Replacements, Quality adjustments and Sales Prices Jörgen Dalén Oxana Tarassiouk
Background to paper First version was a report commissioned by Statistics Sweden for proposing alternatives to its currently most used quality adjustment (QA) method: direct valuation by the price collectors. Report presented to CPI Board (Indexnämnden) of Statistics Sweden in October 2012 Slightly revised, tables updated, summarizing section added for more general discussion This version to be presented at Ottawa Group meeting in Copenhagen, May 1-3, 2013
Currently (2012) used QA/replacement methods Price collectors estimate the value of a quality difference at replacements Used for most CPI goods (not services) Monthly chaining and re-sampling Used for computers and mobile phones Various explicit adjustments for cars (+ motor cycles and boats) Option prices Judgemental adjustment by central Deletion of non-matching product offers for daily necessities and alcohol Effectively using only product offers available from previous December
QA in the presence of sales prices Immediately realised that QA problem in practice is intertwined with the pattern of sales prices Analytical tool: Product offer life= the series of prices for a certain product in a certain outlet Very often a product offer life ends with a sales/reduced price after which a new product offer comes in with a high price What is the right quality adjustment between a product on a sales price and a new product offer?
Sales prices and price reductions Stylized example Month 0 1 2 3 4 5 6 Price 195 195 195 99 99 225 225 One could generalise this example to call it a regular price-sales price-replacement(rsr) cycle. The reason that this pattern creates great problems for price measurement is its asymmetrical nature. Price decreases typically occur within the life-time of one product offer but the price increase occurs at the replacement. Two examples Table 2 and 3
Table 2: Example of price observations in one outlet 2010-2012, curtain (cloth)
Table 3: Example of price observations in one outlet 2010-2011, large TV Raw price change= 6990/17632=39.6 %
Implicit quality indexes IQI = APC API APC = change in average prices API = Actual price index (CPI numbers) Shows the quality change that is implied by the methods and procedures used in the index IQI>100 means quality increase. Almost all IQI in Table 4>100 Question: Is it reasonable to assume that quality has increased for all products? Back to Tables 2 and 3 Price collectors normally attribute part of price increase to quality change
IQI Table Product group 2010, actua l 2010, bridged overlap 2011, actual 2011, bridged overlap 2012, actual 2012, bridged overlap Furniture 100.5 100.9 100.9 103.3 100.8 102.4 Household textiles 104.5 115.5 103.7 122.0 101.0 120.9 Household appliances 101.0 108.3 105.8 106.5 103.2 105.4 Household equipment 103.9 110.8 103.9 109.9 100.1 98.0 Tools for home and garden 104.5 107.8 104.4 103.5 101.6 102.4 Recreational goods 102.2 112.7 100.9 108.5 103.8 109.1 Home electronics 107.7 113.0 111.8 114.2 111.0 119.1 Large TVs 111.0 117.1 113.7 115.3 111.5 116.2 Digital cameras 97.8 100.1 110.9 112.1 121.0 147.4 DVD players 107.2 114.8 106.6 113.1 111.3 117.2
Comments on IQI Table Almost all IQI>100 implies quality improvement (or underestimating bias ) Bridged overlap increases IQI, often by a lot. Since the price increase at replacement is eliminated and replaced by averages of other movements, often small,
