Managemen School
Ins i u o Supe io de Es a ís ica e Ges ão de In o mação
Uni e sidade No a de Lisboa
MEGI
i
NOVA In o ma ion Managemen School
Ins i u o Supe io de Es a ís ica e Ges ão de In o mação
Uni e sidade No a de Lisboa
MULTI-COUNTRY ANALYSIS OF UNEMPLOYMENT RATE
NOWCASTING DURING COVID-19 WITH SEARCH QUERY DATA
By
A hu Hen ique Fe nandes Campos
Mas e Thesis p esen ed as pa ial equi emen o ob aining he Mas e ’s deg ee in S a is ics and
In o ma ion Managemen , wi h a specializa ion in Risk Analysis and Managemen .
Co-Supe iso : B uno Damásio
Co-Supe iso : Flá io Pinhei o
Feb ua y 2023
ii
STATEMENT OF INTEGRITY
I he eby decla e ha ing conduc ed his academic wo k wi h in eg i y. I con i m ha I ha e no used
plagia ism o any o m o undue use o in o ma ion o alsi ica ion o esul s along he p ocess leading
o i s elabo a ion. I u he decla e ha I ha e ully acknowledge he Rules o Conduc and Code o
Hono om he NOVA In o ma ion Managemen School.
São Paulo, 27 de e e ei o de 2023.
iii
ABSTRACT
Nowcas ing me hods aim o p edic he p esen and he e y nea u u e and pas o ci cum en da a
lag. As in e ne usage becomes ubiqui ous, mo e and mo e indi iduals use in e ne sea ch engines as
decision-making ools; consequen ly, sea ch que y da a may be good p oxies o indi idual beha io ,
and hus a use ul nowcas ing p edic o a iable o many mac oeconomic indica o s. This s udy
examines he po en ial o using Google T ends da a o nowcas unemploymen a e du ing he yea s
o he Co id-19 pandemic ac oss six een coun ies by compa ing he pe o mance o ou al e na i e
models wi h Google T ends da a agains a base au o eg essi e model, conside ing wo modelling
aining windows, one limi ed o p e-Co id da a and he o he including 2020 da a. The esul s show
ha sea ch que y da a lack obus ness and ha e a ying p edic i e powe , wi h he inclusion o 2020
da a in o he aining se p o iding a signi ican imp o emen o ou -o -sample o ecas ing accu acy.
These indings indica e ha sea ch que y da a may ha e good p edic i e powe in some scena ios, bu
may no be obus enough o eal-li e applica ions.
KEYWORDS
Economic o ecas ing; Nowcas ing; Unemploymen ; Google T ends.
i
INDEX
1. In oduc ion .................................................................................................................. 1
2. Li e a u e Re iew ......................................................................................................... 2
3. Da a............................................................................................................................... 5
4. Me hodology ................................................................................................................ 8
5. Resul s and Discussion .................................................................................................. 9
5.1. Scena io A: 2015–2019 .......................................................................................... 9
5.2. Scena io B: 2016–2020 ........................................................................................ 13
6. Conclusion .................................................................................................................. 19
Bibliog aphical Re e ences .............................................................................................. 20
Appendix A: Pas and Cu en Google T ends Sampling ................................................. 23

LIST OF FIGURES
Figu e 5.1 – AIC and RMSE alues o 2020 and 2021 o he p e e ed models, Scena io A .. 13
Figu e 5.2 – AIC and RMSE alues o 2021 o he p e e ed models, Scena io B ................... 15
Figu e A.3 – Mean, 10 h and 90 h pe cen ile alues o A s & En e ainmen in he Uni ed
S a es unde pas beha io , n=9 ...................................................................................... 23
Figu e A.4 – Mean, 10 h and 90 h pe cen ile alues o A s & En e ainmen in he Uni ed
S a es unde cu en beha io be o e no icing any change, n=9 .................................... 24
Figu e A.5 – Mean, 10 h and 90 h pe cen ile alues o A s & En e ainmen in he Uni ed
S a es unde cu en beha io a e p uning, n=79 ......................................................... 25
i
LIST OF TABLES
Table 3.1 – Da a sou ces and ype o ime se ies a ailable o unemploymen a e, pe
coun y………………………………………………………………………………………………………………………….6
Table 5.1 – AIC alues pe model pe a iable pe coun y, Scena io A .................................... 9
Table 5.2 – Ou -o -sample o ecas RMSE alues and pe cen age di e ences o 2020, Scena io
A ........................................................................................................................................ 10
Table 5.3 – Ou -o -sample o ecas RMSE alues and pe cen age di e ences o 2021, Scena io
A ........................................................................................................................................ 11
Table 5.4 – AIC alues pe model pe a iable pe coun y, Scena io B .................................. 14
Table 5.5 – Ou -o -sample o ecas RMSE alues and pe cen age di e ences o 2021, Scena io
B ........................................................................................................................................ 15
Table 5.6 – Ou -o -sample o ecas RMSE di e ences and pe cen age di e ences be ween
Scena io B and Scena io A equi alen models o 2021. ................................................. 17
1
1. INTRODUCTION
F om pe sonal budge ing and business planning o go e nmen policymaking, h oughou many ace s
o mode n socie y, indi iduals and ins i u ions alike ely on o icial s a is ics o decision-making.
