Evaluating Deflators and Price Indexes in National Economic Accounts

BEA Price Index Initiative: Plans and Challenges Ana Aizcorbe Fifth World KLEMS Conference 6/4/2018

Introduction The BEA is currently evaluating how well the deflators we use in the national accounts account for quality improvements The results of some of this work is being folded in to the national accounts in our comprehensive update I will briefly discuss other projects that are underway, using our work on Medical Equipment to illustrate some of the challenges 2 3/22/2025

Scope of project Renewed interest in price indexes has generated work in three broad areas: Future measurement: new methods made possible by arrival of Big Data We are exploring these areas with Bajari (Amazon); Klenow and Goolsbee (Adobe) Current estimates: refining BLS methods We are working with the BLS to find data sources that may be used to refine their current production methods (hedonic models) Historical trends: we are using those data sources to make any necessary adjustments to historical estimates Traditionally done by external researchers 3 3/22/2025

Sectors that we have targeted Guided by studies that have made judgmental assessments of the potential biases Consumption: results from Boskin Commission; Lebow, Roberts and Stockton; Lebow and Rudd. Medical care services Residential communication services Investment: Byrne, Fernald and Reinsdorf Medical equipment and supplies Also interested in price measurement problems with digital platforms (Uber, AirBnb, e.g.) We monitor work by external researchers to avoid duplication of effort 4 3/22/2025

Pilot study for medical equipment and supplies Found a data source for medical equipment and supplies (ECRI) Large sector (~$200b) marked by rapid rates of product innovation; our deflators are flat or rising Data are contract-level data for purchases by private health providers over the period 2002-present Wide coverage (diagnostic imaging, surgical robotics, electric beds; programmable pacemakers, stents) We used two segments as a pilot study to Assess the usefulness of these data for price index measurement Develop a set of standard tools that can be applied to similar data for other sectors 5 3/22/2025

Results of pilot study Two segments: Electromedical diagnostic imaging (334511) Irradiation Ionizing Radiation Equipment (334517) Results: Price indexes based on matched model and fixed effect regression methods fall substantially faster than the BLS- based price deflators in the NIPAs Price indexes for Diagnostic Imaging Equipment Electromedical Irradiation Matched Model Fisher Geomean -11.1% -13.1% -10.4% -11.2% Fixed Effect Regressions Weighted Unweighted -10.4% -14.4% -8.9% -14.1% 6 BEA deflator -1.5% 0.4% 3/22/2025

Challenges in standardizing tools Given data on prices, quantities and unique product identifiers, we have code to generate matched-model and fixed-effect regression indexes and to help us assess if results make sense. We are working on ways to automate Obtaining, matching and using attributes for hedonic analysis the building of concordances (to align ECRI categories with NAICS product codes) 7 3/22/2025

Challenges with standardized tools: Attributes Missing data: ECRI data contained attributes for only some models We learned how to scrape these from websites that sell used equipment Many different categories of equipment: Attributes that are relevant for one category are not necessarily relevant for others Many models in each category Electromedical: 500+ models Irradiation: 700+ 8 3/22/2025

ECRI categories SSegment Number of ECRI categories Equipment and supplies 1,694 Electromedical imaging (334511) 73 Irradiation and ionizing radiation (334517) 49 Other 1,572 Irradiation ionizing radiation equipment Radiographic Systems, Digital Radiographic/Fluoroscopic Systems, General-Purpose Radiotherapy Simulation Systems, Computed Tomography-Based Radiotherapy Systems, Cobalt Radiotherapy Systems, Linear Accelerator Radiotherapy Systems, Proton Beam Scanning Systems, Computed Tomography Scanning Systems, Computed Tomography/Gamma Camera, Single Photon Emission Tomography Scanning Systems, Computed Tomography/Positron Emission Tomography Scanning Systems, Computed Tomography/Single Photon Emission Computed Tomography Scanning Systems, Gamma Camera Scanning Systems, Gamma Camera, Single Photon Emission Tomography Scanning Systems, Laser, Optical Coherence Tomography 9 3/22/2025

