Internet Rising, Prices Falling

InternetRising,PricesFalling: MeasuringInflationinaWorldofE-Commerce AustanD.GoolsbeeandPeterJ.Klenow∗ May18,2018 Abstract We use Adobe Analytics data on online transactions for millions of products in many different categories from 2014 to 2017 to shed light on how online inflation compares to overall inflation, and to gauge the magnitude of new product bias online. The Adobe data contain transaction prices and quantities purchased. We estimate that online inflation was about 1 percentage point lower than in the CPI for the same categories from 2014–2017. In addition, the rising variety of products sold online, implies roughly 2 percentage points lower inflation than in a matchedmodel/CPI-styleindex. ∗Goolsbee: University of Chicago Booth School of Business, Chicago IL 60637, and NBER,[email protected]. Klenow: Department of Economics, Stanford University, Stanford,CA94305,andNBER,[email protected]. MathiasJimenezandMatteoLeombroni provided superb research assistance; Alberto Cavallo, Charles Hulten, and Leonard Nakamura helpful comments; and Luis Maykot and Siddharth Kulkarni assistance in accessing and understanding the Adobe data. Goolsbee thanks the Initiative on Global Markets at the University of Chicago Booth School of Business for financial support. Klenow is grateful to the StanfordInstitute for EconomicPolicyResearchforfinancialsupport. Thisisamoreexpansive versionofanAEAPapersandProceedingspaper. 1

2 GOOLSBEEANDKLENOW 1 Introduction The e-commerce share of retail spending in the U.S. has almost tripled in the last 10 years to 10% overall and more than 50% in several major categories, according to the U.S. Census Bureau (2018). See Figure 1. If online pricing is fundamentally different than traditional retail, its spread could have a rising 1 impactontheoverallConsumerPriceIndex(CPI)and,potentially,biasit. WeuseAdobeAnalytics dataononlinetransactionsformillionsofproducts in many different categories from 2014 to 2017 to shed light on how online inflation compares to overall inflation, and to gauge the magnitude of new product bias online. The Adobe data is similar to the Billion Prices Project of Cavallo and Rigobon (2016), which scrapes list prices from the web, but the Adobe data also contains the quantity purchased for the products in addition to prices. Table 1 provides a quick comparison between the CPI, the Adobe dataset, andthescrapeddatafromtheBillionPricesProject. We follow two literatures. One uses detailed scanner data from grocery stores to analyze new product introductions, such as Broda and Weinstein (2010). Another studies consumer surplus from the internet and e-commerce in particular — e.g., Brynjolfsson, Hu and Smith (2003), Goolsbee and Klenow (2006), Brynjolfsson andOh(2012),andVarian(2013). Wedocument1.3percentagepointsperyearlowerinflation online than in the CPI for the same categories. The data also show that the entry of new products and the exit of old products is extremely important for most categoriesofgoods. Thenetentryofnewgoodsduringthesampleimpliesthat matched-modelpriceindicesoverstatetrueinflationbyanadditional1.5to2.5 percentagepointsperyear. 1Economists have long known about the potential for new products to bias upward the inflation measured in the CPI. See Boskin et al. (1996) and, more recently, Groshen et al. (2017). Recent business press articles have argued that online commerce may be leading to growing problems in measuring inflation and complicating monetary policy decisions at the Federal Reserve, as in Cohan (2017), Torrey and Stevens (2017) and Gross (2017). Meanwhile, GorodnichenkoandTalavera(2017)documentthatonlinepricesaremoreflexibleandexhibit higherexchangeratepass-throughthanofflineprices.

INTERNETRISING,PRICESFALLING 3 Figure1: E-commerceShareofRetailSales 8% 6% 4% 2% 1999 2001 2002 2004 2005 2007 2008 2010 2011 2013 2014 2016 2017 Source: U.S.CensusBureau(2018). Table1: ComparingtheDPItotheCPIandBPP DPI CPI BPP Quantities Yes No No #ofitems 2.1 M 140K 500K Offlineprices No Yes No Longhistory No Yes No All categories No Yes No MerchantIdentities No No Yes Notes: DPI = Digital Price Index (Adobe), CPI = Consumer Price Index from the BLS, and BPP = Billion Prices Project scrapeddatafromCavalloandRigobon(2016).

