Presentation is loading. Please wait.

Presentation is loading. Please wait.

Big Data Briefing on BEA Activities

Similar presentations


Presentation on theme: "Big Data Briefing on BEA Activities"— Presentation transcript:

1 Big Data Briefing on BEA Activities
Abe Dunn May 5th 2016 Reporting on BEA some current and future projects BEA has worked on several big data projects, such as our health work that some of you may be familiar with. What I’ll talk about today are more preliminary projects that are ongoing or at their initial stages. The overview will primarily cover work on Palantir, but will also discuss some recent exploratory work with Premise and Zillow.

2 Outline Palantir/FirstData (Abe Dunn, Ledia Guci, Mahsa Gholizadeh, Bryn Whitmire) Exploratory work Zillow Premise Current method for adjusting for across border spending is limited. We must rely on economic reasoning and judgement and limited information for Consumer Expenditure Survey. Basically looking for strong evidence of outliers as a basis for adjustment. Use survey based consumer expenditure share information from CES for adjustment purposes. First Data information uses the location of establishment and the consumer to adjust for flows. In other words, rather than relying on judgement and limited information as a basis for adjustment, we are able to adjust all geographic areas automatically based on the flow information.

3 Help to improve state level estimates of Personal Consumption Expenditures (PCE), and may help generate MSA level estimates But the real advantage would be to expand geographically – and between Econ Censuses Obtaining better changes in PCE (and maybe quarterly changes) or county level

4 Adjusting Establishment Estimates from Census to Construct Regional PCE Statistics
Current process Criteria for adjustment Sufficient evidence of out-of-state spending Economic reason for adjustment A good category match available in consumer expenditure survey data Method Adjust Census-based share with survey-based share Rescale to national accounts totals First Data spending flows allows for a new approach Current method for adjusting for across border spending is limited. We must rely on economic reasoning and judgement and limited information for Consumer Expenditure Survey. Basically looking for strong evidence of outliers as a basis for adjustment. Use survey based consumer expenditure share information from CES for adjustment purposes. First Data information uses the location of establishment and the consumer to adjust for flows. In other words, rather than relying on judgement and limited information as a basis for adjustment, we are able to adjust all geographic areas automatically based on the flow information.

5 New Opportunity: Flows from First Data
Accommodation flows (NAICS 721) To NV From NV

6 Preliminary Estimates
Food Services and Accommodations, 2012 Geography Per Capita Difference from U.S. Value Per Dollar of Disposable Income United States $2,181 0.0% 0.055 Illinois $2,193 0.6% 0.054 -2.2% Hawaii $5,807 166.2% 0.144 159.9% Nevada $3,992 83.0% 0.111 101.5% Initial estimates Incorporating FD flows Geography Per Capita Difference from U.S. Value Per Dollar of Disposable Income United States $2,181 0.0% 0.055 Illinois $2,359 8.2% 0.058 5.2% Hawaii $2,763 26.7% 0.068 23.7% Nevada $1,578 -27.7% 0.044 -20.3% Similar example for food service and accomodation. Use of FD flows automatically corrects outlier problem.

7 Food Services and Accommodations, 2012
Initial estimates Incorporating FD flows Before and after adjustment

8 Consumption Flows and MSA Level PCE Estimates

9 Preliminary Estimates
Food Services and Accommodations, 2012 Geography Per Capita Difference from U.S. Value Per Dollar of Personal Income United States $2,181 0.0% 0.049 Kansas City, MO-KS $2,117 -3.0% 0.047 -4.2% Kahului-Wailuku-Lahaina, HI $9,597 340.0% 0.251 409.3% Las Vegas-Henderson-Paradise, NV $7,707 253.4% 0.199 304.1% Initial estimates Geography Per Capita Difference from U.S. Value Per Dollar of Personal Income United States $2,181 0.0% 0.049 Kansas City, MO-KS $2,231 2.3% 0.050 1.0% Kahului-Wailuku-Lahaina, HI $3,443 57.8% 0.090 82.7% Las Vegas-Henderson-Paradise, NV $2,832 29.8% 0.073 48.5% Incorporating FD flows

10 Food Services and Accommodations, 2012
Initial estimates Per Capita Spending Incorporating FD flows

11 Spending Flows for PCE by State
Opportunities Considerations Flow shares can be readily incorporated and simplify the current methodology Spending and consumption flows across areas provide a unique view of geography of consumption Varying data quality and coverage by industry and by geography Imputation of consumer location Opportunity to improve current methodology for calculating PCE by state, but we may also be able to provide unique consumption flow information. There is still a considerable amount of work to be done. For instance, issues of industry coverage and consumer location must be addressed.