Howe e , gi en he wo k in ol ed in da a collec ion, ea men , and analysis, all o icial eleases a e
bound o be eleased wi h a delay (i.e., esul s o Feb ua y may only be a ailable in Ma ch). When
accu a e p esen in o ma ion is impe a i e, i p o es necessa y o o ecas he p esen — he “now”.
Nowcas ing aims o o ecas no he medium- o long- e m u u e, bu wha is happening igh now.
O en, s a is icians include mo e ecen da a in o a amewo k; o ins ance, o p edic a mon hly
indica o , i may be use ul o include esul s om a weekly su ey, o e en a co a ia e (i.e., a p edic o )
wi h he same elease equency bu di e en elease window. O e all, he mo o is: imely
in o ma ion is s a egic.
Wi h he widesp ead adop ion o he In e ne and he echnological ad ancemen s ha ollowed he
digi al age, a as amoun o public da a became eadily a ailable online. One such da a sou ce, he
sea ch que y agg ega o Google T ends, allows use s o ack he in e es in speci ic opics and
ca ego ies ac oss geog aphy and ime. Unde he hypo hesis ha sea ch engine que ies a e a
easonable p oxy o ac ual beha io s (e.g., pu chasing habi s), moni o ing changes in hese que ies
migh gi e insigh in o cu en e en s and condi ions ha migh o he wise ake weeks o mon hs o
be e lec ed in o icial s a is ics. Thus, should said hypo hesis p o e co ec , hen he po en ial exis s
in using Google T ends (GT) da a o e ine nowcas ing esul s and make be e decisions as e .
Since he seminal wo ks on he opic by Choi and Va ian (2009, 2012), many esea che s ha e explo ed
he idea u he , o a ying deg ees o success. One cons an , howe e , has been he p e alence o
s udies ocused on he de eloped wo ld, e en hough de eloping coun ies o en su e om longe
elease lags o less eliable co a ia e da a al oge he .
This s udy ocus on explo ing he po en ial o GT da a o nowcas ing unemploymen a e in 16
coun ies including high- and low-income economies. These a e Aus alia, B azil, Canada, Chile,
Ge many, I aly, Japan, Mexico, he Ne he lands, Po ugal, Sou h Ko ea, Swi ze land, Tu key, he Uni ed
Kingdom, he Uni ed S a es, and U uguay. These coun ies ha e easy- o-access mon hly
unemploymen da a eleases om p ima y sou ces and, apa om U uguay, a e among he 50 la ges
economies by nominal GDP.
The disse a ion is o ganized as ollows. Chap e 2 con ains an o e iew o cu en li e a u e on he
opics o nowcas ing and o ecas ing wi h sea ch que y da a. Chap e 3 desc ibes he da a and i s
collec ion. Chap e 4 goes in o he nowcas ing me hodology. Chap e 5 p esen s and discusses he
nowcas ing esul s pe scena io. Finally, Chap e 6 concludes he wo k wi h a b ie e lec ion upon he
esul s and sugges ions o u he explo a ion o he heme.
2
2. LITERATURE REVIEW
While back a i s launch Google T ends was pe cei ed as a ool o webmas e s and ma ke e s alike o
sea ch engine op imiza ion pu poses, s a ing in 2008 Google Inc. and independen au ho s published
pape s on applying sea ch engine agg ega ed da a o scien i ic esea ch in epidemiology (Polg een e
al. 2008) and economic nowcas ing (Choi & Va ian, 2009).
In Sep embe 2008, Google Inc. eleased Google Flu T ends (GFT), a lu nowcas ing se ice ueled by
Google T ends que ies. Un il i s shu down on Augus 09, 2015, GFT was he opic o many discussions
ega ding i s p edic i e powe and u ili y as an ou b eak p edic ion ool. No ably, Olson e al. (2013)
ound ha , e en wi h he 2009 e ised me hodology, GFT ell sho by 52% in i s in luenza-like
in ec ions p edic ion o New Yo k Ci y du ing he 2009 A/H1N1 pandemic. Bu , in a ecen u n o
e en s, Kandula and Shaman (2019) look a new su eillance da a o ee alua e he GFT es ima ion
e o s and c ea e a andom o es eg ession model wi h GFT a es ha see an e o educ ion o 80%
o he 2012/13 season o iginal p edic ion, sugges ing a ee alua ion o sea ch que y usage as p edic o
a iables in in luenza o ecas sys ems.
In economics and business, howe e , ecep ion o GT da a-powe ed o ecas s and nowcas s has been
mo e a o able, seeing use o ins ance in p edic ing ou ism in lows (A ola, Pin o & de Ped aza
Ga cía, 2015) and demand (Sili e s o s & Wochne , 2018), suicide occu ences (K is ou ek, Moa &
P eis, 2016), and ashion consume beha io o a big playe in he indus y (Sil a e al., 2019). In he
opic o inancial ma ke s, GT da a ha e been used o p edic ing down u n s ock ma ke mo es (P eis,
Moa & S anley, 2013), o eign exchange a es (Bulu , 2017), di ec ion o opening s ock p ices (Hu e
al., 2018) and acc uable e u ns on p ecious me als (Salisu, Ogbonna & Adewuyi, 2020).
