Real-time economic activity indicators

--- breaks: false --- # Real-time economic activity indicators ## Notable indicators ### AIS (aka Ship GPS) Ship data (estimated that virtually *all* vessels engaged in int'l trade has this). Different papers use different algorithms to estimate shipment. Looks pretty promising if we really want export data with very small lags. * [Furukawa and Hisano (2022)](https://www.boj.or.jp/en/research/wps_rev/wps_2022/data/wp22e19.pdf) for Japan * Data from [VesselFinder](https://www.vesselfinder.com/) * Car export fits very well * Container ships not so much, possibly due to different value per weight; but can estimate per-port value * [Verschuur et al. (2021)](https://typeset.io/papers/global-economic-impacts-of-covid-19-lockdown-measures-stand-4b5wgres6y) * Uses UN AIS data to estimate the pandemic's effect on world trade ### Trucks GPS * Many detailed papers on how to estimate origin and destination, [how to identify activity/industry](https://typeset.io/papers/truck-industry-classification-from-anonymous-mobile-sensor-x8z14s0h8f), due to its application in route planning, etc. #### Related * [Paez (2017)](https://www.tandfonline.com/doi/abs/10.1080/10106049.2015.1120358) in Southern Spain * Usage of construction truck tracking service correlates with construction indicators ### Telco data Japan has hourly data that says how many people are in each 100x100m area. Once this is combined with POI data (what businesses are in that "mesh"), one can track economic activities. * [Matsumura et al. (2021)](https://www.bis.org/ifc/publ/ifcb59_36_rh.pdf): number of people in a factory/shopping centers, etc. using cell phone data * [Furukawa et al. (2022)](https://www.boj.or.jp/en/research/wps_rev/wps_2022/data/wp22e16.pdf) * Short-term fluctuations should follow labor and capital utilization rate * Proxy labor with GPS data, and CapU with electricity usage ### Payment data * [Okubo et al. (2022)](https://www.boj.or.jp/en/research/wps_rev/wps_2022/data/wp22e08.pdf) * Nowcast consumption * Compile 3 data sources * credit card transaction data (JCB Consumption NOW) * point-of-sale (POS) data (METI POS and GfK) * spending records obtained from a personal financial management service (Money Forward) * Note that each data has limitations, so combining them make sense * [Galbraith and Tkacz (2015)](https://www.econstor.eu/bitstream/10419/154645/1/ecbsp10.pdf) for Canada * Use credit/debit/cheque transactions to nowcast GDP * [Chapman and Desai (2021)](https://www.mdpi.com/2571-9394/5/4/36) for Canada * Predict GDP/wholesale/retail sales * Gathers all payment systems in Canada * Noted that using only some might lead to errors due to change in market share * Use with gradient boosting * [Duarte et al. (2017)](https://www.sciencedirect.com/science/article/abs/pii/S0169207016300899) in Portugal * Use ATM/POS data * Quite country specific in that the ATM/POS network is very extensive * Other works * Dunn et al. (2020, 2021) * Bounie et al. (2020) * Carvalho et al. (2021) ### Google Trends data * [Choi and Varian (2009)](https://static.googleusercontent.com/media/www.google.com/en//googleblogs/pdfs/google_predicting_the_present.pdf) * [Nymand-Andersen and Pantelidis (2018)](https://www.econstor.eu/bitstream/10419/194013/1/ecb.sps30.en.pdf) * Car sales in the Euro area * [Nakazawa (2022)](https://www.boj.or.jp/en/research/wps_rev/wps_2022/data/wp22e09.pdf) for Japan * POS data + internet search data -> predict GDP ### Other data * [Lewis et al. (2020)](https://www.nber.org/system/files/working_papers/w26954/w26954.pdf) * Weekly Economic Index (WEI) based on various data available at weekly frequency, tracks GDP * Data: same-store retail sales, an index of consumer sentiment, initial and continued claims for unemployment insurance, an index of temporary and contract employment, tax collections from paycheck withholdings, a measure of steel production, a measure of fuel sales, a measure of railroad traffic, and a measure of electricity consumption ## Notes * Not sure how these measures are actually used in policy making by other central banks * Many of these applications noted that, due to incomplete data coverage (especially payment data), including just one indicator might not make sense. Will need to include other things to ensure coverage if we're trying to forecast/nowcast macroeconomic variables. * Similarly, using "payment data" by themselves might not add much. Many of these add these alternative data into existing, larger, models and show improvement. The data by itself might not track the variable of interest well, but does it "fill the gap" that the current model cannot deal with? * Practical notes: Comparing performance across models is easy enough; but since most of our models include some kind of judgment/tuning, do we have that data of what the "final" model tells us?