4 GOOLSBEEANDKLENOW 2 AdobeData Adobe Analytics provides a variety of services to e-commerce merchants who share their transaction data for Adobe to analyze. Adobe clients include 20 of the30largestemployersinthenationand80%ofFortune500retailers. Its underlying data are the quantities and revenue from individual transactions (not including taxes or shipping costs). Product codes are merchant-specific,soourdefinitionofaproductwillbetheproduct-merchant combination. We use Adobe’s data aggregated up to the monthly level: total quantitiesandaveragetransactionpricesforeachgoodforeachmonth. Adobe anonymizesthedatasowecannotidentifyanyretailersorcustomers. Weuseasubset of the categories and merchants from the full Adobe data set. Table 2 shows the number of products in the Adobe data we use (overall and by CPI Major Group), averaged over the January 2014 to September 2017 period. It contains over 2 million products in the average month from January 2014 through September 2017, vs. about 140,000 per month in the entire CPI. Thereare211CPIcategoriesknownasEntryLevelItems(ELIs),andtheAdobe datacovers65ofthem. Thecategoriescoveredmakeup19%oftheCPIrelative importanceweightsinBureauofLaborStatistics(2018). Revenue in our Adobe dataset amounts to about 15% of all retail e-commercetabulatedbytheU.S.CensusBureau(2018). Table3showshowit is distributed across CPI Major Groups. ComparedtothefullCPI,thedatasetis definitely tilted toward tangible goods like appliances, furniture, clothing, electronics, and toys. Within some broad CPI categories, it covers only certain typesofproducts. WithinHousing,forexample,theAdobedatadonotinclude rent or owner’s equivalent rent, only specific products hence we label them “Household Goods.” Similarly, we refer to “Information Technology” as the goodswithinEducationandCommunicationthattheAdobedatacovers.

INTERNETRISING,PRICESFALLING 5 Table2: AdobenumberofproductsandCPIcoverage #ofProducts CPICoverage Headline 2.1M 19 Foodandbeverages 1M 49 Educationandcommunication 404K 9 Recreation 202K 32 Apparel 130K 100 Transportation 125K 3 Housing 92K 7 Othergoodsandservices 92K 42 Medicalcare 23K 9 Themiddlecolumngivestheaveragenumberofproductsfrom2014through 2017. Headline is all CPI categories. The next rows are CPI Major Groups. The last column gives the monthly average percent of CPI category (ELI) weightcoveredbytheAdobedata. Source: Authors’calculationsusingAdobe AnalyticsandBLSdata.

6 GOOLSBEEANDKLENOW Table3: DistributionofAdobeRevenue(%) Householdgoods 27 Apparel 27 ICT 19 Recreationgoods 14 Foodandbeverages 7 Othergoodsandservices 3 Transportationaccessoriesandparts 2 Medicinesandmedicalsupplies 1 Notes: Entries are the percent of total Adobe revenue in each of the CPI Major Groups, averaged from 2014 through 2017. Source: Authors’ calculations using Adobe Analytics and BLSdata. 3 DPIvs.CPIInflation Weconstruct a matched-model price index using the Adobe data, and call it the Digital Price Index (DPI) to distinguish it from the CPI. We start with price changes for overlapping products in months t−1 and t. These are products selling positive quantities in both months. We take log first differences of average unit prices. To aggregate price changes across products within an ELI, weuseTornqvistweights. Thesearetheaveragespendingshareoftheproduct in the ELI in months t−1 and t. The spending shares are based on Adobe data 2 for overlappingproducts. Tofacilitate comparison with the CPI, we aggregate the Adobe ELI inflation ratesusingtheCPIrelativeimportanceweightsforeachELI-month. Weusethe samesetofELIstoconstructbothourcomparisonCPIandtheDPI.Inthisway 2Like the BLS, we do something special for apparel. We construct a simple index of average unitprices. Thisistoavoidextremedeflationfromfashionandseasonalcyclesforclothing.

INTERNETRISING,PRICESFALLING 7 Figure2: CumulativeInflation,DPIvs. CPI. 102 102 100 100 CPI 98 98 x Inde96 96 94 DPI 94 92 92 Jan−14 Jul−14 Jan−15 Jul−15 Jan−16 Jul−16 Jan−17 Jul−17 Notes: For the 65 ELIs covered by the Adobe Digital Price Index (DPI). Uses CPI relative importance weights for each ELI. Source: Authors’ calculations usingAdobeAnalyticsandBLSdata. wecanruleoutthatdifferencesbetweenthetwoindexesarisefromcategories whicharenotcoveredbytheDPIortheweightingofcategoriescovered. WeplotthetwoindicestogetherinFigure2. TheDPIexhibitsnotablymore deflation over the period than the CPI for the same categories, ending up 4% lowerbytheendofthesample.Table2showstheaverageannualinflationrates from 2014–2017. Overall (headline) DPI inflation is 1.3 percentage points per yearlowerthanCPIinflationfortheequivalentproducts. BreakingoutbyMajor Groups,inflationislowerintheDPIthanintheCPIineverycategoryotherthan medicine&medicalsupplies. Now, excess deflation in high frequency, chain-weighted price indices can result from oscillating prices due to recurring discounts. This phenomenon is knownas“chaindrift.” Evenif the prices and quantities revert to their starting levels, a chained price index may not revert to 1. This has been documented in grocerystorescannerdatabyIvancic,DiewertandFox(2011)anddeHaanand VanderGrient(2011).