Some au ho s a e less en husias ic abou he p ospec o Google T ends as a p edic o . Nagao, Takeda
and Tanaka (2019) sugges ha GT da a-d i en models may lack obus ness and a e dependen on
da a equency and seasonali y adjus men s wi h no consis ency ega ding whe he hey would
imp o e o educe accu acy. Schae , Kou en zes and Fildes (2019) ind ha es ablished o ecas ing
benchma ks ou pe o m hose wi h GT da a and social ne wo k in o ma ion in o ecas ing ideo game
sales and co po a e online ideo iews, al hough he au ho s acknowledge hey a e limi ing he
analysis o linea models.
When i comes o a ocus on mac oeconomic a iables, he e seems o be a na u al endency o s udy
dependen a iables closely ela ed o indi idual beha io , namely p i a e consump ion, and
unemploymen ; Vosen and Schmid (2011), Choi and Va ian (2012), Vosen and Schimid (2012),
Ca iè e-Swallow and Labbé (2013), and Woo and Owen (2019) look a he o me , while Choi and
Va ian (2009), Ba ei a, Godinho and Melo (2013), Fondeu and Ka amé (2013), Vicen e, López-
Menéndez and Pé ez (2015) and Nacca a o e al. (2018) a he la e . As GT da a pe ains mos ly o
indi idual sea ch que ies, i is expec ed ha o ecas ing models o hose a iables would bene i om
such da a. Fewe s udies, such as Ma cellino and Schumache (2010), Kuzin, Ma cellino and
Schumache (2011), and Ban is, Clemen s and U quha (2021), look a GDP g ow h a e.
As o loca ion, he economies s udied, o he mos pa , a e pa o he geopoli ical so-called
de eloped wo ld; he excep ions being B azil (Ban is, Clemen s & U quha , 2021), Chile (Ca iè e-
Swallow & Labbé, 2013) and Hong Kong, China (Choi & Va ian, 2012). Ca iè e-Swallow and Labbé
(2013) poin ou ha good nowcas ing me hods a e e en mo e impo an in de eloping coun ies as
9
5. RESULTS AND DISCUSSION
5.1. SCENARIO A: 2015–2019
This scena io sepa a es p e-Co id and Co id-19 da a in o aining and es ing se s espec i ely. Should
he al e na i e models be obus , hen hey should p oduce mo e accu a e 2020 and 2021 o ecas s
han he base model.
Fi e ou o 12 coun ies wi h unadjus ed ime se ies and six ou o 12 coun ies wi h seasonally
adjus ed ime se ies had al e na i e models wi h lowe AIC han he base model, o a o al o se en
coun ies: I aly, Japan, Po ugal, Sou h Ko ea, he Uni ed Kingdom, he Uni ed S a es, and U uguay.
Table 5.1 p esen s he AIC alues o each model.
Table 5.1 – AIC alues pe model pe a iable pe coun y, Scena io A
Coun y
Va iable
Model 1
Model 2
Model 3
Model 4
Model 5
Aus alia
U
-196.56
-190.26
-191.17
-192.77
-193.96
SA
-212.16
-208.35
-209.86
-208.76
-210.35
B azil
U
-223.78
-185.65
-181.83
-220.90
-217.50
Canada
U
-175.88
-172.53
-168.55
-172.79
-168.84
Chile
U
-178.74
-178.07
-175.41
-176.11
-173.48
SA
-178.97
-178.34
-175.85
-176.43
-173.97
Ge many
U
-328.09
-322.74
-318.86
-325.59
-321.66
I aly
SA
-112.84
-99.29
-101.13
-114.48
-116.60
Japan
U
-211.99
-201.84
-199.11
-212.59
-209.27
SA
-212.64
-209.97
-208.47
-216.23
-212.65
Mexico
SA
-168.22
-148.83
-144.94
-164.38
-160.62
Ne he lands
U
-232.18
-228.24
-226.55
-228.70
-226.78
SA
-235.63
-233.99
-232.99
-233.53
-232.91
Po ugal
U
-175.49
-175.65
-172.03
-173.99
-170.35
SA
-173.91
-176.00
-173.43
-174.58
-171.78
Sou h Ko ea
U
-145.08
-138.84
-137.06
-148.73
-145.23
SA
-149.41
-147.39
-145.69
-153.10
-149.67
Swi ze land
SA
-211.66
-190.85
-187.97
-208.51
-206.92
Tu key
SA
-116.72
-97.09
-95.54
-113.22
-110.50
Uni ed Kingdom
U
-279.04
-276.83
-278.94
-275.13
-277.84
SA
-280.08
-277.59
-280.09
-276.12
-279.33
Uni ed S a es
U
-210.75
-205.17
-207.09
-211.44
-211.54
SA
-190.77
-191.32
-195.56
-192.48
-195.01
U uguay
U
-17.71
-13.36
-11.72
-22.68
-20.77
U and SA indica e unadjus ed se ies and seasonally adjus ed se ies espec i ely.
Model numbe ing ollows he numbe ing om he me hodology sec ion.
Bold and unde sco ed alues indica e al e na i e models ha a e a p e e ed model.