Table 5 product offer live spans P r o d u c t M o n t h s P r o d u c t M o n t h s P r o d u c t M o n t h s P l a t e 1 2. 9 F l o w e r, p l a n t 1 5.7 V i d e o c a m e r a 4. 8 C o ff e e c u p 1 0. 8 S k i e q u i p m e n t 3. 8 A u d i o 6. 0 s y s t e m s G l a s s 1 0. 7 S p o r t e q u i p m e n t 7. 5 H o m e c i n e m a 6. 1 s y s t e m E a t i n g k n i f e 1 9. 2 O u t d o o r 1 0.4 C D r a d i o 6. 0 re c r e a t i o n e q u i p m e n t S a u c e p a n 1 4. 2 T o y 1 1.6 M P 3 p l a y e r 6. 8 K i t c h e n k n i f e 1 7. 2 K i t c h e n t ab l e 1 5.5 G a m e c o n s o l e 9. 0 K i t c h e n s c a l e 1 4. 4 U n u p h o l s t e re d 1 4.4 D i g i t a l 5. 9 c h a i r c a m e r a S a l a d b o w l 1 2. 7 U p h o l s t e re d 1 4.3 C o m p u t e r 3. 5 c h a i r B a b y c ar r i a g e 1 1. 8 B e d 1 6.9 M o b i l e p h o n e 5. 6 B a b y c ar s e a t 1 7. 2 C e i l i n g l a m p 1 4.3 W a s h i n g 6. 0 m a c h i n e B a g, c a s e, 6. 6 A rm c h a i r 1 5.5 D i s h w a s h e r 7. 0 p u rs e T o w e l 1 3. 5 S o f a 1 3.0 V a c u u m 7. 4 c l e a n e r D u v e t c o v e r 9. 7 S h e l f, c ab i n e t 1 7.0 R e f r i g e r a t o r 6. 3 s e t C u rt a i n 1 0. 0 C a r p e t 1 2.8 M i c r o w a v e 8. 9 (c l o t h ) o v e n Q u i l t 1 4. 0 M a t t re s s 2 3.2 C o f fe e m a k e r 9. 6 C a r t y r e 1 5. 3 M i rr o r 1 6.9 W a t e r b o i l e r 9. 0 C a r a c c e s s o r y 2 0. 2 T V, s m a l l 4. 3 W a t c h 1 9.0 B i c y c l e 1 1. 0 T V, l a r g e 4. 3 J e w e l l e r y 2 0.8 M u s i c a l i n s t ru m e n t 1 1. 6 D V D p l a y e r 5. 2
Product offer lives Most POs live about a year in the CPI (Table 5) PO for home electronics live only 4-6 months
Table 6 (part) Group/ product group (w = weight, n = total number of lives) Household equipment Rsr, pure price decrease Pure price increase Other patterns Median price change per Median price change per Median price change per % month % month % month Plate (w = 0.69, n = 346) 8.7, 22 0.92, 0.99 9.4 1.01 59.9 1 Coffee cup (w = 0.74, n = 205) 13.6, 6.6 0.91, 0.98 8.7 1.01 71.2 1 Glass (w = 0.67, n = 59) 7.1, 7.3 0.82, 0.97 29.2 1.01 56.3 1 Eating knife (w = 0.34, n = 27) 4.2, 7.3 0.91, 0.99 23.5 1.01 64.9 1 Saucepan (w = 1.06, n = 97) 7.1, 6.2 0.97, 0.99 17.0 1.00 69.7 1 Kitchen knife (w = 1.06, n = 77) 7.4, 17 0.95, 0.99 18.0 1.01 57.6 1 Kitchen scale (w = 1.06, n = 82) 0.4, 0.0 0.95,. 14.2 1.01 85.3 1 Plastic food container (w = 1.06, n = 103) 1.2, 12.9 0.96, 1.00 26.5 1.01 59.4 1 Salad bowl (w = 1.06, n = 78) 5.5, 0.0 0.94,. 38.7 1.01 55.7 1 Other personal goods Baby carriage (w = 0.31, n = 193) 15.9, 10.1 0.98, 0.99 38.8 1.00 35.2 1 Baby car seat (w = 0.32, n = 94) 7.3, 5.5 0.97, 0.99 45.1 1.00 42.1 1 Bag, case, purse (w = 1.92, n = 388) 12.3, 3.5 0.89, 0.97 20.9 1.01 63.2 1 Household textiles Towel (w = 0.41, n = 305) 18.3, 16.5 0.94, 0.99 22.9 1.01 42.2 1 Duvet cover set (w = 1.15, n = 526) 26.2, 4.6 0.93, 0.96 9.4 1.02 59.8 1 Curtain (cloth) (w = 2.89, n = 548) 20.6, 7.1 0.9, 0.98 20.5 1.01 51.8 1 Quilt (w = 1.19, n = 171) 13.6, 4.2 0.98, 0.98 30.8 1.01 51.3 1
Table 6 Different types of product lives Rsr = regular price-sales price-replacement Pure price decrease ( rsr) Pure price increase Others (one month, no price change...) Most product offer lives involve no price change at all or very small ones. Price change essentially estimated from the first three types of lives and from the replacements between two lives. For high-tech products, the price decreases far outnumber the increases. For other products the patterns are more mixed and in most cases price increases are more numerous than decreases. But price decreases tend to be larger than price increases
How to interpret results? Normal market assumptions: For traditional products no or small quality change on average Still IQI>100 in most cases For high-tech products there is real quality change But how big and how rapid? Look at Tables 7 and 8 for actual results Rather extreme index movements Sweden extreme in Europe for home electronics Swedish krona stronger Cut-throat competion in Swedish market But could this explain everything??