8 GOOLSBEEANDKLENOW Table4: AverageAnnualInflation DPI CPI Headline –1.6 –0.3 Recreationgoods –6.1 –3.0 Householdgoods –4.8 –1.9 ICT –6.6 –3.7 Foodandbeverages –0.9 0.3 Apparel –0.1 0.8 Othergoodsandservices 0.8 1.7 Transportationaccessoriesandparts –1.2 –0.4 Medicinesandmedicalsupplies 1.3 –0.2 Notes: Entries are percentage points per year in annual averageinflationfor2014–2017. Source: Authors’calculation usingAdobeAnalyticsandBLSdata. To gauge the sign and magnitude of chain drift in the Adobe data, for each year we added an artificial “13th month” in which all prices and quantities are identical to first month’s prices and quantities. When then asked whether price index returns to 1 in the 13th month. As shown in Table 5, we found positive chain drift on average and in 6 of the 8 Major Groups. Chain drift was most positive for ICT and apparel items. Thus, chain drift if anything reinforces our findingthatinflationisloweronlinethanoffline. 4 ProductEntryandExit Because the Adobe dataset include quantities as well as prices, we are able to look at spending on entering and exit products. The CPI does not have quantities for items sold within ELIs, so it can see the frequency of products exiting but it cannot tell the market share of exiting products. And, since the

INTERNETRISING,PRICESFALLING 9 Table5: ChainDriftinAdobeData Headline 1.2 ICT 5.2 Apparel 2.3 Foodandbeverages 0.9 Othergoodsandservices 0.5 Recreationgoods 0.3 Transportationaccessoriesandparts 0.2 Medicinesandmedicalsupplies –0.2 Householdgoods –0.7 Notes: Entries are 2014–2017 average annual “excess”inflationduetochaindrift. Source: Authors’ calculationusingAdobeAnalyticsdata. BLS samples only a small fraction of products at a given merchant, it cannot assess the frequency of product entry within merchants, much less the market share of entrants. The AC Nielsen scanner dataset also contains quantities sold, but this dataset is heavily tilted toward food and beverages in grocery stores — see Kaplan and Schulhofer-Wohl (2017). The Adobe data allow us to quantifytheimportanceofnewvarietiesoutsideofgrocerystores. Weclassify a product as new if the product-merchant combination did not exist in the data in the previous calendar year. Analogously, we classify a product as exiting if it does not appear in the following calendar year. We present the entry and exit rates of products by category, weighting by sales of 3 eachproductinTable6. Inapparel,fashionandseasonalcyclesdepresssales of outgoing products and inflate sales of new products. We therefore report results with and withoutapparel. 3We weight by the average monthly sales of a product during the calendar year across the monthstheproductwasavailable.

10 GOOLSBEEANDKLENOW Table6: AdobeProductEntryandExitRates Entry Exit Headline 51.4 24.3 Headlineex.Apparel 43.7 21.9 Apparel 70.8 30.3 Recreationgoods 61.1 20.7 ICT 60.8 31.7 Othergoodsandservices 49.9 13.4 Householdgoods 30.5 19.0 Transportationaccessoriesandparts 24.6 16.9 Foodandbeverages 15.7 9.2 Medicinesandmedicalsupplies 11.1 7.6 Notes: Wesales-weightwithinELIs,anduseCPIrelativeimportance weightsacrossELIs. Entriesareaverage%pointsover2014–2015and 2015–2016. Source: Authors’ calculation using Adobe Analytics and BLSdata.