10
A leas one al e na i e model bea s he base model in ou -o -sample pe o mance in eigh coun ies
o 2020 da a and in nine coun ies o 2021 da a: Aus alia (2020, 2021), Canada (2020, 2021), Chile
(2020, 2021), Ge many (2020, 2021), Japan (2020), Mexico (2020, 2021), he Ne he lands (2020, 2021),
Tu key (2020, 2021), Sou h Ko ea (2021), and U uguay (2021).
When limi ing he da a o p e e ed models, he e a e wo cases in which he p e e ed model
ou pe o ms he base model in ou -o -sample pe o mance: he seasonally adjus ed se ies o Japan
in 2020 and he unadjus ed se ies o U uguay in 2021, wi h espec i e changes in RMSE o -1,6% and
-3,9%.
On he o he end o he scale, he p e e ed model in he seasonally adjus ed se ies o I aly p oduces
e y poo o ecas s compa ed o he base, wi h inc eases in RMSE o 193,2% in 2020 and 154,4% in
2021. Meanwhile, ano he al e na i e model wi h lowe AIC han he base a es be e han he
p e e ed model, educing he di e ence in RMSE e sus he base model o 11,7% and 2,5% in 2020
and 2021 espec i ely.
Excluding I aly as an ou lie , in 2020 da a he pe cen age di e ence in RMSE goes om -1,6% o 27,5%,
while in 2021 da a i anges om -3,9% o 24,1%. Table 5.2 and Table 5.3 p esen he ou -o -sample
o ecas RMSE alues o 2020 and 2021 espec i ely.
Table 5.2 – Ou -o -sample o ecas RMSE alues and pe cen age di e ences o 2020, Scena io A
Coun y
Va iable
Model 1
Model 2
Model 3
Model 4
Model 5
Aus alia
U
0.46729
0.41738
0.44490
0.46807
0.50334
-10.68%
-4.83%
0.17%
7.72%
SA
0.46722
0.43739
0.46905
0.47175
0.50447
-6.38%
0.39%
0.97%
7.97%
B azil
U
0.30873
0.64435
0.70025
0.37633
0.38997
108.71%
126.82%
21.90%
26.31%
Canada
U
2.11369
1.91580
1.89169
2.12371
2.13443
-9.36%
-10.50%
0.47%
0.98%
Chile
U
0.82059
0.81744
0.77748
0.83507
0.80104
-0.38%
-5.25%
1.76%
-2.38%
SA
0.78695
0.77988
0.74368
0.80371
0.77472
-0.90%
-5.50%
2.13%
-1.55%
Ge many
U
0.31225
0.28212
0.28065
0.31491
0.31277
-9.65%
-10.12%
0.85%
0.17%
I aly
SA
0.94008
1.82280
3.15852
1.05023
2.75601
93.90%
235.98%
11.72%
193.17%
Japan
U
0.14343
0.13911
0.14172
0.14585
0.13323
-3.01%
-1.19%
1.68%
-7.11%
SA
0.16064
0.14468
0.16700
0.15801
0.16106
-9.94%
3.96%
-1.64%
0.26%
Mexico
SA
0.52435
0.64092
0.63790
0.52890
0.52415
11
Coun y
Va iable
Model 1
Model 2
Model 3
Model 4
Model 5
22.23%
21.65%
0.87%
-0.04%
Ne he lands
U
0.31453
0.278252
0.279894
0.31076
0.31044
-11.53%
-11.01%
-1.20%
-1.30%
SA
0.29893
0.27186
0.28046
0.28904
0.29979
-9.06%
-6.18%
-3.31%
0.29%
Po ugal
U
0.56934
0.60007
0.61227
0.59183
0.59781
5.40%
7.54%
3.95%
5.00%
SA
0.55292
0.60245
0.63681
0.59573
0.62312
8.96%
15.17%
7.74%
12.69%
Sou h Ko ea
U
0.34768
0.49142
0.52016
0.43853
0.45705
41.34%
49.61%
26.13%
31.46%
SA
0.28848
0.40153
0.42750
0.36783
0.38137
39.19%
48.19%
27.50%
32.20%
Swi ze land
SA
0.13944
0.21358
0.24823
0.18268
0.23603
53.17%
78.03%
31.01%
69.27%
Tu key
SA
0.76704
0.74723
0.88406
0.86881
0.97673
-2.58%
15.26%
13.27%
27.34%
Uni ed Kingdom
U
0.17210
0.18153
0.17881
0.17353
0.17261
5.48%
3.90%
0.83%
0.30%
SA
0.17141
0.17992
0.17615
0.17162
0.17233
4.96%
2.77%
0.12%
0.54%
Uni ed S a es
U
3.44635
4.12511
3.96150
4.04447
4.07108
19.69%
14.95%
17.36%
18.13%
SA
3.33147
4.19825
4.07052
4.04691
4.05539
26.02%
22.18%
21.48%
21.73%
U uguay
U
0.74317
0.92234
1.01664
0.77248
0.82870
24.11%
36.80%
3.95%
11.51%
U and SA indica e unadjus ed se ies and seasonally adjus ed se ies espec i ely.
Model numbe ing ollows he numbe ing om he me hodology sec ion.
Pe cen age di e ence alues a e in ela ion o he base model.
Bold and unde sco ed alues indica e al e na i e models ha a e a p e e ed model.
