Predicting Initial Claims for Unemployment Benefits Hyunyoung Choi, Hal Varian July 5, 2009

What would you search for if you thought you might lose your job? Typical searches might be queries like [unemployment office], [apply for unemployment], [jobs], and so on. Google Search Insights and Google Trends classifies search queries like these into two categories, ‘Local/Jobs’ and ‘Society/Social Services/Welfare & Unemployment’. It is tempting to think that Google searches in these topics may be related to filings for unemployment benefits. In our earlier work([Choi and Varian(2009)]), we described how Google Trends data could help predict economic time series of interest. Here we apply the methodology developed in that paper to U.S. unemployment time series. The “Initial Jobless Claims” is a weekly report issued by US Department of Labor.i.

It

tracks the number of people filed for the unemployment benefits and it is considered a leading indicator of labor market. The claims are reported at the national and state level and the data data is available from first week of 1987. Advance initial claims weekly reports at the national level data are released 5 days after the week ends. Table 1 shows a sample of release schedule for June 2009 at national level along Initial Claims and Google Trends data. When the data is released at 7/2/09, Google Trends data is available to the week of release, which means Google Trends is 7 days ahead of government’s release schedule. Week Initial Claims Continued Claims Jobs Welfare & Unemployment

5/31 - 6/6 605K 6.71M 3% 38%

6/7 - 6/13 612K 6.76M 1% 36%

6/14 - 6/20 630K 6.70M 0% 40%

6/21 - 6/27 614K -2% 44%

6/28 - 7/4 Release at 7/9/09 -4% 40%

Table 1: Initial Claims and Google Trends Schedule from 5/31/09 to 7/4/09 i.

http://www.dol.gov/opa/media/press/eta/ui/current.htm

1

Figure 1(a) and 1(b) depict Initial Claims vs. two labor market related categories in Google Trends from 2004. Initial Claims are plotted against Google Trends to compare the trends over time. These three time series show the similar cyclical pattern and up trends since 2008. A committee of the National Bureau of Economic Research has declared that US has been in recession since December 2007.ii. We note that initial claims for unemployment benefits have been increasing since mid 2007(Figure 1). We could utilize the time series since 2004 under the assumption of no structural change for the past 5 years. Or we could use the data since December 2007 under the assumption that there is structural difference since recession started and there is consistent seasonality in labor market. Initial claims have a very good record as a leading indicator. Macroeconomist Robert Gordon points out that there is a “surprisingly tight historical relationship in past US recessions between the cyclical peak in new claims for unemployment insurance (measured as a four-week moving average) and the subsequent NBER trough.”iii. Other economists, such as James Hamilton, have also documented this relationship.iv. Denote yt , Jobst and Welfaret as the time series of Initial Claims, ‘Jobs’ and ‘Welfare & Unemployment’ categories in Google Trends respectively. We applied standard ARIMA model selection procedures and selected AR(1) model as a baseline. Then we added the Google Trends series to see how much this improved prediction. This procedure follows the logic described in “Predicting the Present with Google Trends.”v. The baseline and alternative model with Google Trends data for the Long term (all data) and Short Term (recession only) model are as follows: Baseline Model : log(yt ) = Intercept + φ log(yt−1 ) + et Alternative Model : log(yt ) = Intercept + αJobst + βWelfaret + φ log(yt−1 ) + et Table 2 summarizes the model fit. There is positive correlation between initial claims and the search related to ‘Jobs’ and ‘Welfare & Unemployment’. Even though the time series plot suggest a change in Initial Claims after 2008, the model predicts better with the longer time series. With the Google Trends series, the model fit is improved significantly and the out-ofsample mean-absolute-error estimated with the rolling window for the past 24 weeks is decreased by 15.74% for the long term model and 12.90% for the short term model. The national level initial claims are released up to the week of 6/21/09-6/27/09 and the figure is 613K, unexpectedly smaller than the consensus, 619K. Decrease in Google Trends ii. iii. iv. v.

http://www.nber.org/cycles/recessions.html http://www.voxeu.org/index.php?q=node/3524 http://www.econbrowser.com/archives/2009/04/another_green_s.html http://google.com/googleblogs/pdfs/google\_predicting\_the\_present.pdf

2

0 2004

2005

2006

2007

2008

2009

2008

2009

−40

−20

5e+05 3e+05

Initial Claims Google Trends

(a) Jobs

3e+05

0

5e+05

50

Initial Claims Google Trends

2004

2005

2006

2007

(b) Welfare & Unemployment

Figure 1: Initial Claims vs. Labor Market related Google Trends

implies decrease in Initial claims. The Google Trends index for the weeks ending 7/4/09 were down and this could indicate a continuous decrease in Initial Claims for the coming weeks.