Each category has many, many attributes (80+ for CTSCANs, for example) GENERATOR Output, kw (X-RAY GENERATORS) Kvp range (GENERATOR) Ma range (IMAGING SYSTEM) SCORING OF CORONARY ARTERY CALCIFICATION (IMAGE STORAGE) DICOM 3.0 COMPATIBLE (POWER NEEDED, VAC) GANTRY Geometry (GANTRY) DETECTOR (SCATTERED LIGHT) Rows (GANTRY) Elements/row (GANTRY) # detection channels (GANTRY) Rotation times, sec 360 (GANTRY) Partial (GANTRY) Slice thickness, mm (GANTRY) X-ray fan beam angle, (GANTRY) Gantry angle deg (GANTRY) Gantry size, hxwxd,cm (GANTRY) Gantry weight, kg (GANTRY) Gantry opening, cm (GANTRY) Scan localizer (GANTRY) OTHER ATTRIBUTES (Interference compensation) FDA CLEARANCE (Interference compensation) CE MARK (MDD) (Interference compensation) MARKETING REGION (Interference compensation) URL MODEL X-RAY TUBE X-RAY TUBE ANODE Heat storage, hu (X-ray tube anode) Heat dissipation rate, hu/min (X-RAY TUBE) Tube cooling (X-ray tube anode) Tube focal spot, mm (X-ray tube anode) POWER NEEDED Type (CONFIGURATION) # of slices (X-ray tube anode) DISPLAY MONITOR Matrixes, pixels (DISPLAY) Range of ct numbers (DISPLAY) Image enlarging scale (DISPLAY) Max # slices displayed simultaneously (DISPLAY) PERFORMANCE Minimum interscan time, sec (PERFORMANCE) Dynamic scan pace (PERFORMANCE) High-contrast spatial resolution 0% mtf, lp/cm (PERFORMANCE) 50% mtf, lp/cm (PERFORMANCE) Low-contrast resolution, mm at % at <=4 rads (PERFORMANCE) Sound, % at <= 2.5 rads (PERFORMANCE) IMAGE PROCESSING (MAXIMUM DISPLAY) Cpu (NETWORK/DATABASE) Scan fovs, cm (IMAGE PROCESSING) Image reconstruction matrixes (IMAGE PROCESSING) Reconstruction time Per slice, sec (Reconstruction time) For localization scan, sec (Reconstruction time) TABLE FEATURES Range of movement Vertical, cm (Range of movement) Longitudinal, cm (Range of movement) Scanning range,cm (Range of movement) Max. patient weight, (precision), kg (Range of movement) HELICAL SCANNING (DISPLAY) Max scan time, sec (DISPLAY) Max scan volume, cm (DISPLAY) Spatial resolution, lp/cm (DISPLAY) Pitch (DISPLAY) Reconstruction time per image, sec (DISPLAY) FLOOR SPACE NEEDED Size, m2 (FLOOR SPACE NEEDED) Image storage Hd capacity, gb (IMAGE STORAGE) No. online images (IMAGE STORAGE) Archive (IMAGE STORAGE) We are considering machine learning techniques (Bajari) to identify which attributes are most relevant for price 10 3/22/2025

Challenges in developing standardized tools: Concordances Electromedical Diagnostic Equipment Medical therapy and patient monitoring equipment Surgical systems and other, exc diagnostic and therapeutic Irradiation ionizing radiation equipment 2016 BEA Domestic Use 4,228,158,091 $ 12,689,706,385 $ $ 1,604,222,853 4,596,210,193 $ ECRI data purchased by BEA percent of domestic use 773,666,001 $ $ 301,064,810 $ 459,574,752 1,308,684,235 $ 18.3% 55.1% 93.5% 28.5% New equipment, after exclusions percent of domestic use 385,869,234 $ $ 98,316,541 $ 192,760,133 $ 868,441,129 9.1% 0.8% 12.0% 18.9% ECRI provided guidance on our pilot segments We are working with Census to obtain a comprehensive (hopefully machine-readable) list of products included relevant NAICS codes Computer code that can be used to match individual products with NAICS codes (e.g., LEHD program) 11 3/22/2025

Next Steps Complete price indexes for medical equipment and devices Begin work on residential communications equipment Explore which sectors to target for future work. 12 3/22/2025

Thank you