Table 5.3 – Ou -o -sample o ecas RMSE alues and pe cen age di e ences o 2021, Scena io A
Coun y
Va iable
Model 1
Model 2
Model 3
Model 4
Model 5
Aus alia
U
0.70517
0.60934
0.66429
0.71141
0.77040
-13.59%
-5.80%
0.88%
9.25%
SA
0.70060
0.62743
0.68573
0.69659
0.75177
-10.44%
-2.12%
-0.57%
7.30%
B azil
U
0.32060
0.84588
0.90028
0.44096
0.45682
163.84%
180.81%
37.54%
42.49%
Canada
U
2.01147
1.79371
1.76694
2.02401
2.03338
-10.83%
-12.16%
0.62%
1.09%
Chile
U
0.99230
0.97980
0.97706
1.00165
1.00449
12
Coun y
Va iable
Model 1
Model 2
Model 3
Model 4
Model 5
-1.26%
-1.54%
0.94%
1.23%
SA
0.96462
0.94404
0.93942
0.97539
0.97671
-2.13%
-2.61%
1.12%
1.25%
Ge many
U
0.35236
0.31285
0.31122
0.35408
0.35163
-11.21%
-11.68%
0.49%
-0.21%
I aly
SA
1.04137
1.70160
2.99813
1.06516
2.64925
63.40%
187.90%
2.28%
154.40%
Japan
U
0.19812
0.21534
0.22989
0.20430
0.21398
8.69%
16.04%
3.12%
8.01%
SA
0.17263
0.18672
0.19648
0.17883
0.18322
8.16%
13.82%
3.59%
6.13%
Mexico
SA
0.56578
0.54537
0.54495
0.56812
0.57234
-3.61%
-3.68%
0.41%
1.16%
Ne he lands
U
0.37383
0.36204
0.32729
0.39018
0.36053
-3.15%
-12.45%
4.37%
-3.56%
SA
0.37494
0.41206
0.39819
0.42284
0.41850
9.90%
6.20%
12.78%
11.62%
Po ugal
U
0.56963
0.61858
0.63844
0.60836
0.61886
8.59%
12.08%
6.80%
8.64%
SA
0.59537
0.66631
0.71772
0.65725
0.69840
11.92%
20.55%
10.39%
17.31%
Sou h Ko ea
U
0.52175
0.47096
0.47508
0.52715
0.53720
-9.74%
-8.95%
1.03%
2.96%
SA
0.44550
0.41665
0.43138
0.44863
0.45105
-6.48%
-3.17%
0.70%
1.24%
Swi ze land
SA
0.15071
0.22922
0.26545
0.18945
0.23728
52.09%
76.13%
25.70%
57.44%
Tu key
SA
1.04308
0.90335
1.17248
1.05918
1.28279
-13.40%
12.41%
1.54%
22.98%
Uni ed Kingdom
U
0.25535
0.27667
0.28343
0.25747
0.26028
8.35%
10.99%
0.83%
1.93%
SA
0.24850
0.27240
0.27934
0.24881
0.25431
9.62%
12.41%
0.13%
2.34%
Uni ed S a es
U
3.41673
4.11576
3.95774
4.03864
4.07116
20.46%
15.83%
18.20%
19.15%
SA
3.26239
4.16751
4.04854
4.01225
4.02966
27.74%
24.10%
22.98%
23.52%
U uguay
U
1.06546
0.94425
0.98781
1.02420
1.11378
-11.38%
-7.29%
-3.87%
4.54%
U and SA indica e unadjus ed se ies and seasonally adjus ed se ies espec i ely.
Model numbe ing ollows he numbe ing om he me hodology sec ion.
Pe cen age di e ence alues a e in ela ion o he base model.
Bold and unde sco ed alues indica e al e na i e models ha a e a p e e ed model.
13
Compa ing he ou -o -sample pe o mance o 2021 agains 2020, Canada, I aly, Mexico, Sou h Ko ea,
he Uni ed S a es, and U uguay see a model ge mo e accu a e. In pa icula , he coun ies ha ha e
hei p e e ed model imp o e in 2021 e sus 2020 a e Canada, I aly, and he Uni ed S a es, he las
being he only o he h ee wi h a p e e ed al e na i e model. Figu e 5.1 plo s he AIC o he p e e ed
models and hei espec i e RMSEs o 2020 and 2021 o ecas s.
Figu e 5.1 – AIC and RMSE alues o 2020 and 2021 o he p e e ed models, Scena io A
Wi h mo e han hal o he coun ies no ha ing sea ch que y da a as pa o hei p e e ed model, i
sugges s ha sea ch que y models do no impa much in o ma ion ha is no al eady con ained in a
simple au o eg essi e componen . To make ma e s wo se, only in wo cases he p e e ed al e na i e
models ha e be e ou -o -sample pe o mance han he base model. E en hough hal he coun ies
ha e a leas one al e na i e model ou pe o m he base in 2020 o 2021, sea ch que y da a may no
be a eliable solu ion o imp o e nowcas ing accu acy. This can be u he seen in he gene al dec ease
in o ecas ing powe in 2021 e sus 2020, which also pu s in o ques ion he obus ness o he da a as
p edic o a iables.