Summary We have shown how Google Trends data can help predict initial claims in initial claims for unemployment benefits in the United States. [Askitas and Zimmerman(2009)] and [Suhoy(2009)] have examined similar unemployment data for Germand and Israel, respectively, and also found significant improvements in forecasting accuracy by using Google Trends. We hope that other researchers will continue to examine these data.

3

LT ST

Est SE Est SE

Intercept 0.1269 0.1618 0.2174 0.2632

Baseline Model φ σ 0.9902 0.0443 0.0126 0.9839 0.0432 0.0202

MAE 3.24% 3.10%

Intercept 1.6498 0.3754 1.792 0.5541

φ 0.8727 0.029 0.8632 0.0427

Alternative Model Jobs Welfare 0.0014 0.0003 0.0014 0.0010 0.0006 0.0004

σ 0.0429

MAE 2.73%

0.0398

2.70%

Table 2: Summary of Long Term Models and Short Term Models

References [1] Nikos Askitas and Kalus F. Zimmerman. Google econometrics and unemployment forecasting. Applied Economics Quarterly, 55:107.120, 2009 URL: http://ftp.iza.org/dp4201.pdf. [2] Hyunyoung Choi and Hal Varian. Predicting the present with google trends. Technical report, Google, 2009 URL:

http://google.com/googleblogs/pdfs/google\_predicting\_the\_present.

pdf. [3] Tanya Suhoy. Query indices and a 2008 downturn: Israeli data. Technical report, Bank of Israel, 2009.4. URL: http://www.bankisrael.gov.il/press/eng/090709/090709e.htm

4

500

0

550

600

5

650

Initial Claims

450

−5

Prediction Error in Percent

Reference Model (MAE = 3.24%) w/ Google Trends (MAE = 2.73%)

08/03

08/31

09/28

10/26

11/23

12/21

01/18

02/15

03/15

04/12

05/10

06/07

07/05

(a) Long Term Model

Initial Claims

600 500

550

0

450

−5

Prediction Error in Percent

5

650

Reference Model (MAE = 3.1%) w/ Google Trends (MAE = 2.7%)

08/03

08/31

09/28

10/26

11/23

12/21

01/18

02/15

03/15

04/12

05/10

06/07

(b) Short Term Model

Figure 2: Prediction MAE from LT and ST Model

5

07/05

Predicting Initial Claims for Unemployment ... - Research at Google

Jul 5, 2009 - ... consensus, 619K. Decrease in Google Trends ii. http://www.nber.org/cycles/recessions.html iii. http://www.voxeu.org/index.php?q=node/3524.

73KB Sizes 1 Downloads 334 Views

Recommend Documents

An Argument for Increasing TCP's Initial ... - Research at Google
3rd Quarter 2009. http://www.akamai.com/stateoftheinternet, 2009. [5] M. Allman, S. Floyd, and C. Partridge. Increasing TCP's. Initial Window. RFC 3390, 2002.

Predicting Electricity Distribution Feeder ... - Research at Google
Using Machine Learning Susceptibility Analysis. Philip Gross, Albert Boulanger, .... feeder section) and dynamic data (e.g. electrical load data for a feeder and its ...

RFC 6928 - Increasing TCP's Initial Window - Research at Google
improves the overall performance of many web services without resulting in a congestion ..... 2 in [Duk10]), effectively limiting how much window a remote host may use. In order to ... initial window will help to stop this trend and provide the best.

Predicting Accurate and Actionable Static ... - Research at Google
May 10, 2008 - many potential signaling factors in large software development set- tings can be expensive, we use a screening methodology to quickly.