INTERNETRISING,PRICESFALLING 11 AsshowninTable6,roughlyhalfofonlinesalesareonproductsthatdidnot exist in the previous year. Even without apparel, the figure is 44%. Entry rates are particularly high for ICT products and for recreational items such as toys. Productsthatdisappear,meanwhile,totaledabout24%ofsalesbeforetheyleft the market (22% excluding apparel). Note that, if all that was happening in the datawasrelabelingofthesameproductseachyear,thenwewouldexpectboth entry and exit rates to be inflated by equal amounts. Such relabeling cannot explain the high share of entering relative to exiting products. Product entry andexitratescovarypositivelyacrosstheMajorGroups. The food and beverage category shows much less dynamism than other categories. Entry andexitratesforthesecategoriesarelessthanhalfthatforall Adobe products (even excluding apparel). Thus previous studies finding substantial new product bias in grocery stores, such as Broda and Weinstein 4 (2010), mayhaveactuallyunderstatedtheimportanceofnewproducts. We close this section by asking whether entry and exit rates vary with a product’s price or revenue. To the best of our knowledge, little is known about this question empirically, despite the prominent role of new products in 5 growth theory. Figure 3 shows that entry rates are higher for high price products. This is consistent with products entering with above-average prices, as has been documented in the CPI for apparel, electronics and appliances by Bils (2009). Entry rates are lower for high revenue products, perhaps suggesting it is harder to create high quality products and/or it takes time for products to accumulatesalesasinHottman,ReddingandWeinstein(2016). Figure 3 shows that exit rates are higher for high price products. This is more surprising, as the aforementioned studies found steep price discounts preceding product exit for apparel and electronics. Exit rates are lower for higher revenue products. This could be because such products are harder to 4Bils and Klenow(2004)reportamarkedlylowerexitrateforfoodthanforotherCPIitems. 5Classic references include Romer (1990) for expanding varieties, and Aghion and Howitt (1992) plus Klette and Kortum (2004) for rising quality through creative destruction. See Acemoglu(2008)foratextbooktreatment.

12 GOOLSBEEANDKLENOW creatively destroy by competitors, as hypothesized by Garcia-Macia, Hsieh and Klenow (2018). It lends support to the idea that firms can escape from competitionbyinnovatingasinAghionetal.(2005). 5 TheImpactofNewProductsonInflation Feenstra (1994) showed that a direct way to gauge the importance of new productsinaCESframeworkistolookatthegrowthrateofoverallspendingin a category minus the growth rate of spending for products that exist in both time periods. The higher this net growth rate, the lower the true inflation rate relative to the matched model inflation rate. As shown in Table 3, entering products do tend to have significantly bigger market shares than outgoing 6 products in the Adobe data, even outside apparel. Feenstra (1994) showed that the reduction in true inflation equals the net growth in spending on new varieties times 1/(σ − 1), where σ is the elasticity of substitution between varieties. We use a baseline value of σ = 4 based on Hottman, Redding and Weinstein (2016). We also consider a higher value of σ = 6 for robustness — a moreconservative value given new varieties are less valuable if they are closer substitutesforexistingvarieties. Table7presentsestimatesofnewgoodsbiasintheAdobeonlinedata. Even excluding apparel, the arrival of new goods is equivalent to 1.5 to 2.5 7 percentagepointslowerinflationthanwhatamatched-modelwouldindicate. This is much higher than the 0.6% per year new product bias estimated by the Boskin Commission, though that was for the CPI as a whole. The Adobe data may cover items with larger-than-average new goods bias. Outside apparel, new goods bias looks largest for recreation and ICT products, and lowest for medicineandfood. 6This should capture improvements in product quality in addition to brand new types of products,becausebothareassociatedwithnewproductIDcodesintheAdobedata. 7Weexcludeapparelbecauseincomingitemsmaysellalotmorethanoutgoingitemspurely duetoseasonal/fashioncycles.

INTERNETRISING,PRICESFALLING 13 Figure3: Entryratebyproductrevenueandprice 60% ● ● ● ● ● ● 50% ● ● Revenue 40% Price 1st 2nd 3rd 4th Quartile Notes: Entry rates are for 2015–2016. Products are sortedintoquartilesbyrevenueorpricewithinELIs. Source: Authors’calculationusingAdobedata. Figure4: Exitratebyproductrevenueandprice 60% ● 50% ● ● ● ● ● ● 40% ● Revenue 30% Price 1st 2nd 3rd 4th Quartile Notes: Exit rates are for 2015–2016. Products are sortedintoquartilesbyrevenueorpricewithinELIs. Source: Authors’calculationusingAdobedata.

14 GOOLSBEEANDKLENOW ThevitalrolefornewgoodsintheAdobeonlinedatacallsformoreresearch on new varieties in traditional retail, preferably outside of just the items with 8 UPCcodesthataremostly confined to food, beverages, and drugstore items. If offline sales are similar to online sales, as suggested by Cavallo (2017), new productsmaybeevenmoreimportantthanpreviouslythought. Table7: NewGoodsBiasBasedontheAdobeData σ = 4 σ = 6 Headline 3.5 2.1 Headlineex.Apparel 2.5 1.5 Apparel 7.3 4.4 Othergoodsandservices 5.9 3.9 Recreationgoods 5.4 3.2 ICT 4.1 2.5 Householdgoods 0.9 0.5 Transportationaccessoriesandparts 0.7 0.4 Foodandbeverages 0.4 0.2 Medicinesandmedicalsupplies 0.0 0.0 Notes: Entries are percentage points per year, averaged over 2014– 2015and2015–2016. Source: Authors’calculationusingAdobedata. 6 Conclusion Using a new dataset on e-commerce transactions in many categories of goods fromAdobeAnalytics,wecalculatedmatched-modelinflationandexploredthe importanceofnewproducts. Combiningthetwo,thetrueAdobeDPIinflation rate — adjusted for new goods — was more than 3 percentage points per year lowerthantheCPIinflationrateforthesamecategoriesfrom2014–2017. 8Aghionetal.(2017)studytheentirenonfarmbusinesssector,butonlyattheestablishment level rather than at the detailed product level.