Results from present methods Price collector valuation TV, large TV, small Mcr Compute rs Coicop 09.1 2010 53,1 55,3 64,0 72,8 70,8 2011 54,9 61,0 60,6 72,7 70,8 2012, July 62,1 66,4 68,0 72,9 73,8 DVDplayer 2010-2012, July 18,1 22,4 26,4 38,6 37,0 Extreme results for TVs!
International comparison from 2005) Table 8: Sweden in extreme low end in a European context 09.1.1 Equipment for the reception. recording and reproduction of sound and pictures 09.1 Audio-visual. photographic and information processing equipment Country HICP (2005=100) Country HICP (2005=100) Sweden 27.6 Sweden 40.3 France 37.4 Estonia 44.0 Czech Republic 39.4 Ireland 45.5 Estonia 39.6 Spain 46.0 Belgium 43.7 Latvia 46.8 Latvia 41.8 Czech Republic 47.6 Slovakia 42.3 Switzerland 47.6
Conclusion Risk for downward CPI bias apparent New methodology needed! Two main methods proposed: Direct price comparison (no quality adjustment) for most products Monthly chaining and replacement for high-tech products
From EU Commission Regulation (EC) No 1334/2007 (4) elementary expenditure categories (weights) consist of explicitly stated consumption segments distinguishable by consumption purpose.. Consumption segments are relatively stable over timealthough the product-offers comprising a consumption segment will change as markets evolve. (5) The notion of consumption segments by purpose is therefore central to sampling and to the meaning of quality change and quality adjustment. (6) The range of product-offers will change over time as products are modified or replaced by retailers and manufacturers. The HICP requires the representation of all currently available product-offers within the consumption segments by purpose selected in the reference period in order to measure their impact on inflation. This applies particularly to new models or varieties of previously existing products. So the idea is that a replacement PO has to represent the reference period PO.
New methodology proposals: Direct price comparison Product specifications fixed in base period for a year or more. Principle of base period representativity can be read from EU regulations Replacements are essentially equivalent (sufficiently similar) for direct price comparisons to be used. As a second replacement criterion the replacement should be well sold. If no essentially equivalent product offer is available the product offer is deleted and price change computed over the remaining product offers. In case a supported QA method is available for a certain product the criterion essentially equivalent is changed to comparable under the QA method.
More on direct comparisons Applied to most products It should be noted that today the interviewers assign a quality difference of zero to 57 % of all replacements The interviewer could thus be instructed to choose a replacement where she would feel comfortable to assign a quality difference of zero. Sales prices in the base period still a problem. Need to be same share of sales prices in December t-1 and t One possibility: Start price measurement in October but prices in CPI only from December
New methodology proposals: MCR method Now used for computers and mobile phones Other home electronics products are similar: Length of product lives Share of price decreasing cycles Expected rapid technological and quality change Method proposed for all machine-type products in home electronics
New mcr method Move to a more rigorous mcr methodology Monthly chaíning Genuine re-sampling every month Ascertain that a price represents a normal sales quantity Central staff needs to be involved in price collection Limit price collection to Stockholm + websites at least for a transitional period?