5.2. SCENARIO B: 2016–2020
Gi en ha in his scena io he aining window includes he spikes in unemploymen o en seen a he
beginning o he Co id-19 pandemic, i is easonable o expec a la ge numbe o coun ies wi h
al e na i e models as bes pe o ming han in Scena io A, as a pu ely au o eg essi e model is poo a
o ecas ing ou lie s.
Aus alia
B azil Mexico
Po ugal
Tu key
Uni ed S a es
Japan (U)
Japan (SA)
Uni ed Kingdom
I aly
Canada
Sou h Ko ea
Ne he lands
Swi ze land
Ge many
U uguay
Chile
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
-350 -300 -250 -200 -150 -100 -50 0
RMSE
AIC
Jan/20–Dec/20 Jan/21–Dec/21
14
Indeed, eigh ou o 12 coun ies wi h unadjus ed ime se ies and se en ou o 12 coun ies wi h
seasonally adjus ed ime se ies had al e na i e models ou pe o m he base model wi hou GT da a,
o a o al o 10 coun ies: Canada, Chile, Ge many, I aly, Japan, he Ne he lands, Po ugal, he Uni ed
Kingdom, he Uni ed S a es, and U uguay. In hose coun ies, models wi h AR(1) componen s and GT
da a, wi h o wi hou lagged ca ego ies, had he lowes AIC alues in 11 ou o 15 cases, and lowe AIC
alues han he base model in 14 ou o 15 cases. Table 5.4 p esen s he AIC alues o each model.
Table 5.4 – AIC alues pe model pe a iable pe coun y, Scena io B
Coun y
Va iable
Model 1
Model 2
Model 3
Model 4
Model 5
Aus alia
U
-167.68
-165.46
-164.57
-164.44
-165.00
SA
-166.97
-165.06
-163.65
-163.72
-163.35
B azil
U
-191.61
-149.59
-147.75
-189.93
-187.76
Canada
U
-22.41
-14.69
-78.14
-37.52
-123.71
Chile
U
-131.43
-102.15
-110.42
-127.57
-145.26
SA
-134.47
-107.33
-115.62
-130.63
-148.69
Ge many
U
-234.21
-229.96
-242.08
-230.36
-246.62
I aly
SA
-83.27
-111.38
-109.39
-120.22
-116.28
Japan
U
-215.15
-206.85
-206.62
-216.42
-215.44
SA
-215.13
-213.60
-212.84
-217.77
-215.93
Mexico
SA
-138.48
-117.65
-115.64
-135.83
-134.35
Ne he lands
U
-205.40
-204.18
-200.86
-202.72
-199.53
SA
-205.23
-206.45
-203.08
-204.50
-201.14
Po ugal
U
-133.68
-131.17
-137.72
-130.49
-136.65
SA
-134.93
-132.39
-138.23
-131.34
-136.83
Sou h Ko ea
U
-131.52
-118.98
-115.15
-128.56
-124.62
SA
-140.35
-132.19
-128.60
-138.33
-134.33
Swi ze land
SA
-209.13
-186.06
-182.86
-205.50
-203.30
Tu key
SA
-82.29
-72.42
-71.34
-80.56
-78.38
Uni ed Kingdom
U
-261.38
-257.39
-261.36
-257.65
-261.90
SA
-261.25
-259.61
-263.16
-258.27
-261.97
Uni ed S a es
U
49.93
-25.16
-43.81
-30.45
-65.32
SA
50.19
-24.25
-43.04
-29.06
-64.06
U uguay
U
-15.45
-7.74
-4.88
-19.09
-15.25
U and SA indica e unadjus ed se ies and seasonally adjus ed se ies espec i ely.
Model numbe ing ollows he numbe ing om he me hodology sec ion.
Bold and unde sco ed alues indica e al e na i e models ha a e a p e e ed model.
Ou -o -sample o ecas s o 2021 a e also be e han in Scena io A. The base model has he lowes
RMSE in only h ee coun ies: B azil, Japan, and he Ne he lands, he las being one o he coun ies
wi h an al e na i e model su passing he base in AIC. O he o he hi een coun ies, only he Uni ed
Kingdom has jus one al e na i e model wi h lowe e o han he base; meanwhile, all al e na i e

15
models ou pe o m he base model o Sou h Ko ea, he Uni ed S a es, and U uguay. The bes
imp o emen is seen in he seasonally adjus ed ime se ies o he Uni ed S a es wi h a d op o 62,3%
in RMSE compa ed o he base model when using GT da a wi h lagged ca ego ies and an AR(1)
componen .
When looking only a he p e e ed models pe a iable pe coun y, 10 ou o 24 ou pe o m he base
models while 5 unde pe o m agains hem, wi h he pe cen age di e ence in RMSE anging om -
62,3% o -4,0% and +1,5% o +29,3% espec i ely. Figu e 5.2 plo s he AIC o he p e e ed models and
hei espec i e RMSEs o 2020 and 2021 o ecas s, and Table 5.5 p esen s all ou -o -sample o ecas
esul s o 2021.