INTERNETRISING,PRICESFALLING 15 7 Appendix 7.1 SummaryoftheComputations We calculate inflation in the Adobe Analytics e-commerce data dataset to facilitate comparison to the CPI in the same categories. Our procedure consistedbroadlyinthefollowingsteps: 1. WefirstmatchedasmanyofAdobe’scategoriesaspossiblewithcategories usedbytheCPI.Afterthisstep,wecontinuedworkingonlywithmatched data and used CPI category names, which are formally called Entry-Level items(ELIs). 2. Since the CPI is computed on a monthly basis, we aggregated the daily Adobe data on revenue and quantities by month for each product. We computedtheaveragepriceforproductiincategory(ELI)j by: 30 XR i,j pi,j,t = 30 Xqi,j wherefromnowontreferstothemonth. 3. For our baseline we do not trim the data at all. As a robustness check, wegaugetheeffectoftrimmingonthepricelevel. AsshowninTableA1, trimmingonthepricelevelhaslittleeffectonAdobeinflationrates. 4. Next, to compute the price index for every ELI at a given month, we first find the products which were sold both in the last and present month (adjacentproducts). 5. Again, our baseline does not trim at all. But Table A1 shows robustness to trimmingonextremepricechangeswithinELI’s.

16 GOOLSBEEANDKLENOW 6. For each ELI we then computethepriceindexforELIj foragivenmonth byusingageometricaverageof“pricerelatives”:  w Y p i,j,t Π = i,j,t j,t p i∈j i,j,t−1 v +v p q wherew = i,j,t−1 i,j,t and v =Pi,j,t i,j,t . i,j,t 2 i,j,t p q i i,j,t i,j,t 7. WethenaggregatetheELIindicesusingmonthlycPIrelativeimportance weightstoarriveatourDigitalPriceIndex(DPI): Π = XW Π t j,t−1 j,t j wheretheW aretheCPIweightsforeachELI-month.9 j,t−1 8. We follow the BLS and do something special for apparel. Apparel prices exhibit a sawtooth pattern over seasonal and fashion cycles. We compute theaverageunitpriceforeveryclothingELI,ratherthancreateamatched- modelindexwhichwouldexhibitsharpdeflation. TableA1: RobustnesstoTrimming 2014 2015 2016 2017 Trim=0/Leveltrim=0 Headline -2.17 -1.58 -2.36 0.20 HeadlineexApparel -2.70 -0.82 -1.17 1.46 Trim=0/Leveltrim=0.01 9https://www.bls.gov/cpi/tables/relative-importance/home.htm

INTERNETRISING,PRICESFALLING 17 Headline -2.80 -1.52 -1.99 0.09 HeadlineexApparel -3.19 -1.00 -0.95 1.49 Trim=0/Leveltrim=0.05 Headline -3.70 -1.05 -1.93 0.12 HeadlineexApparel -3.40 -0.77 -0.69 1.50 Trim=0.01/Leveltrim=0 Headline -4.58 -3.69 -4.42 -2.07 HeadlineexApparel -2.10 -0.97 -0.51 1.55 Trim=0.01/Leveltrim=0.01 Headline -4.72 -3.20 -3.90 -1.43 HeadlineexApparel -2.42 -0.84 -0.43 1.58 Trim=0.01/Leveltrim=0.05 Headline -4.56 -1.76 -2.87 -0.84 HeadlineexApparel -2.81 -0.51 -0.43 1.70 Trim=0.05/Leveltrim=0 Headline -4.44 -3.45 -3.77 -3.17 HeadlineexApparel -1.31 -0.74 -0.19 1.53