12 Next Steps Refine the home location algorithm and further evaluate flow information Investigate e-commerce data Explore alternative uses of Palantir platform and FirstData Currently our estimates are based on

13 Exploratory Work with Zillow
Zillow data includes 374 million detailed public records over two decades Housing accounts for about 15 percent percent of PCE National and regional expenditures and prices for owner occupied housing are imputed using BLS micro data and Census data Potential work Compare our regional and national estimates with Zillow data Potentially improve BEA’s annual estimates of owner-occupied housing costs at the regional and at the national level 15.3 percent of PCE 74 percent of housing expenditures are owner occupied. This value is imputed. Zillow BEA will assess the feasibility of using housing sale transactions to augment and improve BEA’s annual estimates of owner-occupied housing costs at the regional (Metropolitan Statistical Areas –MSAs, and States) and at the national level. At the regional level, BEA currently imputes owner-occupied rent expenditures using microdata from the Bureau of Labor Statistics’ Consumer Price Index division on renters and owner-occupied costs for 38 areas in the U.S., plus rental microdata from the Census Bureau’s American Community Survey for smaller geographies. At the national level, a rent-to-value ratio is estimated together with information on number of renters and owners to obtain a total value of residential housing expenditures. Access to Zillow’s public records on sales transactions of the US housing market (ZTRAX) would enable us to compare our estimates with Zillow’s, and to determine if there are inconsistencies, and where and why they may be arising. A related project is to collaborate with Zillow on analyzing if there are discrepancies between the rental data from public records and Zillow’s rent estimates and actual respondent data from the surveys of the Census Bureau to which BEA has access. This renters’ analysis would be done at an aggregate level without disclosure or sharing of individual observations. Both projects will provide a robustness check on the value of the housing stock in the United States and its regions, an indicator that is a large component of Gross Domestic Product, Consumer Expenditures and Personal Income statistics.

14 Exploratory Work with Premise
New technology for gathering price data: “Human-directed and machine-refined, Premise indexes and analyzes millions of observations captured daily by our global network of contributors, unearthing connections that impact global decisions.” Potential work with BEA to improve regional price parities (RPP) estimates Gather price data to check for consistency with existing RPPs Use as a basis for potentially strengthening hard-to-price CPI categories (e.g.., Medical, Transportation, Education and Food Away from Home) Enable international comparison of prices The premise technology is basically to equip people with smartphones to gather price information in a directed manner across areas. Individuals are paid per quote typically. Where people go and how they gather prices may be highly customized. They have conducted work internationally, but have not conducted work in the U.S. They are looking to enter the U.S. market and they think building regional RPPs would be a good place to start. They gave us an initial cost of the pilot in the $50k-$100k range. This would obviously not cover their cost. Aten Bettina has indicated interest in working with them. We are all hoping for BLS’s involvement. Premise BEA currently uses microdata from the Consumer Price Index (CPI) and the Consumer Expenditure (CE) survey of the Bureau of Labor Statistics to estimate spatial price variation with the United States. This is done for consumption expenditure categories such as Food, Apparel, Transportation, Housing (excluding Rents), Education, Recreation, Medical and Other goods and services. The Rents and Owner-Equivalent Rents components of Housing are estimated separately, using microdata from the American Consumer Survey (ACS) of the Census Bureau and imputed owner-occupied housing costs (see notes on Zillow below). The results are termed Regional Price Parities (RPPs) and consist of multilateral price indexes for Metropolitan Statistical Areas (MSAs) and States within the U.S., and are analogous to the Purchasing Power Parities estimated by the International Comparison Program (ICP). The ICP is a joint international statistical effort that collects prices across nearly 200 countries and works with Eurostat-OECD, the Asian Development Bank and other regional and national statistical offices on a continuous basis. Premise is currently pricing a subset of the ICP products in several countries outside the U.S. There is overlap between the ICP and CPI products, and as a pilot project, Premise will price these products across a variety of markets and regions in the U.S. This will enable BEA to test for consistency with existing RPPs, as well as to form the basis for possibly strengthening the estimates of hard-to-price categories in the CPI, particularly Services such as Medical, Transport, Education and Food Away from Home services. In turn, BEA will help Premise identify the price-determining characteristics of many of these products, especially in the context of spatial differences, which is a less common exercise than models testing time-to-time differences such as those used in the CPI. Premise’s efforts will also serve the international community by enabling comparisons between the national U.S. average prices for some of the broad categories of consumption with those that are currently entered into the ICP.

15 Food Services and Accommodations, 2012
Initial estimates Per Capita Spending Incorporating FD flows


Download ppt "Big Data Briefing on BEA Activities"

Similar presentations


Ads by Google