Figu e 5.2 – AIC and RMSE alues o 2021 o he p e e ed models, Scena io B
Table 5.5 – Ou -o -sample o ecas RMSE alues and pe cen age di e ences o 2021, Scena io B
Coun y
Va iable
Model 1
Model 2
Model 3
Model 4
Model 5
Aus alia
U
0.57642
0.60310
0.58848
0.56031
0.52376
29.06%
25.93%
19.91%
12.09%
SA
0.59580
0.61586
0.60052
0.58119
0.54731
31.81%
28.53%
24.39%
17.14%
B azil
U
0.31901
0.65181
0.64579
0.32765
0.32206
111.13%
109.18%
6.13%
4.32%
Canada
U
1.53422
1.62754
0.99586
1.21164
0.73224
-23.00%
-52.89%
-42.68%
-65.36%
Chile
U
0.64865
1.02168
1.04790
0.64939
0.57271
Aus alia
B azil
Mexico
Po ugal
Tu key
Uni ed S a es
Japan
Uni ed Kingdom
I aly
Canada
Sou h Ko ea (U)
Sou h Ko ea (SA)
Ne he lands (U)
Ne he lands (SA)
Swi ze land
Ge many
U uguay
Chile
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
-300 -250 -200 -150 -100 -50 0
RMSE
AIC
16
Coun y
Va iable
Model 1
Model 2
Model 3
Model 4
Model 5
24.50%
27.70%
-20.86%
-30.21%
SA
0.62036
0.99149
1.01037
0.61401
0.54275
25.99%
28.39%
-21.98%
-31.03%
Ge many
U
0.29249
0.31276
0.23685
0.29117
0.17960
0.16%
-24.14%
-6.75%
-42.48%
I aly
SA
0.87485
0.38721
0.42414
0.49770
0.49429
-58.81%
-54.88%
-47.06%
-47.42%
Japan
U
0.20656
0.21932
0.24769
0.21183
0.23429
52.91%
72.69%
47.69%
63.35%
SA
0.18359
0.19177
0.21013
0.18630
0.20218
19.38%
30.81%
15.98%
25.86%
Mexico
SA
0.57279
0.55165
0.57220
0.56889
0.58640
5.21%
9.13%
8.49%
11.83%
Ne he lands
U
0.33971
0.39808
0.36674
0.38872
0.35175
26.56%
16.60%
23.59%
11.83%
SA
0.35287
0.45624
0.43432
0.45518
0.43263
52.62%
45.29%
52.27%
44.73%
Po ugal
U
0.55916
0.55789
0.42579
0.53686
0.41498
-2.01%
-25.21%
-5.71%
-27.11%
SA
0.58790
0.58865
0.46397
0.57013
0.45418
6.46%
-16.09%
3.11%
-17.86%
Sou h Ko ea
U
0.52426
0.45848
0.45894
0.52203
0.52000
31.87%
32.00%
50.15%
49.56%
SA
0.45577
0.42055
0.42900
0.45457
0.45475
45.78%
48.71%
57.57%
57.64%
Swi ze land
SA
0.15228
0.23391
0.23125
0.14400
0.13403
67.76%
65.84%
3.27%
-3.88%
Tu key
SA
0.98884
0.94397
0.92488
1.05171
1.01818
23.07%
20.58%
37.11%
32.74%
Uni ed Kingdom
U
0.24321
0.28529
0.30061
0.24122
0.26242
65.77%
74.67%
40.16%
52.48%
SA
0.26007
0.28228
0.29711
0.25876
0.27594
64.68%
73.33%
50.96%
60.98%
Uni ed S a es
U
3.17296
1.64167
1.39472
1.54176
1.22023
-52.37%
-59.53%
-55.26%
-64.59%
SA
3.14488
1.60175
1.37722
1.52074
1.18515
-51.92%
-58.66%
-54.35%
-64.43%
U uguay
U
1.09094
0.96543
0.94779
1.04696
1.06386
29.91%
27.53%
40.88%
43.15%
U and SA indica e unadjus ed se ies and seasonally adjus ed se ies espec i ely.
Model numbe ing ollows he numbe ing om he me hodology sec ion.
Pe cen age di e ence alues a e in ela ion o he base model.
Bold and unde sco ed alues indica e al e na i e models ha a e a p e e ed model.
17
Shi ing he aining window as o include 2020 da a has a e y posi i e impac on he o e all
pe o mance o al e na i e models, be i du ing aining— om se en o 10 coun ies wi h p e e ed
al e na i e models—o ou -o -sample o ecas ing in 2021— om nine o 13 coun ies wi h al e na i e
models ou pe o ming he base.
A signi ican imp o emen is seen when ocusing on he pe o mance o p e e ed models in 2021:
only one model has lowe RMSE han he base in Scena io A, while in Scena io B ha numbe jumps
o en. This imp o emen can also be seen in he pe cen age di e ence in RMSE as he lowes dec ease
in e o in Scena io B (-4,0%) is al eady highe han he single dec ease in Scena io A (-3,9%).
Fu he mo e, when compa ing he ou -o -sample o ecas pe o mance o 2021 be ween models
om Scena io A and Scena io B, a leas one al e na i e model om he la e has lowe RMSE han
i s equi alen in he o me in 19 ou o 24 se ies; in 11 o hose, all Scena io B al e na i e models
ou pe o m hei Scena io A equi alen s. The only coun ies wi h Scena io A models ha bea hei
Scena io B equi alen s a e Japan, he Uni ed Kingdom, and U uguay. The pe cen age di e ence in
RMSE anges om -85,85% o +12,05%. Table 5.6 p esen s he RMSE alue di e ence and pe cen age
di e ence be ween Scena io A and Scena io B equi alen models o 2021.