18 GOOLSBEEANDKLENOW Trim=0.05/Leveltrim=0.01 Headline -4.26 -2.49 -2.89 -2.48 HeadlineexApparel -1.40 -0.65 -0.21 1.61 Trim=0.05/Leveltrim=0.05 Headline -3.21 -0.07 -1.19 -1.23 HeadlineexApparel -1.55 -0.46 -0.23 1.70 7.2 MatchingCategories Wewere able to match 63 CPI ELIs and 2 CPI Strata to Adobe categories and aggregatesofthem. Wewillrefertotheseas65ELI’sforshort. TableA2provides averageannualinflationratesandothersampleinformationfortheDPIvs. CPI at the ELI level. Wecanalsoaggregateuptothe8CPIMajorGroups,sincethe65ELI’shave some coverage in all of them. Figures A1 through A8 plot the cumulative inflationratesfortheDPIvs. theCPIattheMajorGrouplevel. Figure A9 shows that revenue weighting in the DPI does not affect cumulative inflation relative to weighting each product within ELIs equally as in the CPI. Figure A10 shows that inflation would be even lower if one used Paasche or Fisher instead of Laspeyres, as we have done to facilitate comparisonwiththeCPI.

INTERNETRISING,PRICESFALLING 19 FigureA1: FoodandBeverages 104 102 x 100 Inde 98 96 DPI CPI 94 Jan−14 May−14 Sep−14 Jan−15 May−15 Sep−15 Jan−16 May−16 Sep−16 Jan−17 May−17 Sep−17 Dates FigureA2: Householdgoods 100 95 x Inde 90 85 DPI CPI Jan−14 May−14 Sep−14 Jan−15 May−15 Sep−15 Jan−16 May−16 Sep−16 Jan−17 May−17 Sep−17 Dates

20 GOOLSBEEANDKLENOW FigureA3: Apparel 110 105 100 x 95 Inde 90 85 DPI 80 CPI Jan−14 May−14 Sep−14 Jan−15 May−15 Sep−15 Jan−16 May−16 Sep−16 Jan−17 May−17 Sep−17 Dates FigureA4: ICT 100 95 90 x Inde 85 80 75 DPI 70 CPI Jan−14 May−14 Sep−14 Jan−15 May−15 Sep−15 Jan−16 May−16 Sep−16 Jan−17 May−17 Sep−17 Dates

INTERNETRISING,PRICESFALLING 21 FigureA5: Medicinesandmedicalsupplies DPI 108 CPI 106 x Inde 104 102 100 Jan−14 May−14 Sep−14 Jan−15 May−15 Sep−15 Jan−16 May−16 Sep−16 Jan−17 May−17 Sep−17 Dates FigureA6: Transportationaccessoriesandparts 100 99 98 x 97 Inde 96 95 DPI 94 CPI Jan−14 May−14 Sep−14 Jan−15 May−15 Sep−15 Jan−16 May−16 Sep−16 Jan−17 May−17 Sep−17 Dates

22 GOOLSBEEANDKLENOW FigureA7: Recreationgoods 100 95 x 90 Inde 85 80 DPI CPI Jan−14 May−14 Sep−14 Jan−15 May−15 Sep−15 Jan−16 May−16 Sep−16 Jan−17 May−17 Sep−17 Dates FigureA8: Othergoodsandservices 106 104 x Inde 102 100 98 DPI CPI Jan−14 May−14 Sep−14 Jan−15 May−15 Sep−15 Jan−16 May−16 Sep−16 Jan−17 May−17 Sep−17 Dates

INTERNETRISING,PRICESFALLING 23 TableA2: ELIsummarystatistics ELI DPI CPI #ofProducts CPIWeight Entry Exit Beer, ale, and other malt -0.4 1.1 4188 0.27 16 5 beveragesathome Wineathome -0.7 -0.1 26619 0.25 27 23 Distilled spirits at home -2.4 0 2604 0.07 5 1 Bakeryproducts -1 0.5 11934 0.74 20 7 Beverage materials including 0.1 0 9210 0.27 14 11 coffeeandtea Breakfastcereal -0.2 -0.6 3465 0.19 13 9 Flourandpreparedflourmixes -0.5 -1.6 777 0.05 19 4 Rice, pasta, cornmeal -1.1 -0.7 4754 0.13 12 7 Cheeseandrelatedproducts -1 0.9 5479 0.27 14 10 Ice creamandrelatedproducts -1.3 -0.1 2148 0.12 24 10 Milk -3.8 -2.7 2521 0.25 6 5 Foodandbeverages Other dairy and related 1.8 0.9 2291 0.2 12 7 products Eggs -6.1 -4.5 706 0.12 9 3 Fatsandoils 0.3 0.2 3744 0.24 8 4 Fruits andvegetables -2.1 0.4 14540 1.35 11 5 Juices andnonalcoholicdrinks -0.2 0.1 14017 0.68 12 7 Meats,poultry,fish,andeggs -0.2 0.8 13962 1.87 20 9 Sugarandsweets -2.3 0.7 7097 0.29 15 10 Majorappliances -5.1 -5.8 14892 0.1 27 22