MCR procedures, more detail Product offers are selected in the base period on the basis of being among the n most sold in the outlet according to a wide product specification. Most sold should ideally relate to a full month and not a much shorter period and to the price as recorded by the interviewer. They are kept in the index as long as they are well sold. This fact has to be established each month. A proxy for well sold may be that it is displayed as a volume product. In each month new product offers are included regardless of whether the old ones have terminated. They are included on the basis of being among the n most sold. The sample size in each outlet will vary over months but should on average be n. The index is computed through monthly links (m-1 to m) including only identical product offers, no essentially equivalent ones.
Note on scanner data We need to study to which extent there are rsr cycles in scanner data! If rsr cycles are frequent then a methodology based on matching EAN codes could fail seriously and create underestimating biases. Instead replacement methods need to be forced on scanner data when ean codes are discontinued.
Summary of proposals for Sweden No more quality adjustment by price collectors Large systematic bias risks in current procedures, The method amounts to unsupported adjustments which is a C-method (not allowed) in EU parlance Many products are of a high-tech nature which the price collectors lack competence in valuing and they also feel uneasy about this task Central explicit QA maintained but is not expected to play major role Two standard methods proposed Direct price comparisons for traditional products with normal life spans of about a year Mcr-method for high-tech products with short life span in the market. Details have to be deleoped futher and be tested in praktice, partly during the course of work The cost of the new procedures are expected to be lower allowing for a high ambition in the detailed execution of the new methods. In practice Statistics Sweden is now taking cautious steps in the direction proposed in the paper.
Reserve slides To be shown only if needed
Products in scope for study Category Weight, All Int.judg. High-tech products 22.7 10.4 Recreational goods 22.2 11 Other personal goods 6.4 4.2 Household equipment 7.7 3.7 Restaurants 51.3 24.8 Furniture 22.3 22.3 Household textiles 5.6 4.5 Household goods 7.7 7.7 Recreational and cultural 1.6 0 services Tools and garden equipment 4.0 4.0 Media for music, photo and 3.8 3.8 film Lodging 5.2 5.2 Personal hygiene 18.1 3.1
Representativity and comparability Basic goal conflict for CPI replacements Representativity - sample of products and outlets should reflect the universe. Comparability replace between identical or equivalent product offers. Reference period representativity can be read from EU Regulation, is inherent in Laspyeres logic (fixed basket representing the reference period) So representativity relates to product offers in the reference period. Criteria for replacements in order of priority: Replacements shall be equivalent or comparable. Replacements shall be the most sold among equivalent or comparable product offers. Deletion (like for food) if no comparable replacement
Evaluating QA methods Intuitive ideas among price statisticians Overlap paradigm: Price differences reveal consumer valuations so not a quality difference Pitfall: Assumes market equilibrium. Short-lived clearance sales violate this. Quality improvement for tech products replacements often have higher quality Price reductions can reflect relative quality decline Still short-lived clearance sales should not influence the index Traditional products tend to have smaller changes so QA should not reduce price change on average Changes may occur but clear evidence needed for accepting average quality change
Explicit quality adjustments Hedonic methods considered best by many Resource-demanding Requires a strictly limited scope within which hedonic model applies Re-estimation needed frequently due to market changes Option prices Information needed from previous period Only known to work for cars and transport equipment Other supported adjustments Examples fuel cost for cars Not frequently available Unsupported adjustments Judgemental without objective model in support Unsupported adjustments Advised against by CENEX Interviewer adjustments example Also central judgemental adjustments would be unsupported