Table 5.6 – Ou -o -sample o ecas RMSE di e ences and pe cen age di e ences be ween Scena io B
and Scena io A equi alen models o 2021.
Coun y
Va iable
Model 1
Model 2
Model 3
Model 4
Model 5
Aus alia
U
-0.12876
-0.00625
-0.07581
-0.15109
-0.24664
-18.26%
-1.03%
-11.41%
-21.24%
-32.01%
SA
-0.10481
-0.01157
-0.08520
-0.11540
-0.20445
-14.96%
-1.84%
-12.43%
-16.57%
-27.20%
B azil
U
-0.00159
-0.19407
-0.25448
-0.11332
-0.13476
-0.50%
-22.94%
-28.27%
-25.70%
-29.50%
Canada
U
-0.47725
-0.16617
-0.77108
-0.81237
-1.30114
-23.73%
-9.26%
-43.64%
-40.14%
-63.99%
Chile
U
-0.34365
0.04188
0.07084
-0.35226
-0.43177
-34.63%
4.27%
7.25%
-35.17%
-42.98%
SA
-0.34426
0.04745
0.07095
-0.36138
-0.43396
-35.69%
5.03%
7.55%
-37.05%
-44.43%
Ge many
U
-0.05987
-0.00009
-0.07436
-0.06292
-0.17203
-16.99%
-0.03%
-23.89%
-17.77%
-48.92%
I aly
SA
-0.16652
-1.31439
-2.57398
-0.56746
-2.15496
-15.99%
-77.24%
-85.85%
-53.27%
-81.34%
Japan
U
0.00844
0.00398
0.01779
0.00754
0.02031
4.26%
1.85%
7.74%
3.69%
9.49%
SA
0.01097
0.00505
0.01366
0.00748
0.01896
6.35%
2.70%
6.95%
4.18%
10.35%
Mexico
SA
0.00702
0.00629
0.02726
0.00077
0.01406
1.24%
1.15%
5.00%
0.13%
2.46%
Ne he lands
U
-0.03412
0.03605
0.03944
-0.00146
-0.00878
18
Coun y
Va iable
Model 1
Model 2
Model 3
Model 4
Model 5
-9.13%
9.96%
12.05%
-0.37%
-2.43%
SA
-0.02207
0.04418
0.03613
0.03233
0.01414
-5.89%
10.72%
9.07%
7.65%
3.38%
Po ugal
U
-0.01047
-0.06069
-0.21265
-0.07151
-0.20388
-1.84%
-9.81%
-33.31%
-11.75%
-32.94%
SA
-0.00747
-0.07767
-0.25374
-0.08711
-0.24422
-1.25%
-11.66%
-35.35%
-13.25%
-34.97%
Sou h Ko ea
U
0.00250
-0.01248
-0.01614
-0.00511
-0.01721
0.48%
-2.65%
-3.40%
-0.97%
-3.20%
SA
0.01027
0.00390
-0.00238
0.00593
0.00370
2.30%
0.94%
-0.55%
1.32%
0.82%
Swi ze land
SA
0.00157
0.00469
-0.03421
-0.04545
-0.10326
1.04%
2.05%
-12.89%
-23.99%
-43.52%
Tu key
SA
-0.05424
0.04061
-0.2476
-0.00746
-0.26461
-5.20%
4.50%
-21.12%
-0.70%
-20.63%
Uni ed Kingdom
U
-0.01214
0.00862
0.01718
-0.01625
0.00214
-4.75%
3.12%
6.06%
-6.31%
0.82%
SA
0.01157
0.00988
0.01778
0.00995
0.02163
4.66%
3.63%
6.36%
4.00%
8.50%
Uni ed S a es
U
-0.24378
-2.47409
-2.56301
-2.49688
-2.85093
-7.13%
-60.11%
-64.76%
-61.82%
-70.03%
SA
-0.11751
-2.56577
-2.67132
-2.49151
-2.84451
-3.60%
-61.57%
-65.98%
-62.10%
-70.59%
U uguay
U
0.02548
0.02118
-0.04002
0.02276
-0.04992
2.39%
2.24%
-4.05%
2.22%
-4.48%
U and SA indica e unadjus ed se ies and seasonally adjus ed se ies espec i ely.
Model numbe ing ollows he numbe ing om he me hodology sec ion.
All alues a e in ela ion o he Scena io A equi alen model.
While hese esul s a e p omising, hey come a he expense o inco po a ing s ess da a in o he
aining. I is s ill unclea wha d i es he imp o emen s seen; jus as hey may come om he models
being ained on he same s ess scena io hey a e o ecas ing, i may also be he case ha sea ch
que y da a is g owing in o ecas ing powe as mo e indi iduals sea ch o and apply o jobs h ough
he In e ne .
25
Figu e A.5 – Mean, 10 h and 90 h pe cen ile alues o A s & En e ainmen in he Uni ed S a es
unde cu en beha io a e p uning, n=79
40
50
60
70
80
90
100
In e es o e ime
Da e
10 h o 90 h
pe cen ile ange
Mean