24 GOOLSBEEANDKLENOW Otherappliances -6.2 -1.4 15261 0.12 35 17 Otherfurniture -8.9 -2.2 33925 0.13 30 20 Bedroomfurniture -4.7 -0.9 18674 0.27 32 20 Living room, kitchen, and -5 -1.6 33647 0.36 35 22 diningroomfurniture Householdpaperproducts 0.2 -0.1 1565 0.24 7 6 Householdcleaningproducts -1.2 -0.8 15285 0.34 14 11 Miscellaneous household -4.5 -0.6 11102 0.27 21 13 products Clocks, lamps, and decorator -15.5 -7 61068 0.25 42 23 items Nonelectric cookware and -3.8 -2.5 22898 0.07 33 20 Householdgoods tableware Outdoor equipment and -2.6 -1 11653 0.35 24 16 supplies Tools, hardwareandsupplies -3.3 -0.6 29154 0.18 32 16 Windowcoverings 0.2 -3.1 7783 0.05 23 18 Otherlinens -9.5 -3.9 58445 0.16 35 15 Floorcoverings -5.1 0.3 69407 0.05 34 18 Boys’apparel 0.7 0.4 28826 0.17 74 22 Footwear -4.1 1.5 167897 0.71 65 33 Girls’ apparel 6.4 1.2 37444 0.22 81 28 Infants’ and toddlers’ apparel -0.6 1.5 46536 0.14 81 33 Jewelry 3.8 1.5 70874 0.16 60 46

INTERNETRISING,PRICESFALLING 25 Apparel Watches -0.4 4 19841 0.06 59 35 Men’sapparel -2.8 -0.3 217435 0.65 62 23 Women’sapparel -2.7 0.7 448006 1.15 75 32 Educationalbooksandsupplies 3.7 3.4 13893 0.18 26 33 Personal computers and -12.3 -6.9 56358 0.28 60 39 peripheralequipment Computer software and -2.7 -3.4 199 0.08 33 31 ICT accessories Telephone hardware, -28.2 -10.1 21699 0.08 63 14 calculators, and other consumerinformationitems Medical equipment and -0.3 -0.2 6319 0.07 9 8 supplies Medicinesandmedicalsupplies Nonprescriptiondrugs 0.2 -0.5 17183 0.35 16 10 Eyeglassesandeyecare NA 0.9 98 0.3 1 1 Vehicle accessories other than -0.5 0.8 77236 0.15 38 32 tires Transportationaccessoriesandparts Tires -1.7 -1.1 14416 0.26 18 10 Toys -12.1 -7.8 69568 0.29 73 18 Petsandpetproducts 0.3 -0.4 9552 0.62 10 5 Photographic equipment and -9.2 -0.6 6522 0.05 39 12 supplies Newspapersandmagazines 5.8 2.7 466 0.1 18 10 Recreationalbooks -5.1 -1.5 16534 0.08 51 35 Sports vehicles including -4.8 -1 1770 0.2 50 16 bicycles

26 GOOLSBEEANDKLENOW Recreationgoods Sportsequipment -4.3 -1.9 15074 0.2 34 19 Audioequipment -16.2 -6.3 7421 0.06 58 31 Othervideoequipment -11 -2.4 868 0.03 57 22 Televisions -21.1 -16 4388 0.12 70 28 Video discs and other media, -25.5 -1.1 70218 0.1 56 13 includingrentalofvideo Hair, dental, shaving, and -1.2 -0.3 36886 0.37 29 13 miscellaneous personal care products Cosmetics, perfume, bath, nail -4 -0.1 86014 0.33 59 14 preparationsandimplements Tobacco products other than 4.5 2.7 765 0.05 3 2 Othergoodsandservices cigarettes Cigarettes 4.1 3.8 1134 0.63 11 5 FigureA9: CumulativeInflation,DPIvs. UnweightedDPI 100 100 98 98 Unweighted DPI x 96 96 Inde 94 94 DPI 92 92 Jan−14 Jul−14 Jan−15 Jul−15 Jan−16 Jul−16 Jan−17 Jul−17

INTERNETRISING,PRICESFALLING 27 FigureA10: CumulativeInflation,MethodologyComparison. 100 100 95 95 90 90 x Inde85 85 80 80 75 Laspeyres 75 Fisher 70 Paasche 70 Jan−14 Jul−14 Jan−15 Jul−15 Jan−16 Jul−16 Jan−17 Jul−17 References Acemoglu, Daron. 2008. Introduction to modern economic growth. Princeton UniversityPress. Aghion, P, and P Howitt. 1992. “A Model of Growth through Creative Destruction.”Econometrica,60(2). Aghion, Philippe, Antonin Bergeaud, Timo Boppart, Peter J Klenow, and HuiyuLi.2017.“Missinggrowthfromcreativedestruction.”NationalBureau of EconomicResearch. Aghion, Philippe, Nick Bloom, Richard Blundell, Rachel Griffith, and Peter Howitt.2005.“Competitionandinnovation: Aninverted-Urelationship.”The QuarterlyJournalofEconomics,120(2):701–728. Bils, Mark. 2009. “Do higher prices for new goods reflect quality growth or inflation?” TheQuarterlyJournalofEconomics,124(2):637–675.

