Secular Stagnation or Secular Deflation? 8 October 2019 Rebecca Riley (Chair, ESCoE and NIESR) Ana Aizcorbe (U.S. Bureau of Economic Analysis) Leonard Nakamura (Federal Reserve Bank of Philadelphia)
Constant-Quality Price Indexes for Smartphones: What do they measure? Ana Aizcorbe ESCoE/NIESR London October 8, 2019
Renewed interest in measurement issues Interest was prompted by questions like: – What sorts of measurement problems has the arrival of the digital economy raised? • Disruptive nature of some industries • How to value “free stuff” • Welfare gains associated with arrival of new goods – Is the slowdown in measured productivity an artifact of measurement problems? • Possibility that growth in prices is overstated with a corresponding understatement of real GDP growth. – And that these gaps have grown since the early 2000’s – This hinges on how deflators do the price vs quantity split 2 10/2/2019
BEA is contributing to this agenda by…. • Partnering with academics to address some of the particularly thorny measurement problems: – “free stuff” -- Nakamura, Samuels, Soloveichic – Cloud price indexes -- Sichel – Smartphone price indexes – - Aizcorbe, Byrne, Sichel • Exploiting alternative data sources to improve price deflators – Insurance claims data for medical care price indexes -- Dunn et al – Administrative data for medical equipment indexes -- Aizcorbe – Ride-level data for UBER/Lyft and traditional rides to correct potential outlet substitution bias problems in transportation services -- Aizcorbe and Chen • Exploring extensions of the national accounts that could address welfare and well-being – BEA has organized a “Beyond GDP” panel discussion at the 2020 American Economic Association meetings to obtain guidance on setting priorities from experts • https://www.aeaweb.org/conference/2020/preliminary/1262?q=eNqrVipOLS7OzM8LqSx IVbKqhnGVrAxrawGlCArI 3 10/2/2019
Today’s focus: the price vs quantity split • Goal is to provide a clear explanation of: – The role that national accountants see for price indexes -- constant-quality indexes • Not a discussion of how the official price indexes are actually constructed, more a focus on the ideas – Questions these measures can address: • Productivity growth • Inflation – Questions they cannot address: • Welfare gains from the introduction of new goods • Illustrate these points with a study on smartphone prices (Aizcorbe, Byrne and Sichel, 2019) 4 10/2/2019
Fundamental decomposition Crude decomposition: CHANGE IN SPENDING = CHANGE IN + CHANGE IN (or REVENUES, or COST) AVERAGE PRICE QUANTITY Accounting for Quality: CHANGE IN = CHANGE IN + CHANGE IN AVERAGE PRICE “CONSTANT - “QUALITY” QUALITY” PRICES Better decomposition: CHANGE IN SPENDING = CHANGE IN + CHANGE IN “CONSTANT - QUANTITY QUALITY” PRICES + CHANGE IN “QUALITY” 5 10/2/2019
Fundamental decomposition Crude decomposition: CHANGE IN SPENDING = CHANGE IN + CHANGE IN (or REVENUES, or COST) AVERAGE PRICE QUANTITY Accounting for Quality: CHANGE IN = CHANGE IN + CHANGE IN AVERAGE PRICE “CONSTANT - “QUALITY” QUALITY” PRICES Real output Better decomposition: CHANGE IN SPENDING = CHANGE IN + CHANGE IN “CONSTANT - QUANTITY QUALITY” PRICES + CHANGE IN “QUALITY” Inflation 6 10/2/2019
Splitting out changes in quality from pure price change • Matched model methods measure changes in constant- quality prices directly and relegate any remaining changes in average prices to quality change. • Hedonic techniques do the opposite: directly measure changes in quality and relegate remaining changes in average price to C-Q price change. 7 10/2/2019
Matched model indexes require granular product-level data Model-level prices for IT equipment typically have downward-sloping contours Chip-Level Data for DRAM Chips 1000.0 100.0 4K 16K 64K Dollars 256K 10.0 1M 4M 16M 64M 1.0 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 0.1 Assume: --model is defined such that the attributes of the model are constant over time. --prices from entry to exit are observed in the data (e.g., as with scanner data) 8 10/2/2019
How do matched-model methods split out price vs quality change? Chart 1. Simple Example of Quality Measurement 25 P 2,1 20 P 1,0 Dollars 15 10 P 1,1 P 2,2 5 0 0 1 2 Time Chip 1 Chip 2 P 2,2 / P 1,0 = ( P 2,2 / P 2,1 ) ( P 2,1 / P 1,1 ) ( P 1,1 / P 1,0 ) change in value of average change in price quality 9 10/2/2019