Trust, but Verify: Predicting Contribution Quality ... - Research at Google
Freebase2 is one of the most popular knowledge bases (as evident by its use by major commercial search engines such. 2http://www.freebase.com/. Figure 1: ...

Predicting Bounce Rates in Sponsored Search ... - Research at Google
Among the best known metrics for these pur- poses is click ... ments in response to search queries on an internet search engine. ... ing aggregate data from the webserver hosting the adver- tiser's site. ..... The top ten scoring terms per source ...

PLUMS: Predicting Links Using Multiple Sources - Research at Google
Jan 27, 2018 - Link prediction is an important problem in online social and col- laboration ... puter science, such as artificial intelligence, machine learning ... We show this in our experiments — several of these feature-based classifiers per- f

Instant Foodie: Predicting Expert Ratings From ... - Research at Google
to infer expert scores using “grassroots” data. ... We performed our analysis using data ..... The statistical model for inferring ratings for both GP and Zagat.

Ad Quality On TV: Predicting Television ... - Research at Google
Jun 28, 2009 - ABSTRACT. This paper explores the impact of television advertisements on audience retention using data collected from television set-top ...

Thwarting Fake OSN Accounts by Predicting ... - Research at Google
accounts in online social networks are victim-agnostic. Even though victims of .... Our main insight herein is that infiltration campaigns run by fake accounts are ...

Semantic Vector Products: Some Initial ... - Research at Google
Mar 28, 2008 - “Red apples” can be interpreted as a Boolean conjunction, but the meaning of “red wine” (more like the colour purple) and. “red skin” (more like ...

Exaggerated Claims for Interactive Stories
interactive stories, in the context of the theory of advertising. Results from a hu- man user study show that female players find linear and branching stories to be.

Mathematics at - Research at Google
Index. 1. How Google started. 2. PageRank. 3. Gallery of Mathematics. 4. Questions ... http://www.google.es/intl/es/about/corporate/company/history.html. ○.

Simultaneous Approximations for Adversarial ... - Research at Google
When nodes arrive in an adversarial order, the best competitive ratio ... Email:[email protected]. .... model for combining stochastic and online solutions for.

Asynchronous Stochastic Optimization for ... - Research at Google
Deep Neural Networks: Towards Big Data. Erik McDermott, Georg Heigold, Pedro Moreno, Andrew Senior & Michiel Bacchiani. Google Inc. Mountain View ...

SPECTRAL DISTORTION MODEL FOR ... - Research at Google
[27] T. Sainath, O. Vinyals, A. Senior, and H. Sak, “Convolutional,. Long Short-Term Memory, Fully Connected Deep Neural Net- works,” in IEEE Int. Conf. Acoust., Speech, Signal Processing,. Apr. 2015, pp. 4580–4584. [28] E. Breitenberger, “An

Asynchronous Stochastic Optimization for ... - Research at Google
for sequence training, although in a rather limited and controlled way [12]. Overall ... 2014 IEEE International Conference on Acoustic, Speech and Signal Processing (ICASSP) ..... Advances in Speech Recognition: Mobile Environments, Call.

UNSUPERVISED CONTEXT LEARNING FOR ... - Research at Google
grams. If an n-gram doesn't appear very often in the training ... for training effective biasing models using far less data than ..... We also described how to auto-.

Combinational Collaborative Filtering for ... - Research at Google
Aug 27, 2008 - Before modeling CCF, we first model community-user co- occurrences (C-U) ...... [1] Alexa internet. http://www.alexa.com/. [2] D. M. Blei and M. I. ...

Quantum Annealing for Clustering - Research at Google
been proposed as a novel alternative to SA (Kadowaki ... lowest energy in m states as the final solution. .... for σ = argminσ loss(X, σ), the energy function is de-.

Interface for Exploring Videos - Research at Google
Dec 4, 2017 - information can be included. The distances between clusters correspond to the audience overlap between the video sources. For example, cluster 104a is separated by a distance 108a from cluster 104c. The distance represents the extent to

Voice Search for Development - Research at Google
26-30 September 2010, Makuhari, Chiba, Japan. INTERSPEECH ... phone calls are famously inexpensive, but this is not true in most developing countries.).