28 GOOLSBEEANDKLENOW Bils, Mark, and Peter J Klenow. 2004. “Some evidence on the importance of sticky prices.” Journal of political economy, 112(5): 947–985. Boskin, Michael J, Ellen R Dulberger, Robert J Gordon, Zvi Griliches, and Dale W Jorgenson. 1996. Toward a more accurate measure of the cost of living: Final report to the Senate Finance Committee from the Advisory Commission to Study the Consumer Price Index. Advisory Commission to StudytheConsumerPriceIndex. Broda, Christian, and David E Weinstein. 2010. “Product creation and destruction: Evidence and price implications.” The American economic review, 100(3): 691–723. Brynjolfsson,Erik,andJooHeeOh.2012.“Theattentioneconomy: measuring thevalueoffreedigitalservicesontheInternet.” Brynjolfsson, Erik, Yu Hu, and Michael D Smith. 2003. “Consumer surplus in the digital economy: Estimating the value of increased product variety at onlinebooksellers.”ManagementScience,49(11):1580–1596. Bureau of Labor Statistics. 2018. https://www.bls.gov/cpi/tables/ supplemental-files/home.htm. Cavallo, Alberto. 2017. “Are online and offline prices similar? Evidence from large multi-channel retailers.” The American Economic Review, 107(1): 283– 303. Cavallo, Alberto, and Roberto Rigobon. 2016. “The billion prices project: Usingonlinepricesformeasurementandresearch.”TheJournalofEconomic Perspectives, 30(2): 151–178. Cohan,Peter.2017.“HowAmazonOverpowerstheFed.”Forbes. deHaan,Jan,andHeymerikAVanderGrient.2011.“Eliminatingchaindriftin priceindexesbasedonscannerdata.”JournalofEconometrics,161(1):36–46.

INTERNETRISING,PRICESFALLING 29 Feenstra, Robert C. 1994. “New product varieties and the measurement of internationalprices.”TheAmericanEconomicReview,157–177. Garcia-Macia, Daniel, Chang-Tai Hsieh, and Peter J Klenow. 2018. “How destructiveisinnovation?” NationalBureauofEconomicResearch. Goolsbee, Austan, and Peter J Klenow. 2006. “Valuing Consumer Products by the Time Spent Using Them: An Application to the Internet.” American EconomicReview,96(2):108–113. Gorodnichenko, Yuriy, and Oleksandr Talavera. 2017. “Price setting in online markets: Basic facts, international comparisons, and cross-border integration.” AmericanEconomicReview,107(1):249–82. Groshen,EricaL,BrianCMoyer,AnaMAizcorbe,RalphBradley,andDavidM Friedman.2017.“HowGovernmentStatisticsAdjustforPotentialBiasesfrom Quality Change and New Goods in an Age of Digital Technologies: A View fromtheTrenches.”JournalofEconomicPerspectives,31(2):187–210. Gross,Daniel.2017.“HowAmazonFuelsDeflation.”Strategy+Business. Hottman, Colin J, Stephen J Redding, and David E Weinstein. 2016. “Quantifying the sources of firm heterogeneity.” The Quarterly Journal of Economics,131(3):1291–1364. Ivancic,Lorraine,WErwinDiewert,andKevinJFox.2011.“Scannerdata,time aggregation and the construction of price indexes.” Journal of Econometrics, 161(1): 24–35. Kaplan, Greg, and Sam Schulhofer-Wohl. 2017. “Inflation at the household level.” Journal of Monetary Economics,91:19–38. Klette, Tor Jakob, and Samuel Kortum. 2004. “Innovating firms and aggregate innovation.”Journalofpoliticaleconomy,112(5):986–1018.

30 GOOLSBEEANDKLENOW Romer,PaulM.1990.“Endogenoustechnologicalchange.”Journalofpolitical Economy,98(5,Part2):S71–S102. Torrey, Harriet, and Laura Stevens. 2017. “As the Fed Deliberates Amazon is Makingit’sJobMoreDifficult.”WallStreetJournal. U.S.CensusBureau.2018.https://www.census.gov/retail/index.html. Varian, Hal. 2013. “The value of the internet now and in the future.” The Economist.