Once you strip out the value of quality improvements, the price index falls rapidly • Once you strip out the value Chip-Level Data for DRAM Chips of quality change, the 1000.0 resulting price index drops rapidly. 100.0 4K • The increase in DRAM 16K 64K Dollars revenues over this period is 256K 10.0 1M more than explained by 4M increases in quality-adjusted 16M 64M 1.0 quantities. 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 • NOTE: This technique cannot 0.1 be applied in all cases (e.g., Housing, custom software) Price Deflator for DRAM • Technique uses incremental 100 improvements to existing 10 goods and doesn’t handle really new goods (e.g., light 1 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 bulb) 0.1 10 0.01 10/2/2019 0.001
Smartphones: data from IDC • IDC “Worldwide Quarterly Mobile Phone Tracker”, available at https://www.idc.com/getfile.dyn?containerId=IDC_P8397&atta chmentId=47322790. • Quarterly-frequency data for 2010-2017 • IDC estimates revenue, units, and prices by model for the U.S. market using public and proprietary information from phone manufacturers, component suppliers and distribution channel companies (e.g. retailers and wholesalers) • For each model, IDC also provides an extensive list of attributes that may be used to estimate hedonic price indexes • We constructed both matched-model and hedonic indexes for these phones 11
Smartphones: matched model methods will likely understate quality changes • For Apple iPhones, prices are flat over the product cycle and new, better phone models are often the same price as older models • This suggests gaps in prices do not reflect differences in “quality” 12 10/2/2019
Tornquist index for smartphones 20.0 • Tornquist index falls 15.0 about 10% over the 10.0 entire period, a little avereage percent changes less than one would 5.0 expect given the 0.0 rapid pace of innovation -5.0 -10.0 • We attribute this to the inability of -15.0 matched-model -20.0 method to properly account for quality -25.0 2011-2013 2014-2017 improvements Change in average prices Matched-model tornquist index Implied quality change 13 10/2/2019
Rapid product improvements in smartphones Note: • Characteristics used in the hedonic regression are attributes of the 14 equipment (not the apps, e.g.) 10/2/2019
One hedonic regression: Time-dummy hedonic price index ln 𝑄 𝑗 , 𝑢 = 𝛽 + Σ 𝑙 𝛾 𝑙 𝑌 k , 𝑗 , 𝑢 + Σ 𝑢 𝜀 𝑢 𝐸 𝑗 , 𝑢 + 𝜁 𝑗 , 𝑢 Where: • 𝑄 𝑗 , 𝑢 is the price of smartphone i in period t , • 𝑌 k , 𝑗 , 𝑢 is the value of characteristic or performance metric k for smartphone i in quarter t (measured in logs or levels, as appropriate), • 𝐸 𝑗 , 𝑢 is a time dummy variable (fixed effect) that equals 1 if smartphone i is observed in quarter t and zero otherwise, and • 𝜁 𝑗 , 𝑢 is an error term. • 𝜀 𝑢 Provide the hedonic price indexes for price change from first period to time t. 15 10/2/2019
Hedonic regression: Time-dummy hedonic price index ln 𝑄 𝑗 , 𝑢 = 𝛽 + Σ 𝑙 𝛾 𝑙 𝑌 k , 𝑗 , 𝑢 + Σ 𝑢 𝜀 𝑢 𝐸 𝑗 , 𝑢 + 𝜁 𝑗 , 𝑢 Where: • 𝑄 𝑗 , 𝑢 is the price of smartphone i in period t , • 𝑌 k , 𝑗 , 𝑢 is the value of characteristic or performance metric k for smartphone i in quarter t (measured in logs or levels, as Constant- quality quality price appropriate), change • 𝐸 𝑗 , 𝑢 is a time dummy variable (fixed effect) that equals 1 if smartphone i is observed in quarter t and zero otherwise, and • 𝜁 𝑗 , 𝑢 is an error term. • 𝜀 𝑢 Provide the hedonic price indexes for price change from first period to time t. 16 10/2/2019
Hedonic index for smartphones 25.0 20.0 avereage percent changes 15.0 10.0 • Hedonic index falls 5.0 about twice as fast as 0.0 -5.0 the Tornquist index, -10.0 about 20% per year. -15.0 -20.0 -25.0 2011-2013 2014-2017 Change in average prices Time dummy hedonic price index Implied quality change Estimates of quality change 25.0 • With substantially faster rates of quality 20.0 improvement. 15.0 10.0 5.0 0.0 17 2011-2013 2014-2017 10/2/2019 Matched model Hedonic
What can constant-quality indexes tell us about economic activity? • In a national accounting context, the role of price deflators is to hold quality constant • The resulting measures are used to make inferences about – inflation – relevant for monetary policy, and – real output – relevant for productivity measurement 18 10/2/2019
Recommend
More recommend
