How CPI is contructed Flashcards
Products and items?
- below item is product
- a product is an example of an item – something you acc buy – product described by item
- chosen to be representative of the item
- the actual p quotes are collected for specific products – specific brand, type or size
- Item – large white loaf unsliced 800g
- products may be Sainsbury, Tesco, Asda own brand white bread
- a product is a precise spec – GITN bar code
Stratification?
How you divide up data geographically
12 regions (London, Wales, Scotland etc.) and shop type (multiple more than 10 outlets, independents 9 or less)
For each product up to 24 ‘cells’ (strata) 12x2 - for some products, can find price quotes for each cell, for others not enough info or stratification not believed to be relevant (e.g., car licence)
Coding of stratum type?
0 - not stratified
1 - stratified by region
2 - stratified by region and shop type
3 - stratified by shop type
white bread 800g stratified region only so just 12 strata
Price quotes?
an observation, a product at a particular shop at particular location – we don’t observe exact location in microprice data
* actual location is not public – within each region, may be several locations – each shop has a code
* since Sainsbury, Tesco, Asda follow largely national pricing policy and are nationwide, same price seller index will appear in each region under multiple category w same p
There are also shop weights - mostly 1, can be 2, 3, 4… - representing importance in the region
Validity of price quotes?
In any month some price quotes are not available/valid
Only codes 3, 4, 53, 54 are used - all others discarded
Indicator - indicates something non-standard
Indicators of price quotes - coding?
S Sale price or special offer
R Recovery of price at end of sale or special offer
T Temporarily out of stock
M Missing – not sold at shop
C Comparable – change in product being priced, new product is similar to the previous product
N Non-comparable – change in product being priced, new product is not comparable to the previous product
Q Additional comments have been supplied to assist with internal validation
W Change in the weight or volume of the product
X Comparable item on sale
Z Non-comparable item on sale
Indicators coding explained?
T,M. No price, invalid code.
C price of comparable product.
Q-Z. May require judgement about whether to include or not.
price quotes coding?
0 Price is outside the min-max range
1 Zero price or failed credibility check
2 Rejected by user
3 Validated
4 Accepted by user
5 Price change failed % test
7 Unknown Indicator Code
8 Ind.= Q/C/N/W but no message exists
9 Price is 0 but Ind. is NOT T or M
10 Ind. is T or M but Price is NOT 0
11 Quote is valid but Ind. is Q/W
15 Scotland eye test charges 2006 - free
51* Zero Price, temporarily out of stock or missing
52* Rejected by user
53* Valid, quote contributed depending on indicator box
54* Valid, quote contributed depending on indicator box
99 Used during end-of-year processing
Central v Local collection?
- 200,000 p quotes collected – mostly monthly
- local – Kantar TNS contracted to collect ~100,000 p for around 580 items in around 150 locations each month
- 2 types of centrally collected –
- Regional Central Shop Collection (CES) – internal internet collection ~20,000 across 429 items around 46 retailers – collected directly from retailer web sites
- further 7000 collected from outlets wo a shop e.g., electricians
- Central Spreadsheet Items - ~80,000 p collected via central spreadsheet mechanism for around 150 items
- CES combined w locally collected p quotes used to generate local p
- Central spreadsheet – p not combined w local or CES – go directly to prod item level indices
- over time more p collected centrally
From price quotes to elementary aggregates?
Have price quotes
Compare to base price, usually Jan
Using price relative loses some info - all you need for inflation, but if looking at cost of living will need to know actual price
Need to integrate with the constant of integration, being the base price
Elementary Aggregates?
Combining price quotes to get item indices
* an elementary aggregate is the index for a stratum/cell
* each shop has relative R – weighted by shop weight
* elementary for CPI, CPIH and RPIX is Jevons mean
So, let’s look at region 3, using sheet 1 (valid).
1. Total number of shops by weight (sum of shop weights in that region). Region 3 = 20.
2. Sum of log-relatives for each shop weighted by shop weight, divided by 1.Region 3 = -0.0348.
3. Take exponential. Region 3 = 0.965798557
- can do unweighted w excel
To get the Item index, take an arithmetic average of elementary indices using stratum weights.
For December 2017 for white bread 800G it was 1.008.
Across the UK, the price of white bread, unsliced 800 grams, coded as item id: 210102 was 0.8% higher than it had been in January. Different means?
Carli
Elementary aggregates bigger, item index 1.014. That is 1.4% inflation for this item sinceJanuary 2017. “Formula effect” of 0.6% over 11 months.
Dutot.
Computes average price level in base month relative to average in December.
Details?
- prices used in calc CPI should be p actually paid in cash and these prices should be prices universally available to all
- exc discounted and subsidised prices not inc unless available to all – coupons, loyalty card discount, multiple item discount
- in addition, sales prices only inc of temporary and likely to be reversed – not end of stock, damaged, sell by date
- variety of checks to identify outliers
Details - stage 1 - on the field?
- made at point of collection
- price change check – bounds set change each month – range varies across items – can be as large as 50%, most common 33% - if change outside the range – validity code of 5 – fresh veg and fruit don’t have this test
- min/max range for item – dep on previous p across all regions and shop types – if outside, collector asked to check if p correctly entered – some shops may charge crazy p for marketing purposes
Details - stage 2 - ONS checks data in field?
- decision made whether to accept
- sales most important thing here
- if item on sale or recovering last month and change is less than 55%, accept
- if recovering from sale in previous month, and increase less than 110%, accept
- if item on sale this month and previous w no price change, accept
- Phase 2 – TUKEY algorithm used to remove outliers – look at p relatives relative to previous month – ignores 1s – remove top and bottom 5% -
- prices failing any ONS checks are invalid unless revalidated
From month to month, about 80% of prices remain the same across all COICOP. Tukey has been in use since 1987 and ONS are happy with it….
Phase 3: Q code checks (written comments etc.), indices above 180 or below 60 looked into..
Item indices and upwards?
- just a matter of taking weighted arithmetic averages of indices using COICOP weights – basically CPI is weighted avg of item indices
- CPI weights for classes and upwards, updated every jan
- items updated every feb along w item weights
- ONS reveals our changing shopping habits
- Elementary aggregates in Jan are prod using p relatives w previous jan
- using new item weights, used to calc class indices based on previous jan base p
- values of the class indices also calc using old item weights
- will in general yield diff val for some of the class indices
- these are chained – val under the new weights is adjusted to be equal to val under old weights
- since class indices are chained, all higher level CPI indices will also be chained inc CPI and CPIH
- same process in Feb
- chaining done at class and above – not item level
- From feb to Dec, p relatives are all to previous jan w same weights
